Universals in Comparative Morphology
i
ii
Universals in Comparative Morphology
Suppletion, superlatives, and the structure of words
Jonathan David Bobaljik
University of Connecticut
revised: July 2011
iii
Contents
Acknowledgments xi
Abbreviations xv
1 Introduction 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Distributed Morphology . . . . . . . . . . . . . . . . . . . . . 10
1.3 Constructing the Database . . . . . . . . . . . . . . . . . . . . 24
1.4 Comparative Typology . . . . . . . . . . . . . . . . . . . . . . 28
2 Comparative Suppletion 39
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.2 *ABA - Explaining a gap . . . . . . . . . . . . . . . . . . . . . 46
2.3 UG vs. the European Sprachbund . . . . . . . . . . . . . . . . 58
2.3.1 The insignificance of the evidence . . . . . . . . . . . . 58
2.3.2 Counting cognates . . . . . . . . . . . . . . . . . . . . 61
2.3.3 (In)stability and change . . . . . . . . . . . . . . . . . 65
vii
2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3 The Containment Hypothesis 73
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.2 Transparent Containment . . . . . . . . . . . . . . . . . . . . 74
3.2.1 Transparent Nesting . . . . . . . . . . . . . . . . . . . 74
3.2.2 The Fennic Superlative: Branching Affixes? . . . . . . 82
3.2.3 Aside: More than anything . . . . . . . . . . . . . . . . 88
3.2.4 Interim Conclusion . . . . . . . . . . . . . . . . . . . . 96
3.3 Comparison and the Synthetic/Analytic Divide . . . . . . . . 97
3.3.1 Periphrasis, Suppletion, and Locality . . . . . . . . . . 98
3.3.2 Aside: Romance suppletion and Poser blocking . . . . 106
3.3.3 Periphrastic superlatives again . . . . . . . . . . . . . . 113
3.4 The Synthetic Superlative Generalization . . . . . . . . . . . . 116
3.4.1 Armenian superlatives . . . . . . . . . . . . . . . . . . 128
3.4.2 SSG: Loose ends . . . . . . . . . . . . . . . . . . . . . 135
3.5 Containment and semantic considerations . . . . . . . . . . . . 138
3.6 Chapter summary . . . . . . . . . . . . . . . . . . . . . . . . . 151
4 CSG: The Data 153
4.1 Adjectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
4.1.1 The Basque problem . . . . . . . . . . . . . . . . . . . 165
4.2 Adverbs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
4.3 Quantifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
viii
4.3.1 Many – more counterexamples . . . . . . . . . . . . . . 186
4.3.2 Coda: Bulgarian, Macedonian and the RSG . . . . . . 192
4.3.3 Afterthought: What’s more? . . . . . . . . . . . . . . . 195
4.3.4 Chapter summary . . . . . . . . . . . . . . . . . . . . . 199
5 Theoretical Refinements 201
5.1 Introduction: Taking Stock . . . . . . . . . . . . . . . . . . . . 201
5.2 Conditions on Suppletion: exponence versus readjustment . . 205
5.3 Adjacency, ABC, *AAB . . . . . . . . . . . . . . . . . . . . . 211
5.3.1 Getting worse: portmanteaus and locality . . . . . . . 214
5.3.2 Aside: A cyclic alternative? . . . . . . . . . . . . . . . 222
5.3.3 Adjacency - Outstanding Issues . . . . . . . . . . . . . 226
5.4 AAB ablaut . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
5.5 Merger, Rule Ordering, Diacritics and Acquisition . . . . . . . 240
6 Getting better: Comparison and deadjectival verbs 247
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
6.2 Preliminary remarks . . . . . . . . . . . . . . . . . . . . . . . 253
6.3 Deadjectival Degree Achievements: Doubting Dowty . . . . . . 258
6.4 To good, to badden, and to many . . . . . . . . . . . . . . . . 275
6.4.1 Ancient Greek goods . . . . . . . . . . . . . . . . . . . 285
6.4.2 Baddening (up) . . . . . . . . . . . . . . . . . . . . . . 287
6.4.3 Moral goodness and ambiguity in Serbian . . . . . . . . 289
6.4.4 Russian: bad feelings . . . . . . . . . . . . . . . . . . . 292
ix
6.4.5 Multiplicity . . . . . . . . . . . . . . . . . . . . . . . . 295
6.4.6 Compounds and roots . . . . . . . . . . . . . . . . . . 297
6.4.7 Outstanding examples . . . . . . . . . . . . . . . . . . 302
6.5 Summary: What’s the Difference? . . . . . . . . . . . . . . . . 304
7 Complexity, Bundling, and Lesslessness 307
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
7.2 Lesslessness . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
7.3 Conservative decomposition: Adjacency and Bundling . . . . . 325
7.4 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . 330
Appendices 334
A The Broad Sample 337
B The Focussed Survey 357
C Principal Sources 373
Notes 393
Bibliography 455
x
xiv
Abbreviations
1/2/3 s/p first/second/third person singular/plural
a.sprl absolute superlative
abl ablative
acc accusative
add additive
adess adessive
adj adjective
adnom adnominal
adv adverb
Anc. Greek Ancient Greek
CGEL Cambridge Grammar of the English Language (Huddleston and
Pullum 2001)
Cl. Greek Classical Greek
cmpr comparative
COCA Corpus of Contemporary American English http://corpus.
byu.edu/coca/
xv
conv converb
cop copula
CSG The Comparative-Superlative Generalization
C∆G The Comparative-Change-of-State Generalization
dat dative
decl declarative
def definite
DM Distributed Morphology
dwb Das Deutsche Wörterbuch (Grimm and Grimm 1854-1961)
elat elative
emph emphatic
fact factive
fs feminine singular
gen genitive
incept inceptive
infin infinitive
inst instrumental
intns intensifier
intr intransitive
loc locative
lv linking vowel
ms masculine singular
neg negative
xvi
neu neuter
nml nominalizer
nom nominative
npi negative polarity item
ns neuter singular
obj object
OCS Old Church Slavonic
OED Oxford English Dictionary
part partitive
U.R. Underlying Representation
PIE Proto Indo-European
pl plural
pos positive
prd predicator
pref prefix
pres present
prog progressive
pst past
ptcp participle
refl reflexive
rev reverse
RSG The Root Suppletion Generalization
Scand. Scandinavian
xvii
sg singular
sprl superlative
SSG The Synthetic Superlative Generalization
subj subject
top topic
tv karel. Tikhvin Karelian
UG Universal Grammar
v verb
VIP The Vocabulary Insertion Principle
vm verb marker
wdg Wörterbuch der deutschen Gegenwartssprache http://www.
dwds.de/
xviii
Chapter 1
Introduction
1.1 Introduction
Morphology is sometimes characterized as the domain of the lawless, and
among the miscreants, no process epitomizes irregularity more than suppletion
— the wholesale replacement of one stem by a phonologically unrelated
stem — as in the comparative and superlative degree of adjectives (good –
better – best). I argue, based on a large, cross-linguistic survey (just over
300 languages), that there are nevertheless strikingly robust patterns in this
domain, robust enough to be solid contenders for the status of linguistic
universals.
One goal of this study is to offer theoretical explanations for these generalizations.
In the course of developing an analysis, I will argue that the
assumptions needed bear on choices among theoretical frameworks, with the
1
framework of Distributed Morphology having the right architecture to support
the account. In addition to the theoretical implications of the generalizations,
I suggest that the striking patterns of regularity in what otherwise appears
to be the most irregular of linguistic domains provides compelling evidence
for Universal Grammar (UG). The core theoretical claim will amount to saying
that certain types of meaning cannot be expressed monomorphemically.
A central example is the superlative, as in English biggest. I contend that
no language has a true superlative morpheme that attaches to adjectival
roots. Apparent examples, such as English -est, have in fact a richer structure,
where the superlative-forming element always embeds a comparative,
(roughly): [[[ adjective ] comparative ] superlative ]. This structure
is transparent in many languages (see Chapter 3). The argument for UG
here constitutes in some ways a twist on the familiar logic of the povertyof-the-stimulus.
Within any one language, the evidence from suppletion is
far too scant for any observed patterns to emerge as significant. Thus, if
there are universal generalizations to be had (as I contend there are), the
significance of these can only be appreciated in their cross-linguistic scope.
Thus a second goal of the book is a contribution to the search for the basic
building blocks of grammatical meanings — morphological primitives.
Over the course of the book I contend that the following generalizations
have the status of linguistic universals. Various clarifications are discussed as
we proceed, and apparent counter-examples are addressed, with alternative
accounts provided. The first generalization is the Comparative-Superlative
2
Generalization (CSG), in two parts, extending observations in Ultan (1972).
(1) The Comparative-Superlative Generalization, part I (CSG1):
If the comparative degree of an adjective is suppletive, then the superlative
is also suppletive (i.e., with respect to the positive).
(2) The Comparative-Superlative Generalization, part II (CSG2):
If the superlative degree of an adjective is suppletive, then the comparative
is also suppletive (i.e., with respect to the positive).
The two parts of the CSG require that an adjective that is suppletive in
gradation will be suppletive in both the comparative and superlative grades
(relative to the basic, positive root). The CSG allows for patterns such as bad
– worse – worst (henceforth, an ABB pattern), with the a suppletive root
common to both comparative and superlative, and for patterns such as Latin
bonus – melior – optimus ‘good – better – best’ (ABC), with distinct roots in
all three grades. What is disallowed (and virtually unattested) is a pattern
in which only the comparative *good – better – goodest (*ABA) or only the
superlative *good – gooder – best (*AAB) shows suppletion. Two important
terminological clarifications should be made here. First, by superlative here
and throughout I refer only to relative superlatives, meaning ‘more X than
all others’. The generalizations do not extend to what are sometimes called
absolute superlatives (also called elatives in some traditions). The latter
do not have a strictly comparative sense, and mean instead “adj to a very
high or excessive degree”. For example, in Italian, there is a distinction
3
between relative superlatives on the one hand (formed by adding the definite
article to the comparative): il migiore ‘the better’ = ‘(the) best’, la più bella
‘the more beautiful’ = ‘the most beautfiul’, and absolute superlatives on
the other (marked by the suffix -issim-) buon-issim-o ‘very, extremely good’,
bell-issim-a ‘very, extremely beautiful’. Other affixes that form absolute
(rather than relative) superlatives include the Slavic prefix pre- (as Slovenian
pre-lép ‘too/very beautiful’ < lép ‘beautiful’), the suffix -ejš-ij in Russian
(as in vern-ejš-ij drug ‘very/most loyal friend’ < vern-yj drug ‘loyal friend’),
and Modern Greek -tat-os.)1
Absolute superlatives are not subject to the
generalizations laid out here, evidently because they lack the comparative
component of meaning (and hence structure).2
A second clarification regards the term suppletion itself. I take suppletion
not to represent a point on a cline of irregularity (contrast Wurzel 1985), but
instead see a categorical divide between suppletion (a relation holding among
distinct exponents, or vocabulary items, as in go – went) and other forms of
irregularity such as ablaut (which involve phonological changes to a single
underlying exponent, as in tell – told). I return to these points in more
detail below, in particular presenting arguments for a distinction between
suppletion and readjustment rules in Chapter 5. For now, I simply mention
these clarificatory points as they are important to understanding the scope
of the generalizations under scrutiny. On suppletion generally, as well as
differing uses of the term, see Mel’čuk (1994) and Corbett (2007).
Two further generalizations to be discussed are the following:
4
(3) The Synthetic Superlative Generalization (SSG)
No language has morphological superlatives (X-est), but only periphrastic
comparatives (more X ).
(4) The Root Suppletion Generalization (RSG)
Root suppletion is limited to synthetic (i.e., morphological) compar-
atives.
The SSG constitutes the claim, independent of suppletion, that no language
shows a regular pattern analogous to *long – more long – longest, in
which the comparative is exclusively periphrastic, where the superlative is
morphological. (The terms periphrastic and morphological in this sense are
interchangeable with analytic and synthetic, respectively, and I will alternate
among these at points below.)
The RSG adds to this the claim that no language shows suppletion of
the root in periphrastic comparatives: *good – more bett. There is a subtlety
to the formulation of the RSG that should be mentioned here, in light
of Modern Romance languages such as French and Italian, in which regular
comparatives are all periphrastic, but which have a handful of suppletive
forms. The key point to keep in mind is that suppletion, in the normal
case, involves a substitution of roots, but leaves functional morphology intact.
Thus bett-er in place of *good-er, preserves the comparative affix -er,
and this is characteristic of the significant majority of examples of suppletion.
While suppletion does appear to compete or alternate with periphrastic
5
constructions in some languages — thus Italian: buon-o ‘good’ may be compared
periphrastically: più buono or suppletively migliore (both ‘better’) —
no language shows root suppletion with (obligatory) preservation of the comparative
morpheme, when the latter is free standing. The introduction is not
the place to explore the subtleties here, but I signal the point here since the
examples are likely to be familiar to many readers (Romance superlatives are
the focus of section 3.3.2).
Yet another generalization concerning the morphological expression of
degree is the following.
(5) Lesslessness
No language has a synthetic comparative of inferiority.
Comparison of superiority (‘more X’) is affixal in many languages, as in
long – long-er, but comparison of inferiority (‘less X’) never is; schematically:
more X : X-er :: less X : *. This last generalization is the most unquestionably
robust of the lot.3
For the generalizations in (1)-(4), there are apparent
challenges which must be (and can be) addressed. But (5) is surface true
without even a hint of a problematic case in the 300 languages examined for
the present study. The generalization is discussed in Chapter 7.
The hypothesis that lies at the core of the proposals in this work can be
phrased as the Containment Hypothesis, given in (6):
6
(6) The Containment Hypothesis
The representation of the superlative properly contains that of the
comparative.
Subject to some qualifications, the central claim is that (7a) is a possible
representation, but that (7b) is universally disallowed.
(7) a. [ [ [ adjective ] comparative ] superlative ]
b. * [ [ adjective ] superlative ]
The Containment Hypothesis stands at the heart of the account of (1),
(2) and (3). These generalizations are shown to follow from (6) on a small
set of independently motivated, general theoretical assumptions. The most
important among these are given in (8), discussed in more detail below:
(8) a. Late Insertion (Realization) — the (abstract) morpho-syntactic
representation is the input to a morphological component, characterized
in part by rules of exponence (vocabulary insertion)
which assign phonological realizations to the terminal nodes.
b. Underspecification, Elsewhere Ordering — the rules of exponence
(vocabulary insertion) may be underspecified, and thus
may compete to realize a given node; such competition is resolved
by the Elsewhere Condition (Subset Principle, P¯an.inian)
ordering, in which more specific rules take precedence over more
general ones.
7
c. Locality — morphological rules are constrained to operate under
strict conditions of locality; in certain configurations, an erstwhile
trigger of a given rule may be too far away to trigger that rule.
Structural adjacency may be a condition on certain types of rule
(rules of exponence, readjustment rules).
How these are to be made precise, and how they interact with the Containment
Hypothesis to derive the generalizations given above, is the focus
of Chapters 2-5.
An important further question addressed here is how (6) is to be formalized,
and relatedly, why it should hold. Though somewhat tentative, I will
suggest that the impossibility of (7b) is a consequence of a general limit on
the complexity of individual morphemes. That is, at least for the functional
or grammatical (as opposed to lexical) vocabulary, there are intrinsic limits
on possible morpheme meanings. I suggest that the meaning ‘more than all
others’ is, by this criterion, too complex to be expressed monomorphemically,
and must therefore be split into (at least) a piece meaning ‘more’ and
another meaning (roughly) ‘than all (others).’ The containment hypothesis
in (6) and (7) is thus not itself a part of UG, but rather a consequence of a
far more general condition. Some rather speculative remarks on this theme,
including how it may underlie an account of (5), and how it connects to
proposals regarding impossible morphemes in other empirical domains (see
Kayne 2005, Bobaljik 2008, Hackl 2009), occupy Chapter 7.
In addition to the generalizations just considered, the following emerges
8
as a strong trend in the data, with surprisingly many (but not all) prima
facie counter-examples readily accounted for in other terms.
(9) The Comparative-Change of State Generalization (C∆G):
If the comparative degree of an adjective is suppletive, then the corresponding
change-of-state verb is also suppletive (i.e., with respect
to the positive adjective).
In a manner strongly reminiscent of the missing *good – better – goodest
patterns with superlatives, the more spotty domain of deadjectival verbs is
well populated with patterns such as bad – worse – (to) worsen, but precious
few examples show the corresponding verb derived from the positive root:
good – better – *to gooden. Though it is less robust that the other generalizations
considered here, its formal parallel to (1) warrants consideration. I
examine the C∆G in Chapter 6, considering in particular the implications
of a parallel to the containment hypothesis (6) in this empirical domain, as
against more traditional proposals for the lexical semantics of deadjectival
degree achievements (cf. Dowty 1979).
The study itself begins in earnest in Chapter 2. In the following sections,
I present some additional background remarks which some readers may find
useful. Section 1.2 discusses some core theoretical assumptions that will come
up in the course of the work. Section 1.3 discusses the nature of the datasets
used for this study, and some of the choices made in aiming for a balance
between diversity of languages and comprehensive coverage of the data. The
9
phenomena of core interest show a significant areal concentration, with comparative
suppletion largely unattested outside of a Greater European area,
a fact that poses a special challenge to what is fundamentally a quantitative
claim about the significance of gaps in the data. Section 1.4 provides
an extremely brief review of prior literature on the typology of comparative
constructions.
1.2 Distributed Morphology
One contribution of the present study is intended to be empirical, and to the
extent possible, I have characterized core descriptive results in a relatively
neutral vocabulary. In addition, I have taken pains to present the theoretical
argument, in particular in Chapter 2, in terms that are as general as
possible, in order to permit fair comparison with alternatives across theoretical
frameworks. That said, one of the goals of the current work is to argue
that a particular set of assumptions is required in order to explain the generalizations
above. I take it that as uncontroversial that a key component
of explanation in this sense is not only to be able to describe the attested
patterns, but also to exclude the unattested patterns; all six generalizations
presented in the previous section describe gaps in the data, which, I argue
at length below, are systematic, rather than accidental. I contend that the
framework of Distributed Morphology (DM) has the right general architecture
to support the assumptions needed to derive these generalizations. It is
10
not clear that competing morphological frameworks do. If correct, then this
book constitutes an argument for that general theoretical framework. With
that in mind, it seems of use to lay out here some of the relevant aspects of
DM (and associated terminology) that I will appeal to in the course of the
accounts to be developed below. This section does not aim for a comprehensive
overview of the framework (see Halle and Marantz 1994, Harley and
Noyer 1999, Chapter 2 of Embick 2010, and Bobaljik In preparation), but
aims only to present some key notions that play a role in what follows.
The most important aspect of DM for the theory of comparative suppletion
to be developed here is the treatment of competition and blocking
in allomorphy. Distinct phonological matrices compete to realize (combinations
of) grammatical features, in different contexts. To take a well-worn
case, there are in English, as in many languages, a variety of exponents of the
grammatical feature plural, with irregular (lexically restricted) exponents
taking precedence over the regular, default spelling out of the plural feature,
wherever available: thus ox-en, sheep-Ø rather than *ox-es, *sheep-s. As
has been widely noted, such competition effects imply that morphology is realizational,
incorporating some version of the Separation Hypothesis (Beard
1995, with many antecedents noted therein), which holds that the derivation
of the morphological representation of complex words is separate from (and
prior to) the spelling out or realization of those representations.
In the DM instantiation of this general approach, Rules of Exponence
(equivalently, of Vocabulary Insertion) provide the phonological realization
11
for morphosyntactic representations, held to be derived by the syntax (hence
insertion is late, as opposed to the early, pre-syntactic insertion characteristic
of Lexicalist frameworks. Examples of some rules of exponence for English are
given in (10); collectively, the rules of exponence for a language are referred
to as the vocabulary of the language.
(10) a. pres 3 sg → -s =/-z/
b. pres → Ø
c. pl → Ø / ]N where N = sheep, foot, ...
d. pl → -en / ]N where N = ox
e. pl → -s =/-z/ / ]N
f. cmpr → -er
g.
√
good → be(tt)- / ] cmpr
h.
√
good → good
i.
√
big → big
. . .
A few words on notational conventions here: the rules relate a grammatical
feature matrix (in caps) to a phonological representation. For ease of
exposition, I use standard orthography as a proxy for the phonology throughout
this book. Rules of exponence may be context-free or context-sensitive,
where the context may include idiosyncratic lexical restrictions (as in (10cd)).
In my presentation of the rules of exponence, I use the symbol → to
express the relationship these rules represent, taking it to be one of rewriting
of features with their exponents (see Halle 1990, Trommer 1999, Bobaljik
12
2000b). Another formalism within DM uses ⇔ expressing correspondence,
rather than rewriting (see Halle and Marantz 1993, Embick 2010). Nothing
of consequence turns on this distinction in the present work.
The entries in (10a-f) provide exponents for the functional (i.e., grammatical)
vocabulary of the language, constituted by grammatical features
(these features should be understood as shorthand for a more sophisticated
feature analysis, or decomposition into grammatical primitives). By contrast,
the vocabulary entries for Roots (10g-i), indicated as
√
root, are not
to be understood as features, but rather as identifiers for individual roots,
abstracting away from allomorphy. It will be crucial to the analysis of competition
in suppletion that good and bett- be seen as manifestations of a single
abstract root; hence, rules of exponence must apply to roots (Beard’s 1995
Lexemes; l-morphemes in the terminology of Harley and Noyer 1999) as well
as to grammatical morphemes, an assumption that is incompatible with the
framework of Beard (1995).4
Beyond that simple point, I take no stand on
the correct representation of these identifiers; for the vast majority of lexical
roots, the identifier may just as well be the (or a) phonological representation,
making exponence for these roots trivial. Where there is no risk of
confusion, I will often omit the root symbol in vocabulary fragments given
below.
Given representations such as (11), the rules in (10) will apply to provide
exponents to the various nodes (for (11a) I have given a labeled partial tree
diagram; for the others only more abbreviated representations with informa-
13
tion omitted not relevant to the point):
(11) a. [ [
√
play ]V pres,3,sg ] V+Infl
V
√
play
play
Infl



pres
3 sg



/z/
b. [ [
√
dog ]N pl ] = dog-z
c. [ [
√
ox ]N pl ] = ox-en
d. [ [
√
big ]A cmpr ] = big-er
e. [ [
√
good ]A cmpr ] = bett-er
Where more than one rule is compatible with a given node, competition
is regulated by the Elsewhere Condition (Kiparsky 1973), which may be
formulated as in (12):5
(12) If two (incompatible) rules R1, R2 may apply to a given structure,
and the context for application of R2 is contained in that of R1,
then, R1 applies and R2 does not.
The Elsewhere Condition, as widely discussed, applies to force the choice
of an irregular allomorph over a competing regular one, as in, for example
14
the plural ox-en in (11c). Rule (10d) bleeds application of rule (10e) in (11c),
by the Elsewhere Condition. (It is understood that rules are disjunctive, and
only one rule of exponence may apply to a given node in the general case.
Disjunctivity of rules satisfies Kiparsky’s incompatibility condition on the
application of (12).)6
In a similar manner, (12) ensures that (10a) and not
(10b) is inserted in the inflectional position in (11a), yielding (She) play-s,
rather than play-Ø, even though the description for (10b) is met. Finally, in
the realm of root allomorphy, and directly relevant to the rest of this book, the
Elsewhere Condition forces a contextually-restricted allomorph (10g) to block
insertion of a context-free allomorph of the same root (10h), when the context
for insertion is met, as in (11e). This derives bett-er, rather than *good-er
as the comparative of good. See Embick and Marantz (2008) for further
discussion of blocking in DM, and comparison to alternative conceptions.
Working backwards through the derivation, the representations that are
the input to vocabulary insertion are taken to be derived, in the first instance,
in the syntax, though they are subject to additional post-syntactic
manipulations in some cases. We may take one example that will figure
in the discussion of the RSG and SSG in Chapter 3. Among the abstract
morphemes of English, along with a slew of adjectival roots, is a morpheme
cmpr which combines with adjectives to yield the comparative. As is well
known, English comparatives have both a periphrastic and a synthetic guise:
more polite and polit-er, respectively. In a DM account (see Embick and
Marantz 2008), a single syntactic structure underlies both the periphrastic
15
and synthetic expressions, namely, that in (13a).
(13) a. CmprP
cmpr AdjP
adj
b. CmprP
cmpr AdjP
adj
c. c
a
adj
cmpr
Sticking to the basic case, a periphrastic comparative like more polite
arises when (13a) is subject to vocabulary insertion, with the comparative
element (pronounced as more) and the adjective in separate maximal projections.
A synthetic construction arises when some operation M indicated
by the arrow in as in (13b), combines the two terminal nodes into a single
complex head yielding (13c), cf. (11d)-(11e). The theory provides a variety
of options for the identity of M, including Head Movement (either in, or after,
the syntax) or the operation Morphological Merger (Marantz 1989)7
, or perhaps
others (cf. Embick and Noyer 1999, for an array of options). Very little,
if anything, in the current study hinges on what M is in any given example,
and so I will remain generally agnostic throughout. For concreteness, and for
consistency with the analysis of reinforcing adverbs in section 3.3.1, I will use
a downward-pointing arrow in trees such as this, thus taking M as (Lowering
via) Merger. (The various operations that are candidates for M may differ in
the labels they would assign to the nodes in (13c), for example, in whether
cmpr is adjoined to adj or vice versa; I intend to remain agnostic here and
use lowercase letters simply as mnemonic devices to refer to various nodes.)
16
In the view that the input to morphology is a syntactic representation,
DM stands in contrast to other realizational frameworks such as versions of
Word-and-Paradigm models (Matthews 1972, Anderson 1992, Stump 2001)
as well as Aronoff (1994) and Beard (1995). In many of these frameworks,
which reject the idea that syntax is the source of the concatenation and arrangement
of grammatical features, the morphosyntactic feature bundle is
unstructured. Effects of affix order are the product of stipulated rule ordering,
for example, by ordering rules of exponence into disjunctive rule blocks
(Anderson 1992). In the course of this book, I contribute to the argument
for a syntactically structured morphosyntactic representation. Specifically,
there appear to be locality conditions on allomorphy that require the kind
of hierarchically structured input representation that DM posits but which
competing theories reject.
The argument begins from the assumption that rules of exponence operate
cyclically, beginning with the root (cycle-based theories of locality within DM
are offered in Bobaljik (2000b) and Embick (2010)). A first consequence of
such cyclic approaches is that some form of No Lookahead Condition will hold
(e.g., Simpson and Withgott 1986, 155). To start with their example, the
derivation of the English word cliticization [[[ clitic ] iz(e) ] ation ] proceeds in
steps, adding -ize to the stem clitic on the first cycle, and adding -ation to the
result on a second cycle. In the first cycle, when -ize is added, information is
available about the stem (e.g., that it is a noun), and hence this may constrain
the process (for example, triggering allmorphy). However, no information is
17
available about what will happen in the second cycle: thus information about
the more peripheral suffix may not condition processes such as allomorphy
for the first suffix. To do so would involve ‘looking ahead’ to a subsequent
cycle. In other words, as the complex word is constructed, one condition on
allomorphy for a given affix is that it may only be sensitive to information
already present in the morphological structure at the time that affix is added.
In the domain of inflection, No Lookahead in its strongest form is too
strong. As (Carstairs-McCarthy 1992, 214) puts it: “[i]t is as if inflectional
realisation operates on the basis of precise information about what has already
been spelled out . . . , but only vague information about what has yet to
be spelled out.” Though differing in detail, cyclic approaches which posit a
structured morphosyntactic representation as the input to rules of exponence
derive something weaker than Simpson and Withgott’s No Lookahead, and
more in line with Carstairs-McCarthy’s description: a rule of exponence at
the root may be conditioned by information that is part of the representation
at that stage of the derivation, namely, the morphosyntactic properties of the
higher nodes, but not their phonological form.8
In addition to the cyclicity condition, two other locality conditions play
a role in the current work. One condition, at least as a working hypothesis,
is that a morpheme (or feature) β may condition allomorphy for morpheme
α only if the the two are in the same morphological “word” (i.e., complex
X0
): β may condition allomorphy for α in the environment in (14a) but not
that in (14b), where a maximal projection intervenes (abstracting away from
18
linear order). (An alternative formulation, in line with Embick (2010), would
make reference to phases or cyclic nodes, in place of maximal projections, in
(14b), relating the condition more closely to the discussion of cyclicity in the
preceding paragraph.)
(14) a. α . . . ]X0 . . . β
b. * α . . . ]XP . . . β
The RSG (4) falls out as a special case of this broader condition. The
Italian root
√
good mentioned above has two allomorphs: the comparative
miglior- and the elsewhere form buon-. To a first approximation, Italian
grammar treats operation M in (13b) as optional — if M applies, the suppletive
root is required, but if M does not apply, then the head cmpr is
insufficiently local to the root, and suppletion is not triggered, giving: piú
buono ‘more good’.
Evidently, (14) is a necessary, but not a sufficient condition on locality
for contextual allomorphy. Even within a complex X0
there appear to be
locality conditions at work. Thus, in the structure in (15), for at least some
values of X, Y is unable to condition root allomorphy.
19
(15) Y
X
√
√
root
X
Y
A strong proposal is that any X serves as an intervener between Y and
the root in (15), in essence, a condition of structural adjacency on root allomorphy,
as proposed in one form or another by various authors (Siegel 1978,
Allen 1979, Embick 2003, 2010).9
A structural adjacency condition appears
to be somewhat too strong as a general condition on contextual allomorphy
(see Carstairs 1987, Bobaljik 2000b, Chung 2007b) but in the current study,
adjacency makes the right divide in a number of interesting cases. I will
therefore tentatively adopt the adjacency condition as part of the theory of
locality for root allomorphy, leaving it to a future project to understand why
adjacency might not restrict affixal allomorphy (Bobaljik 2000b), as well as
reconciling this with the sporadic apparent counter-examples in the literature
just mentioned (see the exchange in Bobaljik 2000b, Carstairs-McCarthy
2001, 2003, Adger et al. 2003 for pertinent discussion).
It is in this last, and narrowest sense of locality, that the DM approach
clearly parts company with competing realizational theories that assume no
hierarchical structure among the abstract, grammatical features. In theories
such as those in Anderson (1992), Stump (2001), where the morphosyntactic
20
representation is unstructured, with word structure given by the rules of exponence,
there is no obvious sense of ‘higher nodes’ in the morphosyntactic
representation — no sense in which ‘adjacency’ can be defined or evaluated
relative to the abstract features that serve as the input to the rules of exponence
(as opposed to adjacency among the exponents, which is readily
definable). If an adjacency condition in this sense (or something like it) is
substantiated (as argued for in this work), this serves as a further argument
for word-internal hierarchical arrangement of grammatical features prior to
vocabulary insertion.
A final aspect of the framework that should be mentioned here, and one
that has received less attention to date, is the treatment of portmanteau
morphology — cumulative exponence in the terminlogy of Matthews (1972,
1991). In some instances, a single phonological string appears to correspond
to multiple terminal nodes of the (morpho-)syntactic representation. Contrast
bett-er, the comparative of
√
good, which appears to contain the regular
comparative suffix -er alongside the suppletive root allomorph, with
worse, the comparative of
√
bad, which appears to express simultaneously
the root meaning and the comparative. The theoretical framework adopted
here presents at least two ways of approaching portmanteaus (and no reason
to suspect that all will fall to the same treatment).
On the one hand, the general structure of comparatives [[ adj ] cmpr ]
can be maintained even for the worse case, but with mutually-conditioned
contextual allomorphs: bad → worse / ] cmpr, and cmpr → Ø / ]A
+
],
21
where A+
= worse, less. This approach would preserve the structure for
English comparatives unchanged even for the portmanteaus, and essentially
recapitulates history, where the deletion of -er had a phonological motivation.
Note that the overwhelming majority of root suppletion is of the bett-er type,
preserving derivational and inflectional morphology beyond the root.
On the other hand, one could treat the exponent worse as a true portmanteau,
spelling out both the root and the comparative elements: [bad,
cmpr] → worse, inserted at the top node in (13c). Radkevich (2010) proposes
to allow insertion at non-terminal X0
nodes in exactly this way (see
also Neeleman and Szendrői 2007, Caha 2009), with a specific proposal as
to how to restrict application of this mechanism to avoid overapplication.
Equivalently, an operation of Fusion may be invoked (Halle and Marantz
1993, see also Bobaljik 1997, Chung 2007b) to combine the two terminals in
(13c) into a single node prior to the application of rules of exponence.
As it happens, there is perhaps an argument in the domain of comparatives
for treating (some) portmanteaus as complex exponents (either with
fusion, or insertion at complex nodes) rather than a conspiracy of mutually
conditioned allomorphs. The argument has to do with the locality conditions
mentioned above. Recall from section 1.1 that alongside suppletive patterns
like good – better – best (with a common allomorph in the comparative and
superlative), patterns are also attested with three distinct root allomorphs,
such as Latin bonus – melior – optimus ‘good – better – best’. The Latin pattern
poses a prima facie challenge to the adjacency condition on allomorphy
22
in that the superlative is, by the Containment Hypothesis, never structurally
adjacent to the root, yet there is clearly root allomorphy conditioned by the
superlative in this example. However, combining the non-terminal-insertion
or fusion approach to portmanteaus with the adjacency condition on allomorphy
allows for a narrow loophole to adjacency, with a clear prediction,
stated here:
(16) The superlative may only condition root allomorphy (distinct from
the comparative), when the root and comparative are expressed by a
portmanteau.
The prediction is indeed borne out (see Chapter 5 for a discussion of why
this should hold), although the number of relevant examples is small enough
that we cannot be confident this is not an accident. It is worth mentioning
here, though, as it illustrates on the one hand how the various assumptions
interact, and on the other, how surface violations of conditions (such as
adjacency) may arise as the product of the interaction of various principles
at some level of abstraction.
While there is far more that should be said about the theoretical framework,
the discussion above will I hope provide more than sufficient background
for the material to be discussed in the next chapters. Chapter 5
returns to more of the fine detail of the theory, after Chapters 2 and 3 which
are presented in broader terms.
23
1.3 Constructing the Database
The results reported here are drawn from the investigation of just over 300
languages, representing an attempt to strike a balance between the opposing
demands of representing areal and genetic diversity on the one hand, and
achieving a comprehensive catalogue of attested examples of suppletion in
comparative and superlative morphology. The assembly of the data may be
thought of as a three-stage procedure, and this approach is reflected in the
presentation of the results in Appendixes A-B, as well as Chapter 4.
The first stage, reported in Appendix A, is a broad sample of 148 languages,
representing approximately 40 families and a dozen or so isolates (using
the classifications at Bickel and Nichols (1996ff) and http://multitree.
linguistlist.org/ as a rough guide, with no implied commitment to these
classificatory schema). Languages were included in this sample on a quasirandom
basis, with availability of descriptions leading to some skewing.10
Multiple languages were chosen from most families, as there is variation in
the expression of comparison even within a single family, sometimes striking
variation among closely related languages.
While the broad sample is used for verifying the empirical generalizations
discussed above, the broad sampling method proves to be not particularly
illuminating in this regard as the majority of the world’s languages satisfy
the interesting generalizations for rather trivial reasons. Most languages, for
example, lack ABA pattens in superlative adjectival suppletion because they
24
lack suppletion in gradation, lack superlatives, or lack adjectives altogether.
Thus, the primary function of the broad sample lies in identifying regions
and families with the morphological properties of interest to the present
study (suppletion in adjectival gradation, morphological comparatives or superlatives).
Taking the results of the broad sample (along with a LinguistList
query and a literature review) as a starting point, I assembled a second survey,
whose aim was for comprehensiveness, rather than breadth — to have
as many data points as possible from languages with comparative suppletion
and/or morphological superlatives. To this end, I telescoped in (as it were)
on languages which have evidence of morphologically marked grades of comparison
(from the first survey and the literature review), and then expanded
outwards to languages closely related to the core sample, to the extent grammatical
descriptions were available. The results of this focused survey are
presented in Appendix B. This second sample includes 173 languages (of
which 20 are from the broad sample). Approximately 2/3 of these languages
have morphological comparative forms (somewhat fewer have morphological
superlatives), and of these, approximately 70 show examples of comparative
suppletion. A striking areal skew emerges at this point: morphological marking
of comparative and superlative is attested (though not common) around
the globe, but (with perhaps the sole exception of Cherokee) the confluence
of both suppletive comparatives and morphological superlatives (central to
investigating the CSG) appears to be limited to languages historically from
Europe and its closest neighbours — a Greater European Sprachbund.
25
Despite the concerns this areal concentration raises for a universalist perspective,
there is sufficient variation in the data from this group of languages
for interesting and significant patterns to emerge. In light of the fact that
many of these languages are closely related, however, counting languages in
a survey of this type becomes increasingly problematic (see Bickel 2008 and
references therein). To see the problem, consider the English triple good –
better – best, a pattern consistent with the CSG. German also has a CSGconsistent
ABB pattern for the corresponding adjective: gut – besser – (am)
besten, indeed so do almost all known Germanic languages present and past,
cf. Gothic goþs – batiza – batists. Yet surely the facts from each of these
individual languages are contingent facts of their history; ultimately, the
entire Germanic ‘good’ paradigm represents but a single innovation, somewhere
in the prehistory of Northern Europe. In assembling the actual data
presented below, in order to mitigate against the influence of borrowing and
common inheritance, I took the approach of counting cognate sets, rather
than languages, counting only one exemplar for each cognate triple of the
form positive – comparative – superlative. A comprehensive listing of attested
suppletive triples (positive – comparative – superlative) is given in
Chapter 4. In order to be counted there, each pattern must differ in at least
one of the roots from a pattern already in the dataset (see section 2.3.2 for
ellaboration). This avoids questions of defining and counting languages. We
need not take a stand on whether Norwegian and Swedish, or Serbian and
Croatian, are distinct languages, or how many stages in the history of En-
26
glish, we should recognize. The Germanic ‘good’ triple contributes a single
data point. Some triples like this thus represent numerous languages over a
wide time period, while other triples may be contributed to the data set by
a single dialect, for example the Giazza Cimbrian (German) guot – ăeg-ur –
ăeg-ur-ste ‘good – better – best’, for which the comparative and superlative
have a different etymological source from German besser and English better
(see Schweizer 2008, 397, and section 2.3.2 below).
Approached this way, the generalizations remain robust, despite the concern
of an areal concentration. As regards the CSG, there are more than 100
distinct cognate triples reported in Chapter 4 — some 70 of these are for qualitative
adjectives (with only one potentially problematic example). Quantifiers
such as ‘much/many’ and ‘few’ (which add additional complexities) contribute
nearly 40 additional triples, with a handful of apparent challenges.
The full dataset is presented in Chapter 4, with problems and qualifications
discussed in the relevant chapters below. Moreover, although comparative
suppletion is unquestionably an areal phenomenon, there is sufficient variation
within the languages that have it to raise the questions addressed here
— why are some patterns attested and others not? Simply noting that abstract
patterns may be resilient to change over time provides an insufficient
answer. The patterns that are attested do change – suppletion arises and is
lost, doublet patterns come into being or fall out of use, etc. Of the various
ways in which one could characterize patterns in the data, most are in fact
not stable over time — what is stable is precisely the generalization that the
27
attested (or reconstructible) changes do not yield AAB and ABA patterns in
this domain. What this study aims to provide is an account of why, in a sea
of irregularity, these particular generalizations constitute islands of stability.
More discussion of these issues occupies the latter part of Chapter 2, after
more details of the theory and its empirical basis are presented.
1.4 Comparative Typology
A final array of useful background information for this study concerns the
typology of the expression of comparison, and the place of this study in relation
to the previous work. In broad terms, an expression of comparison has
three principal parts: a predicate denoting a gradable property, the subject
of comparison, and the standard against which it is contrasted; see the English
(17a) and synonymous Russian (17b) examples here. Two additional
elements in these particular examples are: comparative morphology (either
free or bound), glossed cmpr, expressing that the utterance is comparative,
and a special marker of the standard of comparison, the particle than in
English and the genitive case in Russian.
(17) a. The bear is larg -er than the dog.
subject property cmpr std. mkr standard
b. medved’ bol’š -e sobak -i.
bear big -cmpr dog -gen
subject property cmpr standard std. mkr
28
Previous typological studies focus largely or exclusively on the broad syntax
of these constructions, and/or on the morphology of standard marking
(Andersen 1983, Stassen 1985, 2008). Only Ultan (1972) discusses the morphology
of comparative marking, and his study does not focus on the contrasts
that are of primary interest in the present work (although he remarks
briefly on suppletive patterns, and notes the essential content of the CSG).
In terms of coarse-grained morphosyntax, three broad types of comparative
construction can be identified cross-linguistically (my classification is in part
different from prior authors in ways that will become clear presently): the
conjoined comparative, the ‘exceed’ comparative, and the standard comparative.
All three are illustrated here, although most of this book will be
concerned with the third type, and then, only with a subset thereof.
In the conjoined comparative, illustrated in (18)-(20), a simple positive
expression such as The bear is large is juxtaposed with a contrasting expression,
for example, an antonymous predicate (as in (18)), with negation (as
in (19), or by the juxtaposition of a plain and intensified predicates (20).
(18) mosbi
Port Moresby
ó=le
neu-top
sum
big
eka
and
banimo
Vanimo
ó-ta
neu-emph
gwˇaab=o=be
small-prd-decl
‘Port Moresby is bigger than Vanimo.’ (Mian, Fedden 2007, 122)
29
(19) tinuPn
these
LeNu-Pn-č
berry-pl-dim
č’eBuz-laX-aPn,
sweet-adj-pl,
a
but
xaNnaPn
those
qaPm
not
‘These berries are sweeter than those.’ (Itelmen, field notes [SB14A])
(20) poka
stilt
fain
this
maala,
long
ne
add
oko
other
maala
long
akena.
very
‘This stilt is long but the other one is longer (lit: very long).’ (Mauwake,
Berghäll 2010, 272)
Roughly one out of every four to five languages makes use of a conjoined
comparative as the primary or exclusive means of expressing comparison
(34/167 languages in Stassen 2008, and 37/148 in my broad sample).11
In the ‘exceed’ comparative type, a verb meaning ‘exceed’ or ‘surpass’
(either as the main predicate, or as a modifier) expresses the meaning of
comparison. This type may be expressed by the loose English paraphrases
The bear is big, exceeding the dog or The bear exceeds the dog in tallness
(height). Exceed comparatives are particularly common among (though not
limited to) languages with serial verb constructions, and are widespread in
Sub-Saharan Africa and Southeast Asia. Examples illustrating variants of
this type are given in (21)-(22).12
(21) Uqa
3sg
cecela.
long
Uqa
3sg
ija
1sg
wol-te-na.
surpass-1sg-3sg-pres
‘He is taller than me.’ (Amele, Roberts 1987, 91)
30
(22) t¯a
he
bi
exceed
n¯ı
you
g¯ao
tall
‘He is taller than you’ (Mandarin, Li and Thompson 1981, 564)
(23) Ø-j`ær-\a-hi
3m.sg.s-surpass-\O-1sg
t-à-y@tte
fem-sg-intelligence
‘He is smarter than I’ lit: ‘He surpasses me in smartness.’ (Tamashek,
Heath 2005, 244)
On the whole, this type is roughly as common as the conjoined type,
comprising 33/167 languages in Stassen (2008) and 39/148 languages in my
broad sample.
The remainder (better than half) of the world’s languages make use of
some version of the standard construction. In this construction, comparison
is (superficially at least) monoclausal, with a special marker for the standard
of comparison, often a locative case-marking (or adposition), or sometimes
another element specific to comparatives, such as the English particle than.
In English and Russian (exemplified above), the standard is marked and
there is also analytic or synthetic marking of the property-denoting predicate.
Marking of the predicate is often optional, as in Modern Hebrew (24), and
in fact the most common strategy cross-linguistically has no (obligatory)
marking of the predicate, with overt comparative morphosyntax thus marked
only on the standard, as in the Japanese example in (25) and many other
languages.13
31
(24) Dan
D.
gavoha
tall
(yoter)
more
mi-meri
from-Mary
‘Dan is taller than Mary.’
(25) Sally-wa
Sally-top
Bill-yori
Bill-from
kasikoi
smart
‘Sally is smarter than Bill.’ (Beck et al. 2004, 327)
Whether languages of the Japanese type have a null comparative element
is a significant issue in the semantics literature (see Beck et al. 2004, 2009,
Kennedy 2007a, Oda 2008, Hayashishita 2009, among many others, and I
take up some discussion of this in light of Armenian in section 3.4.1).
In the most comprehensive typology work on comparison to date, Stassen
(1985, 2008), the morphological marking of the predicate is not considered.
Stassen’s typology focusses instead on the nature of the marking of the
standard. For example, Stassen draws a distinction between case-marked
standards, and particle-marked standards, such as English than, as different
major types of comparative construction. However, this distinction appears
to cross-classify with the type of morphological marking of the predicate of
interest here, and I have thus not followed this classification. All of the languages
in Appendix A and Appendix B, constituting the main data set for
the study of the CSG in particular, are drawn from the standard type, for
the reason that languages of this type were the only ones to exemplify the
phenomena of interest. By way of elaboration on that last observation, a few
points of general interest arise, which I mention briefly here.
32
For the conjoined and exceed-type comparatives, it is a matter of debate
whether expressions of this sort should be considered to constitute a
grammatical(-ized) comparative in any meaningful sense (Sapir 1944, Ultan
1972, Seuren 1984, Steele 1987). Is their interpretation simply the sum of
their (overt) parts, or do they involve any hidden grammatical comparative
element? Only recently has the semantics of these constructions come under
scrutiny (see Beck et al. 2009, Kennedy Forthcoming), with no conclusive
evidence of any hidden elements of comparative pieces, over and above what
translation equivalents in languages like English have. Regarding conjoined
comparatives, it may be noted that the languages of this type investigated for
the current study also typically lacked any clear overt comparative marker as
an obligatory component of these constructions. For a few such languages,
an element glossed ‘more’ or ‘comparative’ is given in examples — six of the
20 conjoined comparative languages in Stassen (1985) (namely Maori, Menomini,
Miskitu, Motu, Nahuatl, Samoan) have such elements in the glosses,
but as far as I have been able to determine from the grammatical descriptions,
these elements are general-purpose intensifiers, rather than comparative adverbs.
For example, Stassen (1985) provides the Miskitu example in (26a)
with an element glossed as a comparative marker, based on the description of
this element in Conzemius (1929, 80): “The adjective is compared by placing
before it the words kau, k¯ara, kanra, or kanmapa . . . ‘more’ for the comparative.
. . ”. Yet elsewhere, Conzemius (1929, 106-109) discusses the same
elements that occur in comparatives (in Miskitu, and in the other Chibchan
33
languages) as having a broader range of intensifying meanings, for example,
glossed as ‘still’ in (26b), where a comparative sense is clearly precluded by
the context. Examples in other languages tend to be of a similar nature.14
(26) a. YaN
I
kau
more
tukta,
young
man
you
almuk.
old
‘I am younger than you.’
b. . . . yapti-k¯ı
mother-my
kau
still
raya
alive
sa
cop?
‘. . . my mother is still alive.’
In any event, suppletion was not indicated in any description of the conjoined
comparative languages.
Like the conjoined comparatives, it is not clear that there is any hidden
comparative semantics, that would necessitate a grammatical comparative
marker other than the verb in ‘exceed’ comparatives. No language in
my sample has an identifiable comparative marker on the property-denoting
predicate in such a comparative construction; to the extent this predicate has
morphology different from what big would have in The bear is big, the morphology
is what is expected from the general syntactic environment (e.g.,
a nominalization, as in (23)), or is a general (non-comparative) emphatic
marker (see note 12).
Interestingly, a handful of ‘exceed’ comparative languages do show (apparent)
suppletion.15
Wolof and Fulfulde/Pulaar, two related Niger-Congo
languages lack a grammatical category of adjective, with quality-denoting
34
predicates patterning as verbs. These languages are typified by ‘exceed’ comparative
strategies, however, they do have a handful of verbs with inherently
comparative meanings, such as Wolof sut ‘be taller than’, used without the
‘exceed/suprass’ verb, as in the Wolof examples in (27a-d) from Mc Laughlin
(2004 and personal communication); (see Labouret 1952 for comparable Fulfulde
examples). Note that the comparative verb, unlike canonical examples
of suppletion, does not block the regular ‘exceed’ comparative for njool ‘tall’,
as in (27e), from Munro and Gaye (1991, 49).
(27) a. Ibu
Ibu
moo
3s.subj.focus
nay
be.miserly
‘Ibu is miserly’
b. Ibu
Ibu
moo
3s.subj.focus
gën
surpass
a
lv
nay
be.miserly
Faatu
Faatu
‘Ibu is more miserly than Faatu.’
c. Faatu
Faatu
dafa
3s.V focus
njool
be.tall
‘Faatu is tall.’
d. Faatu
Faatu
dafa
3s.V focus
sut
be.taller.than
Ibu
Ibu
‘Faatu is taller than Ibu.’
e. Moo
3s.subj.focus
gën-a
surpass-lv
njool
be.tall
Aamadu
Amadou
‘He’s taller than Amadou.’
35
Similarly, in the unrelated Berber language Tamashek, the verb -Vjvr‘surpass’
is used (with a nominalization of the property predicate) to form
comparatives and superlatives (see (23), but there are also verbs with intrinsically
comparative meaning, including at least -VfV- ‘be better than’ (Heath
2005, 245).16
As far as I can determine, these languages do not have a distinct
superlative construction for which the CSG would make predictions,
and since the languages do not draw a distinction between adjectives and
verbs, it is not clear what expectations may hold for an analogue to the
C∆G (cf. remarks on inceptives in section 6.4.4). The apparent suppletion
in these language types is noted here, but not explored any further.
Returning to the standard comparative languages, and in particular those
with morphological marking, by way of a passing etymological observation,
it is perhaps worthy of note that there appears to be a recurring source
for comparative affixes, namely, in morphemes with a meaning like ‘rather’,
‘more or less’, or ‘in contrast to others.’ Benveniste (1948, 124-126) suggests
that the original meaning of the Proto-Indo-European *-i
“
os was to defuse the
force of the positive. Across Turkic the suffix -rak (in its many surface forms)
forms variously comparatives, as in Karaim (Musaev 1966, 267) and Uzbek
(Reshetov 1966, 346), or diminutives (cf. English -ish, Russian -ovat-), as
in Khakass (Karpov 1966, 434) or Nogaj (Baskakov 1966, 284). Similarly,
the reflexes of Finno-Ugric -*mp are comparative affixes in some languages
(Finnish, Hungarian), but have meanings like ‘rather’ in others (Nenets); see
Fuchs (1949). Among adverbs as well, in Lahu, the adverb a-cí has the basic
36
meaning ‘a little, sowewhat’, but when used with a property-denoting stative
verb, serves to form the comparative (Matisoff 1973, 273-274). An exception
to this pattern is Chukchi, where the comparative affix appears to be an
oblique case marker (see Skorik 1977, 334).
This completes the review of some theoretical, methodological and typological
background that informs the remainder of this book. On now, to the
phenomena of interest.
37
38
Chapter 2
Comparative Suppletion
2.1 Introduction
In many languages, a handful of adjectives form their comparative grade via
a root (or base) that is etymologically unrelated to the positive root. It is
for this phenomenon that the term ‘suppletive’ (German: suppletorisch) was
originally coined in Osthoff (1888, 1899). A sample of examples is given in
(28), with roots boldfaced:1
39
(28) pos cmpr
a. English: good bett-er
b. Latin: bon-us mel-ior ‘good’
c. Icelandic: gamall eld-ri ‘old’
d. Georgian: k’argi-i u-mˇob-es-i ‘good’
e. Upper Svan: žG@d xo-š-a ‘big’
f. Abkhaz: a-cw
gj
a j-ejcw
o-w ‘bad’
g. Basque: on hobe ‘good’
h. Estonian: hea pare-m ‘good’
In most such examples, the suppletive comparatives have regular comparative
morphology but the root is not supplied by or phonologically derived
from the positive base. Pairs of suppletive and non-suppletive comparatives
in (29) illustrate this. (In a small minority of examples, such as English bad
– worse, and the Basque example in (28g), the comparative morpheme is
missing, a situation which we return to in section 2.2 and 5.3.1.)
40
(29) pos cmpr
a. English: good bett-er
long long-er
b. Latin: bon-us mel-ior ‘good’
dur-us dur-ior ‘hard’
c. Icelandic: gamall eld-ri ‘old’
stór stær-ri ‘strong’
d. Basque: asko gehi-ago ‘most’
zoro zoro-ago ‘crazy’
e. Votic üvä par@-pi ‘good’
süvä süve-pi ‘deep’
Adjectives in some languages also have a superlative grade, with the
meaning “more adj than all others.” (Recall from the introduction that
throughout this work, unless otherwise noted, I will restrict the term superlative
to the sense of relative superlatives, excluding the “absolute” superlatives
such as Italian bell-issim-a ‘very beautiful’.) Suppletion also extends to the
superlative grade. In the majority of cases, the superlative and comparative
forms share a common root, distinct from that of the positive (Ultan 1972,
144), as in (30). I will refer to this as an ABB pattern, to indicate that the
shared root (B) in the comparative and superlative is distinct from that of
the positive (A).
41
(30) pos cmpr sprl
a. English: good bett-er be-st
b. English: bad worse wor-st
c. Danish: god bed-re bed-st ‘good’
d. Czech : špatn-ý hor-ší nej-hor-ší ‘bad’
e. Georgian: k’argi-i u-mˇob-es-i sa-u-mˇob-es-o ‘good’
f. Estonian: hea pare-m par-im ‘good’
g. Kildin Saami: šig’ pEr’-am pEr’-mus ‘good’
h. Basque: asko gehi-ago gehi-en ‘many’
i. Kabardian: kwad, ba nax nax-deda ‘many’
j. Cherokee: osda dajehla wi-dajehl-˜2Pi ‘good’
Another suppletive pattern that is rare, but nevertheless attested, is the
ABC pattern, in which each grade is built on a distinct root. The Latin,
Welsh, Old Irish, and Middle Persian triples meaning ‘good – better – best’
constitute ABC patterns, as shown in (31):
(31) pos cmpr sprl
a. Latin: bon-us mel-ior opt-imus ‘good’
b. Welsh: da gwell gor-au ‘good’
c. Old Irish: maith ferr dech ‘good’
d. M. Persian: x¯ob weh/wah-¯ıy pahl-om/p¯aš-om ‘good’
I will argue below that ABB and ABC are the only attested suppletive
patterns, and that a handful of apparently divergent patterns should be reanalyzed.
The state of affairs regarding comparative suppletion may then
42
be schematized as in (32). Of five logically possible patterns of root identity
and suppletion, only three are attested.
(32) pos cmpr sprl
a. regular A A A big – bigger – biggest
b. suppletive A B B good – better – best
c. doubly-suppletive A B C bonus – melior – optimus
d. unattested A B A *good – better – goodest
e. unattested A A B *good – gooder – best
This patterning was noted briefly in the closing paragraphs of the only
previous study of the morphology of comparison (Ultan 1972). For a 20language
sample, Ultan notes that “suppletive paradigms in the comparison
of adjectives almost always imply formal identity or near-identity of the bases
shared by the comparative and superlative vis-à-vis those shared by the positive
and equative” (Ultan 1972, 144). Ultan’s generalization is robustly
supported in the larger survey reported here. Indeed, I submit that (a slight
reformulation of) Ultan’s generalization is a strong contender for the status
of a linguistic universal.2
For reasons that will become clear as we proceed,
I suggest breaking the generalization into two pieces, to cover (32d) and (e)
separately. We may thus formulate the two-part generalization in (33) and
(34), referring to the whole generalization as the CSG:
43
(33) The Comparative-Superlative Generalization, part I (CSG1):
If the comparative degree of an adjective is suppletive, then the superlative
is also suppletive (i.e., with respect to the positive).
(34) The Comparative-Superlative Generalization, part II (CSG2):
If the superlative degree of an adjective is suppletive, then the comparative
is also suppletive (i.e., with respect to the positive).
The CSG (in both parts) ranges over synthetic, i.e., morphological, comparative
and superlative grades. Analytic (periphrastic) constructions show
different behavior (I return in the next chapter, section 3.4, to why this is
so). The immediate goal of this chapter is to propose an explanation of this
apparent universal and to consider some consequences of that explanation.
The explanation I offer has two key components (both qualified below). First,
the representation of the superlative must properly contain (be derived from)
that of the comparative. Second, suppletion constitutes a special case of rules
of exponence (also called vocabulary insertion), introducing distinct root formatives
into specific contexts. Such rules are most properly formulated in
terms of a realizational theory of morphology (in the sense of Stump 2001),
in which morphological rules spell out a derivationally prior complex (i.e.,
syntactic) structure. Part 1 of the CSG in (33) follows essentially from these
two assumptions. An additional assumption, that there is an adjacency condition
on context-sensitive rules of exponence, extends the theory to account
for the CSG2.
44
The remainder of this chapter is organized as follows. First, I will sketch
the bare bones of the analysis, presenting the leading ideas, but leaving details
and motivation of assumptions aside. I then turn to a type of poverty-ofthe-stimulus
argument that this data raises. The key generalizations concern
gaps, and since suppletion is such a marginal phenomenon in any one language,
there is no way in which the data available to any given learner could
be robust enough to distinguish accidental from systematic gaps in their language.
I elaborate on this argument for the hand of Universal Grammar,
arguing against an appeal to historical accident, in section 2.3. Once the
reader can see where we are headed, I turn in Chapter 3 to a discussion of
independent evidence for one key assumption, and then a prevent a refinement
of some details of the theory, necessitating many asides (some of which
reveal fruitful additional generalizations).
In Chapter 4 I will lay out in more detail the results of a cross-linguistic
investigation of (33). In a relatively comprehensive survey of comparative
suppletion encompassing more than a hundred distinct examples (there are
some tricky issues in counting, discussed below), there are but a handful of
prima facie counter-examples, for which alternative analyses are proposed
in sections 4.1 and 4.3. If the apparent counter-examples can indeed be
explained away, as I suggest, then the ABB and ABC patterns are indeed the
only attested patters — no adjective shows an unambiguous ABA pattern,
i.e., hypothetical *good – better – goodest or *bonus – melior – bonissimus in
which the comparative alone is suppletive, with the positive and superlative
45
sharing a common root.
2.2 *ABA - Explaining a gap
The account of the CSG begins by assuming that the representation of the
superlative properly contains that of the comparative in all languages. I call
this the Containment Hypothesis. Note that this embedding is transparent
in the overt morphology in many languages, as in the Czech and Georgian
examples in (30), and in the non-suppletive examples in (35):
(35) pos cmpr sprl
a. Persian: kam kam-tar kam-tar-in ‘little’
b. Cimbrian: šüa šüan-ar šüan-ar-ste ‘pretty’
c. Czech: mlad-ý mlad-ší nej-mlad-ší ‘young’
d. Hungarian: nagy nagy-obb leg-nagy-obb ‘big’
e. Latvian: zil-ais zil-âk-ais vis-zil-âk-ais ‘blue’
f. Ubykh: nüs◦
@ ç’a-nüs◦
@ a-ç’a-nüs◦
@ ‘pretty’
Further evidence for the Containment Hypothesis (and some qualifications)
will be provided in the next chapter. The most straightforward (but
not the only) means of representing containment is via a nested structure,
in which the superlative is derived from the comparative by the addition of
some morpheme. Thus we have the Nesting Hypothesis in (36a) as a special
case of the Containment Hypothesis.3
Under certain assumptions, this can
be seen as a cashing out of the markedness hierarchy positive < compara-
46
tive < superlative proposed in Greenberg (1966), Canger (1966) and Ultan
(1972), a topic to which we return.4
On the Nesting Hypothesis, despite
appearances, the representation in (36b) is thus incorrect for languages like
English; there must be a “hidden” comparative element even in forms such
as biggest.5
(36) a. [ [ [ adjective ] comparative ] superlative ]
b. * [ [ adjective ] superlative ]
The structure in (36a) can of course be represented as a tree diagram, as in
(37). I use the convention of allcaps to represent the abstract morphemes,
and put aside the questions of the proper labels for non-terminal nodes. Thus
a, c, s are simply mnemonics to refer to specific nodes in the diagrammes.
Also, as in (36a), I represent this as a suffixing structure, though what is of
interest is the hierarchy (constituency), not linear relations (for example, the
superlative exponent is a prefix in Czech).
(37) a. positive b. comparative c. superlative
a
adj
c
a
adj
cmpr
s
c
a
adj
cmpr
sprl
47
Consider now the nature of the rules of exponence that will insert phonological
material at the nodes in (37). The first case to consider is Czech,
a language in which the nesting structure in (37) is morphologically transparent,
as noted in descriptive grammars of the language (e.g., Janda and
Townsend 2000). The relevant forms of two Czech adjectives are given in
(38).
(38) pos cmpr sprl
a. ‘young’: mlad-ý mlad-ší nej-mlad-ší
b. ‘bad’: špatn-ý hor-ší nej-hor-ší
A fragment of the Czech rules of exponence (the Vocabulary) is given in
(39).6
(39) a. sprl → nejb.
cmpr → -ší
c. young → mladd.
bad → hor- / ] cmpr
e. bad → špatnThe
rules in (39a-c) account for the three forms of the regular adjective
mlad-ý (additional rules insert inflectional exponents, including the masculine,
singular, nominative -ý in the citation form). The exponents simply
realize the corresponding terminal nodes in the structures in (37).
Where a regular adjectival root such as mlad- ‘young’ has only a single
form, suppletive roots have, by definition, multiple allomorphs. The Czech
48
root meaning ‘bad’ has two: one (hor-) introduced by the context-sensitive
rule in (39d), and the other (špatn-) introduced by the context-free rule
in (39e). The logic of elsewhere ordering ensures that the more specific
allomorph is chosen whenever possible. Thus, both allomorphs compete for
insertion in the comparative structure in (37b), and the allomorph hor-, being
the most specific form available, wins. In the positive structure (37a), the
environment for hor- is not met, and thus the elsewhere allomorph špatn- wins
by default. Now notice what happens in the superlative. By assumption, the
representation of the superlative (properly) includes that of the comparative.
Therefore, the context for the “comparative” allomorph hor- is met, and that
allomorph is selected. The rules in (39a-b) operate as before, supplying
exponents to the comparative and superlative nodes. The superlative, with
the proper exponents, is shown in (40):
(40) s
sprl
nej-
c
a
adj



hor-
*špatn-



cmpr
-ší
49
This constitutes the first key result: the combination of assuming a nested
structure and applying the elsewhere logic to root allomorph selection yields
an ABB pattern as an automatic consequence whenever an adjective has a
suppletive comparative and nothing further is said. Because the comparative
is contained in the superlative, the comparative allomorph (of an adjectival
root) will automatically be compatible with the superlative context as well,
and will necessarily block the positive (default) allomorph of that root. Thus
the ABA pattern (a return to the positive root in the superlative) is effectively
excluded, yielding the core content of part 1 of the CSG.
Now, although the logic just sketched correctly excludes ABA patterns,
it nevertheless does allow for patterns beyond ABB. In particular, the ABC
pattern is readily describable. The comparative root allomorph will necessarily
block the positive in the superlative context, but the comparative
allomorph can in turn be blocked by an even more highly specific exponent,
as in the allomorphs of the Latin root for ‘good.’ The Vocabulary fragment
in (41) provides the rules of exponence that will derive the ABC pattern for
this adjective (in Chapter 5, I will revise the somewhat clunky formalism for
the superlative context in (41a)):
50
(41) Latin a. good → opt- / ] cmpr ] sprl ]
b. good → mel- / ] cmpr ]
c. good → bond.
sprl → -imus
e. cmpr → -ior
As it happens, one additional clean up is needed for the Latin examples.
The rules in (41) provide the correct root allomorph in each context, but will
generate an overt comparative suffix inside the superlative. While this was
correct for Czech (and many other languages, see Chapter 3), the rules as
stated will incorrectly give *opt-ior-imus in place of opt-imus for Latin. To
derive the surface forms in which the comparative marker is not visible in
the superlative, I posit that the comparative morpheme has a phonologically
null allomorph that occurs in the context of the superlative, as in (42). The
standard elsewhere logic will ensure that the null allomorph wins out over
the regular comparative (43b) in the superlative, just as it selects the correct
root exponents in the suppletive cases so far examined.
(42) cmpr → Ø / ] sprl ]
The revision to the context for (41a) to be offered in Chapter 5 will in
fact avoid the need for (42) for the Latin examples just given, but the issue
is a more general one, and a null allomorph has broader application. For
example, adding (42) to the English rules of exponence in (43) will correctly
51
generate bigg-er, as opposed to *bigg-er-est.
(43) a. big → big
b. cmpr → -er
c. sprl → -est
The assumptions above serve to effectively derive the CSG1 from the
Containment Hypothesis and the assumption that rules of exponence are
subject to Elsewhere ordering: the ABA pattern is unstatable as a formal
pattern. If there are only two distinct listed root forms in a language’s
vocabulary, no ordering of the rules introducing these roots will lead to an
ABA pattern. But the attested patterns, ABB and ABC are readily derived,
with ABB in some sense being the unmarked case of comparative suppletion,
as against the ABC pattern, which requires an additional rule.
Now, there is a loophole, in the sense that it is in principle possible to
generate a surface ABA pattern by appeal to accidental homophony. The
grammatical pattern would necessarily be an ABC pattern, with three formally
distinct root allomorphs, but where (the analogue of) (41a) would
introduce the same phonological matrix as (41c), sandwiching the comparative
allomorph between them. I suggest that the grammar per se does not
exclude this possibility, but invoke the well-worn idea that there is a general
anti-homophony bias in acquisition, of which (44) is a special case:
(44) Anti-Homophony:
Learners avoid positing a contextual allomorph of a morpheme µ that
is homophonous with the default exponent of µ.
52
If one thinks of rules of exponence as overwriting or rewriting rules, the
essence of (44) is that learners will posit no rule of exponence of the form
X → X, which effects no change but serves only to block some more general
rule. Irregular forms that appear to have no change from the base, such as
the English irregular plural sheep (singular sheep) or past tense hit must thus
involve a zero affix (contrast Anderson 1992). The proposal in (44) closes the
loophole that would otherwise allow for false ABA patterns, arising as ABC
patterns in disguise, despite the grammar universally excluding true ABA
patterns. This leaves some wiggle room; one could consider the homophony
strategy as a retreat of last resort with the implication perhaps that observationally,
the CSG1 should emerge as a trend, rather than an absolute
(compare Pertsova 2007). Accidents do happen after all. I will nevertheless
resist this, and at least as a working strategy pursue the stronger claim
that (44) is never violated, and thus that apparent ABA patterns must find
alternative explanations. I return to this point in section 4.2 and Chapter 3.
Up to the loophole just mentioned, the preceding paragraphs have served
to derive the first part of the CSG as a straightforward consequence of two
main assumptions: the Containment Hypothesis and Elsewhere ordering.
Another component of morphological theories is the question of locality. Various
authors across a variety of frameworks have argued for an adjacency (or
contiguity) condition on allomorphy, such that the trigger for suppletive allomorphy
must be adjacent to the root that undergoes allomorphy (see Siegel
1978, Allen 1979, Embick 2003, 2010; see Hay 2000 for a brief review of the
53
earlier proposals). Adopting this assumption will serve to extend the account
above to the second part of the CSG, excluding the unattested *AAB pattern
*good – gooder – best, as in (32e). Under the Containment (specifically, the
Nesting) Hypothesis, an attempt to derive an AAB pattern via the vocabulary
items in (45) will fail, as the context in (45a) falls afoul of the adjacency
condition.
(45) a. good → be(tt)- / ] sprl ]
b. good → good
Of course, care must be taken in formalizing this such that whatever device
is used to condition the superlative root allomorph in the ABC cases (see
(41a)) must be unavailable for putative, but unattested, *AAB cases in (45).
I postpone further discussion of this point until section 5.3, concentrating on
the *ABA cases until then.7
Consider now what happens if the Nesting (or Containment) Hypothesis
were not adopted. Standard descriptions of English (including the majority
of formal treatments of the superlative, such as Szabolcsi 1986, Heim 2000,
Hackl 2009) posit two morphemes, the comparative and the superlative, both
of which attach to the adjective directly, as in (46).
(46) a. [ [ adjective ] comparative/-er ]
b. [ [ adjective ] superlative/-est ]
A variation of this description posits a single degree head (deg) as in
(47), of which both the comparative and superlative are possible values.
54
(47) [ [ adjective ] degree ]
Theories starting from these assumptions will be able to describe the attested
patterns of root suppletion (although something additional needs to be
said in order to insert both comparative and superlative morphology in languages
with overt nesting, like Czech). Under (47), the ABB pattern among
the roots in Czech could be described as in (48) (compare to (39) above), by
making the degree head (common to both comparative and superlative) the
context for the allomorphy.
(48) a. bad → hor- / ] deg
b. bad → špatnAlthough
reference to deg may look like a convenient means of representing
the ABB patterns, it appears to fail for languages that have more
degree affixes than just comparative and superlative.8
Ultan (1972) noted
that the common pattern of shared suppletive roots groups comparative and
superlative together, to the exclusion of the equative degree, even where this
can be affixal, as in Welsh (49).
(49) pos cmpr sprl equative
drwg gwaeth gwaeth-a cyn-ddrwg ‘bad’
The same point can be made with reference to the intensified or absolute
superlative degree, marked by the prefix pře- in Czech (as in (50a-b)), or
other intensifiers in the language, such as the suffix -ink- (50c). In both these
cases, positing a suppletive allomorph in the context ] degree ] would
55
incorrectly overgeneralize that allomorph to the equative and intensified or
absolute superlative forms, respectively.9
(50) pos cmpr sprl a.sprl/intns
a. dobr-ý lep-ší nej-lep-ší pře-dobr-ý ‘good’
b. špatn-ý hor-ší nej-hor-ší pře-špatn-ý ‘bad’
c. mal-ý men-ší nej-men-ší mal-ink-ý ‘small’
Even if we restrict the discussion to languages without these extra degrees,
the ABC patterns establish that reference to degree alone is insufficient, and
that statements of contextual allomorphy must be able to distinguish between
comparative and superlative. Thus, under (46) or (47), Latin would require
a treatment as in (51).
(51) a. good → opt- / ] sprl ]
b. good → mel- / ] cmpr ]
c. good → bonAnd
therein lies the rub. Consider a pattern exactly like Latin, but
without the rule in (51a). In a theory with the Containment Hypothesis,
elsewhere ordering ensures that the allomorph in (51b) will automatically
extend to the superlative environment, yielding the ABB pattern. But in a
theory lacking the Containment Hypothesis, no such extension is guaranteed.
The comparative environment is just that — the comparative — and the basic
pattern is thus the unattested *ABA (*bonus – melior – bonissimus). Put
differently, without the Containment Hypothesis, nothing precludes direct
56
reference to the comparative context alone, and it is this possibility which
prevents the theory from deriving (hence explaining) the CSG. At the risk of
redundancy: the argument here is not that the theories differ in their abilities
to describe the attested patterns; what is at stake is whether the unattested
*ABA pattern is excluded by the theory for principled reasons.
This concludes the presentation of the core analysis. There are numerous
refinements and clarifications to be presented, and it is these, as well as
more careful discussion of the empirical basis for the claims, and apparent
counter-examples, that will occupy the next chapters. One obvious question
is why the Containment Hypothesis would hold, and relatedly, how it might
be formally cast within UG. I will postpone discussion of this until Chapter
7, offering only a rather incomplete speculation even there. As noted
in the introduction, the suggestion I will make in that chapter is that UG
imposes limits on the complexity of functional morphemes — the combination
of pieces needed to compose a superlative (namely, the comparative and
something yielding “than all (others)”) is simply too big to fit into a single
functional head. The Containment Hypothesis is thus not formally a part
of UG, but rather a corollary of the complexity of its meaning. Before addressing
the details of the morphological theory, and the empirical minutiae,
I wish to step back to consider (and reject) an alternative approach to the
CSG which would treat is as a historical accident, seeking an externalist
explanation rather than the internalist mechanism of UG.
57
2.3 UG vs. the European Sprachbund
In the preceding section, I have sketched a preliminary account of the CSG.
The account derives this generalization from the Containment Hypothesis
and from the assumption that suppletion is to be modeled as contextual allomorphy,
implemented by Rules of Exponence (Vocabulary Insertion), subject
to elsewhere ordering. Arguments from gaps (of the sort characterized by the
CSG) rest implicitly on the premise that the missing pattern is unlikely to
be merely an accidental gap. This is essentially a statistical argument, even
if it is not formalized as such, and this raises two points within the current
context.
2.3.1 The insignificance of the evidence
The first point is a newish twist on the familiar poverty-of-the-stimulus type
of argument for Universal Grammar. If the CSG is a valid generalization, as
I contend here, then it is not something that can be learned from the data
available to a normally developing child. Since suppletion is such a marginal
phenomenon internal to any one language, with so few relevant forms (if any
at all), the Primary Linguistic Data to which any one child is exposed is far
too sparse to warrant any conclusions about impossible suppletive patterns.
In other words, the absence of the ABA pattern in any given language is
in and of itself insignificant. The generalization is only significant, and thus
worthy of attention, in its cross-linguistic context.
58
This point can be appreciated with reference to English, which has a
handful of suppletive comparatives (better, worse, more, less), all of which
participate in ABB patterns. The ABA pattern is indeed absent from English,
but with only four triples (two of which have further irregularities),
it would seem rash to infer that this is a significant absence, indicative of
a general property of UG. Indeed, the ABC pattern is just as absent from
English as the ABA pattern is, yet it would be simply false to conclude that
the ABC pattern is disallowed by UG, as it is indeed attested in other languages
(Welsh and Latin examples were given above). Other languages show
a similar situation, many with but a single element entering into suppletion
(and the vast majority having of course none whatsoever).
The explanation of the contingent fact that Modern English good compares
on an ABB pattern good – better – best, and not some other pattern,
certainly lies squarely in the history of the language: this pattern is inherited
from previous stages of English (ultimately from the earliest common
Germanic ancestor). If the ancestor of modern English had had a different
pattern (say along the lines of good – better – finest), then that pattern would
presumably have been the one Modern English inherited. The existence of
ABB and ABC patterns demonstrates that children are equipped to acquire
those patterns on the basis of evidence in the input, and the account of what
any one child acquires needs no appeal to UG beyond the general ability to
learn these patterns from the input. Where UG is invoked instead is at the
level of the broad, cross-linguistic generalization. UG explains (in the man-
59
ner laid out in the previous section) why, from among the many suppletive
patterns that have arisen in a variety of languages, no genuine ABA or AAB
patterns are to be found. In an important sense, then, UG places bounds
on possible language change (cf. Kiparsky 2008): no language can undergo
a change that yields a genuine ABA or AAB pattern, since no learner could
posit a grammar that would accommodate such a pattern.
In sum, the logic here is that the CSG cannot be learned from the data.
It must therefore either be spurious, holding true only accidentally (I argue
against this Chapter 4), or it must be attributed to some more general constraint
imposed on possible grammars that a learner may posit. In principle,
this yields two options: either an inherent grammatical constraint (UG), or a
language-external consideration, such as some general principle of cognition.
So far as I can see, there are no current candidates for a general cognitive
or functional principle that would exclude the ABA or AAB patterns. For
example, the markedness hierarchy positive < comparative < superlative is
discussed in the functionalist literature by Heine (1997, 124-126), yet rather
than providing independent evidence for some language-external consideration
that will derive the hierarchy, Heine approaches the matter from the
other direction, arguing that the evidence for this hierarchy is (thus far)
purely linguistic, and suggesting therefore that some (as yet unknown) functional
pressure must exist. Absent a cognitive account, the best alternative is
that some a fortiori constraint on mental grammars derives the CSG. I have
offered the Containment Hypothesis (in tandem with elsewhere ordering) as
60
the relevant constraint, as a property of Universal Grammar.10
2.3.2 Counting cognates
This brings us to the second point in connection with taking the the evidence
for the CSG to be significant only in the broader, cross-linguistic context, and
this is a more narrowly methodological issue. While there are numerous examples
of suppletion in the material assembled here, and while the handful
of apparent counter-examples seem likely to fall to an alternative account
(see Chapter 4), there is nevertheless a very restricted distribution (in both
geographic and genetic terms) to the component phenomena of the CSG
(suppletion in adjectival gradation and morphological comparatives and superlatives).
Outside of the Greater European Sprachbund (Indo-European,
Finno-Ugric, Kartvelian, Northwest Caucasian and Basque), the CSG is trivially
true. With the notable exception of the Cherokee example in (30j), the
languages that supply the key evidence represent a large, but contiguous,
geographic area. This raises the concern that the CSG, even if descriptively
valid over hundreds of examples from numerous languages, could at its core
be an effect of shared vocabulary, to be explained by a mixture of common
inheritance and borrowing (or calquing) under contact, rather than through
an appeal to a universal structural characteristic.
I suggest that this worry can be mitigated by looking more closely at the
individual examples and being more careful about what we count in constructing
the quantitative argument that the gap is not accidental. First,
61
we must give up counting languages. In large geographically and genetically
diverse samples, counting languages (as commonly understood) may be reasonable
at getting a rough measure of the distribution of some phenomenon
(see Bickel 2008 and references therein for relevant discussion). But in the
case at hand, language-counting faces intractable problems. Most obvious
among these is definitional: are Norwegian and Swedish (or Serbian and
Croatian) one language, or two, or more? How many varieties of English are
there? Even if we settle on some (essentially arbitrary) measure, the fact
remains that many of the attested suppletive paradigms are not independent
of one another. Consider, for example, the following Germanic paradigms
for ‘good’:
(52) pos cmpr sprl
a. English good bett-er be-st
b. German gut bess-er (am) be-st-en
c. Gothic goþ-s bat-iz-a bat-is-t-s
d. Afrikaans goed bet-er be-ste
e. Swedish god bätt-re bä-st
f. Cimbrian guat pez-ar pez-ar-ste
Almost all Germanic languages, spanning the globe from Norwegian to
Afrikaans, from the oldest attested (Gothic) to the present, share a paradigm
whose pieces are cognate to English good – better – best. Undoubtedly, this
pattern arose exactly once, at some point in the dawn of Germanic (if not
before), and the explanation of all of the attested triples is that they have
62
inherited it from a common source. In weighing the evidence for the CSG,
then, these cannot be counted as independent data points, and must collectively
be counted as a single data point. To abstract away from the surface
variation, I will use all caps in representing the roots in the cognate sets; (52)
is thus a single cognate set: GOOD – BE(TT) – BE(TT).
This same consideration must apply in evaluating the other examples
brought to bear on the evidence for the CSG. In light of the close genetic and
areal affinities in the data pool, it is clear we are not drawing from a random
sample, but must aim for a comprehensive survey of the individual attested
data points, where each data point is a distinct cognate triple (positive –
comparative – superlative). As illustrated with Germanic, whenever there is
a common cognate triple, shared vocabulary is the most ready explanation.
It is only where the triples are distinct that there must have been some
linguistic change, and it is only in the changes that common inheritance
cannot be the entire account.
Pursuing this further, I take it that two cognate sets are distinct from one
another, if (and only if) they differ in at least one of their component roots,
though they need not differ in all. Thus, while (52) represents the majority
of Germanic, there are two or three additional cognate sets for Germanic
‘good’, illustrated in (53):
63
(53) pos cmpr sprl
a. Swedish bra bätt-re bä-st
b. Giazza Cimbrian guot ăeg-ur ăeg-ur-ste
c. Old English gôd sêl-(ra) sêl-ost
Swedish (53a) (also Norwegian) overlaps with (52) in the comparative and
superlative, but differs in the positive root, thus we have BRA – BE(TT) –
BE(TT). Another triple in (53b) is GOOD – WAEH – WAEH from the Cimbrian
German variety spoken in Giazza (Italy), which retained the positive
root (cognate to) GOOD, but forms comparative and superlative grades on a
root descended from Middle High German wäh(e), ‘artful, fine, dainty, good,
etc.’ (Schweizer 2008, 397 and DWB, vol. 27). Old English gód ‘good’ had
two comparatives and superlatives, with the forms in (53c) attested alongside
a GOOD – BE(TT) – BE(TT) triple (Sievers 1882, 108, Bosworth and Toller
1898, 858). Although these triples overlap in part with GOOD – BE(TT) –
BE(TT), each is counted as a distinct triple, as each represents a separate
innovation, and thus, in a sense, a distinct chance for an ABA pattern to
have arisen.
Approaching the matter in this way, I count a total of somewhere just over
one hundred distinct suppletive triples; see Chapter 4. (Providing an exact
figure still requires some arbitrary choices, such as in cases where the relation
between positive and comparative forms is not 1:1.) Of these triples, there
are a handful of potential problems for the CSG: the Basque word for ‘good’
which shows a suppletive comparative, but a doublet in the superlative: on
64
– hobe – hobe-ren/on-en; and apparent ABA patterns in the ‘many – more
– most’ triples in Karelian, Aremian and Bulgarian/Macedonian. I discuss
these examples in more detail in Chapter 4, offering tentative alternative
explanations for each case. The remainder are mostly ABB patterns, with a
few ABC cases like Latin bonus – melior – optimus. No putative AAB cases
are attested in the data.
In sum, despite the areal and genetic limitations, there is sufficient variation
within the data to be reasonably confident that the *ABA and *AAB
gaps are not merely the accidental result of shared vocabulary. It is a real
and systematic gap, in need of an explanation.11
2.3.3 (In)stability and change
A further argument against attributing the absence of an ABA pattern to
shared vocabulary, whether through inheritance or borrowing, is that the
vocabulary itself is quite typically not shared across languages, even where
the overall pattern is. This can be seen by perusing the cognate lists in
Table 4.1 in Chapter 4. While there is certainly an areal affect of having
suppletive patterns for core adjectives (GOOD, BAD, BIG, MANY...), there
is no evidence to support the claim that the patterns arise from borrowing
of adjectives (as lexical items) across languages. Similarly, it is not the
functional vocabulary as such (the morphological exponents of comparative
and superlative) that are borrowed among languages leading to the areal
diffusion of the nesting pattern.12
Languages with transparent nesting vary
65
widely in the resources they draw on to derive the nested structure (see the
next chapter).
Even within a single family, there is significant variation in the actual
lexical items (exponents) that make up a given suppletive pattern. Where
Germanic shows a remarkable stability in the GOOD – BE(TT) – BE(TT)
pattern, not all patterns are nearly as stable. The Slavic ‘good’ patterns
show remarkable volatility. In the earliest documented language, Old Church
Slavonic (OCS), the form for ‘good’ had a regular comparative, and alongside
this, had additional suppletive comparative options.13
(54) pos cmpr
a. dobr˘u dobr-ěi
b. — luč-˘ıi
c. — unj-˘ıi
(d. — sul-˘ıi)
Bulgarian and Macedonian alone among the daughter languages preserve
a regular comparative for ‘good’. Exclusively suppletive patterns have arisen
in all the other languages, most resolving to a single option, rather than the
one:many situation attested in OCS.14
66
(55) pos cmpr sprl
a. Bulgarian dob@r po-dob@r naj-dob@r
b. Czech dobr-ý lep-ší nej-lep-ší
c. Sorbian (%) dobr-y redl-iši
d. Serbian dobar bol-ji naj-bol-ji
e. Ukranian dobr-yj krašč-yj naj-krašč-yj
f. Ukranian harn-yj krašč-yj
g. Russian xoroš-ij luč-še (nai-luč-š-ij)
The consistency of the Slavic patterning (all suppletive patterns are ABB)
is not the result of shared vocabulary, since the vocabulary items (roots)
themselves are not shared. None of the suppletive patterns in (55) reflects
roots in both the positive and comparative from the OCS patterns in (54).
The Ukranian and Czech comparative roots were not (so far as available
sources indicate) comparatives of ‘good’ in OCS, and bol-ji (55d) was one of
two suppletive comparatives for ‘bigger’ in OCS (Vondrák 1900, 160, Lunt
1959, 66). Russian retains an OCS comparative root, but the positive root
dobr- has been usurped by xoroš-. What the languages have in common is not
the specific vocabulary, but rather the more abstract ABB pattern (Mel’čuk
2006, 456).
Similarly, one finds a common ABB pattern for ‘small – smaller – smallest’
in a variety of Indo-European languages, with a shared comparative root minbut
variation in the positive roots:15
67
(56) pos cmpr sprl
a. Latin parv-us min -or min-imus
b. OCS mal-˘u m˘ınj-˘ıi
c. Czech mal-ý men-ši nej-men-ši
d. Gothic leit-il-s minn-iza minn-ists
e. Danish lille / små mind-re mind-st
f. Ancient Greek m¯ıkr-ós meí-on me˜ı-st-os
On the basis of such evidence, one might contend, then, that it is not vocabulary
that is shared (by borrowing and/or inheritance) but rather something
more abstract, such as the ABB pattern. The account offered here
explains why that pattern is shared: it is precisely the pattern that arises by
default, whenever the comparative is suppletive, if (i) the Containment Hypothesis
holds, and (ii) rules of exponence are subject to elsewhere ordering.
Simply declaring that abstract patterns may be inherited or borrowed does
not provide any clear alternative – it merely restates the description which
the UG-based account aims to explain. Moreover, there is clear evidence
that the overall patterning is susceptible to change, inasmuch as suppletive
patters rise and fall.
The Slavic examples for ‘good’ illustrate this point. OCS had a one:many
pattern, with the suppletive comparatives forming ABB doublets alongside a
regular AAA pattern (54a). Yet, none of the daughter languages retains the
OCS pattern. Simply saying that abstract patterns are retained strikingly
fails in this case to explain the facts.16
68
Similarly, adjectives that show a suppletive pattern in one language may
undergo regularization in daughter languages (ABB → AAA), sometimes
giving rise to doublets in the daughter languages, the opposite of the change
in Slavic ‘good’ paradigms. Compare thus Swedish god – god-are – god-ast,
a regular triple for ‘good’ in the sense of ‘pleasant-tasting’, or colloquial
English bad – badder – baddest, (the only morphological comparative form
available for ‘bad’ in the sense of ‘cool, hip’; see section 6.4, below). In
Homeric Greek, agath-os ‘good’ formed only suppletive comparatives, but
the adjective regularizes in Old Testament Greek (comparative: agath¯o-teros,
superlative: agath¯o-tat-os; Liddell and Scott 1996).
In the other direction, innovative suppletive patterns arise where none existed
before (AAA → ABB). Swedish bra (in (53a)) is one such example (the
SAOD gives 17th century braf – brafv-are – braw-ast-e). Another may be
provided by OCS m˘unog˘u ‘many’, which is given as undergoing regular comparison
(množai ‘more’, Diels 1963, 201, n.2; Lunt 1959, 66), but which forms
suppletive patterns in the daughter languages (although Krause and Slocum
2002-2004 give a suppletive comparative for this even in OCS). Another innovated
suppletive paradigm is the Old Icelandic paradigm for ‘old’ — this
adjective is regular (more accurately, non-suppleting) throughout Germanic,
but the positive root underwent lexical replacement in Old Icelandic yielding
a new ABB paradigm (retained in the daughter languages).
69
(57) pos cmpr sprl
a. Gothic alþ-eis alþ-iz-a
b.Old English eald ield-ra ield-est
c. Old Icelandic gamall ell-re ell-ztr
In sum, it is not the case that suppletive patterns as such are immutable.
The ABB pattern (and the absence of an ABA pattern) holds a special place
in the array of reconstructable changes, and the theory here explains why this
is so. By hypothesis, the superlative is always derived from the comparative.
Hence, if a suppletive pattern arises, for example, by replacement of the
positive root (as in (57c)), then the superlative will not follow suit, but will
instead remain tied to the comparative. Likewise, if the comparative changes,
then the superlative will change in tandem, tied as it is to the comparative.
This holds whether the relevant change is the resolution of doublets (as in
(55)), the innovation of a new comparative root (as possibly in Slavic words
for ‘many’ mentioned above (57)), or as in the change by which little in Old
English was compared by less ((58b), compare (56)):
(58) pos cmpr sprl
a. Gothic leit-il-s minn-iza minn-ists
b. Old English lýtel læss-a læs-(as)t
The only permitted exception to the default suppletive pattern of ABB
is the possibility for a marked ABC pattern to arise, as in Latin, where three
separate roots are listed. ABA does not arise, not because the vocabulary of
suppletive patterns is generally inherited, nor because the abstract suppletive
70
patterns are passed down intact from mother to daughter languages as such,
but rather, because UG excludes a means to derive a pattern with two roots
organized in such a way that the comparative is the odd one out, with a
single root shared between the positive and the superlative.
2.4 Summary
In this chapter, I have argued that there is an intimate relationship between
comparatives and superlatives, and that this relationship is manifest
as proper containment of the former in the representation of the latter. There
is no (relative) superlative morpheme that attaches directly to adjectives,
despite appearances to the contrary in languages like English. The primary
evidence for this conclusion, as discussed in this chapter, was the CSG, the
relationship among root suppletion in comparatives and superlatives, discussed
in the abstract in section 2.2, with a more careful discussion of the
data (and apparent counter-examples) in Chapter 4. Because the primary
evidence for the generalization is from a striking gap, and moreover because
the strikingness of the gap is apparent only in large, cross-linguistc comparison,
the results point to the hand of UG in accounting for this state of affairs.
I will return in Chapter 7 to a tentative proposal about why this might be
the case, but first, we turn in the next chapter to further evidence for the
Containment Hypothesis, as well as some related considerations that arise in
that discussion.
71
72
Chapter 3
The Containment Hypothesis
3.1 Introduction
The previous chapter laid out the core of a theoretical account of a gap in
suppletive alternations, codified as a linguistic universal in the form of the
CSG in (33) in the previous chapter. The key premise in this account is the
Containment Hypothesis (59), of which nesting (see (36a)-(37) in Chapter 2)
was a specific structural instantiation.
(59) The Containment Hypothesis
The representation of the superlative properly contains that of the
comparative
[in all languages that have a morphological superlative].
This hypothesis, together with a theoretical model of morphosyntax that
allows the Elsewhere principle to apply to suppletive stem alternations, yields
73
the important result that the ABA pattern *good – better – goodest is unstatable
(see section 2.2). Since the pattern is virtually unattested, this appears
to be a welcome result, explaining this gap as a consequence of a structural
universal. The success of that explanation thereby constitutes an argument
for the Containment Hypothesis, and we now turn to the question of whether
there is additional, independent evidence for that hypothesis. This chapter
evaluates the strength of such evidence, from two morphological domains
(affix co-occurrence and affix inventories) and, briefly, from semantics. I also
discuss the interaction of morphological and periphrastic superlatives, where
some additional generalizations, but also some new puzzles, arise.
3.2 Transparent Containment
3.2.1 Transparent Nesting
I noted in the previous chapter that the containment relation is morphologically
transparent in Czech, and indeed, it is transparent in a wide variety
of languages that have morphological comparative and superlative degrees,
including most Slavic languages (but not Russian or Bulgarian), also Lithuanian,
Latvian, Persian, Hungarian, one variety of Saami, Georgian (to the
extent there are morphological comparatives and superlatives, see n. 18 in
Chapter 4), Batsbi, Ubykh, Chukchi, and Cherokee. Examples are given in
(60) (where ‘X’ represents the adjectival root; these forms are not in all cases
the only, or the most productive, exponents of the relevant degree morphol-
74
ogy):1
(60) cmpr sprl
a. Persian: X-tær X-tær-in
b. Lithuanian: X-iau X-iaus-ia
c. Cimbrian German: X-ar X-ar-ste
d. Batsbi: X-vx X-vx-č
e. Latvian: X-âk vis-X-âk
f. Czech: X-ši nej-X-ši
g. Hungarian: X-bb leg-X-bb
h. Chukchi: X-@N @nan-X-@N
i. Cherokee: X-ka/ya/... w-X-k˜2Pi/y˜2Pi/...
j. Ubykh: ç’a-X a-ç’a-X
In Paiwan (Austronesian, Formosan), the superlative circumfix tjala-...an
also appears to contain the comparative particle tja, although Egli (1990,
149) does not offer a segmentation of the superlative. In American Sign
Language as well, the morphological superlative appears to contain the morphological
comparative (although only a small set of adjectives have morphological
grades): the sign glossed -EST (a suffix) consists of the -ER sign
but uses a greater extent of movement, suggesting the superlative could be
analyzed as the comparative plus an intensifier (D. Lillo-Martin, personal
communication, 2010).
Ultan (1972, 140-1) noted that shared morphology (whether affixal or periphrastic)
between the comparative and superlative is exceedingly common
75
in his sample, and that while this sharing often takes the form of embedding
just illustrated, the reverse embedding is unattested: superlatives are
often derived from comparatives, but comparatives are never derived from
superlatives.2
A nested structure is also attested for comparatives in older Indo-European
languages, and is reconstructed for Proto-Indo-European (Seiler 1950, 6;
Bopp 1856, 374-415; Kuryłowicz 1964, 227-239; Weiss 2009, 357). Even Modern
English superlative -est is historically transparent, with the -es- portion
sharing an origin with the comparative -er; compare the Gothic forms in
(61a).3
(61) cmpr sprl
a. Gothic: -iz-a -is-ts
b. Sanskrit: -(i)y¯as -is.-t.has
c. Latin: -ior < -ios -issimus < -is-m.mo-s
d. P-IE: *-i
“
os-, *-is- *-is-to-s
Note that what is common in (60) and (61) is the structural relation
among the grades; there is wide variation in how the superlative is formed
from the comparative. Superlatives may be suffixal (as in Lithuanian, Persian,
and Saami) or prefixal (as in Hungarian, Latvian and Czech), and the etymological
source of the superlative morphology varies. The superlative prefixes
in (60) all show different origins: in Latvian, the prefix is the root meaning
‘all’; in Czech, it is etymologically a preposition and pronoun (roughly,
‘on it’); in Hungarian an intensifier of sorts;4
in Chukchi an emphatic pronoun
76
(roughly ‘self’, Russian sam; Skorik 1977, 334); and in Ubykh, the prefix is
the definite article. The generally accepted origin for the PIE suffix *-tois
also something like a marker of definiteness or individuation (see Cowgill
1970 for extensive discussion).
The derivation of superlatives from comparatives is also evident in many
languages in which the (relative) superlative is formed periphrastically. One
periphrastic superlative-forming strategy involves the addition of a definite
article (or other definiteness marker) to the comparative (whether the comparative
is itself formed morphologically or periphrastically). Some examples
are given in (62) (another strategy will be discussed in section 3.2.3). This
is the normal pattern in all the Modern Romance languages (exemplified
by French in (62a) and, with a definite pronoun/demonstrative-like element
rather than the article, Romanian in (62b)) and in Modern Greek, in which
morphological and periphrastic comparatives are in free variation, and both
form superlatives with the definite article (as in (62c)). The pattern is attested
as well in some varieties of Austrian German, including Upper Austrian
(Martin Prinzhorn, Martin Hackl, personal communication 2010), and
also in Maltese (in some cases with a change in word order), Neo-Aramaic
(Arsanis 1968, 496), Middle Armenian (Karst 1901, 395), and, alone among
the Fennic languages, Livonian (Nau 1992, 17).5
77
(62) pos cmpr sprl
a. French: gros plus gros le plus gros ‘big’
b. Romanian: bun mai bun cel mai bun ‘good’
c. Greek: psil-ós pjo psil-ós o pjo psil-ós ‘tall’
or: psiló-ter-os o psiló-ter-os
d. Maltese: kbir ikbar l-ikbar ‘big’
e. Livonian: van¯a van¯ım se van¯ım ‘old’
Compare also in this light some of the Celtic languages. Standard descriptions
of (Modern) Irish, Manx and Scottish Gaelic characterize these
languages as lacking a morphological distinction between comparative and
superlative, with the distinction indicated by context and/or syntax (Breton
and Welsh retain a morphological superlative grade). For example, Phillips
(2004, 29) writes: “[t]hough Manx does not make the formal distinction that
English does between comparative and superlative meanings, the definite
article can be used to express uniqueness, as in [(63)].”6
(63) yn
the
boayl
place
share
better/best
‘the best place’ (Phillips 2004, 28)
Arabic is also described as having no morphological distinction between
comparative and superlative, with a single “elative” form used in both contexts.
Grammars note that a variety of syntactic devices can be used to draw
the distinction between comparative and superlative meanings, including the
78
use of the definite article on (or defined form of) the adjective (following a
definite noun), as illustrated by this pair from Gulf Arabic (the comparative
is formed with the template (P)aCCaC ):7
(64) a. il-banaat
the-girls
PashTar
clever.cmpr
bi
by
kathiir
much
min
than
al-awlaad
the-boys
‘The girls are much cleverer than the boys.’ (Holes 1990, 91)
b. haadha
this
huwa
he
(l-walad)
the-boy
il-ashTar
the-clever.cmpr
fi
in
S-Saff
the-class
‘This is the cleverest (boy) in the class.’ (Holes 1990, 231)
Other languages in which a single form may serve either as a comparative
or superlative include Klon, a Papuan language of Alor, Indonesia, as described
in Baird (2008, 116) and Misantla Totonac, an indigenous language
of Mexico (MacKay 1999, 413).
It is not clear to me whether there is a meaningful line to be drawn
between languages described as drawing no distinction between comparative
and superlative (Celtic, Arabic, Klon, Totonac), and those described as
deriving superlatives from comparatives with the aid of a definite article (Romance,
Greek, Maltese etc.). A case in point is Vlach Romani. Descriptions
of this collection of dialects differ on exactly this point. Hancock (1995, 77)
describes Vlach as being like Modern Romance: “Maj alone plus the adjective
gives the comparative, ... while with the appropriate definite article it
translates the superlative” (the comparative particle maj is a Romanian borrowing);
Katsánis and Dínas (1990, 58) describes a similar formation with
79
the enclitic article for a Vlach variety spoken in Greece.8
Describing the
Lovari variety, Pobożniak (1964, 45-46) treats it in Celtic-like terms, stating
that there is no special exponent of the superlative, distinct from the comparative,
but that the article may be used (embedding the comparative) “if it
is necessary to make a distinction between the comaprative and the superlative.”
In yet a third description, Boretzky (1994, 48) specifically contrasts
Kalderaš (Vlach) Romani with Romance: “The placement or non-placement
of the article may not serve for the distinction of the two grades of comparison
(as is partially the case in the Romance languages)...”.9
Weigand (1888,
73) gives a comparative + all construction for the superlative in another
variety, with no mention of the definite article; while an Arumanian variety
uses a borrowed superlative particle (see n. 29 in Chapter 4). By and large,
it appears we may well be dealing here with a difference in descriptive traditions,
rather than a grammatical distinction between language types. It
may be that the meaning of a comparative plus a definite article comes close
enough to rendering the superlative meaning that it frequently serves as such,
while not rendering all nuances of a grammatical superlative, thus leaving
room for the variation in descriptive perspectives. In this regard, note that
the pattern of (apparently) deriving superlatives by means of a comparative
plus a marker of definiteness seems to have arisen independently on many
occasions, suggesting something rather basic about this pattern. Within
Indo-European, the Modern Romance languages, Greek, Vlach Romani, and
those Modern Celtic languages that have this pattern (if that is indeed what
80
(63) shows), as well as Upper Austrian, all developed from ancestors with a
distinct morphological superlative grade.
Another question is whether the combination of a definite marker and a
comparative, with a superlative meaning, should be treated as containing a
formal (but unpronounced) superlative element (i.e., a superlative node, as in
(37)) or whether the superlative meaning can be derived from the semantics
of the comparative and definiteness alone. If there is no covert superlative
element in such representations, then they cannot be taken as support for
the Containment Hypothesis. They are consistent with the hypothesis, but
only trivially so, in the sense that such languages may lack a grammatical
superlative category altogether. On the other hand, if the combination of a
definite article plus the comparative were sufficient to derive a superlative
meaning (with no null elements, or equivalent semantic devices such as typeshifting
operations or other postulates), then this reading should be routinely
available in other languages, such as English, where it is not (except in the
superlative comparing two items: (Of the two books), the shorter one is on
the table.). I therefore leave this question open.
Either way, there is abundant morphological and morphosyntactic evidence
that the comparative is properly contained inside the superlative in
many languages. The CSG effectively follows from the assumption that this
widely observable structural relationship is indeed present in all languages
with comparative and superlative grades of adjectives.
81
3.2.2 The Fennic Superlative: Branching Affixes?
Returning to morphological formations, Finno-Ugric languages also provide
evidence that the superlative properly contains the pieces of the comparative,
but Finnish (and some varieties of Karelian) provide a wrinkle in terms of
affix order.10
The containment relation, and thus the affix order, are not
obvious in the citation (nominative) forms (as in in (65)):
(65) pos cmpr sprl
a. paksu paksu-mpi paksu-in ‘thick’
b. uusi uude-mpi uus-in ‘new’
c. hyvä pare-mpi parha-in ‘good’
In nominative, complex morphophonology obscures the underlying forms
of the affixes—comparative -mpi and superlative -in are in fact derived from
-mpa and -i-mpa, respectively (see Hakulinen 1957). The containment relationship
is transparent in oblique cases, such as the illative, shown in (66):
(66) pos cmpr sprl
a. paksu-un paksu-mpa-an paksu-i-mpa-an ‘thick’
b. hyvä-än pare-mpa-an parha-i-mpa-an ‘good’
The issue raised by this pattern is that the superlative does indeed seem
to contain the pieces of the comparative, plus something in addition (namely,
the vowel -i-),11
but with respect to a nesting structure as in (37) of Chapter
2, repeated here as (67), the linear position of this superlative element (if
82
that is in fact what it is12
) is surprising, sandwiched between the base and
the comparative suffix.
(67) a. positive b. comparative c. superlative
a
adj
c
a
adj
cmpr
s
c
a
adj
cmpr
sprl
One possible avenue of analysis is to reconsider whether a strictly nested
structure as in (67) is the only way to represent the Containment Hypothesis
that underlies the account of the CSG. If nesting is the only option, then the
Finnish facts pose a challenge. On the other hand, there are other structural
relationships to consider, among these, structures with a branching affix, as
in (68a-b), which are hierarchically identical but differ only in the linear order
of the comparative and superlative nodes under the placeholder node label
x.13
(68) a. s
a
adj
x
cmpr sprl
b. s
a
adj
x
sprl cmpr
83
The tree in (68b) would readily accommodate the morpheme order in
the Fennic examples. The theoretical question, then, is: does this structure
contain the comparative in the right way to force the comparative root allomorph
to be selected in the superlative (in (65c) and (66b))? To make
this concrete, the relevant fragment of the Finnish Vocabulary is given here,
modeled on previous examples.14
(69) a. good → par(e)- ] cmpr ]
b. good → hyvä
c. sprl → -id.
cmpr → -mpa
Broaching the question in any detail goes beyond what I wish to address
in this chapter, and I will return to the question in Chapter 5. For now, I note
that there are a variety of assumptions on the market that could yield the
answer ‘yes’, in which case the account remains unchanged. For example, if
the node x in (68) bears the label cmpr – either because [comparative] is
a feature subject to percolation, or because cmpr is the head of the subtree x
and thus projects – then the adjectival root is both structurally and linearly
adjacent to a cmpr node, even in (68b), satisfying the context for (69a). An
empirical argument for an analysis along these lines can perhaps be made
from Basque.
Basque comparatives are formed with the suffix -ago (as in (70a-b)).
Basque also has a morphological means for expressing a slight degree of
superiority (‘a little more X’), namely the suffix -xe (de Rijk 2008, 711).
84
This suffix occurs between the adjective root and the comparative (as in
(70c)):
(70) a. high gora
b. higher gor-ago
c. a little higher gora-xe-ago
On semantic grounds, it seems most plausible to assume that -xe combines
first with -ago, as it modifies the degree of comparison, and not the adjective
root. That is, the meaning appears to be [ [ a little more ] X ] and not [ [ more
] a little X ]. If this assumption is correct, then xe-ago represents a complex
affix, like (68b). In particular, the element xe is semantically a modifier, not
the head of the complex affix, and hence the whole element is a (species of)
comparative, even though linearly, xe intervenes between the adjectival root
and the (regular) comparative exponent:
(71) s
a
adj
cmpr
xe cmpr
These considerations predict that -xe-ago should behave as -ago alone for
the purposes of suppletion triggered by the comparative. And this is indeed
correct, as (72) shows (de Rijk 2008, 710-711):15
85
(72) a. much asko
b. more gehi-ago
c. a little more gehi-xe-ago
It appears, then, that both on theoretical and empirical grounds, branching
affixes constitute a viable structure, and that moreover, in at least some
configurations of the form in (68b), a sub-part of a branching affix may stand
in the right configuration to govern root allomorphy.16
I conclude, then,
that there are no particular hurdles to positing a branching affix structure
for the Fennic superlative. This jibes with Ultan’s observation, mentioned
above, that no language derives a comparative from an independent superlative
stem. Finnish (and its close relatives) are no exception: although the
comparative affix appears to be peripheral to the superlative, the superlative
consists of all the morphemes in the comparative, plus an additional
morpheme, and not the other way around.17
The structure in (68a) is also an alternative to consider for languages like
Persian and Lithuanian in (60) (where both comparative and superlative are
suffixal). For that matter, structures such as (68) provide a possible alternative
to positing a zero allomorph of the comparative (as in (42)) for languages
in which the nesting is not transparent, and in which comparative and superlative
are both suffixes (as in English) or both prefixes (as in Bulgarian
— but see section 4.3). In such languages, the superlative could be analyzed
as a portmanteau affix, inserted at node x, as in (73) (a-b are English, c-d
Bulgarian; on the formal treatment of portmanteaus, see Chapter 5):
86
(73) a. sprl, cmpr → -est
b. cmpr → -er
c. sprl, cmpr → najd.
cmpr → poA
final remark on the branching affix structure is that a structure like
(68) may actually provide a more seamless integration with the semantics of
the comparative, at least on many approaches. In various current semantic
theories, the comparative takes as its first argument the (degree corresponding
to the) standard (i.e., the than-clause in English), and only then does
it combine with the adjective (see Heim 2000, 1985, Lechner 2004 among
others).18
If the superlative means something like ‘than all others’, then a
branching affix structure would in fact represent the semantics transparently,
while a more canonical nested structure (as in (67)) would require additional
operations to yield the correct interpretation. Pushing this further, it is then
languages like Latvian and Czech, in which the comparative and superlative
affixes are on opposite sides of the root, that are difficult to accommodate
under a branching structure, and suggest that the nesting stucture is also
needed.
For the remainder of this chapter, I will set aside the possibility of branching
affixes, using nested structures in general to illustrate containment relations,
but it is important to keep in mind that we now understand this to be
an expository simplification — there are multiple means of satisfying containment
structurally, including branching and nesting. For a class of languages,
87
including English, we are left with two possible analyses of some cases (such
as the English superlative), but with no pressing need to decide between
them.
3.2.3 Aside: More than anything
In section 3.2.1, we considered periphrastic superlatives formed by a combination
of the comparative and a marker of definiteness, with no overt superlative
morpheme. There are other means of forming periphrastic superlatives from
comparatives. Prominent among these is the use of a universal quantifier in
the form that normally marks the standard of comparison; another is the
use of an intensifier, along with the comparative. Both types are attested in
Russian, as in (74).19
(74) pos cmpr sprl
a. xoroš-ij luč-še luč-še vse-go/-x
good-m.sg better-cmpr better-cmpr all-gen.sg/-gen.pl
b. xoroš-ij luč-š-ij sam-yj luč-š-ij
good-m.sg better-cmpr-m.sg same-m.sg better-cmpr-m.sg
Like the constructions involving a marker of definiteness forming a superlative
in combination with the comparative, these raise the question of
what constitutes a grammatical superlative (as opposed to an expression with
superlative meaning, but no grammatical superlative element). In this section,
I discuss some observations regarding the construction type in (74a),
88
but ultimately this section is somewhat of an aside, as nothing in this discussion
bears directly on the main arguments of the book, to which we return
below.
The combination of a comparative (or sometimes positive) adjective with
a universal quantifier, as in (74a), is in fact the most common pattern of
forming superlatives cross-linguisticaly, as noted by Ultan (1972, 123) and
confirmed in my sample. Some examples are given in (75):
(75) a. vanj
mïz-lešj
everything-abl
vïlj
new
(Udmurt)
‘newest’ (‘new than everything’) (Csúcs 1998, 286)
b. irFeli-ši
all-gen
u-F@ra-š-i
cmpr-bad-cmpr
(Mingrelian)
infl
‘worst’ (‘worse than all’) (Kipshidze 1914, 34)
c. iNkaraka-Na
everything-abl
kNar-alkura
large-cmpr
(Arrente)
‘greatest’ (‘greater than all’) (Strehlow 1942, 87)
d. . . . upkat-tuk
. . . all-abl
engesi-tmer
strong-cmpr
(Evenki)20
‘strongest’ (‘stronger than everybody’) (Nedjlakov 1997, 120)
This ‘more than all’ strategy is also widely attested in languages without
a morphological comparative. Representative examples from a variety of
languages are given here, drawn not only from a variety of regions and genetic
stocks, but also from among the various ways of representing comparison.
89
The Tungusic language Even has no special marker of comparison, and
comparatives are formed with the positive degree of the adjective, with the
standard of comparison in a locative case (ablative), as in (76a). One means
of forming the superlative in this language is to use a universal quantifier in
the position of the standard of comparison, as in (76b):
(76) a. Neeluki
wolf.nom
Nin-duk
dog-abl
eNi-dmer
strong-intens
‘The wolf is much stronger than the dog.’ (Malchukov 1995, 12)
b. erek
this
oran
reindeer.nom
čele-duku-n
all-abl-3sg
gud
high
‘This reindeer is the highest (of them all).’ (Malchukov 1995, 12)
The same pattern is found in Bagri (Indo-Aryan), where an element meaning
‘more’ is optional.
(77) a. gita
Gita
mira
Mira
sy˜u
from
mot
˙
i
old
hE
be-pres.3fs
‘Gita is older than Mira.’ (Gusain 2000, 38)
b. rajes
Rajesh
sE/sara/s@gla
all
sy˜u
from
kabil
competent
hE
be-pres.ms
‘Rajesh is the most confident of all.’ (Gusain 2000, 38)
In Tümpisa Shoshone (Uto-Aztecan), the standard of comparison is marked
by one of three postpositions, all meaning (roughly) ‘more than’, while the
90
adjective is unmarked. In this language as well, superlatives are formed by using
the word for ‘anyone/everyone’ as the standard of comparison (the order
between the adjective and postpositional phrase is reported to be flexible).
(78) a. nüü
I
yuhupi
fat
üng
you.obj
kawiki
more.than
...
‘I am fatter than you.’ (Dayley 1989, 291)
b. satü
that
noohakka
anyone.obj
kawi
more.than
yuhupi
fat
‘That one is the fattest.’ (Dayley 1989, 295)
Comparatives in Rapanui, a Polynesian VSO language, are formed with
the main predicate ’ilhau ‘more’ which takes three arguments: the subject,
the property or attribute being compared, in nominalized form, and then the
standard of comparison, introduced by dative ki. A very loose paraphrase in
English might be something like: ‘X is more than Y in height’ to mean ‘X is
taller than Y.’ As (79) shows, the superlative is rendered via comparison to
all.
(79) ’Ihau
more
a
prs
Ari
Ari
i
rel
te
+spe
roa
long
roa
red
ki
dat
te
+spe
ta’ato’a
every
‘Ari is the tallest.’ (lit: ‘taller than all’) (Du Feu 1996, 73)
Zulu (Bantu) provides an example of the use of the universal quantifier
as the object of the verb meaning ‘surpass’ to express the superlative in an
‘exceed’ comparative construction:
91
(80) mude
he.is.tall
ukwedlula
suprass.infin
bonke
all
esikoleni
in.the.school
‘He is tall above [lit. surpassing] all (others) in the school.’ (Poulos
and Msimang 1998, 403)
Such examples also show up in language where ‘exceed’ comparatives occur
in verb serialization contexts. The published description of Ogbronuagum
(Bukuma), a Niger-Congo language, does not explicitly discuss superlative
formation in this language, but gives the following text example, which appears
to illustrate this point:
(81) @lugulu
tortoise
k´@-´@rísíi
fact-to.say
mú
that
j´O
he
jóo-âeeri
fact-know
n´E´Ema
knowledge
j´O
he
tE
surpass
ikpom
all
ína
animals
wa
the
‘Tortoise said that he was the wisest of all the animals.’ (Kari 2000,
53)
The strategy is even attested in juxtaposed, ‘conjoined’-comparatives, as
in the following example from Sinaugoro:
(82) mota
snake
mabara-ri
all-3pl
tu
top
kei,
small
avaro
[snake.name]
tu
top
barego.
big
‘Avaro is the biggest of all snakes.’ [lit. ‘All the snakes are small,
avaro is big.’] (Tauberschmidt 1999, 38)
92
What is interesting about all of these examples is that the universal quantifier
has the morphosyntax of the standard of comparison for the language.
Literally, these correspond to ‘bigger than all’ and not ‘biggest of all’. Russian
examples draw out this difference clearly. As in English, the morphosyntax
of the two constructions is different, and a phrase such as ‘tallest of the
students’ is rendered by the superlative form of the adjective, along with a
PP expressing the comparison set, as in (83). In this sentence, it is clearly
implied that Vanya is one of the students.
(83) Vanja
Vanja
samyj
most
vysokij
tall
iz
from
student-ov.
student-gen.pl
‘Vanja is the tallest of the students.’
This contrasts with the construction in which the adjective stands in the
comparative form, and the standard of comparison is expressed in the genitive,
with no preposition (84a). In this example, as in its English translation,
Vanya is being compared to the students, and there is a strong implication
that Vanya himself is not one of the students. If Vanya is one of the students,
then a contrastive word is required, as in (84b), again, just as in English.
(84) a. Vanja
Vanja
vyš-e
tall-er
student-ov.
student-gen.pl
‘Vanja is taller than the students.’
b. Vanja
Vanja
vyš-e
tall-er
drug-ix
other-gen.pl
student-ov.
student-gen.pl
‘Vanja is taller than the other students.’
93
The construction with the universal quantifier in (85) clearly has the
morphosyntax of the comparative, not that of the ‘tallest of all’ frame in
(83). However, unlike (84a-b), there is no need for a separative word such as
‘other’ or ‘else’.
(85) Vanja
Vanja
vyš-e
tall-er
vse-x.
all-gen.pl
‘Vanja is the tallest.’ (lit: ‘Vanja is taller than everyone.’)
In English, a construction of this sort occurs, most prevalently with any,
in casual contexts, as in the following examples drawn from the Corpus of
Contemporary American English (COCA).21
(86) a. the thing that is going to guide this administration more than
anything is the safety of the American people...
b. We love two things more than anything: Family and music.
c. The smoking ban hurt more than anything...
d. Perhaps more than any state in America, California represents
the end of the rainbow.
e. At twenty-nine, he thought he knew more than anybody.
f. . . . the former chief knew more than everyone . . .
Read literally, an example like (86d) should imply that California is not
an American state, or else it should be inherently contradictory: if California
is an American state, then California represents the end of the rainbow
more than California does. Likewise, (86f) should entail that the chief
94
knew more than himself. As Barbara Partee notes in an informal comment
on LanguageLog (http://itre.cis.upenn.edu/~myl/languagelog/
archives/004925.html), it appears that in this usage, we accommodate a
silent else or other in understanding these. To English ears, examples of this
sort feel like an imprecision, an intuition supported by the observation that
the majority of relevant COCA hits for “more than any/all/every(one)” have
an overt contrastive word.
And yet, the cross-linguistic evidence suggests that such an accommodation
is routine. The literal meaning of the construction appears to yield a
contradictory entailment, particularly clear in examples like (82), which is
simply ignored in interpreting these sentences.22
Of the many grammatical
descriptions that describe superlatives as being formed by the comparative
construction with ‘all’ as the standard, only one gives an ‘other’ or ‘else’
word in describing the relevant construction, namely Nandi, an ‘exceed’comparative
language (Creider and Creider 1989, 151).23
The widespread
distribution of the ‘more than all’ superlative, across language families and
more importantly, across comparative types, suggests that this accommodation
strategy may be available as a part of Universal Grammar (or universal
pragmatic strategies, if these are something distinct). And yet, this leaves
the English situation, in which such an accommodation is not routine, as an
oddity.
Stepping back from these details, there is something else at stake here,
which (as noted at the outset of the discussion) may make the entire dis-
95
cussion beside the point for the main themes of this book, and that is the
question of whether any of these examples are ‘superlatives’ in a grammatical,
as opposed to functional sense (see also the remarks on the definite article
in section 3.2.1). An English expression such as John is taller than everyone
else is grammatically a comparative — there is no reason (that I know of)
to suppose that its morphosyntactic or semantic representation contains any
superlative element. If there is an (extra-grammatical) pragmatic accommodation
at play in the ‘more than all’ type of “superlative” then it would seem
that these too should be treated as grammatical comparatives. On this view,
languages with only ‘more than all’-type superlatives would in fact lack superlatives
as a grammatical category altogether. The examples in this section
would thus be neither here nor there in their bearing on the Containment
Hypothesis. It is only if these examples require a special superlative element
in their representation that they would bear on the Containment Hypothesis.
Either way, there is no challenge to the general thesis advanced in this
chapter: the ‘more than all’ examples either support the hypothesis, or are
(at worst) neutral with respect to it.
3.2.4 Interim Conclusion
Taking stock, thus far I have shown that nesting relations (or more complex
containment structures) are plausible on cross-linguistic morphological
grounds. It is crucial to the account of the CSG that these relations hold of all
languages that have morphological comparative and superlative grades, and
96
thus that they hold also of languages where the relation is not transparent
on the surface. Although there is an affix-ordering wrinkle in the Fennic languages
for the (stricter) Nesting Hypothesis, the more general Containment
Hypothesis is consistent with a broad array of data.
3.3 Comparison and the Synthetic/Analytic Di-
vide
The discussion of superlatives formed with the definite article or a universal
quantifier have brought us into the realm of periphrastic expressions of the
superlative. In discussing the CSG, I set these aside, noting that the scope of
the CSG is limited to morphological expressions of the comparative and the
superlative. In fact, even where there is suppletion in the comparative, when
the superlative is periphrastic, it sometimes shares the suppletive root of the
comparative (as in Modern Greek and Votic, (87a-b)), and sometimes does
not (as in Tikhvin Karelian (Rjagoev 1977, 96-97) and Russian (87c-d)):
(87) pos cmpr sprl
a. M. Greek kak-ós cheiró-ter-os o cheiró-ter-os ‘bad’
b. Votic üvä par@-pi k@ikkia par@-pi ‘good’
c. Tv Karel. hüvä pare-mbi {ülen/suamo} hüvä ‘good’
d. Russian plox-oj xuž-e samyj plox-oj ‘bad’
In this section, I will investigate the structure of periphrastic constructions,
arguing in the first place that the limitation of the CSG, as an absolute,
97
to morphological superlatives, has a principled basis, and that moreover, the
alternation seen in (87) is a predictable consequence of the interaction of
the Containment Hypothesis with the general theoretical mechanisms that
derive periphrastic and morphological constructions from a single underlying
syntactic representation. There are a number of steps to this argument, but
along the way, I will derive and defend two further candidate universals:
(88) The Root Suppletion Generalization (RSG)
Root suppletion is limited to synthetic (i.e., morphological) compar-
atives.
(89) The Synthetic Superlative Generalization (SSG)
No language has morphological superlatives (X-est), but only periphrastic
comparatives (more X ).
We start with a general consideration of locality in contextual allomorphy,
of which suppletion is a special case.
3.3.1 Periphrasis, Suppletion, and Locality
As noted in the introduction, an important area of inquiry within Distributed
Morphology has been on the locality of morphological interactions, such as
allomorphy (see Embick 2010 for an extensive survey, and specific proposals).
A morpheme (or feature) β may condition allomorphy for morpheme α
only if the the two are sufficiently local. At least as a working hypothesis,
assume that a head or feature that conditions root allomorphy must be in
98
the same morphological word (complex X0
) as the root. Put differently, β
may condition allomorphy for α in the environment in (90a) but not that
in (90b), where a maximal projection intervenes (abstracting away from linear
order). (An alternative formulation, in line with Embick (2010), would
make reference to phases or cyclic nodes, in place of maximal projections, in
(90b), although this does not differ in any way that is relevant to the present
discussion, and I will make use of (90) in what follows.)
(90) a. α . . . ]X0 . . . β
b. * α . . . ]XP . . . β
With this in mind, we turn to the morphosyntax that underlies the comparative.
I assume that the morphological and syntactic comparative have
the same underlying syntax, at least at an initial level of representation (Dstructure
in the GB framework, or its analogue in subsequent versions of
the theory). Simplifying for the sake of exposition, let us assume that the
syntactic structure underlying a comparative is (in part) as in (91), where a
comparative head takes an adjectival phrase as its complement.24
(91) CmprP
cmpr AdjP
adj
99
If nothing further happens, this yields a periphrastic comparative, like
more intelligent. A morphological comparative is derived from the structure
in (91) by means of an operation M which combines the two heads. This
operation could be head movement in syntax, or the post-syntactic operation
Morphological Merger (see section 1.2). The choice does not matter at this
point, and for expository convenience I will present it as Merger, a process
adjoining the comparative head to the adjective as in (92a), yielding (92b)
(=(37b) above).
(92) a. CmprP
cmpr AdjP
adj
b. c
a
adj
cmpr
Now, given the locality condition in (90), we make an immediate prediction.
Where the structure that is subject to morphological exponence is
that in (91), comparative suppletion will be blocked, but where the structure
that feeds exponence is as in (92b), suppletion will be permitted (and hence
required, by the Elsewhere Principle). Let’s take a concrete example.
In Modern Greek, both periphrastic and (for some adjectives) morphological
comparatives exist. In Greek, these two constructions are generally
in free variation (as they are in English for a small class of adjectives, such
as polite – politer/more polite). Modern Greek also has one suppletive adjec-
100
tive, kak-ós ‘bad’ with comparative root cheiró-. Thus, we have the partial
Modern Greek vocabulary in (93):
(93) a. bad → cheiró / ] cmpr ]
b. bad → kakc.
cmpr → -ter- / ]ADJ ]
d. cmpr → pjo
The restriction in (90) ensures that the two root allomorphs and two
comparatives do not combine freely. If Merger (or head movement) applies,
yielding (92b), then the affixal comparative (93c) is required, and, by the
Elsewhere Principle, forces the comparative root allomorph in (93a). On the
other hand, if there is no amalgamation of the two heads, and (91) is the
input to the rules of exponence, then a periphrastic construction emerges.
The comparative will be realized as pjo, since it is not affixal. At the same
time, the comparative allomorph of bad in (93a) is unavailable, since the
comparative element is not within the same complex X0
as the root. The
periphrastic output is necessarily pjo kak-ós, and not *pjo cheir-ós. The
prediction of the locality condition is borne out (I include superlative forms
here for reference though they are not directly relevant):
(94) pos cmpr sprl
a. kak-ós cheiró-ter-os o cheiró-ter-os ‘bad’
b. kak-ós pjo kak-ós o pjo kak-ós ‘bad’
101
The same is true for other languages. Whenever potentially suppletive
adjectives alternate between a periphrastic and a morphological comparative,
suppletion is limited to the morphological construction (but see the discussion
of (101) for an important qualification). Georgian provides a clear illustration,
with four adjectives undergoing suppletion in morphological comparatives
(three of these also have corresponding morphological superlatives,
showing ABB patterns, as expected):
(95) pos cmpr sprl
a. k’arg-i u-k’et-es-i sa-u-k’et-es-o ‘good’
b. k’arg-i u-mˇob-es-i sa-u-mˇob-es-o ‘good’
c. cud-i u-ar-es-i ‘bad’
d. cot’a nak’l-eb-i ‘few’
e. bevr-i met’-i u-met’-es-i ‘many’
All four adjectives also enter into periphrastic comparatives, and in each
case are regular, rather than suppletive:
(96) pos cmpr sprl
a. k’arg-i upro k’arg-i q’vela-ze (upro) k’arg-i ‘good’
b. cud-i upro cud-i q’vela-ze (upro) cud-i ‘bad’
c. cot’a upro cot’a q’vela-ze (upro) cot’a ‘few’
d. bevr-i upro bevr-i q’vela-ze (upro) bevr-i ‘many’
Other than the interesting question of how circumfixes are to be modeled,
the Georgian facts show the same thing as Greek. In the periphrastic
102
construction, the comparative head is not local enough to the adjectival root
to govern suppletion — a phrasal boundary intervenes.
An alternation between a regular periphrastic construction and a suppletive
one is also attested in Abkhaz (Northwest Caucasian, unrelated to
Georgian):25
(97) pos cmpr sprl
a. a-bz@ja jejèa i-bz@j-aw zag’ ra:sta jejèa i-bz@j-aw ‘good’
b. a-bz@ja i-jejKj
-u ‘good’
The same pattern arises in the Italian adjectives for ‘good’ and ‘bad’,
given in (98), with similar patterns throughout Romance wherever a periphrastic
comparative alternates with a suppletive one.
(98) pos cmpr sprl
a. buono migliore il migliore ‘good’
b. buono più buono il più buono ‘good’
c. cattivo peggiore il peggiore ‘bad’
d. cattivo più cattivo il più cattivo ‘bad’
In some cases, the alternation between periphrastic and morphological
comparatives may correlate with a meaning difference. French ‘bad’ (99)
serves to illustrate, with the suppletive, morphological pattern used for abstract
situations and the regular, periphrastic pattern for more concrete properties,
according to Dietiker (1983, 104) and Judge and Healey (1983, 303):
103
(99) pos cmpr sprl
a. mauvais pire le pire ‘bad’ (abstract)
b. mauvais plus mauvais le plus mauvais ‘bad’ (concrete)
There is thus a clear generalization to be stated here. This generalization
is the Root Suppletion Generalization in (88), repeated here. As demonstrated
with Greek, the RSG is simply a corollary of the locality condition in
(90), applied to the structures assumed for the comparative morphosyntax
in (91)-(92). Allomorphy (of which suppletion is a special case) can only be
triggered within a morphological domain.
(88) The Root Suppletion Generalization (RSG)
Root suppletion is limited to synthetic (i.e., morphological) compar-
atives.
What is excluded by (88), i.e., as a violation of (90), is a situation in which
the comparative is formed periphrastically, with an obligatory free comparative
adverb, but requires a suppletive root, a pattern as in (pseudo-)English
or Greek (100):
(100) pos cmpr
a. good * more bett
b. kak-ós * pjo cheir-ós
Note that some care needs to be taken to distinguish the excluded, and
unattested, construction type in (100) from the co-occurrence of the adverb
and affix, which Cuzzolin and Lehmann (2004, 1217) describe as a means
104
to “reinforce or strengthen” the comparative. Cuzzolin and Lehmann (2004)
note that this is widespread among languages that have both analytic and
synthetic comparatives. Examples of such reinforcement occur with regular
(non-suppletive) adjectives, as with suppletive stems, as the following
illustrate.26
(101) pos cmpr
a. English old (more) old-er
b. good (more) bett-er
c. Late Latin fort-is (magis) fort-ior ‘strong’
d. bon-us (magis) mel-ior ‘good’
e. Mod. Greek mikr-os (pjo) mikró-ter-os ‘small’
f. kak-os (pjo) cheiró-ter-os ‘bad’
Such ‘reinforcement’, though proscribed in (Modern) Standard English,
has been attested as long as there has been a periphrastic comparative alternating
with the older affixal comparative in the history of the language (Kytö
and Romaine 1997, González-Díaz 2006). Formally, if Merger is treated as
a lowering operation (as in (92)), then reinforcement involves the redundant
spell-out of the cmpr head in addition to the affixal exponent. This
would limit reinforcement to affixal comparatives, excluding reinforcement of
periphrastic constructions: *more more intelligent. Such doubling of more
would only be possible as a type of (possibly meta-linguistic) double comparison,
which the forms in (101) are not.
105
Reinforcement also occurs in (at least some of) the modern Romance
languages, where only suppletive adjectives are not periphrastic (see below).
Thus in Italian più migliore ‘more better’ is in use alongside migliore ‘bet-
ter’.27
The same is apparently true of Abaza (Tabulova 1976, 71).
In all cases, the reinforcing adverb, obligatory in true periphrastic comparatives,
is optional when it occurs in the presence of a morphological comparative.
This is a key factor distinguishing reinforcement from the excluded
pattern in (100). In the significant majority of examples of root suppletion,
the regular affixal comparative morpheme is obligatorily present (even
though it is functionally redundant, as suppletion alone unambiguously signals
the comparative). Yet when the comparative morpheme is free-standing,
its presence is never obligatory with a suppletive root.
Before moving on, there is a final remark which could be made regarding
the Modern Romance situation as regards the RSG and locality. This is
somewhat of an aside, but has attracted attention in the literature, and is
thus worth discussing here.
3.3.2 Aside: Romance suppletion and Poser blocking
At this point, let us return briefly to the Modern Romance languages, and the
interaction, or rather complementarity, between periphrasis and suppletion.
In the previous section, I noted that whenever there is an alternation between
a periphrastic and a morphological comparative, it is only the morphological
form that may be suppletive. This was stated as the RSG, and derived from
106
the locality condition in (90), with examples from Modern Greek, Georgian,
and Modern Romance languages.
An apparent challenge to this comes from a treatment of the Modern
Romance languages, in particular French, in terms of what Hankamer and
Mikkelsen (2005) refer to as Poser blocking (after Poser 1992, see also Vincent
and Börjars 1996). In French, all comparatives are periphrastic, except
for a small handful of suppletive forms, such as (102). Unlike, say, Greek
discussed in the previous section, there are no non-suppletive, morphological
comparatives in French. Moreover, for bon ‘good’, there is no alternation
with a periphrastic construction, as there is for mauvais ‘bad’ (99), or for
the Italian cognates in (98).28
(102) pos cmpr sprl
a. bon meilleur le meilleur ‘good’
b. *plus bon *le plus bon
Poser (1992) and Vincent and Börjars (1996) contend that the French
example in (102) shows blocking of a phrase by a single word. Cast in
realizational terms, the lexical item meilleur on a Poser-blocking account
replaces the entire phrasal structure in (91), not just heads therein.29
Poserblocking
is at odds with the locality condition in (90). In the configuration
in (91), the comparative head is insufficiently local to admit of interactions
of this sort with the adjective root.
While Modern French seems to display an alternation between a regular,
phrasal expression and a listed, suppletive word, a broader cross-linguistic
107
perspective shows that there are two alternations that are in play here, and
that there is no compelling reason to abandon the locality condition that
underlies the robust RSG (see Embick and Marantz 2008 and Embick 2007
for extensive discussion of this point, on which I draw in what follows).
The first alternation is between periphrastic and morphological comparison.
This alternation is independent of suppletion and irregularity, up to the
RSG. This independence is clear in English (and other languages), where,
when suppletion is set aside, we know that some adjectives form periphrastic
comparatives, some form morphological ones, and some alternate (see section
5.5 for more discussion of the English distinction):
(103) pos morphological periphrastic
a. intelligent *intelligent-er more intelligent
b. polite polite-r more polite
c. smart smart-er * more smart
Similarly, in Modern Greek, regular adjectives alternate freely among the
two comparatives, as already noted:
(104) pos cmpr sprl
a. kal-ós kalí-ter-os o kalí-ter-os ‘good’
b. kal-ós pjo kal-ós o pjo kal-ós ‘good’
Thus, individual grammars must allow for a determination of which adjectives
are subject to Merger, whether obligatorily or optionally, and under
what conditions. As the above examples show, morphological listedness
108
(suppletion) plays no role in that alternation: both the periphrastic and
morphological comparatives are regular.
The competition known as ‘blocking’ that arises between regular (smarter)
and suppletive (bett-er) stems is thus a distinct alternation. Given the
locality condition that derives the RSG, only the output of Merger can be
the input to suppletion. Thus there is a relationship between the two alternations,
but a weak one: suppletion is restricted to a subset of the roots that
undergo Merger. This is patently true of the majority of languages in the survey
that have periphrastic-morphological alternations and suppletion, illustrated
by English and Modern Greek above, but no less true of most of Germanic,
all of Slavic, Georgian, Latin, and even Old French, which had nonsuppletive
morphological comparatives, such as grant – graindre/graignor
‘big – bigger/bigger.obl’ alongside suppletive mieudre – oblique: meillor
‘better’ (Bauer and Slocum 2006).
In these languages, the set of adjectives that undergo suppletion is a
proper subset of those that undergo Merger — in English, a small proper
subset, while in Old French, there are few adjectives like grant. But nothing
requires that the suppletive adjectives be a proper subset of those undergoing
Merger, and a logical possibility is for the set of adjectives undergoing Merger
to be a small set comprised solely of those that also undergo suppletion.
Indeed, this is what Modern French instantiates, with a clear historical record
showing the set of adjectives forming morphological comparatives declining
over time until only the suppletive ones remain (Romanian has gone further,
109
with even ‘good’ and ‘many’ becoming periphrastic only, and thus leaving no
suppletion whatsoever).
In other words, the theoretical tools needed to describe French without
invoking a distinct mechanism of Poser-blocking (and thereby giving up the
otherwise well-motivated locality condition on morphological interactions)
are independently needed for the description of languages like Greek and
Georgian, discussed in the previous section. What is special about French, on
this view, is only that the set of adjectives that undergo Merger is extremely
small, and coextensive with the set that also happen to undergo suppletion.
This account treats French on a par with the other languages discussed in
this section, and indeed, where French does show alternations, rather than
competition, it does behave exactly like the other languages, with suppletion
restricted to the non-periphrastic construction. The following vocabulary
can be taken to underlie the French forms for ‘bad’ in (99), which alternate
between periphrastic and morphological expressions depending on meaning;
the derivation of the forms is in all respects exactly parallel to the Greek
forms in (cf. (93)):
(105) a. bad → pire / ] cmpr ]
b. bad → mauvais
c. cmpr → Ø / ]ADJ ]
d. cmpr → plus
From this perspective, the burden of proof lies squarely with those who
110
would invoke the additional mechanism (and weaker locality conditions) of
Poser-blocking. The intuitive motivation behind such a view appears to be
that French, unlike Greek, has no independently detectable comparative affix,
which forms regular morphological comparatives. Perhaps there is an implicit
appeal to learnability here — proponents of Poser-blocking may worry
about how the Merger operation could be acquired by the French child, if
there are no transparently segmentable instances of synthetic comparatives
instantiating (92b) in the language. But this is a red herring. If the locality
condition in (90) is a part of UG, and thus of the innate knowledge
the child brings to bear on the acquisition process, then the child learning
French is forced by the overt evidence to posit the two-step derivation that is
(more) transparent in other languages. Given the evidence that meilleur is
the comparative of bon ‘good’, the child has no alternative but to posit root
suppletion. Since suppletion is restricted to complex X0
nodes, the child must
postulate an operation, such as Merger, which derives a complex X0
from the
underlying syntax in (91). With no (synchronically) segmentable affix in the
suppletive forms (unless the common -r of meilleur, pire permits of decompostion),
the child must then treat the comparative forms as portmanteaus,
for example, by means of a zero affix (as in (105c), but see section 5.3 for
an alternative). There is no mystery here, nor any conspiracy, but rather
a deterministic acquisition process, in fact the same process as the English
child applies to bad – worse, yielding the same grammatical representation
(up to phonology) for these cases. Invoking Poser-blocking not only threat-
111
ens to undermine the explanation of the RSG as a special case of the locality
condition on morphological interactions (as it rejects such a condition), but
it appears to require a non-uniform treatment of suppletive comparatives
across languages, arguably complicating the acquisition process as against
the deterministic view of acquisition incorporating (90) offered here.
Similarly, from a historical perspective, on the Poser-blocking approach,
to the extent that the difference between lexical blocking (English better
blocks *good-er) and phrasal blocking is a difference in grammatical organization,
it appears that the change in the behaviour of non-suppletive adjectives
in the history of French implies a reorganization of the grammar of the suppletive
adjectives. By contrast, on the view advocated here, the grammatical
derivation of the French suppletive comparatives never changes — meilleur
and pire are derived as indicated above, by (lexically conditioned) Merger
feeding contextual allomorphy of the root. What changes in the grammar of
French when an adjective like grant ceases to form morphological comparatives
is just that — the representation of that adjective changes, in that
it loses the ability (for example, a diacritic [+M]) that allows it to undergo
Merger. Nothing else changes as a consequence of this lexical change. Variation
and gradience in the historical change are thus readily accommodated,
as properties of individual lexical items. The Modern Romance situation is
then merely the endpoint of a series of lexical changes, the relentless march
towards regularization of the vocabulary, with Romanian (lacking suppletion
entirely) at the forefront.
112
3.3.3 Periphrastic superlatives again
With the quibble about Romance comparatives resolved, we may return to
the issue of periphrastic superlatives. We have seen that a basic locality condition
(90) restricts allomorphy, and thus root suppletion, to local, specifically,
word- (i.e., X0
)-internal domains. This derives the RSG. This in turn
now gives us a handle on the cross-lingusitic variation in suppletion in periphrastic
superlatives in (87) repeated here:
(106) pos cmpr sprl
a. M. Greek kak-ós cheiró-ter-os o cheiró-ter-os ‘bad’
b. Votic üvä par@-pi k@ikkia par@-pi ‘good’
c. Tv Karel. hüvä pare-mbi {ülen/suamo} hüvä ‘good’
d. Russian plox-oj xuž-e samyj plox-oj ‘bad’
In all four examples, the superlative is periphrastic. In the Modern Greek
and Votic examples, the superlative inherits the suppletive root from the
comparative, but in Tikhvin Karelian and Russian, it does not.30
The RSG
contributes to the explanation of this difference. They key observation is
that the languages differ independently of suppletion as to whether the periphrastic
superlative transparently embeds the comparative or not, as shown
here:31
113
(107) pos cmpr sprl
a. M. Greek kal-ós kalí-ter-os o kalí-ter-os ‘good’
b. Veps čoma čome-mb ke
“
iki´d čome-mb ‘pretty’
c. Tv. Karel. hoikku suamo hoikkua ‘thin’
d. Russian sux-oj suš-e samyj sux-oj ‘dry’
Combined with the locality restriction, the behaviour of the regular (nonsuppletive)
superlatives in (107) explains the suppletive patterning in (106).
In the Modern Greek and Votic/Veps examples, the periphrastic superlative
transparently embeds the morphological comparative; thus the comparative
morpheme is sufficiently local to the adjective root to trigger allomorphy.
But in the non-suppletive Tikhvin and Russian examples, the periphrastic
superlative embeds the positive, not the comparative, form of the adjective.
Since the comparative is thus not contained in the same word as the adjective
in the superlative in these languages, the comparative root allomorph
is not expected in the superlative, and suppletive allomorphy is limited to
the comparative form. Generalizing: given the locality condition in (90)—
the condition that derives the RSG—the CSG (i.e., CSG1) is predicted to
hold only of those periphrastic constructions in which the superlative marker
combines with a comparative adjective, and the CSG is predicted not to
hold where the superlative element (appears to) select the positive form of
the adjective. Not only is the restriction of the CSG to morphological forms
thus not stipulated, in fact the variation in the behaviour of periphrastic
constructions is predicted by the theory advanced here.32
114
Moreover, Tikhvin and Russian show the same sort of alternation discussed
for Greek, Georgian and Italian in section 3.3.1. The periphrastic superlatives
exist as doublets alongside morphological superlatives. But where
the periphrastic constructions lack suppletion (up to reinforcement), the morphological
alternants are built on the suppletive roots, as predicted, thus
Tikhvin: suamo hüvä ∼ para-š ‘most good ∼ best’ and Russian samyj ploxoj
∼ (nai)-xud-šij ‘most bad ∼ worst’ (the morphological form being rather
literary).
And yet, the success of the predictions regarding (106) raises the specter of
a problem concerning the Containment Hypothesis. Why don’t the Tikhvin
and Russian examples in (106)-(107) pattern with the Greek and Votic/Veps
examples in embedding the morphological comparative? Within the general
framework adopted here, if the Containment Hypothesis holds of anything,
it should hold of the structure that is the input to morphology, namely,
syntax, and thus its effects should be visible in both synthetic and analytic
realizations of that structure. The examples in (107c-d) therefore appear
to challenge the Containment Hypothesis. Ultimately, I will argue that the
forms in (107c-d) do satisfy the Containment Hypothesis (by containing a
‘hidden’ comparative morpheme), and that the difference in (107) is a function
of where that comparative morpheme ends up in the superlative. If it
is part of morphological word containing the adjective root, we see the behaviour
in (107a-b), but where the comparative is instead contained in the
superlative marker, the pattern in (107c-d) emerges. There are, however,
115
some stpes needed in getting to this conclusion, and it is to this we now
turn.
3.4 The Synthetic Superlative Generalization
In the preceding section, I showed that the adoption of a relatively innocuous
looking locality condition derives an apparently valid universal about the independence
of root suppletion and periphrasis. I showed moreover that this
condition interacts with different types of periphrastic superlative formation
to determine whether comparative suppletion will extend to periphrastic superlatives
or not, in any given language. The account was incomplete, and it
is time to look in more detail at the derivation of superlatives. In doing so,
we will develop a preliminary typology of superlative constructions, for which
all of the basic theoretical possibilities are instantiated. In addition, we will
derive another universal, namely, the Synthetic Superlative Generalization,
given in (89). This generalization is almost entirely borne out by the data
in this study, with a handful of potentially problematic cases, none of which
looks to be a compelling counter-example; I return to these at the end of the
section.
(89) The Synthetic Superlative Generalization (SSG)
No language has morphological superlatives (X-est), but only periphrastic
comparatives (more X ).
116
A priori, the SSG is not predicted in any obvious way under theories in
which the superlative and comparative are merely two types of degree head,
which can combine with adjectives (though it may be stipulated, for example,
as part of a markedness hierarchy).33
On the other hand, the SSG may follow
from the Containment Hypothesis, on relatively straightforward (though not
wholly innocuous) assumptions. One way of thinking about this is as follows.
Assume the syntactic structure underlying a comparative is (in part) as in
(91), repeated here as (108), where a comparative head takes an adjectival
phrase as its complement.
(108) CmprP
cmpr AdjP
adj
Under (the most straightforward implementation of) the Containment
Hypothesis, the superlative properly embeds the comparative, thus (109):
117
(109) SprlP
sprl CmprP
cmpr AdjP
adj
As noted above, morphological comparatives are modeled by application
of Merger (or Head Movement), which combines the two heads, as in (92),
repeated here as (110).
(110) a. CmprP
cmpr AdjP
adj
b. c
a
adj
cmpr
As a matter of observation, Merger is available in some languages, though
often, as in English, subject to constraints that restrict its application to certain
classes of adjectives. In other languages, such as Romanian, Albanian
and Turkish, Merger is evidently unavailable, inasmuch as these languages
exclusively permit periphrastic comparatives; even commonly suppletive adjectives
such as ‘good’ and ‘many’ are entirely regular in these languages:
118
(111) pos cmpr
a. Romanian mult mai mult ‘many’
b. Albanian i mirë më mirë ‘good’
c. Turkish iyi daha iyi ‘good’
Now, a morphological superlative, under the containment hypothesis,
must be derived by successive operations of Merger, as in (112a), yielding
the morphological object (112b). This is the basic structure of morphological
superlatives underlying the theory presented in the previous chapter. The
morphological output of this derivation (112b) corresponds to the forms with
transparent morphological embedding as in (60). For languages with a null
allomorph of the comparative (see (42) in chapter 2), this same derivation
yields morphological superlatives of the English long-est type.
(112) a. SprlP
sprl CmprP
cmpr AdjP
adj
b. s
c
a
adj
cmpr
sprl
The SSG is derived if a language that lacks Merger in the comparative
structure (108), also lacks Merger whenever that structure is further embed-
119
ded, as in (109). A further assumption is that Merger cannot skip intervening
heads (this is part of the definition of Morphological Merger in Marantz 1989).
Recall from above that an alternative to Merger for all of these structures
is head movement in the syntax. Analogous considerations apply if head
movement is the device of choice for deriving morphological comparatives
and superlatives; the SSG is derived if head movement must be successive
cyclic and cannot skip intervening heads (see Travis 1984 and much subsequent
work).34
Under these assumptions, a morphological superlative will
only be possible when the operation that creates morphological comparatives
is also possible. A language lacking (110) must also lack (112). As
expected, languages with exclusively periphrastic comparatives have exclusively
periphrastic superlatives:
(113) pos cmpr
a. Romanian mult mai mult cel mai mult ‘many’
b. Albanian i mirë më i mirë shumë i mirë ‘good’
c. Turkish iyi daha iyi en iyi ‘good’
Fully periphrastic constructions like that in Romanian, matching up to
(109) with no movement, are also attested in the other Modern Romance
languages, as well as (Modern) Greek and Georgian (in the latter two, the
periphrastic comparative alternates with a morphological comparative, at
least for some lexemes; we return presently to the less transparent cases such
as Turkish):
120
(114) pos cmpr sprl
a. French: gros plus gros le plus gros ‘big’
b. Greek: psil-ós pjo psil-ós o pjo psil-ós ‘tall’
c. Georgian: lamaz-i upro lamaz-i q’vela-ze upro lamaz-i ‘beautiful’
Now, it is also theoretically possible for the superlative marker to be an
affix, even where the comparative is not. Yet given the derivation of the
SSG above, the only way for this state of affairs to arise is if the superlative
attaches not to the adjective, but to the (otherwise free) comparative
morpheme. The relevant derivation is given in (115a).
(115) a. SprlP
sprl CmprP
cmpr AdjP
adj
b. s
c
cmpr
sprl
This derivation is characteristic of one superlative-forming strategy in the
Slavic languages. In Slovenian, for example, as in English, not all gradable
adjectives permit of a morphological comparative formation, and some adjectives
form only periphrastic comparatives. The superlative is formed in
all instances with the prefix naj-. For morphological comparatives, the pre-
121
fix attaches directly to the comparative adjective, following the derivation in
(112), but for periphrastic comparatives, the prefix attaches to the comparative
adverb bolj, reflecting the derivation in (115). Illustrative examples of
the two patterns are given in (116) (from Herrity 2000, 81-85):
(116) ‘weak’ ‘healthy’
positive slab zdrav
comparative slab-ši bolj zdrav
superlative naj-slab-ši naj-bolj zdrav
Another language which may have derivations of this type is Sinhalese.
Garusinghe (1962, 43) gives the following, noting that the comparative is
formed periphrastically (with vád. ¯a), and the superlative may be formed from
this by adding the emphatic particle -ma.35
(117) a hoñda lámay¯a ‘the good boy’
b vád.¯a hoñda lámay¯a ‘the better boy’
c vád.¯a-ma hoñda lámay¯a ‘the best boy’
more-emph good boy
The derivation in (115) could also underlie periphrastic superlatives where
there appears to be a superlative adverb or particle that combines directly
with the positive form of the adjective. For example, English most could
be (and probably should be) treated as an amalgam of the comparative and
superlative heads (that is, most contains more and is thus derived as in
(115)). Turkish is another case in point, with comparative and superlative
122
adverbs (in Modern Turkish, all comparatives and superlatives are formed
periphrastically). This pattern is consistent with the containment hypothesis,
if, as just suggested for English most, Turkish en can be analyzed as a
portmanteau morpheme, realizing the comparative and superlative heads in
(112). Given the locality condition discussed above, positing that the apparent
superlative adverbs are portmanteaus of [sprl + cmpr] requires that
these heads be combined, prior to exponence, exactly as in (115)).
(118) ‘tall’ ‘bad’
positive uzun kötü
comparative daha uzun daha kötü
superlative en uzun en kötü
Returning to morphological comparatives, nothing said to this point forces
a language with morphological comparatives to have morphological superlatives.
Another way that the structure in (109) may be spelled out is for
Merger to apply only between the comparative morpheme and the adjective,
with the superlative expressed as a free-standing element. We have seen examples
of this type above, for example Modern Greek and Livonian (119a-b),
which form the superlative by means of a definite article or demonstrative.
The pattern is also attested in some of the many languages in which the
superlative is formed with a universal quantifier, as is possible in Veps (c)
and Evenki (d), and in languages with a free-standing superlative marker,
such as Ingush (e).36
These examples may correspond to a derivation in
which the superlative embeds the comparative, but only the lower step of
123
movement/merger applies.
(119) pos cmpr sprl
a. Greek psil-ós psiló-ter-os o psiló-ter-os ‘tall’
b. Livonian van¯a van¯ı-m se van¯ı-m ‘old’
c. Veps: čoma čome-mb keikiš čome-mb ‘beautiful’
d. Evenki: engesi engesi-tmer upkat-tuk engesi-tmer ‘strong’
e. Ingush: doaqqa doaqqa-gh eggara doaqqa-gh ‘big’
A final possibility, and perhaps the least obvious, is another type of situation
in which comparatives are morphological, but superlatives are periphrastic
(as in (119) above), but in which, unlike (119), the superlative does
not embed the overt morphology of the comparative. This is the situation
discussed above in reference to Tikhvin Karelian and Russian in (107c-d),
bringing us back to the beginning of this chapter. Some examples of this
pattern are given in (120). Note that in some of these languages, the pattern
illustrated is not the only one — in some cases there is a morphological
comparative in addition, or a variety of means of forming superlatives.37
(120) pos cmpr sprl
a. Russian: prost-oj prošč-e samyj prost-oj ‘simple’
b. Ossetian: bærzond bærzond-dær iuuIl bærzond ‘high’
c. Udmurt: kužj
kužj
-ges tuž kužj
‘long’
d. Chuvash: layăX layăX-raX či layăX ‘good’
e. Arrente: tjenja tjenj-ulkura tjenj’ indora ‘tall’
124
In line with the assumptions made above, the most straightforward analysis
of these examples would be to assume that the words meaning ‘most’
have the morphosyntax of their counterparts in English. That is, they reflect
portmanteaus of the comparative and superlative heads, derived as in
(115), repeated below as (121a).38
These languages differ, though, from the
pattern in (118) in that the comparative, when it does not combine with the
superlative, combines with the adjective (as in (110) = (121b)), yielding a
morphological comparative form. This paradigm is interesting, since it indicates
that, while a language cannot have Merger in (110) in superlatives if
that operation is not available in comparatives generally, the reverse does not
hold — a language may allow Merger in comparatives generally, yet opt for a
different pattern (namely (115)) when the comparative subtree is embedded
under a superlative. That was somewhat of a mouthful, and is schematized
in (121):
(121) a. SprlP
sprl CmprP
cmpr AdjP
adj
b. CmprP
cmpr AdjP
adj
125
Though the analogy may prove spurious, there is at least one widely discussed
precedent for derivations along these lines, where a given functional
element (or feature) undergoes movement to a higher position in some configurations,
but otherwise surfaces as an affix on the head of its complement.
At least to a first approximation, this parallels the classic Affix-Hopping
analysis of English inflection Chomsky (1957) (recast in the framework of
Distributed Morphology in Halle and Marantz 1993 and Bobaljik 2002a). In
that analysis, English finite inflectional morphology originates in a functional
projection (Infl0
) above the verb. In inversion contexts, such as interrogatives,
inflection raises to the C (COMP) position, spelled out as inflected
do, and the verb is in situ in its bare form (as in (122a)). When there is no
inversion, the inflectional morphology ‘lowers’ to the verb (by Morphological
Merger, in Halle and Marantz 1993, Bobaljik 2002a), as in (122b):
(122) a. CP
C0
IP
Infl VP
V0
b. IP
Infl VP
V0
Coming back to the comparatives, the proposed derivational pairing in
(121) for languages with the pattern in (120) adds the final point to the
126
discussion of the interaction of periphrasis and the CSG, with which we ended
section 3.3. In that section, I noted a split among periphrastic superlatives,
where a suppletive comparative allomorph sometimes was, and sometimes
was not, carried over to a periphrastic superlative. The representative pairs
are repeated here:
(123) pos cmpr sprl
a. M. Greek kak-ós cheiró-ter-os o cheiró-ter-os ‘bad’
b. Votic üvä par@-pi k@ikkia par@-pi ‘good’
c. Tv Karel. hüvä pare-mbi {ülen/suamo} hüvä ‘good’
d. Russian plox-oj xuž-e samyj plox-oj ‘bad’
In section 3.3, I noted that the difference was derivative of an independent
difference, namely, whether the periphrastic superlative transparently
embeds a morphological comparative as its complement or not. The discussion
above contributes the explicit derivations that underlie that difference.
When the morphological comparative head surfaces on the adjective, the
derivation is that in (110), embedded under a superlative, and the comparative
head is local enough to the adjectival root that it governs suppletion.
The derivation of (123c-d) by contrast is that in (121a). The comparative
head, as in (115) is combined with the superlative, and thus is not in the
same local domain as the adjective root and fails to govern suppletion.39
127
3.4.1 Armenian superlatives
Finally, I return to the question of apparent counter-examples to the SSG.
The main case to consider here is from Modern Western Armenian. Sporadic
descriptions in the literature suggest some other cases deserving of further
scrutiny, but none of these are clear-cut, convincing cases of languages with
regular affixal superlatives cooccuring with uniquely periphrastic comparatives.
Brief remarks on these other cases are provided after the discussion
of Armenian. It should be kept in mind on the one hand that the SSG is
independent of suppletion, and therefore holds over a much wider range of
languages than the CSG. On the other hand, it should also be remembered
that here, as throughout, the SSG ranges over relative superlatives; absolute
superlatives are frequently morphological, with no relation to the expression
of the comparative, as, for example, in the Modern Romance languages.
In Classical Armenian, the suffix -goin was added to adjectives to mark
the comparative. In Middle Armenian, this was replaced by a periphrastic
construction ail + adj (Karst 1901, 394), although as late as the 19th century
it was reported that the suffixal comparatives are “occasionally met
with” (Riggs 1856, 19). In Modern (Western) Armenian the comparative
can be formed periphrastically, and there is no (overt) affix that attaches to
adjectives to yield a comparative form. The superlative is generally formed
periphrastically, with the adverb amen-e-n ‘all-abl’, as in (124) (see also
(169b) in section 4.3).
128
(124) pos cmpr sprl
partsrahasag (aveli) partsrahasag amenen partsrahasag ‘tall’
However, there are also two options for superlative formation described in
modern grammars as affixal. One is the prefixed form of the quantifier amena-
and the other is the suffix -kujn, descendant of the classical comparative
suffix (homophonous with the world for ‘colour’), as in lav-a-kujn, ‘best’, from
lav ‘good’ (H. Khanjian, personal communication 2010, see also Bardakjian
and Vaux 2001; Modern Eastern Armenian has analogous comparatives and
employs the amen-a- construction for superlatives, see Kozintseva 1995, 10-
11 and Dum-Tragut 2009). Described in these terms, Modern Armenian
poses a potential counter-example to the SSG, with affixal superlatives, but
apparently no affixal comparative.
But the counter-example may well be only apparent, with at least two
analyses presenting themselves. I will discuss both here, remaining agnostic
however about which (if either) of these is correct.
In the first place, the putative ‘prefix’ amen-a bears the linking vowel
characteristic of a compound structure. Evidently, compounds do not count
as local morphological domains in terms of the ability of the non-head element
to govern suppletion on the head (see also section 4.3). The element
a-kujn, which is moreover reportedly rather unproductive, also appears to
have a compound structure (compare compounds in which it serves as a
colour term: vard-a-guyn ‘rose-lv-colour’ = ‘pink’ Dum-Tragut 2009, 673).
If these are compounds, rather than affixes, then they may not count as
129
affixal superlatives for the purposes of the SSG, and thus not as problems.
If we lay the compounding question aside, there is another reason that
it is not clear that the Armenian examples would count as problems for the
SSG. This second reason is worth discussing in some detail (even if ultimately
inconclusively) as it connects to a point of current debate in the semantic
literature. In Modern (Western and Eastern) Armenian, the comparative
is described (Bardakjian and Vaux 2001, Kozintseva 1995) as periphrastic,
formed with the adverb aveli ‘more’ and a standard in the ablative case, as
in (125):40
(125) a. es
I
qezme’
you.sg.abl
aweli
more
mec’
big
em
be.1sg.pres
‘I am bigger than you.’ (Bardakjian and Vaux 2001, 121)
b. Artak-@
Artak-def
Bagrat-e-n
Bagrat-abl-def
aveli
more
partsrahasag
tall
e
be.3sg.pres
‘Artak is taller than Bagrat.’
On the other hand, as in a number of other languages, the adverb meaning
‘more’ is optional here, and comparatives may consist simply of the unmarked,
positive form of the adjective together with the ablative-marked
standard (Riggs 1856, 19), as in (126):
130
(126) a. qezme’
you.sg.abl
mec’
big
em
be.1sg.pres
‘I am bigger than you. (lit: ‘I am big from you.’) (Bardakjian
and Vaux 2001, 121)
b. Artak-@
Artak-def
Bagrat-e-n
Bagrat-abl-def
partsrahasag
tall
e
be.3sg.pres
‘Artak is taller than Bagrat.’
So the question of whether Armenian constitutes a genuine counterexample
to the SSG hangs on whether (126) is a ‘synthetic comparative’
in the relevant sense, for example with a phonologically null, affixal, exponent
of the comparative. If it is, then (126) has the structure in (91) and
Modern Armenian is consistent with the SSG after all. The optional element
aveli might then involve reinforcement (see section 3.3.2). The superlative
would in turn be built on this structure, just as the superlative contained
a concealed comparative structure in English, for which a null allomorph of
the comparative morpheme, was posited in (42). For English, the null allomorph
was a theoretical device that allowed for the nesting pattern to be
present in English as well, even though the comparative is not visible on the
surface. The only difference between English and Armenian in these terms is
that the zero affix would be the only (affixal) exponent of the comparative in
Armenian, where it is a contextually restricted allomorph thereof in English.
In fact, null comparative morphemes have been considered for languages
that use the strategy in (126) at least since Ultan (1972, 127). The most thor-
131
oughly investigated of such languages is Japanese, which forms comparatives
with no overt marking of the adjective (neither affixal nor periphrastic), as
shown in (127).
(127) nihon-go-wa
Japnese-language-nom
doits-go
German-language
yori
from
muzukashi
difficult
‘Japanese is more difficult than German.’ (Kennedy 2007a, 2)
The proper semantic (and thus syntactic) analysis of Japanese comparatives
is the matter of no small debate (see Ishii 1991, Beck et al. 2004,
Kennedy 2007a, Oda 2008, Beck et al. 2009). At center stage in the debate
is not just the obvious morphological difference between Japanese and
English, but a host of syntactic and semantic differences between comparative
constructions in the two languages, such as whether the language allows
‘comparative subdeletion’ (The shelf is taller than the door is wide) and others
(see Beck et al. 2009 for the broadest current cross-linguistic investigation
of semantic variation in comparative constructions). Of current relevance is
that proponents of both poles of the debate posit a null comparative morpheme
in Japanese, which combines with the adjective (Beck et al. 2004
versus Kennedy 2007a); the differences among the approaches lie in the semantics
assigned to this element and in how it interacts with the semantics
of the language more broadly.41
For example, Kennedy (2007a) explicitly
considers, and rejects, the possibility that Japanese could lack a comparative
morpheme, relying on ‘implicit comparison’ to exploit the inherent vagueness
of gradable adjectives to achieve comparative-like readings in the appropriate
132
contexts.
A preliminary attempt to investigate in more detail the relevant properties
of Armenian suggests that it patterns in relevant respects like the languages
that Beck et al. (2009) treat as having a null comparative affix. For example,
Armenian shows what Kennedy refers to as crisp judgments. In English, a
comparative like Michael is taller than Hillary does not imply the positive
Michael is tall. The same is true in Armenian — example (128b) can be
judged true even in a context where (128a) is judged false (for example,
where both men are short). What makes this observation interesting is that
the Armenian sentence in (128b) has this property even without the adverb
aveli ‘more’. This behaviour contrasts with Motu, a language that uses a
conjoined comparative strategy, as described in Beck et al. (2009).
(128) a. Artak-@
Artak-def
partsrahasag
tall
e
be.3sg.pres
‘Artak is tall.’
b. Artak-@
Artak-def
Bagrat-e-n
Bagrat-abl-def
(aveli)
more
partsrahasag
tall
e
be.3sg.pres
‘Artak is taller than Bagrat.’
Another property that Beck et al. (2009) take to diagnose a true comparative
structure, with an operator over degrees, is the ability to take a
measure phrase, specifying the differential between two objects in a comparative
construction. This too Armenian allows with or without the overt
133
makrer aveli.42
(129) Artak-@
Artak-def
Bagrat-e-n
Bagrat-abl-def
jergu
two
santim
cm.
(aveli)
more
partsrahasag
tall
e
be.3sg.pres
‘Artak is 2cm taller than Bagrat.’
In other properties as well, Armenian seems to pattern with languages
that are analyzed elsewhere in the literature, on semantic grounds, as requiring
a comparative degree operator, which must therefore be a null element
in the sentences lacking aveli.43
Although this discussion has been tentative, if a null comparative affix
is indeed motivated for Armenian, then the language is not in violation of
the SSG. It does have affixal comparatives after all, the affix being simply
inaudible. What the SSG disallows, as stated above, is a language with a morphological
superlative and only periphrastic means of forming comparatives.
Armenian is not such a language, nor is any other that I have encountered.
This result does, however, raise a different question. Since unmarked comparatives
of the Japanese/Armenian are quite common (roughly a third of
the 110 languages in Stassen 1985 have this type of comparative): why is it
that the Armenian pattern of a zero comparative and an overt morphological
superlative is so rare? I leave this question unanswered, noting that it
may detract from the plausibility of the null comparative solution for Armenian,
and suggest instead that pursuing locality restrictions in compounds
134
(as suggested earlier) is the more promising alternative.
Without delving deeper into this developing area, I leave this thread with
the observation that there are at least two reasons to think that Modern
Armenian, despite initial appearances, is not a counter-example to the SSG.
First, the putative affixal superlatives may well not be affixes. Second, the
language may have a (null) affixal comparative, a device widely appealed
to in the analysis of other languages with similar syntactic and semantic
properties. If either (or both) of these considerations prove to be valid, then
the SSG stands unthreatened as a linguistic universal.
3.4.2 SSG: Loose ends
I return here to a few remarks on some additional examples, beyond Armenian,
for which available descriptions appear to suggest morphological superlatives
in the absence of morphological comparatives. None of these seems
particularly compelling, but I note them here for the sake of completeness.
In Classical Nahuatl, one of the adverbial modifiers that marks the superlative
incorporates, but none of the comparative intensifiers do, see Andrews
(1975, 563-566). From the description, however, this appears to be an
intensifier, marking primarily an absolute superlative and hence beside the
point. This may also be true of the Zulu suffix -kazi, cited briefly in Poulos
and Msimang (1998, 403) as an alternative to the regular means of forming a
superlative meaning via an exceed comparative as shown in (80). Poulos and
Msimang do not discuss whether -kazi forms primarily relative or absolute
135
superlatives, although the example they give is glossed as ‘He is the tallest
in the school.’
The description of Paiwan in Egli (1990), mentioned above, gives comparatives
formed with the free-standing particle tja preceding the adjective, and
superlatives formed with the circumfix tjala-...-an. The criteria for treating
the tjala portion as a prefix, rather than a particle, are not clear from the
description; note in particular that the tjala portion of the superlative appears
to consist of an emphatic element -la- that also occurs in the other
superlative formative ka-la-...-an.
Note also short the descriptions of Koryak and (closely related) Alutor in
Zhukova (1968b, 277) and Zhukova (1968a, 299), which give a zero-marked
standard comparative construction for predicate adjectives, as in Koryak:
konja qoja-k n@-mej@N-qin = horse reindeer-loc infl-big-3sg ‘A horse is
bigger than a reindeer (lit: at a reindeer big)’, but a circumfixal superlative
@nan-majN@-čè@n ‘biggest’ among a list of intensified adjectival forms. But
the longer description of Koryak in Zhukova (1972) gives the missing intermediate
forms: the suffix -čè@n forms comparatives, such as @ppulju-čè@n
‘smaller’ (root: @ppulju- ‘small’) from which the superlatives are derived by
the intensifying prefix @nan- (compare Chukchi in (60h). Thus in the Koryak
case, the SSG violation is only an artefact of the brief description.44
Note also in this context that throughout this work I have implicitly
limited the scope of the investigation to adjectival gradation. In some languages,
the comparative and/or superlative morphology can combine with
136
other parts of speech, including spatial expressions such as adverbs or adpositions.
The SSG, at least as phrased in (89a), appears not to hold of this
domain; thus English top – topmost, left – leftmost, etc., likewise Hungarian
alsó – leg-alsó ‘down’ – ‘bottommost’. Note however that in this type of
formation, there appears to be no corresponding comparative at all, neither
affixal nor periphrastic: *topper, *topmore, *more top, and thus these are
consistent with the formulation of the SSG in (89b). In (Modern) English,
where there is an apparently comparative form of the preposition/adverb, the
nesting structure occurs: upper – uppermost; though in such cases, only the
comparative and superlative forms can be used as pre-nominal modifiers: the
upper(most) reaches of the river vs. *the up reaches of the river. I put the
use of apparent comparative and superlative morphology with adpositions
and locational expressions aside, noting that there are interesting questions
to be investigated in this area. (On the history and origin of the English
suffix -most, see the entry for this affix in the OED.)
A final loose end to be considered here is the following: Adam Albright
points out (personal communication 2008) that some adjectives in English
seem to enter into morphological superlative formation, but are strikingly
awkward in a morphological comparative: ?supremest –?*supremer, ?sublimest
– ?*sublimer, ?alertest – ?*alerter. Judgments are slippery here (see
Arnold Zwicky’s discussion at: http://languagelog.ldc.upenn.edu/nll/
?p=1923), and the comparative forms are attested. Here again, the relative
vs. absolute superlative distinction is probably important (even if not always
137
sharp). Consider the apparent SSG-violating mere – *merer – merest mentioned
by a commenter on the LanguageLog post cited; the counter-example is
only apparent, as the meaning of mere is not gradable, and merest is clearly
only an absolute, not relative, superlative in meaning. So also darn(ed) –
*darned-er – darnedest, as in the outmoded cliché Children say the darnedest
things, which again has only an absolute, not a relative, reading. For cases
that truly involve relative superlatives, I suspect that the grammar of English
permits both periphrastic and morphological forms for both superlative
and comparative grades (thus Merger is optional for these adjectives, as for
politer/more polite and politest/most polite), but variability in both attested
distribution and relative acceptability judgments reflect gradient preferences
(i.e., performance, rather than competence), at least in some cases phonologically
driven (for example, avoiding the sequence of -er-final syllables in
?*alerter).45
3.5 Containment and semantic considerations
The account of the CSG presented in section 2.2 above relies on the premise
that the representation of superlatives in all languages (or at least all languages
with suppletion) satisfies the Containment Hypothesis. The preceding
sections have focused on providing morphological evidence, independent of
patterns of suppletion, that the Containment Hypothesis holds true quite
generally, with specific attention to explaining away apparently problematic
138
cases. At this point, I wish to touch briefly on the question of whether there
is any semantic evidence bearing one way or another on this issue. This is
not the place for an in-depth discussion of the semantics of superlatives. My
aim here is instead to show that a compositional semantics for a containment
structure seems attainable (as argued, with important caveats, by Stateva
2002), and that while this is not without challenges, some of the apparent
problems for such an approach identified by Stateva may be orthogonal to
the question of containment.46
In pretheoretic terms, a relationship between comparatives and superlatives
might seem almost to require no special pleading. Relative superlatives
seem by their very nature to be comparative in meaning: biggest means ‘bigger
than all other members of the comparison set’. While this has intuitive
appeal, not all formal treatments of the superlative include an explicitly comparative
semantics. Consider the following two (partial) versions of a lexical
entry for the superlative morpheme, drawn from the recent literature (these
have been simplified in ways that are orthogonal to the point to be made).
(130) -est (R)(x)=1 iff ∀y[y=x → max{d:R(d)(x)=1} > max{d:R(d)(y)=1}]
(131) -est (R)(x)=1 iff ∃d[R(d)(x)=1 & ∀y[y=x → ¬R(d)(y)]
In each of these entries, the superlative takes two arguments: a gradable
property R of type < d, < e, t >> (the adjectival stem) and an individual
x. The first formalism (from Hackl 2009, 79, cf. von Fintel 1999, Heim
2000) is transparently comparative: the resulting expression is true just in
139
case the individual x has the property R to a degree that is greater than
the degree to which any other individual has that property. John is the
tallest is true if the degree to which John is tall is greater than the degree
to which any other (relevant) individual is tall. The ‘greater than’ symbol in
the lexical entry explicitly encodes comparison between (maximal) degrees.
For reference, compare the meaning that Hackl (2000, 50) assigns to the
comparative morpheme, given in (132). This element takes two (maximal)
degree operators and compares them directly. The components of (132) are
contained in the superlative in (130).
(132) -er (d)(d′
)=1 iff max(d) > max(d′
)
By contrast, the formalism in (131) (mentioned in Hackl 2009, n.36 as
a “more standard semantics for the superlative,” see also Gajewski 2010)
achieves essentially the same meaning as (130), but without an explicitly
comparative element. Instead, the expression is true if there is a degree to
which the individual has the property in question, and no other individual
has the property to that degree. Thus: John is the tallest is true if there is
a degree to which John is tall (say 2.5m), and no other (relevant) individual
is tall to that degree. While this paraphrases comparison, the superlative in
(131) does not formally contain (132).47
So our first conclusion from the semantics is this: to the extent that a
representation like (131) is logically possible, it does not follow from logical
considerations alone that the superlative must contain an explicit compar-
140
ative element of meaning. The comparative sense may be achieved in an
indirect manner, and the semantics alone thus does not force the Containment
Hypothesis upon us.
The next observation about the semantics is the following: to the extent
that the superlative does have a comparative element in its meaning, it is
far from obvious from the semantic representation that this element stands
in the right grammatical configuration to trigger the elsewhere reasoning
that automatically extends the comparative stem allomorph to the superlative,
discussed extensively above. Recall that the key reasoning is that the
representation of the superlative satisfies the structural description of the
rule (of exponence) in (39d), repeated here, ensuring that the comparative
stem allomorph, and not the default allomorph, extends to the superlative
environment.
(39) d. bad → hor- / ] cmpr
e. bad → špatnIf
we substitute (132) for cmpr in (39d), will the allomorph hor- be used
in the superlative?
It should be possible to set the theory up in such a way that the answer
will be affirmative, for example, by stipulating that any morpheme that attaches
to adjectives and has the comparative symbol in its representation will
contain the comparative in the relevant sense. However, such a move seems
to be empirically inadequate. Consider in this light Heim’s proposal for other
degree operators, which also involve a comparative component. Heim’s se-
141
mantics for the excessive operator represented by English too (as in too tall,
too good) is given in (133), and it too contains the greater than symbol evaluating
a relation among (maximal) degrees. Heim’s paraphrase for John is
too tall is explicitly comparative: ‘John is taller than it is acceptable for
him to be tall’ (Heim 2000, emphasis added).
(133) too w
= λP<s,dt>. max(P(w)) > max{d:∃w′
∈ Acc(w):P(w′
)(d)=1}
The meaning of ‘too’ in (133) contains the comparative in the same way
that the superlative in (130) does. But as noted above, in languages where
the excessive or equative degree is affixal (illustrated by Czech in (50)) the excessive
fails to trigger comparative stem allomorphs.48
In addition to the evidence
from suppletion, both the evidence from morphological transparency
(nesting) and the SSG support the conclusion that the superlative is distinct
from other degrees such as the equative or the excessive — even when
affixal, excessive and equative degree morphology does not typically embed
comparative morphology, and languages may have a morphological excessive
or absolute superlative without having a morphological comparative.49
The
modern Romance languages are of this latter type, as are the many languages
lacking comparative morphology but with affixal intensifiers meaning roughly
‘very’ or ‘too’ (just as Heim’s too has a comparative component, the meaning
of very could be plausibly rendered as ‘more than the average or normal
degree’). There is no valid analogue of the SSG for degree morphology other
than the relative superlative.50
142
In sum, the superlative morphemes proposed in the formal semantics
literature, with the notable exception of Stateva (2002), attach directly to
the gradable adjective stem, one of many degree heads, instantiating (47),
repeated here, a configuration I have argued must be excluded.
(47) [ [ adjective ] degree ]
But these authors’ concerns were not morphological, and one may ask
whether semantic considerations force an indivisible structure on the superlative
morpheme, or whether representations such as (130) might be seen as a
shorthand for a more complex structure decomposed into comparative and
superlative elements, consistent with the morphological evidence. Stateva
(2002), noting the transparent nesting in Serbo-Croatian and Old Church
Slavonic/Old Bulgarian, argues extensively for just such a decomposition,
proposing a formal semantics for a superlative element that embeds the comparative.
In her theory, the comparative portion of the superlative meaning
is assigned to a comparative head, and the superlative morphology amounts
to, in effect, the meaning component ‘than all others’. Despite the promise
of a dovetailing of considerations from the semantics and the morphology,
Stateva ultimately retreats somewhat from her own conclusion, postulating,
at least for English, that the comparative morpheme that is embedded in superlatives
(her -ER) is semantically, as well as phonologically, distinct from
that heading the true comparative (-er). Specifically, she proposes that the
true comparative is quantificational, while that embedded in superlatives is
143
not. The postulation of two distinct comparatives in this way is worrying
from the theoretical perspective adopted here, since if there is such a difference,
and if the morphology can ‘see’ this distinction, then the result that
excludes the ABA pattern would be lost. A rule of stem allomorphy could (in
principle) make specific reference to the quantificational comparative; said
rule would then not extend to the superlative, and the (unattested) ABA
pattern would be readily derivable.
We could stipulate the problem away, and assert that even if Stateva
is correct about the distinctions, the morphology is blind to the semantic
distinctions that differentiate the quantificational and non-quantificational
versions of the comparative. On the other hand, it behooves us to ask at
this point: how strong are Stateva’s arguments that the two comparative
morphemes are distinct? I do not have the means to explore this thoroughly
here, but wish to point to a direction that an investigation might take. As
noted, Stateva’s arguments for this important difference in the comparatives
are built largely around English. She considers distributional differences between
comparative and superlative forms, and reasons that the superlative
should pattern with the comparative in the relevant construction, if they
contained the same quantificational element. One such contrast she presents
in detail is in the extensively discussed comparative conditional (or comparative
correlative) construction, shown in (134) (for recent discussion of (134a)
see, e.g., Den Dikken 2005 and references therein).
144
(134) a. The faster he drives, the earlier he’ll get there.
b. * The fastest he drives, the earliest he gets there.
Stateva follows Wold (1991), Beck (1997) and others in assuming that the
derivation of (134a) crucially involves movement of the comparative degree
operator, a process licensed by its quantificational nature. She contends that
(134b) is thus incorrectly predicted to be acceptable if the superlative contains
the same quantificational comparative degree marker. The conclusion
she draws is that the phonologically null comparative -ER, which she takes
to be contained in superlatives is crucially different from the phonologically
overt comparative -er, in that only the latter is quantificational. Thus the
explanation that she offers of the contrast in (134) is that the (true) comparative,
but not (the comparative in) the superlative, can undergo movement
(Stateva 2002, 135). This state of affairs, as noted, is potentially problematic
for the Containment Hypothesis developed here, inasmuch as I rely on the
comparative morpheme being the same in both contexts (up to its phonological
nullity). There are, however, lacunae in Stateva’s argument, that
undermine her conclusion, and thus void her argument against having the
same comparative morpheme in both contexts.
In the first place, Stateva limits her discussion to the role of the comparative
degree operator in the licensing of the construction, and does not
consider the role of the additional element that derives superlatives from
comparatives. As Jon Gajewski points out (personal communication, 2010),
this is a potentially crucial omission. Beck (1997) begins with the obser-
145
vation that in comparative conditionals, an overt standard (than-clause) is
disallowed:
(135) The faster he drives (*than me), the earlier he’ll get there (*than
me).
For Beck, the impossibility of (135) follows from a core aspect of her
proposal, namely that a (silent) standard of comparison is present in the semantics
of (the head of) the construction, and therefore “an (additional) overt
[standard] of comparison would be uninterpretable” (Beck 1997, 230). Now,
under Stateva’s theory, the superlative is decomposed into a comparative
element and an additional element that (in combination with a contextual
variable) “suppl[ies] the standard value” (Stateva 2002, 105). For all intents,
the meaning contributed by the superlative element is “than all (relevant)
others.” But if this is semantically contributing a standard of comparison,
then this theory already suffices to explain the impossibility of superlative
conditionals like (134b). The superlative cannot participate in this construction
for exactly the same reason that an overt than-phrase is impossible
(135). The quantificational nature of the (null) comparative is neither here
nor there, and the pair in (134) provides no impediment to the unified theory
proposed here in which there is but a single comparative morpheme in both
comparative and superlative constructions, with mere surface differences in
phonology.
The ability to maintain a unified semantic representation for the com-
146
parative is also a positive result for understanding (134) in cross-linguisitc
perspective. As noted above, there are many languages in which the (regular)
comparative morpheme is transparently contained in the superlative.
For Stateva, where the relevant factor in distinguishing (134a) from (134b)
is the difference between -ER and -er, she would appear to expect that the
contrast would disappear in languages that have but a single comparative
morpheme used in both contexts. Such languages should allow superlative
conditionals, analogous to (134b).
A preliminary investigation does not appear to support this, however.
There are three languages in Den Dikken’s (2005) survey of comparative
correlatives in which the superlative transparently contains the comparative,
namely Hungarian, Russian and French (affixally in the first, and periphrastically
in the other two). So far as I can determine, all three pattern with
English, excluding the superlative from the relevant construction.51
(136) Hungarian (A. Szabolcsi, personal communication, 2010)
a. Amennyivel
A-how.much-INST
magasabb
taller
az
the
apa,
father
annyival
that.much-INST
alacsonyabb
shorter
a
the
gyerek.
child.
‘The taller the father, the shorter the child.’ (Den Dikken
2005, 525)
147
b. * Amennyivel
A-how.much-INST
(a)
(the)
leg-magasabb
sprl-taller
az
the
apa,
father
annyival
that.much-INST
(a)
(the)
leg-alacsonyabb
sprl-shorter
a
the
gyerek.
child.
(137) Russian (Zh. Glushan, N. Radkevich, personal communication, 2010)
a. i
and
čem
what.instr
bolee
more
povtorjal,
repeated,
tem
that.instr
bolee
more
ubeždal-sja.
convinced-refl
‘And the more he repeated (it), the more he became convinced.’
(Russian National Corpus)
b. * i
and
čem
what.instr
nai-bolee
sprl-more
povtorjal,
repeated,
tem
that.instr
nai-bolee
sprl-more
ubeždal-sja.
convinced-refl
(138) French (P. Schlenker, personal communication, 2010)
a. Plus
more
je
I
lis,
read,
plus
more
je
I
comprends.
understand
‘The more I read, the more I understand.’
b. * Le
the
plus
more
je
I
lis,
read,
le
the
plus
more
je
I
comprends.
understand
To extend her theory to account for this range of data, Stateva (see
pp.110ff) has to posit systematic homophony for languages that show the
148
transparent embedding relationship. Such languages have two comparative
morphemes abstractly, corresponding to -ER and -er, but the morphemes
happen to receive the same pronunciation. Such systematic cross-linguistic
homophony suggests a generalization is being missed. The generalization, I
suggest, is that the explanation for the absence of superlative conditionals
does not lie in the (non)-quantificational nature of the comparative morpheme,
but rather, as discussed with respect to (135), in the contribution of
the superlative morpheme (i.e., the -T of -ES-T) to the meaning of a superlative.
By giving the superlative the meaning “than all others”, and allowing
the superlative to combine with the (regular) comparative, the absence of the
superlative conditionals falls out as a special case of the same consideration
that excludes (135).
It awaits future work to see if the other differences that Stateva identifies
between comparatives and superlatives remain constant across the morphological
divide between transparent and non-transparent containment. Of possible
relevance is Herdan and Sharvit (2006), who examine the potential of a
superlative to license a negative polarity item (NPI) in various definite and
indefinite contexts. From the facts they examine, they construct an argument
against quantifier-movement of the superlative. As with the cases Stateva
discusses, this would be an odd restriction if the superlative contains the
comparative, if they admit of quantifier movement of the comparative. However,
Herdan and Sharvit (2006) discuss, in addition to English, Romanian,
which they contend behaves the same as English in the relevant respects, al-
149
though in Romanian, the superlative is transparently formed by embedding
comparatives under a demonstrative(-like element). In the domain of NPIlicensing
too, then, it seems that the differences between superlatives and
comparatives should not be attributed to the absence of the comparative
morpheme in the former.
In any event, it should be noted that (at least as I understand the literature),
Stateva’s conclusion that the superlative is non-quantificational (in
contrast to the comparative), is by no means a consensus view, and a healthy
debate on this point continues (see Hackl 2009, Gajewski 2010 for recent discussion).
Authors on both sides of the debate note that the evidence for or
against -est-movement, the core of the quantificational analysis (Heim 1985,
Szabolcsi 1986), is delicate at best and some key examples are subject to variation
and uncertainty among speakers. Moreover, to date, extremely little
attention has been paid to the relevant semantic properties of superlatives
in languages other than English, and what evidence there is is somewhat
ambivalent.
In sum, I conclude that although the idea of embedding the comparative
in the superlative is currently a minority view in the formal semantics literature,
and while there are some hurdles to implementing this technically, at
the level of current understanding, the project is by no means a lost cause,
and the existing arguments against this decomposition of the superlative are
far from conclusive.
150
3.6 Chapter summary
Despite a few lacunae, I take it that there is overwhelming evidence in favour
of a nested structure (or at least a containment structure) whereby the superlative
properly contains the comparative. Although it is not morphologically
transparent in all languages, it is always there. A corollary of this
conclusion is that UG lacks a “superlative” morpheme, in the standard understanding
of that term, namely, a single morpheme that attaches to an
adjectival stem A, yielding a form meaning ‘more A than all others’. To the
extent there appear to be affixes of this sort, such as English -est, the surface
form masks an underlying grammatical complexity. UG only permits of a
structure in which a comparative morpheme is properly contained within the
representation of superlative adjectives. Extending this to periphrastic constructions,
and adopting a relatively uncontentious view of locality, yielded
two further predicted generalizations, namely the Root Suppletion Generazlization
and the Synthetic Superlative Generalization, both of which appear
to be robustly supported cross-linguistically.
151
152
Chapter 4
CSG: The Data
In the preceding chapters, I have presented some representative examples of
patterns of comparative and superlative suppletion, and sketched a theory
that will account for the CSG, arguing that the account lies in properties of
(universal) grammar, rather than accidents of historical change. With the
theoretical results in mind, we may now examine in detail the empirical basis
for the claimed generalizations, and discuss apparent counter-examples.
Largely for expository reasons, I will break this presentation into three parts,
looking first at qualitative adjectives (where there is but a single problematic
case), then adverbs, and finally, quantifiers (where three apparent problems
lurk in paradigms for many – more – most). I have treated quantity-denoting
modifiers (many, much, few, etc.) separately from property-denoting modifiers,
since it is far from clear that the former are adjectives in all the languages
considered. See section 1.3 for a discussion of the method employed
153
in assembling the data.
4.1 Adjectives
The first range of data to examine is comparative suppletion in qualitative
adjectives. Table 4.1 presents a listing of suppletive cognate triples that is
comprehensive as regards the descriptive material investigated for this book.
Of the more than 300 languages considered, some 70 or so show comparative
suppletion. From among these languages a total of 73 cognate triples are
attested, listed here, grouped by (rough) gloss. For each triple, one example
is given; the fourth column states the language the example is from, then lists
other languages in which cognate triples are found. As discussed in sections
1.3 and 2.3.2, two triples count as distinct if they have non-cognate roots in
at least one grade, even if there is overlap in other grades. Thus a cognate
triple contributes only one data point even if it has reflexes in dozens of
languages, (as in the case of the Germanic cognates to good – better – best).
On the other hand, a single language may contribute multiple, overlapping
triples in the case of suppletive doublets (as in the Georgian comparatives of
‘good’).
Examples of suppletion from languages lacking a morphological superlative,
and thus having only positive and comparative grades, are not included.
For example, Italian cattivo – peggiore and Catalan dolent – pitjor (both
pairs meaning: ‘bad – worse’) involve positive roots that are distinct from
154
the Latin root for ‘bad’ in the table. Similarly, beyond the language families
represented in the table, Abaza and Abkhaz have suppletion for the comparative
of ‘good’. But since these languages lack a synthetic superlative, they
do not bear on the status of the CSG and are thus not included here.
Table 4.1: Suppletive adjectival triples
adj cmpr sprl Language; Cognates
good
good bett-er be-st English; oth. Germanic
bra bätt-re bä-st Swedish; Norwegian
gôd sêl-ra sêl-ost Gothic
guot ăeg-ur ăeg-ur-ste Giazza Cimbrian
dobr-ý lep-ší nej-lep-ší Czech; Polish
dobar bol-ji naj-bol-ji Serbo-Croatian; Slovenian
dobr-y redl-iši Sorbian1
dobr-yj krashch-yj naj-krashch-yj Ukrainian
harn-yj krashch-yj naj-krashch-yj Ukrainian
xoroš-ij luč-še nai-luč-šij Russian
bon-us mel-ior opt-imus Latin 2
da gwell gor-au Welsh
maith ferr dech Old Irish 3
mat gwell-(oc’h) gwell-añ Breton
155
Table 4.1: Suppletive adjectival triples
adj cmpr sprl Language; Cognates
agath-ós ameín-¯on (phér-ist-os) Anc. Greek4
agath-ós areí-¯on ár-ist-os Anc. Greek
agath-ós belt-í¯on bélt-ist-os Anc. Greek
agath-ós kreíss-¯on krát-ist-os Anc. Greek
agath-ós ló-ïon lõ-ist-os Anc. Greek
pra-śásya-s śré-y¯an śré-s.t.has Sanskrit
pra-śásya-s jy¯a-y¯an jy¯e-s.t.has Sanskrit
śśära- has-tara has-tama Khotanese
x¯ob weh/wah-¯ıy pahl-om/p¯aš-om Mid. Persian
xub beh-tær beh-tær-in Persian
on hobe hobe-ren Basque
on hobe on-en Basque
hyvä pare-mpi parha-i-n Finnish; other Fennic
š’ig’ pEr’a-mp pEr’-mus Kildin Saami
k’argi-i u-mˇob-es-i sa-u-mˇob-es-o Georgian 5
k’argi-i u-k’et-es-i sa-u-k’et-es-o Georgian
ezär xo-č-a, xo-č-¯el ma-č-¯ene Svan 6
osda dajehla wi-dajehl-˜2Pi Cherokee
156
Table 4.1: Suppletive adjectival triples
adj cmpr sprl Language; Cognates
bad
bad worse wors-t English
ubil wirsi-ro wirsi-sto OH German; O English
vándr ver-re ver-str O. Icelandic; other Scandinavian
íllr ver-re ver-str O. Icelandic; other Scandinavian
slem vær-re vær-st Danish
ringur ver-ri ver-st-ur Faroese
dålig vär-re vär-st Swedish; Norwegian
dålig säm-re säm-st Swedish
šlext erg-er erg-st Yiddish
zł-y gor-szy naj-gor-szy Polish; other Slavic
špatn-ý hor-ší nej-hor-ší Czech
pohan-yj hir-šyj naj-hir-šyj Ukranian
drènn-y hor-šy naj-hor-šy Belorussian
blah-i hor-šy naj-hor-šy Belorussian
loš gor-i naj-gor-i Serbo-Croatian
plox-oj xuž-e nai-xud-š-ij Russian
malus p¯ej-or pe-ssimus Latin
olc messa messa-m Old Irish
drwg gwaeth gwaeth-af Welsh
157
Table 4.1: Suppletive adjectival triples
adj cmpr sprl Language; Cognates
fall gwash-(oc’h) gwash-añ Breton
kak-ós cheír-¯on cheír-ist-os Anc. Greek 7
kak-ós héss¯on hék-ist-os Anc. Greek 8
big (great)
mikil-s mai-za mai-sts Gothic; Old Germanic, Icelandic
velk-ý vět-ší nej-vět-ší Czech; other Slavic
velyk-yj bil’-šyj naj-bil’-šyj Ukranian
duż-y wie˛k-szy naj-wie˛k-szy Polish
mawr mwy mwy-af Welsh9
wazurg meh/mah-¯ıy mah-ist Middle Persian10
dz<G@d xo-š-a, xo-š-¯el ma-š-¯ene Svan
small
leitils minn-iza minn-ists Gothic; OH Ger, Scandinavian
lýtel læssa læ-st O. English
små mind-re mind-st Danish; Norwegian 11
mali, malen manji naj-manji Serbo-Croatian; oth. Slavic
parvus; paulum min-or min-imus Latin
bach llai llei-af Welsh; Old Irish
158
Table 4.1: Suppletive adjectival triples
adj cmpr sprl Language; Cognates
mikr-ós meí-¯on me˜i-st-os Anc. Greek
mikr-ós eláss-¯on elach-ist-os Anc. Greek
k¯odag keh kas-išts12
Middle Persian
old
gamall ell-re ell-ztr Old Icelandic; Scandinavian
near
agos nes nes-af Welsh; Old Irish
strong
cryf trech trech-af Welsh
As is readily discerned, the overwhelming majority of the triples in Table
4.1 constitute ABB patterns, along with a handful of examples of ABC
triples. There are no AAB patterns, and only one problematic (ABA) case,
a doublet in some varieties of Basque, to which we return below.
Though it plays no role in the account, it may be noted that the adjectives
that show suppletion are drawn from a very limited set, with roots meaning
‘good’ and ‘bad’ together comprising 50 or so of the 73 cognate triples. There
159
are a few cases of suppletion in comparison for ‘big’ and ‘small’, and isolated
examples of suppletion for ‘old’ (Scandinavian), ‘near’ and ‘strong’ (the latter
two in Celtic). This limited distribution is presumably unsurprising, as (true)
suppletion is broadly limited to high-frequency, semantically basic vocabulary
(but see Corbett 2007 for qualifications).
In some cases, the relationship between a positive adjective and the comparative
(resp. superlative) degrees can be one:many, many:one or many:many,
often (but not always) with a regular paradigm alongside a suppletive one.
Some examples are given in (139):
(139) pos cmpr sprl
a. Swedish: god bätt-re bä-st ‘good’
(also): god-are god-ast
b. French: mauvais pire le pire ‘bad’
(also): plus mauvais le plus mauvais
c. Georgian: k’argi-i u-k’et-es-i sa-u-k’et-es-o ‘good’
(also): u-mˇob-es-i sa-u-mˇob-es-o
In some instances at least, the doublets reflect a nuance in meaning.
Holmes and Hinchcliffe (2003, 98) report that the Swedish comparative and
superlative on the stem god is used for ‘pleasant-tasting’, while the suppletive
forms are otherwise the usual ones. Dietiker (1983, 104) notes that the French
doublet in (139b) distinguishes comparison of abstract (pire) from concrete
(plus mauvais) situations. Colloquially at least, English shows a doublet for
bad, with suppletive forms worse – worst for the regular sense (antonym of
160
good), but regular forms badder, baddest in the sense of bad in which it is in
fact a term of approbation: cool, hip, awe-inspiring (senses 12 and 13 in the
Oxford English Dictionary), thus:
(140) a. ... a chance to check out our badder alloy wheels . . .
b. I’m the baddest sonofabitch that ever moved.
The Swedish ‘bad’ paradigm shows a many:many interaction, with two
basic adjectives dålig, ond and an adverb illa corresponding to two suppletive
comparatives sämre, värre and superlatives sämst, värst. Holmes and
Hinchcliffe (2003, 98-99) describe the choice among suppletive forms as denoting,
respectively, ‘more of a bad property’ and ‘less of a good property’.
In addition, they note a regular colloquial pair dålig-are – dålig-ast for the
sense of ‘in bad (worse, worst) health’.
A particularly striking case of multiple comparatives for a single adjective
is provided by the Ancient Greek forms for ‘good’, where five or more distinct
suppletive comparatives or superlatives corresponded to a single positive stem
(see (141) from Seiler 1950, Chantraine 1967).
(141) pos cmpr sprl
Anc. Greek: agathós ameín¯on ‘good’
(also): beltí¯on béltistos
(also): kreítt¯on krátistos
(also): lóïon lõistos
(also): phéristos
161
This situation leaves a degree of uncertainty in quantifying the data
(counting the number of suppletive triples), in particular as there are gaps —
there is no attested superlative corresponding to ameín¯on and no comparative
from root phér-. Moreover, sources differ to some extent on how many of
these are treated as comparatives of agathós. On the other hand, whatever
the answer given to these questions, it is clear that there is no ABA pattern
lurking here, and they may be safely put aside.
Another source of uncertainty connected to doublets arises from examples
where patterns of grammaticality and acceptability might diverge. Thus, for
English, the adjectives small and little (in the same sense, rather than the
sense of ‘few’) appear to be grammatically regular, as in (142).
(142) pos cmpr sprl
a. English: small small-er small-est
b. English: little little-r little-st
However, some sources suggest that the forms littler, littlest are not used.
The CGEL notes “[t]he adjective little has littler as its comparative form,
but this and superlative littlest are rarely used, the corresponding forms
of small generally being preferred.” (Huddleston and Pullum 2001, 1124,
n.15). Even more extreme is the view in the (linguistically conservative)
OED, which states: “[little] has no recognized mode of comparison. The
difficulty is commonly evaded by resort to a synonym (as smaller, smallest);
some writers have ventured to employ the unrecognized forms littler, littlest,
which are otherwise confined to dialect or imitations of childish or illiterate
162
speech.” Whether to count an ABB pattern little – smaller – smallest, at
least for some speakers, thus seems to be a tricky question, but luckily one
on which nothing in particular hangs.
A final remark in this vein concerns the existence of apparent suppletive
doublets in some primary sources. For example, Turkish has two adjectives
meaning ‘bad’: fena and kötü, both of which enter regular (periphrastic)
comparative constructions: (daha) fena ‘worse’ (lit. ‘more bad’). EnglishTurkish
dictionaries also offer beter, a borrowing from Persian, with the
meaning ‘worse’, suggesting suppletion. Yet it is far from clear that beter
is the comparative of either of the adjectives meaning ‘bad’; beter itself can
enter into comparison: daha beter ‘more worse/bad’, and in constructions of
explicit comparison (X is bad, but Y is worse), it appears that an overt comparative
form (not just beter) is needed. I suspect that Turkish beter is like
English optimal — a borrowed form with a meaning that is very similar to a
superlative, but which is not a grammatical superlative. Thus, although optimal
may be glossed “best” in dictionaries, it is subject to further comparison
(more optimal receives 566,000 Google hits, August 10, 2006) and does not
participate in grammatical constructions that characterize superlatives: best,
biggest of all versus *optimal of all. Similarly in Italian, adjectives such as
superiore ‘superior, higher’, maggiore ‘bigger’, are transparent descendants
of Latin comparatives (suffix -ior) but are now simple adjectives, and fail
to participate in comparative constructions (*maggiore di ‘bigger than’, A.
Calabrese, personal communication 2006). One also finds sporadically words
163
glossed as ‘better’ (and sometimes ‘best’) in languages lacking standardcomparative
constructions, such as Nandi kýkáy (Creider and Creider 1989,
63), Swahili bora, heri (Ashton 1947, 202); though like English optimal and
Turkish beter, these enter into comparative constructions as plain adjectives,
and do not appear to constitute grammatically suppletive comparatives as
opposed to lexical forms with comparative-like (or elative) meaning.13
There
is clearly a murky boundary here, though, between such cases and true suppletion.
Although I put these types of examples aside, it should be noted that
none of them would have contributed an ABA pattern, and their exclusion
is thus not a means of sweeping aside inconvenient problems.
As a rule of thumb, in counting the data where the relationship is not
one:one (positive:comparative) I have listed distinct triples separately, even
when they occur in doublets. Thus Georgian in (139c) contributes two ABB
pairs, and Swedish only one — the pan-Germanic GOOD – BETT – BETT
(since the non-suppletive AAA member of the doublet is not counted, of
course). For the unique case of the Greek data in (141) I have counted three
ABB triples, and combined (perhaps unwisely) the two defective triples into
a single ABC count.14
Furthermore as discussed above, I have also counted
distinct patterns where there is a doublet in the positive that resolves to
a single triple for comparative and superlative (as in (53)). This counting
strategy gives six ABB triples for Scandinavian ‘bad’, with Swedish, as it
happens, having all six.
164
4.1.1 The Basque problem
As noted, there is one doublet among the adjectives in the dataset that
appears problematic for the CSG. This is the Basque form for good, given
in (143), see Trask (1997, 100), Trask (2003, 140) (the -(r)en suffix is the
regular superlative suffix). De Rijk (2008, 721) attributes the regular on-en
form (and related adverbial forms) to southern usage, with the triple in (143a)
characterizing northern usage or elevated style.
(143) a. Basque: on hobe hobe-(r)en
b. on-en
In any event, the Basque pattern presents an apparent ABA pattern,
on its own in southern usage, or as part of a doublet — a lone adjectival
counter-example to the CSG.15
Although I can not definitively dismiss the
example, and leave it noted as a potentially important problem, it is perhaps
worth drawing attention here to a curious property of the Basque superlative
that may provide a clue to the analysis. Trask (1997, 210), attributing the
original proposal to Wilhelm von Humboldt, notes that the superlative “is
widely thought to represent a specialization of genitive -en”. According to
the literature, the two are systematically homophonous (see for example,
De Rijk 2008, 720). More to the point, the ‘genitive’ is somewhat of a misnomer,
implying, as it does that the suffix is a case-marker. Rather, the
genitive suffix -(r)en attaches to nominals (with internal syntactic structure,
according to de Rijk), allowing them to serve in adnominal modifier functions,
165
including, but not limited to, possession. Yet while the function is in this
way similar to genitives in other languages, the morphological behaviour is
derivational (De Rijk 2008, 45, citing also Mitxelena 1988); for example the
genitive may be followed by appropriate case suffixes, and indeed, the genitive
itself may iterate in appropriate contexts (with concomitant iteration of the
definite article in both instances, as required). Examples of both follow:
(144) a. gizon-a-ren-gatik
man-def-gen-because
‘because of the man.’
b. katu-a-ren-a-rekin
cat-def-gen-def-with
‘with the one of the cat’
c. alkate-a-ren
mayor-def-gen
alaba-ren
daughter-gen
dirua
money
eta
and
apaiz-a-ren-a-ren-a
priest-def-gen-def-gen-def
‘the money of the daughter of the mayor and (that) of the one of
the priest’
On this approach, an apparent superlative expression as in (145) would
have to derive its meaning from a structure more loosely paraphrased as
something like ‘among the tales in the world, the good one’, or more cumbersomely
‘... the one of the good (ones).’
166
(145) mundu-ko
world-adnom
ipuin-ik
story-part
on-en-a
good-gen-def
‘the best tale in the world’ (cited in De Rijk 2008, 721)
A plausible route from genitive to superlative can be envisaged on comparative
grounds. Cross-linguistically, a common means of expressing a superlative
meaning without superlative morphology is to use the positive or
comparative form of the adjective, with the adjective root also occupying the
position of the standard. Examples from three unrelated languages are given
here (note that the literal paraphrases are not ‘the best of the best’, as in
English, but rather more like ‘bigger than big’):
(146) a. eki-den
good-abl
eki
good
‘(the) best’ (Tyvan, Anderson and Harrison 1999, 33)
b. r1ti
good.abl
khO
more
t1rut
good
kapur
cloth
‘the best cloth’ (Kashmiri, Wali and Koul 1997, 137)
c. te
ˇ
di
ˇ
-le´š
white-abl
no
no
te
ˇ
di
ˇ
white
‘whitest’ (Udmurt, Winkler 2001, 42)
167
In some languages, including Basque, this type of expression serves an
intensifying, rather than strictly superlativizing function:
(147) a. uyúm-˙cum
big-cmpr
uyúm
big
‘very big’ (Burushaski, Berger 1999, 76)
b. ilun
obscure
baino
even
ilun-ago
obscure-cmpr
...
...
‘extremely obscure’ (Basque, De Rijk 2008, 718)
In some languages with superlative constructions of the sort in (146),
the ‘standard’-marking adjective may take the genitive or possessive form.
In Tyvan, for example, alongside (146a) with the ablative, the adjective is
marked genitive when the head adjective is substantivized:
(148) eki-niN
good-gen
eki-zi
good-nml
‘(the) best of all’ (Tyvan, Anderson and Harrison 1999, 33)
Examples such as (148) provide a potential source for a genitive or possessive
marked adjective to become conventionalized as a superlative. Ellipsis
of the head (deadjectival) noun in (148) would yield something that on the
surface is identical to the problematic case from Basque. As there is no
168
comparative morpheme inside the genitive marked adjective, there is no expectation
of suppletion in such a form. It is thus conceivable, though far
from established, that Basque in fact lacks a true (grammatical) superlative,
using an elliptical construction with the genitive instead. Since both positive
and comparative adjectives may be marked genitive, the attested doublet in
(143) is perhaps therefore unsurprising. Having noted the possibility of an
analysis capitalizing on the superlative-genitive homophony, I leave Basque
for now as an unresolved challenge.
Another challenging case that should be mentioned is Old Irish.16
Thurneysen
(1909, 226) gives the following suppletive paradigms (with some further
variations in the forms for ‘smaller, smallest’, not relevant here):
(149) pos cmpr sprl
a. maith, dag- ferr dech, deg ‘good’
b. olc, droch- mess-a * ‘bad’
c. bec(c) lug-u lug-am ‘small’
d. acus, ocus ness-a ness-am ‘near’
The last three of these are unproblematic, but (149a) shows an apparent
alternation in the positive form between maith, which enters into an ABC
pattern, and dag-, which, adhering to the manner of doublet-counting above,
would seem to constitute an ABA pattern if it is the same root as dech.
But there may in fact be no doublet here at all, and hence no problem. In
Old Irish, according to Thurneysen’s description, adjectives entered into one
of two morphosyntactic frames. In one frame, Thurneysen’s ‘inflected ad-
169
jectives’, the adjective could be used predicatively or attributively (in the
latter case following the head noun) and agreed in number, gender, and case
with the noun it modified. The inflected forms of ‘good’ and ‘bad’ were
maith and olc, respectively. In contrast to the inflected adjectives were ‘uninflected’
forms. These latter stood before the noun they modify, forming
a compound-like morpho-phonological unit with it, as diagnosed by stress
and consonant lenition. Not all inflected adjectives had a corresponding uninflected
form, and among those that did, ‘good’ and ‘bad’ had the special
property of using distinct roots for the uninflected construction (dag- and
droch-, respectively). What is important for present purposes is that comparative
and superlative forms, although they did not take morphological
inflection, had the syntactic distribution of ‘inflected’ forms, occurring primarily
in predicate position (Thurneysen 1909, 223). In other words, there
are really two (morpho-)syntactic environments to consider: the compoundinternal
environment, in which there is no gradation and hence no paradigm,
and the compound-external environment, represented by predicative forms.
While the majority of adjectives used the same root in all contexts, the root
meaning ‘good’ occurs as dag- in the compound construction, whereas in the
syntactic construction that allows for the expression of comparison, the only
possibility was maith – ferr – dech. The appearance of a problematic triple in
the doublet in (149) is thus an artefact of the presentation, in particular, the
conflation of two distinct morpho-syntactic constructions or environments.
170
In sum, of the 73 distinct adjectival cognate triples in 72 languages with
comparative suppletion consulted for this project, only a single qualitative
adjective paradigm stands out as potentially problematic, namely the doublet
superlatives in the triple meaning ‘good – better – best’ in Basque, and there
is at least some reason to suspect that this may fall to an alternative analysis.
4.2 Adverbs
In many languages, there are close morphological connections between adjectives
an adverbs, and indeed, suppletive gradation is encountered among
adverbs as well. Some examples are given in (150).
(150) pos cmpr sprl
a. English adj: bad worse wor-st
adv: bad-ly worse wor-st
b. Czech adj: dobř-ý lep-ší nej-lep-ší ‘good’
adv: dobř-e lép-e nej-lép-e ‘well’
c. French adj: bon meilleur le meilleur ‘good’
adv: bien mieux le mieux ‘well’
d. Karelian adj: hyvä pare-mpi ‘good’
adv: hyv-in pare-mm-in ‘well’
e. Georgian adj cud-i u-ar-es-i ‘bad’
adv cud-ad u-ar-es-ad ‘badly’
171
In all of these examples (and in the majority of examples of suppletive
adverbial comparatives), the adjective and the adverb share a root and the
suppletive patterns are also shared, as in English bad-ly – worse – worst.
Such examples thus add no distinct data points, if we count distinct cognate
root triples.
There are however a small number of examples where the roots in the
adverbial triple are not (fully) cognate with the adjectival roots. One such
example is English well (and cognate Icelandic vel), the suppletive adverbial
form of good; some others are given in (151):17
(151) pos cmpr sprl
a. English well bett-er be-st
b. German gern(e) lieb-er am lieb-st-en ‘willingly, rather’
c. Icelandic gjarna held-ur hel-zt ‘willingly, rather’
d. German bald ehe-r am ehe-st-en ‘soon’
I will argue here that at least some of these, in particular English (151a),
should not count as true ABB patterns in the sense relative to the CSG,
though others, such as those in (151b-d) should. The discussion is somewhat
of an aside, in the sense that the patterns under investigation are superficially
consistent with the CSG, so nothing of consequence hinges on whether
they are to be included in or excluded from the range of relevant data. Nevertheless,
discussion of this case will illustrate a more general point about
containment, and in particular about alternate derivational routes that converge
on a similar endpoint. This consideration will play a role in other
172
sections as well.
The question at issue here is whether the containment relation in (36a)
(repeated in (152), with a-root intended to cover both adjectives and adverbs)
holds of adverbs, as well as adjectives.
(152) [ [ a-root ] comparative ]
There is reason to be suspicious of this. In some languages, the overt
morphology seems to indicate that the comparative adverb is derived not
from the positive adverb, but rather from the comparative adjective. That
is, although a comparative adverb like better appears to function as the
comparative of well, its actual derivation may be more circuitous, at least
in some languages. Karelian illustrates this pattern. The suffix -h forms
adverbs from adjectives, as illustrated by the pair rutto, rutto-h ‘quick, quickly’.
In the corresponding comparatives, the adverb-forming morphology is
clearly peripheral to the comparative, thus rutto-mpa- ‘quicker’ (adjective,
oblique stem), rutto-mpa-h ‘quicker’ (adverb). Assuming affix order reflects
derivational history, this implies that the adverbial comparative is derived
from the adjectival comparative, rather than being the comparative of the
adverb as such. This is indicated schematically in (153):
(153) pos cmpr
adj rutto → rutto-mpa↓
↓
adv rutto-h rutto-mpa-h
173
Georgian, with a circumfixal comparative, appears to show the same pattern
as Karelian, though the evidence is perhaps more slender. The adverbial
case suffix -ad derives adverbs from adjectives (and nouns), as in
xšir-i ‘frequent-nom’, xšir-ad ‘frequently’ (Hewitt 1995, also http://www.
translate.ge). As in Karelian, the adverbial case suffix stands outside the
comparative (at least for those adjectives that take a morphological comparative
alongside the periphrastic construction).18
In these languages, where
an adjective takes a suppletive comparative, that suppletion is ‘inherited’ by
the adverb (cf. (150)):19
(154) pos cmpr
a. Karelian adj: hyvä pare-mpi ‘good’
adv: hyv-in pare-mm-in ‘well’
b. Georgian adj k’argi-i u-mˇob-es-i ‘good’
adv k’argi-ad u-mˇob-es-ad ‘well’
c. Georgian adj cud-i u-ar-es-i ‘bad’
adv cud-ad u-ar-es-ad ‘badly’
The proposed derivational ‘path’ to the comparative adverb, supported
by the regular morphology, explains why this is so. The forms that serve the
function of the comparative of the adverb are derived from the comparative
of the adjective, and thus contain that form’s idiosyncratic properties.
Yet if this same derivational path is employed by English (and Icelandic),
then the triple well – better – best is not a genuine triple after all. Although it
appears to be an ABB pattern (consistent with the CSG), in fact there may
174
be no direct relationship between well and better. In other words, assume
that the structure of English adverbial comparatives is as in Karelian and
Georgian, namely:
(155) [ [ [ good ] cmpr ] adv ]
The vocabulary fragment with the exponents of English good may be as
in (156).
(156) a. good → well / ] adv ]
b. good → be(tt)- / ] cmpr ]
c. good → goodNote
that (156a) and (156b) are not ordered by the Elsewhere Condition.
However, there are a number of reasons why (156b) should bleed (156a) in
(155), explaining why the pattern is well – better rather than *well – weller.
In Chapter 5, I explore the possibility that the contexts for root allomorphy
can only make reference to adjacent morphemes (cf. Embick 2010), which
would force this ordering (though note that there are various empirical problems
with a strong adjacency condition, including one mentioned just below).
Even without a strong adjacency condition, an intervention condition of the
familiar type would also be applicable here: the comparative is closer to
the root in (155) than the adverbial marker, and thus (156b) takes priority
over (156a). Either way, considerations of structural locality, rather than the
Elsewhere Condition per se, will serve to determine the correct application
of the rules in cases of this sort.
175
It should be noted that the derivation in (153) does not appear to be
universal. In Standard Basque, comparative -ago attaches outside adverbforming
-ki, with adverbial comparatives apparently derived from adverbs,
rather than from adjectival comparatives, a different derivational history
from that in Karelian and Georgian.
(157) pos cmpr
adj berri → berri-ago ‘new, fresh; adv = recently’
↓
adv berri-ki → berri-ki-ago
Thus the morpheme order evidence (from this small sample)20
points to
the structural variation in (158):21
(158) a. Karelian, Georgian, English? [ [ [ adjective ] comparative ] adverb ]
b. Basque [ [ [ adjective ] adverb ] comparative ]
Because the morpheme order in Basque is the reverse of what is seen
in Karelian and Georgian, the logic that excluded *well-er in (156) (given
(158a)), will make the opposite prediction for Basque: the suppletive root
in the adverbial context should be preserved in the adverbial comparative.
Strikingly, this is correct. Standard Basque has one suppletive adverbial:
the adverb corresponding to adjective txar ‘bad’ is gaiz-ki, rather than txarki
(de Rijk 2008, 236).22
And unlike Karelian, Georgian and English, the
adverbial suppletion is preserved in the adverbial comparative, a difference
we may attribute to the different derivational histories in (158):
176
(159) pos cmpr
adj txar → txar-ago ‘bad’
↓
adv gaiz-ki → gaiz-ki-ago
Before drawing too strong a conclusion from this one example, though, I
note that here, as in the previous section, the Basque root for ‘good’ shows
a puzzling behaviour. In Basque, the adverbs from ‘good’ are regular, and
not suppletive: on ‘good’, on-do, on-gi ‘well’ (the suffixes are underlying -to,
ki; the voicing in ongi is irregular, and the suffix -to is unproductive). The
comparative of the adjective is suppletive hobe ‘better’. For this adjective,
it is the comparative allomorph and not the regular root that appears in the
adverbial comparatives: hobe-to, hobe-ki ‘better (adv.)’. There is, however,
a further difference between the forms for ‘better’ and the other adverbial
comparatives, including gaiz-ki-ago, namely that in addition to the root alternation,
the comparative suffix -ago is missing from the various ‘better’
forms. In investigating portmanteau morphology in Chapter 5, I will suggest
a reason why this might be significant.
Taking stock, the important point that we have seen in this aside is that
there may be more than one derivational path available to a given surface result.
In the case at hand, a form that would appear to be the comparative of a
(deadjectival) adverb, may be derived either as the comparative of the adverb
(as in Basque), or as the adverbialized comparative (as in Karelian). This
subtle difference (for which there may be no clues in the overt morphology,
177
as in English) nevertheless has important consequences for the interaction
of suppletion and comparative formation. This general logic, developed here
for cases such as English well which are innocuous for the purposes of the
CSG, will turn out to have more important applications later in this work.
Returning to the CSG, while well – better – best may end up being irrelevant
for the CSG, there are a handful of additional suppletive paradigms
which are not extensions of corresponding adjectives, and these may indeed
contribute relevant triples. Two such patterns are from German, as characterized
in standard reference works (e.g., WDS), namely gern(e) – lieber –
am liebsten ‘gladly – rather’ and bald – eher – (am) ehesten ‘soon – sooner
– soonest’.23
The former (with correspondents in other Germanic languages,
thus West Frisian: graach/jeerne – leaver – leafst) is a recent innovation according
to Seiler (1950, p.32, n.2), who gives gern – gerner – am gernsten into
the 18th century. Examples illustrating its use are given in (160) – the (b)
example shows an als ‘than’ phrase, the hallmark syntax of a comparative:
(160) a. Tee
tea
trinke
drink
ich
I
gerne,
gladly,
aber
but
Kaffee
coffee
trinke
drink
ich
I
lieber.
rather
‘I’m happy to drink tea, but I would rather drink coffee’
b. Tee
tea
trinke
drink
ich
I
lieber
rather
als
than
Kaffee.
coffee
‘I’d rather drink tea than coffee.’
178
c. Leo
Leo
geht
goes
gerne
gladly
in
in
die
the
Schule,
school,
aber
but
ins
in.the
Sommercamp
summer.camp
ist
is
er
he
noch
even
lieber
rather
gegangen.
go.partic
‘Leo happily goes to school, but he went to summer camp even
more happily.’
As with the English little paradigm discussed above, it is not clear whether
we are dealing here with a suppletive pattern as opposed to a defective one,
in which gern(e) has no comparative, but for which (inherently) comparative
lieber ‘rather’ (lacking a positive) is a close synonym. English rather is transparently
historically a comparative (of the now obsolete rathe), as evidenced
by its ability to license a than-complement (as in the translation to (160b),
but it is hard to see rather as synchronically the suppletive comparative of
an adverb like gladly. In any event, a decision on whether these cases are
suppletive triples or conflations of defective paradigms does not affect the
theoretical questions in this book (it affects only the quantitative aspects
in the number of relevant cases) and I will not pursue the analysis of these
examples further.
Before closing this section, let us return briefly to one other point raised
earlier, which can perhaps be appreciated more clearly in light of the discussion
of (156). In the presentation thus far, I have (tacitly) rejected the assumption,
common in some semantic approaches (see, e.g.. Kennedy 2007b)
that the positive degree of the adjective contains a null morpheme -pos that
179
is not contained in the comparative, as in (161):
(161) a. positive [ [ a-root ] pos ]
b. comparative [ [ a-root ] comparative ]
If this were indeed the correct representation, nothing would go wrong for
any of the examples discussed so far, since there is no reason to assume anything
but the default exponent in the positive slot of the adjectives. However,
a theory incorporating (161) might have the added power needed to generate
ABA patterns. Consider, specifically, the hypothetical Vocabulary fragment
in (162). This incorrectly admits *good – better – good-est.
(162) a. good → good - / ] cmpr ] sprl
b. good → good- / ] pos
c. good → bett- / ( ] cmpr )
Discussing the basic case of *ABA above, I suggested that learners do
not posit contextual allomorphs that are accidentally homophonous with the
default exponent of a given morpheme (see (44)). That assumption prevented
deriving *good – better – goodest as an ABC pattern, with A,C accidentally
homophonous. But (162) subverts that result, since (162b) is not a default,
but is instead a contextual allomorph restricted to the ] pos environment.
Although there are two homophonous allomorphs in (162), they do not stand
in a subset-superset relation to one another, and thus the Antihomophony
bias in (44) from Chapter 2 does not obviously apply. I continue, therefore,
to reject the postulation of a -pos suffix for adjectives.
180
In sum, the few suppletive patterns in adverbs appear to add nothing new
to our understanding of the CSG. The few patterns that are not extensions
of adjectival patters are consistent with the CSG (despite Ultan’s claim to
the contrary, see note 23), but this appears to be a fact that is derivative of
the patterns of the underlying adjectives. It appears that, in many languages
at least, adverbs, by virtue of their structure, fall outside the scope of the
CSG.
4.3 Quantifiers
The next (and last) empirical domain we turn to, regarding the empirical basis
of the CSG, is quantificational elements, namely, words meaning ‘many,
much’ and ‘few, little’. These enter into comparative (and superlative) constructions,
and participate in suppletive patterns, as shown in English:
(163) pos cmpr sprl
a. many more most
b. much/a lot (of) more most
c. (a) few fewer/less fewest/least
d. (a) little less least
Suppletion patterns for ‘many, much’ are in fact among the most common
source of suppletive paradigms, and there are a few cases of suppletion for
‘few’. A list of 39 cognate triples for quantifiers is given in Table 4.2. Note
that for a number languages, such as Kabardian, only the words for ‘much,
181
many’ are reported to be suppletive. Note also that many languages use
related forms for ‘many’ and ‘big’, or for ‘small’ and ‘few’. Where the roots
in the quantifier triples are fully cognate with those in the adjective table, the
triples are listed only in the first table; however, in keeping with the general
counting scheme used throughout, entires are given in the quantifier table
when the comparative (and superlative) are cognate with ‘bigger’ or ‘smaller’,
but the positive root of the quantifier is not cognate with an adjective.
Table 4.2: Suppletive quantifier triples
adj cmpr sprl Language; Cognates
many, much
many mo-re mo-st English; Scandinavian
much mo-re mo-st English; Scandinavian24
(a) lot mo-re mo-st English25
viel me-hr mei-st German; other Germanic
nógv-ur mei-ri me-st-ur Faroese
party me-er mee-ste Afrikaans
baie me-er mee-ste Afrikaans
sakh me-r me-r-ste Yiddish
mange fle-re fle-st Danish; Scandinavian
mnoho víc-e nej-víc-e Czech; Serbo-Croatian
mnogo bol’š-e naj-bo’lš-ij Russian
182
Table 4.2: Suppletive quantifier triples
adj cmpr sprl Language; Cognates
mnogo po-več-e naj-mnogo
wiele wie˛c-ej naj-wie˛c-ej Polish
puno već-e naj-već-e Serbo-Croatian
baháto bil’-šyj naj-bil’-šyj Ukranian
dúže bil’-šyj naj-bil’-šyj Ukranian
daudz vair-âk vis-vair-âk Latvian
mult- pl¯u-s pl¯ur-imu-s Latin
kalz mui-(oc’h) mui-añ Breton
llawer mwy mwy-af Welsh
pol-ús ple-í¯on ple-˜ıst-os Anc. Greek
šat-erë aveli amena-šat Armenian
xele beš-(tar) beš-tar-in Persian
ziad beš-(tar) beš-tar-in Persian
asko gehi-ago gehi-en26
Basque
monet usea-mmat use-i-mmat Finnish
paljon ene-mmän en-iten Finnish
äijä enä-mpi äij-in Karelian
sok tö-bb leg-tö-bb Hungarian
bevri met’-i u-met’-es-i 27
Georgian
kwad, ba nax nax-deda Kabardian
183
Table 4.2: Suppletive quantifier triples
adj cmpr sprl Language; Cognates
few
(a) little less leas-t English
weinig minder mind-ste Afrikaans; Dutch
(en) bytsje minder minst Frisian
troxi men-še naj-men-šij Ukrainian, Slovenian
paucus min-or min-imus Latin
olíg-os héss-¯on hék-ist-os Anc. Greek
olíg-os meí-¯on me˜ı-st-os Anc. Greek
Suppletive paradigms for quantifiers are also attested in many languages
that lack morphological superlatives; some examples meaning ‘many’ or
‘much’ are given in (164).
(164) Many-more examples:
pos cmpr
a. Manx: ram smoo
b. French: beaucoup plus
c. Spanish: mult-o más
d. Mod. Greek: pol-ís perissó-ter-os
184
e. Ossetian: biræ fil-dær
f. Veps: ei ena-mba
g. Abkhaz: šarda ajha
h. Tatar: küp artıq
i. Hindi: kai zyaadaa28
Note the inclusion of Abkhaz (N. W. Caucasian) and Tatar (Turkic) in
this list, which, along with Kabardian (also N. W. Caucasian) in Table (4.2),
are examples of suppletion outside (although not all that far outside) the
Greater European Sprachbund discussed in section 2.3.2. It is not always
clear from grammatical descriptions, though, whether elements glossed as
‘more’ are really the comparative of ‘many/much’, and thus it is not clear
that all such examples involve a suppletive relationship. For Tatar, for example,
the artıq from is given with the meaning ‘more’ in some sources (including
http://www.suzlek.ru/suzlek/), but other sources give a regular küb-r@k,
(Burganova et al. 1969, 173). Similarly in (unrelated) Lezgian, Haspelmath
(1993, 433) gives an example with artux having the meaning ‘more’ (greater
in quantity), but notes that “[t]his could be considered a suppletive comparative
form of gzaf ‘much’, but it is a unique case.’ Other examples given in
the same chapter illustrate regular comparison with gzaf.
185
4.3.1 Many – more counterexamples
As with the adjectives considered above, where there are superlatives, the
overwhelming majority of the 39 or so suppletive patterns are ABB patterns,
but there are three apparent triples that look like ABA patterns, namely, the
words for ‘many’ in Karelian, Armenian, and Bulgarian (Macedonian shares
a cognate triple with Bulgarian):
(165) pos cmpr sprl
a. Karelian: äijä enä-mpi äij-in
b. Armenian: šat aveli amena-šat
c. Bulgarian: mnogo po-veče naj-mnogo
Starting with Bulgarian-Macedonian, it appears that the source of the
problem lies in the classification of these examples as morphological (synthetic)
comparative and superlative constructions. But the available descriptions
suggest that it is not correct to list them among morphological
comparative forms in the first place, and that (despite orthographic conventions)
they are periphrastic expressions. Superlative naj (and the comparative
po-) are not prefixes, but rather clitics or free-standing adverbs, which
do not form a minimal morphological unit with the adjectival root. Since
the CSG ranges over morphological, but not periphrastic, comparatives (see
Chapter 3), these examples are not counter-examples if they are not synthetic
constructions. Evidence that there is more structure in these was mentioned
already in note 21: the superlative and comparative markers in Bulgarian and
186
Macedonian attach not only to adjectives, but also to phrases of a variety of
sizes. Macedonian examples are given here (Friedman 2002, 22):
(166) a. na
to/in
jug
south
‘to the south’
b. po
cmpr
na
to/in
jug
south
‘more southerly’
(167) a. ne
neg
saka
like
‘dislike’
b. naj-ne-saka
sprl-neg like
‘dislike the most’
Orthographic conventions dictate that the comparative and superlative
markers are sometimes written together with the phrase they modify and
sometimes not. Thus, (167b) is written as a single word, but when an object
clitic occurs between the negation and the verb, then the superlative (as well
as the other clitics) are written as separate words: naj ne go saka ‘sprl neg
3sg.obj like’ = ‘he dislikes him the most’ (Friedman 2002, 22). Authors
also note that the comparative and superlative constitute an independent
prosodic domain, having their own stress, independent of that of the stem
(even when written together) (Alexander 2000, 198). Compare also Lunt
(1952, 52): “Po and naj, although they ... are considered so much an integral
part of [the word that they modify] that they are written as prefixes, still are
really adverbs.”29
As analytic (periphrastic) constructions, the Bulgarian and Macedonian
superlatives pattern with periphrastic superlatives in Tikhvin Karelian and
Russian (and Armenian compounds, considered below), which do not inherit
187
the suppletive root of the comparative (see chapter 3; see also (173c) from
Ludian Karelian):
(168) pos cmpr sprl
a. Bulgarian: mnogo po-veče naj mnogo ‘many’
c. Tv Karel. hüvä pare-mbi {ülen/suamo} hüvä ‘good’
c. Russian plox-oj xuž-e samyj plox-oj ‘bad’
Although understanding the Bulgarian-Macedonian superlative as a periphrastic,
rather than a morphological, construction gets us out of the woods
as far as the CSG is concerned, the considerations just raised do not render
the Bulgarian data un-problematic. Rather, the nature of the problem is
shifted, and it is the suppletion in the comparative that becomes an issue.
Bulgarian now looks like a possible counter-example to the claim that suppletion
is restricted to morphological domains, and that periphrastic constructions
cannot be suppletive. I postpone discussion of this issue to the end
of this section, and offer a tentative solution to the Bulgarian-Macedonian
question there.
Turning now to Armenian, we find that the morphological ‘wordhood’ of
the offending example is once again in question. Although some grammars
gloss amena- as a prefix meaning ‘most’, it appears instead to be a compounding
form, consisting of the universal quantifier amen ‘all’ and the linking
vowel -a- used to form compounds when the second member is consonantinitial
(Dum-Tragut 2009, 671). The compound form alternates (freely, so
far as I can tell with limited investigation) with a non-compounded form,
188
where the universal quantifier serves as the standard of comparison (as in
(169)) in a configuration that constitutes a regular (zero-marked) comparative
in the language. As noted above, this is the most common means of
expressing superlatives cross-linguistically (see section 3.2.3). For ‘most’, the
corresponding form is: amenen šat ‘all-abl-def many.’
(169) a. Artak-@
Artak-def
amen-a-partsrahasag
all-LV-tall
e
be.3sg
‘Artak is the tallest.’
b. Artak-@
Artak-def
amen-e-n
all-abl-def
partsrahasag
tall
e
be.3sg
‘Artak is the tallest.’ (lit: ‘Artak is taller than all.’)
What is at issue here is whether the modifer-head relation in a compounding
structure should count as sufficiently local for X0
-internal triggering
of contextual allomorphy. Evidently, it does not, at least for synthetic
compounds such as (169a) in which the modifier in the compound is the
argument of the adjective, in this case the standard of comparison. At the
relevant level of representation, (169a) has the same structure as (169b), and
it is inaccurate in the present context to consider amena to be a superlative
prefix.
This leaves only the Karelian example in (165a). There is no reason to
suspect a hidden periphrastic structure here, but further digging casts some
doubt on this pattern, and it may reflect the conflation of various dialect
patterns. The ABA pattern in (165a) is given in Zajkov (1999, 51) and
189
reaffirmed by P. M. Zajkov, who states that it is the only pattern (personal
communication, 2008). Against this, the online Karelian-Russian-Finnish
dictionary at http://sanakniigu.onego.ru/ [as of October 2010] has no
listing for the problematic äijin, giving instead enin for the superlative, an
ABB paradigm (a shared root with the comparative, and the superlative
suffix -in), as in (170).
(170) pos cmpr sprl
Karelian: äijy enä-mbi en-in
The pattern in (170), and not that in (165a) jibes with the results of
a brief questionnaire with seven Karelian-speaking respondents in 2009 (all
native speakers, and all but one daily users of the language, some working in
Karelian media).30
The questionnaire was designed to elicit comparative and
superlative constructions, with regular adjectives as well as with the quantifier
‘many’. A part of the questionnaire asked for the Karelian equivalents
of the (Russian translations of the) sentences in (171).
(171) a. Pekka has many dogs.
b. Pavel has more dogs than Pekka.
c. Lauri has the most dogs.
The translations offered for (171a) and (171b) conformed to the suppletive
comparative paradigm expected. Representative examples are given in (172)
(glosses are constructed on the basis of information in Zajkov (1999); there
was some variation not relevant to the point made here.)
190
(172) a. Peka-l
Pekka-adess
on
is
äijy
many
koiru-a.
dog-part
‘Pekka has many dogs.’
b. Puavila-l
Pavel-adess
on
is
enä-mbi
more-cmpr
koiru-a
dog-part
migu
than
Peka-l
Peka–adess
‘Pavel has more dogs than Pekka.’
For the superlative, none of the respondents gave the problematic form in
(165a), nor did any recognize it when asked.31
Four of the seven speakers gave
enin as the superlative of ‘many’, as in (173a). Other possibilities offered
were kaiki-s enä-mpi ‘more than all’ (173b) and a periphrastic superlative,
with intensifier ylen ‘very’ combining with the positive root, as in (173c).
(173) a. Lauri-l
Lauri-adess
on
is
en-in
most-sprl
koiru-a.
dog-part
‘Lauri has the most dogs.’
b. Lauri-l
Lauri-adess
on
is
kaiki-s
all-elat
enä-mbi
more-cmpr
koiru-a.
dog-part
‘Lauri has the most dogs. (lit: ‘more than all’)
c. Lauri-l
Lauri-adess
on
is
ylen
emph
äijä
many
koiru-a.
dog-part
‘Lauri has the most dogs.
The strategy in (173b), expressing the superlative meaning via a comparative
plus a universal quantifier is a strategy available in all Fennic languages,
191
except perhaps some varieties of Finnish, according to Nau (1992), and is indeed
the most common superlative forming strategy cross-linguistically (see
section 3.2.3).32
On the strength of this evidence, I will put aside the forms cited by
Zajkov (1999) as perhaps resulting from an admixture of various dialects,
or some other source of interference, acknowledging, of course, that they
cannot be conclusively dismissed at this time. Thus, while questions remain,
the Karelian, Armenian, and Bulgarian-Macedonian patterns do not appear,
at this time, to constitute insurmountable problems for maintaining that the
CSG is indeed an absolute linguistic universal as far as affixal superlatives
are concerned.
4.3.2 Coda: Bulgarian, Macedonian and the RSG
Before moving on, let us return briefly to Bulgarian and Macedonian words
for ‘many’, addressed above as a prima facie problem for the CSG. The
examples are given here.
(174) pos cmpr sprl
a. Bulgarian: mnogo po-veče naj-mnogo
b. Macedonian: mnogu po-ve´ke naj-mnogu
Above, I noted that, despite the orthographic practice of writing po- and
naj- as prefixes, they are in fact free-standing particles, with their own stress
and able to modify full phrases. This resolves the problem as far as the CSG
192
is concerned — in the superlatives, there is a phrasal boundary between the
superlative marker and the root, and thus suppletion is not conditioned. But
the solution to the CSG now shifts the problem to the RSG. If po-, like najis
separated from the root by a phrasal boundary, then it too should fail
to trigger root suppletion, yet there is clearly suppletion in the comparative
forms in (174).
I tentatively suggest that despite appearances it is not the element pothat
triggers the suppletion in the comparative root, but rather a (synchronically
null) comparative affix. This suggestion recapitulates the history of the
aberrant nature of Bulgarian and Macedonian comparatives within Slavic. In
all other Slavic languages, including Bulgarian up to about the 14th Century
(Reiter 1979, 21), the comparative is formed by means of a suffix, typically
-ji- or -ši- or derivatives thereof. The prefix po- occurs across Slavic with a
weakening or reinforcing function, similar to English rather or somewhat, as,
for example, Russian: po-molož-e ‘(somewhat) younger’ < molož-e ‘younger’,
comparative of molod-oj ‘young’. Across Slavic, po- may occur with positive
and comparative forms (though not both in all languages; see Reiter 1979,
22-25), and Reiter notes that both are attested in Bulgarian in the Middle
Ages:
(175) a. po-chraber
po-brave
ot
from
tebe
you.sg
‘braver than you.’
193
b. koja
which
est
is
ot
from
vas
you.pl
. . . po-lěp-š-aa
po-beautiful-cmpr-infl
‘which is more beautiful than you.’
Alone in Bulgarian-Macedonian was the original comparative suffix lost,
and concomitantly, in a manner reminiscent of the Jespersen cycle of negation,
the erstwhile optional reinforcer po- became effectively obligatory as the
sole overt marker of comparison. The analysis of Bulgraian-Macedonian here
treats this as the loss of the overt exponent of the comparative, with a zero
affix retained abstractly. Note that zero-marked comparatives are possible
at least in some varieties of Bulgarian; Reiter cites the following from the
Šumen dialect:
(176) toj
he
e
is
chubav
handsome
ot
from
mene
me
‘He is handsomer than me.’ (Reiter 1979, 23)
Positing a null comparative suffix, with overt po- merely in a reinforcing
function (as po- has in all other Slavic languages) thus permits the solution
to Bulgarian-Macedonian offered in the previous section to be maintained,
without compromising the RSG. It is the null affix that governs suppletion
in the comparative; the elements po- and naj- are phrasal, and too remote
(structurally) from the root to interact with it. As always, extra pleading
is needed in the postulation of invisible elements, yet here the historical
justification, if not the synchronic one, seems well established.
194
4.3.3 Afterthought: What’s more?
A final remark seems worth making concerning the status of the quantifiers,
before bringing this part of the discussion to a conclusion. Patterns like
many – more – most versus many – more – *maniest are relevant to the
account of the CSG only to the extent that the comparatives do in fact
contain the basic quantifier roots. If the element meaning ‘more’ in some
language is an underived (i.e., monomorphemic) item, with a comparative
meaning, not derived from — and thus not containing — a quantifier like
‘many’ or ‘much’ in its representation, then the CSG would be irrelevant
for this language. There would be no morphological relationship between
‘many’ and ‘more’. If UG allowed this, then the paradigms in (164) could be
false paradigms, connected by meaning only. One could analyze French, for
example, as having an incomparable adverb beaucoup ‘many/much’ and an
inherently comparative adverb plus ‘more’, with no derivational relationship
(not even a suppletive one) holding between the two.
We may note off the bat that although an apparently suppletive ‘more’ is
quite common, not all languages are of this sort. In many languages, the word
that translates English more in the sense of ‘greater amount’ is transparently
the comparative of ‘many’ or ‘much’. Some examples are given here:
195
(177) pos cmpr sprl
a. O. Ch. Slavonic: m˘unog-˘u m˘unož-ai
b. Lithuanian daug daug-iau daug-iau-siau
c. Sanskrit bahu bahu-tara bahu-tama
d. Chuvash nummay nummay-taraX
e. Kazakh köp köb-irek
f. Mari šuko šuk-@rlak
g. Khanty ar ar-šăk
h. Guaraní heta heta-ve
i. Klon ubei mi-ubei
j. Evenki kete kete-tmer
Note also that languages which form periphrastic comparatives also often
use the same periphrastic formations to express English ‘more’, in the sense
of the comparative of ‘many’:
(178) pos cmpr
a. Romanian mult mai mult
b. Albanian i shumë më shumë
c. Turkish çok daha çok
d. Mordvin lama sjada lama
The pervasiveness of this pattern is somewhat obscured in dictionaries,
because of the polysemy of English more. Looking up English more in an
English-Turkish or English-Albanian dictionary will yield single words such
as daha, resp. më, but these are limited in their use to the ‘more’ that forms
196
comparatives (as in: more intelligent, more interesting) and cannot be used
to stand on their own as the comparative of ‘many’.
It seems abundantly clear, then, that UG allows for a regular derivation of
comparative ‘more’ from (the root of) ‘many’ or ‘much’. For languages that
have such a derivation underlyingly, suppletion may arise in the comparative
(and superlative, if applicable) and the CSG should hold. The question that
we are considering here is whether UG permits of an alternative relationship,
one in which at least some apparent cases of suppletion are in fact the conflation
of a non-comparable quantifier and an inherently comparative more,
with no derivation linking the two. I can do little more than scratch the
surface of this here, but I note that the very robustness of the CSG for ‘more
– most’ words in Table 4.2 may constitute an argument against this view.
The fact that the CSG holds of the quantifier domain, with a large number
of cognate sets, and no genuine exceptions (if my accounts of Karelian, Armenian
and Bulgarian are correct) suggests itself as a non-accidental fact of
language. If a monomorphemic counterpart to ‘more’ were indeed available,
there would be no reason for the CSG to hold in this domain.
A more subtle version of this question is whether there may be a relationship
between the quantifier and the comparative, but one that is more
indirect than proper containment. It may be that an admissible derivation is
as schematized in (179), modeled after the adverbial derivations in (153).33
197
(179) root → more → most
↓
quantifier
Although I will not pursue this line of analysis in this work, one reason
for thinking that the relationship between the comparative and the quantifier
may be less direct than with adjectives, and that there may be more (or
different) structure in the quantifier words, is that quantifiers are not always
adjectives, morphosyntactically. For example, in English, the quantificational
expressions a lot of, lots of are nearly equivalent to ‘much, many’, and seem
to enter into comparison with more, most (180a). Yet these expressions on
the face of it have some additional (apparently nominal) structure, and thus
seem to reflect a pattern like (179):
(180) root → mo-re → mo-st
↓
(a) lot(s) of
Similarly in Russian, quantifiers mnogo ‘many, much’, malo ‘little’, (also
skol’ko ‘how many’ and derivatives with ne- ‘not’ have adverbial (or nominal)
morphosyntax in nominative and accusative case (mnog- takes adjectival
morphology in oblique environments, while mal-o is excluded from oblique
environments altogether; Garde 1998, 247). In Bulgarian as well, the form
mnogo (the source of the apparent problem in (165b)), fails to inflect as
an adjective. Tasseva-Kurktchieva (2006) provides an analysis of Bulgarian
quantifiers and notes that they fall into two broad categories. The modifying
198
quantifiers, are distinct from (what she calls) proper quantifiers (including
mnogo) not only in that only the former have adjectival (concord) morphology,
but the groups differ with respect to a number of both morphological and
syntactic diagnostics, including co-occurrence with definite articles, possessive
clitics, and demonstratives, as well as extraction patterns. If Bulgarian
mnogo should therefore be described in terms of the derivation in (179),
it could be modeled as an instance of the abstract pattern in (162) in the
previous section. The root may exist only in the abstract, with no default
realization, while quantifier and superlative are accidentally homophonous in
a manner that is in theory consistent with (44) as neither is the default, and
neither properly contains the other.
If there are indeed quantifiers that have additional (possibly covert) internal
structure, and instantiate the pattern in (179) as regards comparison,
then these quantifiers would lie outside the reach of the theory presented
here, by parity of reasoning to the discussion of well – better – best surrounding
example (156). Apparent ABA patterns would not be problematic, but
it would then be a mystery why the overwhelming majority of patterns in
this domain are ABB patterns, entirely as expected under the CSG.
4.3.4 Chapter summary
Taking the adjectives, adverbs and quantifiers together, we find more than
100 cognate triples, some representing a single case in one language variety,
others reflecting broad and historically stable patterns (such as Germanic
199
good – bett – bett). There remain but a handful of apparently problematic
ABA patterns, and these may thus seem quantitatively negligible. But
the account offered above excludes the ABA pattern categorically, attributing
its absence to aspects of UG. And UG brooks no dissent. The account
will ultimately fail if it cannot be shown that these apparent affixal ABA patterns
are just that — apparent — and may fall to an alternative explanation
(perhaps including accidental homophony, see discussion of (44)). While I
have not conclusively demonstrated that the counter-examples are explained
away, I hope to have shown here directions for potential explanations that
have some measure of plausibility.
200
Chapter 5
Theoretical Refinements
5.1 Introduction: Taking Stock
In the theory developed thus far, two assumptions are of paramount impor-
tance.
The first is an assumption about the architecture of the grammar, namely,
Late Insertion, i.e., that morphology is realizational. The syntax generates a
hierarchical arrangement of abstract morphemes, which are then subject to
rules of exponence, which match (underlying) phonological representations to
terminal nodes, as in (181), with the additional operation of Merger included,
which occurs post-syntactically, but feeds vocabulary insertion/exponence:
201
(181) Syntax
⇓
(Merger)
⇓
Exponence ← suppletion
⇓
Phonological Rules
This theoretical architecture allows for competition among exponents (vocabulary
items), regulated by the Elsewhere Condition, and allowing suppletion
to be treated as a special case of contextual allomorphy. The English
exponents good and bett- (with orthography standing in for more accurate
phonological representations, of course) compete to realize the abstract adjectival
root good, with bett- winning out wherever its context is met.
(182) a. good → bett- / ] cmpr
b. good → good
The second assumption is an empirical one, namely the containment hypothesis,
repeated here.
(59) The Containment Hypothesis
The representation of the superlative properly contains that of the
comparative.
Chapter 2 laid out the reasoning that runs from these assumptions to
(one part of) the CSG, discussed at length in the preceding chapters (we will
202
turn presently to the second part, the CSG2).
(33) The Comparative-Superlative Generalization, part I (CSG1):
If the comparative degree of an adjective is suppletive, then the superlative
is also suppletive (i.e., with respect to the positive).
In Chapter 3, I added the assumption that the theory of morphology includes
a locality restriction on possible contexts for suppletion. As a working
hypothesis, I suggested that locality prevents a node β from serving as the
context for allomorphy of a node α if a maximal projection (or other designated
intervenor, see Embick 2010) intervenes between them, as in (90),
repeated here:
(90) a. α . . . ]X0 . . . β
b. * α . . . ]XP . . . β
The assumption in (90) derives the RSG in (88), and also anchors the
account (together with some ancillary assumptions) of the SSG in (89).
(88) The Root Suppletion Generalization (RSG)
Root suppletion is limited to synthetic (i.e., morphological) compar-
atives.
(89) The Synthetic Superlative Generalization (SSG)
No language has morphological superlatives (X-est), but only periphrastic
comparatives (more X ).
203
The two general theoretical assumptions are drawn from the existing literature
and have broad consequences beyond the morphological phenomena investigated
here. The assumption of Late Insertion is at the core of the framework
of Distributed Morphology and is supported by a significant amount of
evidence in that framework (and other realizational frameworks). Almost all
work invoking competition among (potentially) underspecified exponents requires
some version of this hypothesis (although not all frameworks take the
input representation to be the output of the syntax). Similarly, the locality
assumption underlying the RSG is not in any way specific to comparatives,
but must be a special case of a more general theory of locality in morphology.
The RSG should thus find paralles in restrictions on analytic-synthetic
alternations generally. In English, for example, where tense and agreement
are expressed synthetically in a simple declarative, but analytically in negatives
and (non-subject) interrogatives, the locality condition ensures that
suppletion is limited to the synthetic forms ((183) versus (184)).
(183) a. Leo goes swimming on Sundays.
b. Leo went swimming on Sunday.
(184) a. Did Leo go swimming on Sunday?
b. Leo didn’t go swimming on Sunday.
An underlying assumption here is that root suppletion in verbs is formally
characterizable in the same manner as in adjectives, thus:
204
(185) a. go → wen- / ] past
b. go → go
On this assumption, the analysis parallels that of (182) – the suppletive
allomorph is selected when the conditioning feature is an affix (185a), but
not when the conditioning feature is expressed periphrastically, with a word
boundary intervening.
The Containment Hypothesis is, of course, specific to comparatives and
superlatives, by definition, and I have devoted a significant part of Chapter
3 to arguing that it is robustly supported, independently of the CSG.
Having presented and defended the theory in broad outlines, the remainder
of this chapter is devoted to some refinements in detail, and some qualifications,
picking up on some threads left open in previous chapters.
5.2 Conditions on Suppletion: exponence versus
readjustment
In the derivational architecture in (181), rules of exponence provide the
phonological form to the abstract morphemes. Assuming that insertion applies
cyclically, beginning with the root, the architecture derives a version of
the No Lookahead Condition of Simpson and Withgott (1986) or the Peripherality
Condition of Carstairs (1987); see Bobaljik (2000b), Chung (2007b)
and Embick (2010). Consider this with reference to a concrete structure,
205
namely (186), the structure taken to be the input to vocabulary insertion,
after the application of syntax and Merger:
(186) s
c
a
adj
cmpr
sprl
Insertion applies first at the most deeply embedded node, namely, the
adjective root. At this point in the derivation, the morphosyntactic identity
of the more peripheral nodes (cmpr,sprl) is knowable, and (at least the
adjacent node, see below) may thus condition allomorphy, as in (182a), above.
In terms of the morphosyntactic representation, the Containment Hypothesis
ensures that the superlative is, morphosyntactically, but a special case of
the contexts in which this structural description is met, and thus any rule
of exponence specified as occurring ‘in the comparative’ will also apply in
the superlative, unless blocked by a more specific rule. This consideration
anchors the account of the CSG. Subsequent cycles of vocabulary insertion
provide (possibly null) phonological exponents for the the more peripheral
nodes in (186).1
The output of the rules of exponence (and, other morphological operations,
perhaps including bracket erasure and linearization of the structure)
206
constitutes the input to the phonological component. One class of phonological
rules includes the unproductive readjustment rules which effect irregular
changes to the phonology, including unpredictable vowel alternations (tell –
tol-d) and the like (see Halle and Marantz 1993 within the DM framework,
and numerous antecedents). In theory, there is a sharp division of labour
between rules of exponence and readjustment rules. The former introduce
distinct underlying representations, in the case at hand — distinct roots,
into the derivation, while the latter perform manipulations on a previously
introduced root. Of course, there is a difficult grey area for the analyst in
establishing just where the boundary lies — alternations like many – mo-re
could be treated formally as suppletion (with synchronically accidental initial
m- in both allomorphs) or as the output of a very powerful readjustment
rule, rewriting the syllable rime (cf. Halle and Marantz 1993 on English verbal
morphology). If the theoretical distinction is correct, then establishing
the boundary, and determining the limits of the power of readjustment, is
ultimately an empirical matter. For present purposes, I take the conservative
position of counting the examples in the grey area as instances of suppletion,
to ensure that no problematic cases for the CSG are inadvertently swept
under the rug of overly powerful readjustment rules.
Since phonological rules, including readjustment rules, apply after the
rules of exponence (perhaps with some measure of cyclic interleaving, though
this is irrelevant as far as roots are concerned), there is no a priori expectation
that the CSG (or a suitably worded extension thereof) would hold of irregular
207
comparatives. And it is indeed the case that irregulars may display a range of
surface patterns, including patterns not permitted for suppletive derivations.
The Classical Greek examples in (187) serve to illustrate this difference. The
triples for ‘good’ and ‘few’ are suppletive, and constitute ABB patterns (for
other forms for ‘better, best’ see (141)), but in the triples for ‘big’ and ‘swift’,
if one looks only at the surface forms of the root, the comparative constitutes
the odd one out.
(187) pos cmpr sprl
a. Anc. Greek: agath-ós belt-í¯on bélt-ist-os ‘good’
b. Anc. Greek: olíg-os eláss-¯on elách-ist-os ‘few’
c. Anc. Greek meg-as meiz-¯on meg-ist-os ‘big’
d. Anc. Greek tach-us thass-¯on tach-ist-os ‘swift’
The forms in (c) and (d) involve no competition among exponents. The
surface form meiz¯on is derived from underlying meg-i¯on by a rule of palatal-
ization.2
As far as the rules of exponence (and thus the CSG) are concerned,
the phonologically irregular patterns in (187c-d) are morphologically regular,
instances of the AAA pattern, with a single root form occurring underlyingly
in all three grades.
The key work in explaining the CSG is done by the elsewhere ordering,
which relies on subset-superset relations among (the structural descriptions
of) rules. The architectural concerns discussed above effectively limit suppletive
rules (rules of exponence) to morphosyntactic contexts, forcing the
CSG.3
But no such limitation is imposed on phonological (including readjust-
208
ment) rules, and thus whether a rule that effects a change in the comparative
root will extend to the superlative depends on the relation between the structural
description of the rule and the representation of the superlative. The
palatalization rule refers to a (morpho-)phonological context, applying only
when the triggering palatal is prevocalic. Since the underlying palatal was
prevocalic in the comparative *-j¯os, but not in the superlative formatives
-istos (Greek) and -issimus (Latin < -ism
˚
mos), there is no superset-subset
relation between the phonologically-defined environments, and no expectation
that the ‘comparative’ form would spread beyond the comparative.
While (surface) ABA patterns may be derived by readjustment rules applying
to a unique underlying root (as in (187c-d)), irregulars are therefore
also expected to show ABB patterns whenever both the superlative and the
comparative meet the environment for some rule. For example, palatalization
of stem-final consonants is triggered by the initial palatal of the comparative
in Slavic; since the superlative is transparently built on the comparative (except
in Bulgarian-Macedonian), the context for the palatalization rule is met
in both environments, as the following Serbo-Croatian examples illustrate:
(188) pos cmpr sprl
a. dug duž-i naj-duž-i ‘long’
b. brz brž-i naj-brž-i ‘fast’
c. mlad mladž-i naj-mladž-i ‘young’
Consider also German umlaut, which is characterized in the framework
adopted here as a readjustment rule (Halle and Marantz 1993). Though
209
historically phonological, the structural description of this rule is (synchronically)
morphologically defined – certain roots are specified to undergo umlaut
in the presence of the comparative morpheme. If this rule makes reference
to the abstract morpheme cmpr, then the umlaut pattern should extend
to the superlative as well, given the containment hypothesis. This is, of
course, what happens, with adjectives showing a consistent pattern between
comparative and superlative:
(189) pos cmpr sprl
a. hart härt-er (am) härt-est-en ‘hard’
b. zart zart-er (am) zart-est-en ‘tender’
c. schwach schwäch-er (am) schwäch-st-en ‘weak’
d. wach wach-er (am) wach-st-en ‘awake’
The correctness of the CSG, and the lack of an analogous generalization
for irregulars, may thus constitute indirect evidence in favour of a theory
(such as DM) that draws a strict distinction between suppletion and readjustment,
as against frameworks in which all synchronically irregular root allomorphs
are held to be, in essence, special cases of suppletion (Wurzel 1985,
Anderson 1992 and more recently, Bonet et al. 2007 and related work).4
I
do note, though, that the restriction of suppletion to morphosyntactic environments
remains controversial, with some recalcitrant challenges. Carstairs
(1987, 1988) argued that there are cases of phonologically governed suppletion,
most notably the Italian verb andare ‘go’, which is built on the root
va(d)- when stress falls on the root (e.g., 1sg vád-o), but on and- when
210
unstressed (as in 1pl and-iámo). Although the distribution correlates with
stress placement in this case, it is not possible to demonstrate empirically
that the suppletion is conditioned by stress, since the environments in which
the root is or is not stressed, and hence the distribution of suppletion can
also be characterized (albeit with some clunkiness) in morphological terms
(1sg, pl ...), and thus some authors have argued that this case is not phonologically
conditioned after all. Particularly relevant in this regard is Maiden
(2005) and related work, which argues for that the suppletive patterns are a
special case of a broader morphological patterning that is consistent across a
range of Romance varieties, even where the pattern does not have clear-cut
non-morphological correlates like stress. While the majority of cases of apparent
phonologically-governed suppletion fall to alternative analyses in this
way, a more difficult case, which I leave unresolved for now, is the variety of
Surmiran (Rumantsch) discussed in Anderson (2008).5
5.3 Adjacency, ABC, *AAB
Another controversial issue regarding the conditioning environments for suppletion
concerns the postulation of an adjacency condition (see Chapter 2).
In the following subsections, I return to the role that condition plays in the
account, attempting to shed light on a delicate interaction between adjacency
and portmanteaus, and then, for completeness, add a few remarks on
outstanding problems with invoking adjacency.
211
In Chapter 2, I suggested that there is also an adjacency (or contiguity)
condition on allomorphy, such that the trigger for suppletive allomorphy must
be adjacent to the root that undergoes allomorphy. Such a condition would
exclude (190a) (= (45a)) as a possible context for root suppletion in (186)
since the comparative intervenes.6
(190) a. good → be(tt)- / ] sprl ]
b. good → good
The adjacency condition, in tandem with the central assumptions of
Chapter 2 serves to derive the following typology of possible suppletive pat-
terns:
(191) pos cmpr sprl
a. regular A A A big – bigger – biggest
b. suppletive A B B good – better – best
c. doubly-suppletive A B C bonus – melior – optimus
d. unattested A B A *good – better – goodest
e. unattested A A B *good – gooder – best
I left open in Chapter 2 what the formulation of adjacency would need to
be so as to exclude *AAB without incorrectly excluding the attested ABC
patterns such as in Latin and Welsh. Some examples from Chapter 2 are
repeated here:
212
(192) pos cmpr sprl
a. Latin: bon-us mel-ior opt-imus ‘good’
b. Welsh: da gwell gor-au ‘good’
c. Old Irish: maith ferr dech ‘good’
The main import of the ABC patterns is that they illustrate the logic of
the Elsewhere reasoning, in particular, the observation that the ABB pattern
will be the default suppletive pattern unless the comparative root allomorph
is overridden by a more specific exponent in the superlative. It must therefore
be possible for root suppletion to be conditioned by the superlative. Above,
this was formalized as in (193) corresponding to the Welsh example in (192b)
(cf. Latin (41) from Chapter 2).
(193) a. good → gor- / ] cmpr ] sprl ]
b. good → gwell / ] cmpr ]
c. good → da
The challenge here is to ensure that whatever formalism is used in (193a),
allowing the superlative root allomorph to bleed the comparative, is unavailable
in cases like (190a), which would falsely admit AAB patterns. In other
words, in order to derive the CSG2, what we need to ensure is that the presence
of a suppletive allomorph in the superlative (as in (193a)) for a given
root ensures that the same root also has a suppletive comparative allomorph
(193b). In the following paragraphs, I suggest two approaches to deriving
this generalization, one drawing in part on ideas about the treatment of portmanteau
morphemes in Radkevich (2010), and the other a return to early
213
notions of the cycle, and of domain extension. I leave deciding among these
alternatives as a project for future work.
5.3.1 Getting worse: portmanteaus and locality
I start with some theoretical housekeeping, taking the position that adjacency
is the right condition to exclude (190a) and thus that the clunky statement
of the context in (193a), representing the C case of an ABC pattern, needs
to be reconsidered.
Now, it happens that the few clear cases of superlative allomorphs participating
in ABC patterns lack an overt exponent of the comparative affix.
To be sure, the number of ABC cases is too small to be confident that this
is more than an accident, and I have had recourse to a zero allomorph of the
comparative within the superlative (in (42)). Nevertheless, I suggest that the
root allomorph gor- in Welsh (and likewise Old Irish dech and Latin opt-)
are portmanteaus, expressing both the adjective root and the comparative,
simultaneously.
In the previous chapters, I did not dwell at any length on the distinction
between (i) cases of clear contextual allomorphy (such as bett-er) in which
the root is suppletive, but the structure is otherwise as expected, with the
comparative morpheme following the root, and (ii) cases of portmanteau
allomorphy, in which the suppletive form replaces not only the root, but also
the expected affix, as in worse (not *worse-r). I have simply treated the latter
as a special case of contextual allomorphy of the root, with a concomitant
214
lexically conditioned null allomorph of the comparative: worse+Ø. It is now
time to revisit the worse case.
The toolbox of Distributed Morphology contains an alternative means of
characterizing worse-type portmanteau suppletion. Specifically, rather than
posit that worse is the spell-out of the adjective root bad alone, worse may
spell out both root bad and the feature cmpr simultaneously, as in (243):
(194) a. bad, cmpr → worse
b. bad → bad
Compare the representation of (194a) to (182a), repeated here:
(182) a. good → bett- / ] cmpr
b. good → good
The suppletive allomorphs in (194a) and (182a) differ in the location
of the element cmpr. This difference does not affect the applicability of
the Elsewhere reasoning which establishes an intrinsic ordering among these
exponents, and makes no changes to the main theoretical proposals in this
book. However, the vocabulary item in (194a), unlike that in (182a), may
serve as the spell out of the node c in (195) (=(186)) directly, with no appeal
to a zero comparative allomorph (as in (42)) in the comparative or superlative
in this case.
215
(195) s
c
a
adj
cmpr
sprl
The current literature provides at least two ways to flesh this out concretely,
and deciding between them is not relevant for present concerns. Early
DM provides for an operation of Fusion, which joins two sister nodes into a
single node with the combined features of both (see Halle and Marantz 1993,
Bobaljik 1997, Embick and Noyer 1999, Chung 2007a). Fusion of cmpr and a
in (195) is schematized in (196); the fused node is a single locus of vocabulary
insertion, to which the rule of exponence in (194a) may apply.
(196) s
c
a
adj
cmpr
sprl
F usion
−−−−→ s
ac
adj⊕cmpr
sprl
An alternative formalism proposed by Radkevich (2010) is to admit insertion
directly at non-terminal X0
nodes (see also Neeleman and Szendrői
2007 and Caha 2009), i.e., inserting the exponent worse directly at node c in
216
(195) without the intermediate operation of Fusion. In order to prevent many
kinds of unwanted over-application of insertion, Radkevich also proposes to
reformulate the standard version of vocabulary insertion as in (197):7
(197) The Vocabulary Insertion Principle (VIP)
The phonological exponent of a vocabulary item is inserted at the
minimal node dominating all the features for which the exponent is
specified.8
(Radkevich 2010, 8)
For current purposes, the Fusion and VIP proposals are equivalent.9
The
relevant consideration here is that, one way or another, portmanteaus may
be characterized as insertion at a node dominating multiple features, corresponding
to multiple terminal nodes at some stage of the derivation. With
Fusion or the VIP at our disposal for describing portmanteau morphemes,
we may return to the Welsh ABC case, dispensing with the clunky statement
of the context in (193a), and replacing it instead with the clearly adjacencysatisfying
allomorphy in (198a).
(198) a. good, cmpr → gor- / ] sprl ]
b. good, cmpr → gwell
c. good → da
The Elsewhere Condition applies as before, ensuring the ordering among
exponents. Moreover, assuming that adjacency restricts suppletion requires
ABC patterns to involve portmanteau exponence of
√
root + cmpr in the
superlative, since it is only the node dominating both of these that is adjacent
217
to the sprl node. A Vocabulary fragment like (193) is impossible, and
must be represented instead as in (198). As noted above, the consequence
that ABC patterns must involve portmanteau suppletion for the comparative
contained in the superlative grade is consistent with all the known ABC
cases.10
The ramifications of this approach to portmanteau suppletion extend well
beyond grades of comparison, and the approach yields the general prediction
in (199):
(199) If an exponent X expresses
√
root + F1...Fn, for some features
F1...Fn, then F1...Fn must be adjacent to (or contiguous with)
√
root.
The empirical content of this prediction is that, all else being equal, if
there are portmanteaus expressing
√
root1 + F1 in some language, then for
any non-portmanteau root
√
root2,
√
root2 and F1 will be adjacent (and
where the portmanteau includes a group of features, these will be contiguous
with the root).
Radkevich (2010) argues that this prediction is borne out in the large
survey of verbal suppletion in Veselinova (2006). A well-substantiated generalization
in affix order is that (direction of affixation aside) aspect morphemes
are closer to the root than tense morphemes, which are in turn closer than
mood morphemes (Bybee 1985, Cinque 1999, Julien 2002). Structurally, this
can be represented as in (200):
(200) [[[[ root ] aspect ] tense ] mood ]
218
Portmanteau expression of
√
root + aspect is admitted by the theory,
and well attested. By (199), portmanteau expression of
√
root + tense
should be disallowed in the general case, but it is allowed if either (i) there
are no aspect morphemes in the domain (or language) in question (hence
tense is adjacent to the root) or (ii) the portmanteau expresses
√
root +
aspect + tense, i.e., a contiguous sequence of heads, corresponding to a
single (complex) node in (200). Radkevich reports that this is indeed the case
in Veselinova’s survey (and likewise for portmanteaus that include mood).11
Confirmation of this prediction stands as independent support for an approach
to portmanteau suppletion that incorporates a hierarchical arrangement
of morphosyntactic features independent of (and prior to) the rules of
exponence that realize these features.12
This is because the prediction relies
on a hierarchical arrangement of the features, prior to the rules of exponence
that realize these features. Theories that deny such an arrangement (notably
Word-and-Paradigm theories such as Anderson 1992 and Stump 2001) are,
it seems, unable to accommodate generalizations of this sort.
Returning to the issue at hand, although I have just shown how the
ABC pattern may be profitably described while maintaining an adjacency
condition on allomorphy, this alone does not suffice to exclude the AAB
pattern. Consistent with what has been said thus far, it should be possible
to have a Welsh-like system, but with only two root allomorphs for good,
as in (201):
219
(201) a. good, cmpr → gor- / ] sprl ]
b. good → da
This vocabulary will derive the unattested AAB pattern. I suggest that
the following condition holds:
(202) If there is a context-sensitive rule of exponence involving a node α,
then there is a context-free rule of exponence involving α.
For the basic case, this entails that if a root has a suppletive allomorph,
then that root also has a corresponding default form. The vocabulary entry
in (203a) (=(182a)) is permissible, just in case there is also a context-free
form (203b).13
(203) a. good → bett- / ] cmpr
b. good → good
The motivation for (202) may lie in the process of acquisition, in essence,
allowing for the acquisition of a contextually restricted allomorph of some
morpheme α only after α itself has been acquired. The intuition is thus reminiscent
of leading ideas in Pinker (1984), in particular the discussion there
of the acquisition of paradigm structure (see also remarks on markedness in
Noyer 1998 and Calabrese 2005).14
The pair of exponents (198a) and (198b) constitute context-sensitive and
context-free allomorphs, competing for the expression of a single node, node
c in (195). This pair of rules is thus permitted under (202). Now consider the
relation between (198b) and (198c), which is exactly parallel to that between
220
worse and bad in (194), repeated here. The condition in (202) is intended to
cover (194) in the same way it covers (203).
(194) a. bad, cmpr → worse
b. bad → bad
To ensure that this pair satisfies (202), we must understand (194a) to
count as a context-sensitive counterpart to (194b). Rule (194b) applies
to (thus involves) node a [
√
bad ] directly, while the portmanteau rule
(194a) affects (and thus involves) that same node, but only in a contextsensitive
manner, namely, only when a is contained in the cmpr node. Understood
in this way, (202) thus permits (198) (and (194)). By contrast
(201), which would derive an AAB pattern (*da – da-ch – gor-au), is excluded
by (202). Rule (201a) provides a context-sensitive exponent for the
node [ good, cmpr ] (node c above), but there is no corresponding contextfree
rule involving that node [ good, cmpr ] (even though there are entries
for its components).
The markedness condition in (202), when coupled to the VIP treatment
of portmanteaus, and an adjacency condition on allomorphy, makes the right
cut among attested (ABC) and unattested (*AAB) patterns of superlative
suppletion.
Latin makes the same point with a minor twist. The comparative melior
is clearly bi-morphemic, containing the regular comparative affix -ior.
Assuming that this segmentation is synchronically part of Latin grammar,
then opt- can be treated as a portmanteau (resolving the adjacency issue),
221
but mel- cannot be made parallel to (198b), requiring instead (204).
(204) a. good, cmpr → opt- / ] sprl ]
b. good → mel- / ] cmpr ]
c. good → bon
Here again, (202) makes the right cut. The pair of rules in (204b-c) is
exactly parallel to (203) discussed above. In turn, the pair in (204a-b) both
involve the node c dominating the adjective root and the comparative, with
only the rule in (204a) sensitive to an additional context beyond that node.15
In sum, combining the treatment of portmanteaus via Fusion or Radkevich’s
VIP with the assumption in (202) cleans up an issue left unresolved in
earlier chapters, ensuring that the account of the ABC patterns is consistent
with the assumptions that exclude the unattested *AAB pattern.16
5.3.2 Aside: A cyclic alternative?
Before moving on, I make a brief digression here to put on the table a possible
alternative to (202), invoking cyclicity (domains), rather than markedness to
ensure the transitivity effect in (198) and (204).17
Recall that the key work
that (202) does is to ensure that the existence of a suppletive allomorph in
the superlative for a given root (as in (204a)) entails that the same root also
has a suppletive comparative allomorph (204b), excluding the *good – gooder
– best (*AAB) pattern, while permitting the ABC one.
Conceivably, the same work could be done without (202) in a theory
which incorporates some version of the Strict Cycle Condition and/or a Phase
222
Impenetrability Condition (Chomsky 2001), along with a condition that suspends
the effects of the latter in particular instances. Consider, for example, a
cyclic theory that incorporates the following assumptions (cf. Embick 2010):
(205) a. A cyclic node triggers “spell-out” of its complement, where spellout
includes rules of exponence.
b. A node that has been “spelled-out” is inaccessible to further application
of rules of exponence.
If we assume that cmpr and sprl are designated as cyclic nodes, then
the first step to consider in the derivation of a non-suppletive superlative like
biggest is that in (206):18
(206) c
a
√
big
cmpr
At this point, by (205a), the adjective root will be subject to exponence;
for example, the node a could be spelled out as big. The next cycle would
then be as in (207). Note that the innermost node has been spelled out, and,
by (205b), is thus impervious to further rules of exponence. In particular, the
(structural) adjacency condition on allomorphy is here cast as a by-product
of cyclic derivation—when rules of exponence apply to the root, the sprl
node is not visible; and when the sprl node is introduced, the chance to
affect exponence of the root has passed.
223
(207) s
c
big cmpr
sprl
Now, imagine adding to (205) a qualification as follows:
(208) a. Spell-out of a domain D under (205a) is suspended, if a rule of
exponence spans D.
b. A rule spans D if it involves X,Y in the configuration: [ [ X ]D Y ]D+1
This qualification has no effect on the spell out of a non-suppletive adjectival
root in (206), but in the case of a suppletive root, such as
√
good (with
suppletive allomorph bett- / cmpr), the qualification in (208) blocks spellout
of node a on the first cycle. Spell-out is delayed until the next higher
cycle. While a regular adjectival root is thus frozen by the time the superlative
is introduced (205b), a root with a comparative allomorph will remain
‘active’ at the outer cycle, and spell out of the root (or more correctly the
node c containing the root) could potentially be affected by the superlative:
224
(209) s
c
a
√
good
cmpr
sprl
In this way, the Domain Suspension proposal in (208) appears able to
capture the transitivity effect that distinguishes between *AAB and ABC
patterns. The sprl node is only accessible to the root (and thus can only
participate in conditioning root allomorphy) if spell-out of the root on the
first cycle is suspended, effectively expanding the spell-out domain that contains
the root to node c. Crucially, it is the existence of a comparative
allomorph for the root that triggers this domain extension; for this approach
to have the desired effect, (208) must be defined in such a way that a rule
like (201a) does not trigger Domain Suspension, either because it does not
apply to node a as such, or because it cannot be considered on the first cycle
as the sprl in the contextual restriction is too remote.
One reason to entertain a Domain Suspension approach, in place of (202)
is the potential for convergence across theoretical domains. The current
syntactic literature contains a variety of suggestions to the effect that cyclic
domains (phases) are in some sense dynamically defined (Bobaljik and Wurmbrand
2005, Bošković 2010). In some proposals, such as Den Dikken’s (2007)
Phase Extension (and other related suggestions, with antecedents including
225
Baker 1988), Head-Movement plays the role of spanning in (208b) expanding
cyclic domains; while in others, such as Wurmbrand (2011a), closer to
the present proposal, it is the requirement of feature dependencies to span
a domain boundary that suspends the domain effects (see also Despić 2011
and Takahashi 2011 for related proposals). It is not trivial however, at this
juncture, to show that Domain Suspension provides a fully viable alternative
to (202), and I leave the matter open.
5.3.3 Adjacency - Outstanding Issues
The previous aside notwithstanding, an adjacency condition on suppletion is
motivated within the current theory by the work it does in explaining the
non-occurrence of AAB patterns (*good – gooder – best). It is also reasonably
well-motivated in the literature independently of comparatives (see Embick
2003, 2010). In the realm of root suppletion, as mentioned above, Radkevich
(2010, 199-200) evaluated the instances of verbal suppletion discussed in
Veselinova’s 2006 extensive study, and found that all examples of suppletion
in that study conformed to the predictions of an adjacency account — none
of the examples of suppletion presented in Veselinova (2006) involve a nonadjacent
trigger.19
Nevertheless, there are apparent cases in the literature of allomorphy
in non-adjacent contexts, an issue first brought under scrutiny in Carstairs
(1987). For the current project, we may restrict our attention to cases involving
root suppletion, putting aside cases of non-adjacent inwards-sensitivity
226
in affixal allomorphy, such as the Itelmen class and agreement marking discussed
in Bobaljik (2000a,b), as well as examples of non-adjacent application
of readjustment rules.
One case to consider is Korean verbal suppletion, discussed in Chung
(2009). In Korean, some verbs show suppletion for negation (al- ‘know’,
molu- ‘not know’), while others show suppletion for honorification (mek- ‘eat’,
capswu ‘eat [honorific]’). One verb shows three-way suppletion (iss- ‘exist’,
eps- ‘not.exist’, kyey- ‘exist [honorific]’. The negative suppletive forms are
portmanteaus, replacing both the root and the regular marker of negation
(which is prefixal). The honorific forms involve contextual allomorphy and
obligatorily cooccur with the regular honorific suffix:
(210) a. eysute-ka
Esther-nom
ttek-ul
rice cake-acc
mek-ess-ta.
eat-pst-dcl
‘Esther ate rice cake (non-honorific).’
b. apeci-kkeyse
father-hon.nom
ttek-ul
rice cake-acc
capswu-*(si)-ess-ta.
eat.hon-hon-pst-dcl
‘Father ate rice cake (honorific).’
Although the verbal root and honorific suffix are string adjacent, Chung
gives two careful arguments that they are not structurally adjacent. One
argument comes from the interaction of the two types of suppletion. Negative
suppletion in Korean is shown to require structural adjacency (Chung 2007b),
but honorific root allomorphy bleeds negative suppletion. For the one verb
227
that undergoes both types of suppletion, the only possible form for a negative
honorific is to combine regular negation with the honorific root allomorph:
(211) a. apeci-kkeyse
father-hon.nom
silhemsil-ey
lab-loc
an(i)
neg
kyey-si-ta.
exist.hon-hon-dcl
‘Father is not in the lab (honorific).’
b. * apeci-kkeyse
father-hon.nom
silhemsil-ey
lab-loc
eps-u-si-ta.
not.exist-V-hon-dcl
c. * apeci-kkeyse
father-hon.nom
silhemsil-ey
lab-loc
an(i)
neg
iss-u-si-ta.
exist-V-hon-dcl
Including a further argument regarding the presence of a (silent) v head,
Chung argues that the structure of the Korean verb is as in (212):
(212) T
hon
neg
neg v
V
exist
v
hon
T
In this structure, the suppletion in (211a) thus takes place in a context
of structural non-adjacency. Admitting a parallel analysis in the realm of
228
superlatives would open the door to unattested AAB patterns, which could
be characterized as (190a). Despite the elegance of Chung’s solution, it is
conceivable that the alternations characterized as honorific suppletion are
not instances of grammatical suppletion as such, but are rather lexical, with
two verbs in play: capswu- ‘to eat (of an honorific agent)’ and mek- ‘to eat’;
compare, for example, German fressen ‘to eat (of animals)’ versus essen ‘to
eat’ and many similar cases.20
I will thus, tentatively, put this example aside.
A more serious problem arises closer to home, as it were, with Basque
comparatives once again. Basque on ‘good’ has a suppletive comparative
hobe ‘better’, which appears, like English worse to occur without the comparative
suffix (Basque: -ago). As noted in section 3.2.2, Basque has a
morphological means of creating a comparative with the meaning ‘a little
more X’, created by intercalating the morpheme -xe- between the adjective
root and the comparative suffix. As (213) illustrates, there are two points
to notice here. First, suppletion occurs in the -xe-ago forms just as in the
regular comparatives, and second, the regular suffix -ago reappears in the
(c) example even with the otherwise portmanteau-like root hobe (examples
from de Rijk 2008, 710-711).
(213) ‘much’ ‘good’
a. asko on
b. gehi-ago hobe
c. gehi-xe-ago hobe-xe-ago
In section 3.2.2 I argued that semantic considerations here support an
229
analysis of the xe-ago forms invoking a branching affix structure, perhaps
required in any event for a few Fennic varieties, thus (214):
(214) s
a
adj
cmpr
xe cmpr
In this structure, the roots are adjacent to a projection of the node cmpr
and thus these examples are consistent with the adjacency condition on allomorphy.
The reappearance of -ago in (213c) suggests that hobe should be
treated not as a portmanteau, but rather as an instance of garden-variety
contextual allomorphy, with a concomitant zero allomorph of the comparative
lexically conditioned to occur under adjacency with this root, thus:
(215) a. good → hobe / ] cmpr ]
b. good → on
c. cmpr → Ø / hobe ]
d. cmpr → -ago
Now, it should be clear why the zero comparative (215c) is selected in
(213b). But why does -ago show the distribution that it does, surfacing in
(213c), rather than the zero comparative being inserted at the higher cmpr
node in (214)? Even if insertion at non-terminal X0
nodes is possible, as
entertained above, the VIP (197) will ensure that the comparative exponent
230
-ago must be inserted at the lower cmpr node in (214). In that position,
even though the root is adjacent to (a projection of) the cmpr, the cmpr
node at which insertion applies is not in turn adjacent to the root. The
environment for the Ø allomorph is not met, and -ago must be inserted.
Despite the promise of this subtle result, there is a further complication
in Basque. In sec 4.2, I noted variation across languages in the order of
morphemes deriving adverbs from adjectives, with respect to the comparative
marker, as in (216).
(216) a. Karelian, Georgian [ [ [ adjective ] comparative ] adverb ]
b. Basque [ [ [ adjective ] adverb ] comparative ]
Where the pieces are overt, Basque clearly has the structure in (216b); on
semantic grounds there is no reason to consider a branching affix structure
here:
(217) berri
new
‘new, fresh’
berri-ki
new-adv
‘recently’
berri-ki-ago
new-adv-cmpr
‘more recently’
Corresponding forms for ‘good’, resp. ‘well’ are given in (218):
(218) on on-gi hobe-ki
on on-do hobe-to
good good-adv better-adv
‘good’ ‘well’ ‘better (adv)’
231
The occurrence of the suppletive root and the absence of the overt comparative
affix in the adverbial form for ‘better’ both suggest (given the discussion
above) that the comparative is underlyingly adjacent to the adjectival
root, but in non-supletive forms, where the comparative is overt, it occurs
peripheral to the adverbial marker.
In sum, whether adjacency constrains (root) suppletion, and if so in what
form, is thus an important piece of the overall puzzle, which I leave as currently
unresolved. The bulk of the evidence seems to indicate that there
is a role for an adjacency condition, but puzzles like the Basque adverbial
example just discussed remain to be solved before we can consider this point
established.21
5.4 AAB ablaut
In discussing the CSG, I have kept the two pieces of the generalization distinct
(the impossibility of the ABA and AAB patterns). The Containment
Hypothesis and Elsewhere ordering are central to the account of both, while
the Adjacency Condition plays a role only in the CSG2. Nothing beyond
expository convenience is intended by keeping the generalizations distinct
in this manner — they are, if I am right, simply descriptive statements that
enumerate various empirical consequences of the theoretical proposals, which
could be rephrased in a variety of ways.22
One reason for keeping the generalizations distinct, though, is in compar-
232
ing the cases of comparative suppletion to other alternations. As we examine
different feature structures, parallels to the CSG1 and CSG2 are predicted
to emerge, though not always in tandem. In this section, I examine one such
extension, drawing on Wiese’s (2004, 2005) analysis of ablaut in German
verbs.23
In German, the strong verbs are characterized by vowel alternations in
their principal parts, i.e., the present, preterite (simple past), and perfect participle
stems (cf. English sing-sang-sung). These vowel alternations (ablaut)
are largely unpredictable and must therefore be listed in one way or another.
In the framework adopted here, such alternations are treated as readjustment
rules (see above), either lexically restricted as a part of the context of
the rule, or equivalently, restricted via diacritic marking of the root.24
All
tolled, Wiese notes that there are some 40 distinct patterns attested (Wiese
2005, 2). Some verbs show a three-way distinction, with a different vowel in
each principal part, while others show some degree of syncretism. Strikingly,
the attested patterns of syncretism may be summarized as in (219), where
identity refers to identity of the stem (vowel).
(219) a. Distinct for all three principal parts.
b. Past and past participle identical, present distinct.
c. Present and past participle identical, simple past distinct.
d. Unattested: present and past identical, participle distinct.
German examples illustrating the attested patterns are given in (220);
233
forms are listed in the conventional order:
(220) present 1sg preterite 3sg perfect participle
a. ‘speak’ sprech-e sprach-Ø ge-sproch-en
b. ‘water’ gieß-e goß-Ø ge-goss-en
c. ‘give’ geb-e gab-Ø ge-geb-en
The examples in (221) show it may be possible to make the same point
for English, with a pair given for each pattern showing that the distribution
is independent of whether the participle is marked with an additional suffix
or not.25
(221) present past past participle
a. sing sang sung
ride rode ridd-en
b. shine shone shone
tear tore tor-n
c. come came come
give gave give-n
Wiese’s solution to the puzzle of why one of the possible syncretisms is
unattested (namely (219d)) follows the same logic as I have proposed for
the CSG, and thus stands as an important precedent for the general reasoning
pursued here. Specifically, Wiese proposes that there is a markedness
hierarchy characterizing the principal parts, namely that in (222a) (Wiese
2005, 29), and crucially, that the features defining the three forms stand in
a containment relation, as shown in (222b), with Wiese’s feature labels:
234
(222) a. present < participle < preterite
b. i. [ ] “present”
ii. [past] “(perfect) participle”
iii. [past, finite] “finite past tense”
Present is treated here as a default, thus with no defining features; (perfect)
participles share with finite past forms the classificatory feature [past],
with finite forms bearing an additional feature, [finite], lacking in the participles.
Thus, the features characterizing (the suffix node of) the participle
are contained in those for the more marked preterite.26
Note that the order determined by the markedness hierarchy differs from
the traditional textbook presentation of the principal parts (as in (220)) in
that the participle occupies the middle position on the hierarchy, and in terms
of features, is considered to be marked with respect to the present/infinitive
stem, but unmarked among past forms, in contrast to the marked finite past
form (the simple past).27
But the markedness hierarchy in (222a) is consistent
with the distributional evidence across Germanic; a venerable understanding
of such markedness hierarchies holds that if a language lacks one category in
the sequence, it will lack the most marked category (e.g., Jakobson 1941),
and indeed, there are Germanic varieties (Yiddish, Afrikaans, varieties of
German including Austrian) that have a present and a perfect participle,
but lack a simple preterite.
If this treatment of the features is supportable, then the logic of under-
235
specification now applies for any rules that are conditioned by these features.
In the absence of a designated form for any given category, the next most
highly specified form is used. The possible patterns of syncretism, defined
by (222b), are given in (223).
(223) present participle preterite contexts examples
a. A A A (i) (weak verb)
b. A B B (i, ii) gießen
c. A B C (i, ii, iii) sprechen
d. A A B (i, ii) geben
e. * A B A (i, iii) unattested
Just as with the CSG, the *ABA pattern is unstatable. It is not possible
to single out the middle member of the hierarchy (in this case, the participle),
to the exclusion of one of the other columns. To the extent that a particular
ablaut form does in fact only surface in the participle (as in gesprochen in
(220b)), this can only arise when the use of this form in the preterite is bled
by a more highly specified form for that context.
Despite the parallel with the CSG, there is also an important difference.
In the case of comparative suppletion, neither the ABA nor the AAB pattern
was attested, yet in the case of the German ablaut patterns, only one
of these is excluded, the other attested. This difference falls out from the
theory presented here if the features do indeed stand in the containment
relationship indicated in (222b). More specifically, if the features defining
236
the preterite in (222b.iii) constitute a bundle of features, occupying a single
morphosyntactic node, as diagramed in (224a), then the adjacency issue —
central to excluding the *AAB pattern in suppletion — does not arise with
the preterites. Compare the structure assumed for the superlative in (224b)
— the sprl head is non-adjacent to, and thus too far away from, the root to
condition allomorphy, but the past and finite features are both adjacent
to the root.
(224) a. b.
V
V
verb



past
finite



s
c
a
adj
cmpr
sprl
The Elsewhere logic applies when the context for one rule (R2) is a proper
subset of the context of application of a more general competing rule (R1).
When there are three relevant contexts, but only two rules, the Elsewhere
logic prevents extension of the general rule (R1) to a context where R2 may
also apply. This excludes ABA patterns, whether the rules are rules of exponence
or of readjustment, and regardless of the internal arrangement of the
features that make up the contexts in question. Exclusion of the AAB pattern
requires the containment logic, but also a further assumption, namely,
that of an adjacency condition restricting the range of configurations in which
237
a head may serve as a contextual trigger for root allomorphy. With this in
place, the theory predicts differences in the range of ABA and AAB patterns;
not every context that excludes ABA will necessarily exclude AAB. The evidence
from Wiese’s account of German ablaut patterns appears to support
this theoretical position.
Another context where it has been argued that ABA patterns are excluded,
but AAB patterns are attested, is case syncretism. Caha (2009)
proposes to represent case-markedness hierarchies such as (225a) (cf. Blake
2001) as feature-containment structures. Thus the representation of “genitive”
properly contains that of “accusative”, which in turn properly contains
“nominative", etc. Caha implements this as a richly decomposed tree (225b),
with each node corresponding to one feature (see Chapter 7, below).
238
(225) a. Nominative < Accusative < Genitive < Dative < Instrumental
...
b. InstrP
instr DatP
dat GenP
gen AccP
acc NomP
nom NP
. . . N . . .
The interest of Caha’s proposals for the present study is that he builds
on the containment logic developed in Chapter 2 to develop an account of
(mostly) unattested ABA patterns in case syncretism. Thus, for example,
Accusative and Dative cannot be syncretic to the exclusion of Genitive, given
the structure in (225b). Like Wiese’s ablaut phenomena, the case patterns
investigated by Caha differ from the comparative-superlative patterns studied
here in admitting AAB patterns. While the rich tree structure may not be
compatible with the current proposals (to the extent that cases other than
nominative may trigger root allomorphy), Caha’s proposals may show the
239
applicability of the general logic developed here over a wider domain. Like
the ablaut examples, they also demonstrate the (partial) independence of
(*)ABA and (*)AAB generalizations.
5.5 Merger, Rule Ordering, Diacritics and Ac-
quisition
One final housekeeping comment is in order before we move on to the next
empirical domain. Recall from the architectural assumptions at the beginning
of this chapter that Merger is held to apply prior to vocabulary insertion.
This assumption was critical to the discussion of locality in section 3.3.1 —
Merger derives the local configuration in (226) which is the input to rules of
expoenence. As argued in section (3.3.1), without the prior application of
Merger (or an equivalent operation such as Head Movement), the comparative
morpheme could not condition allomorphy of the adjectival root.
(226) a. CmprP
cmpr AdjP
adj
b. c
a
adj
cmpr
However, this ordering appears to raise a paradox if, as is often assumed
(at least for English), it is phonological properties of a root that determine
240
whether or not that root undergoes Merger. One frequently encounters rough
characterizations such as: mono-syllabic, and di-syllabic adjectives in -y,
form synthetic comparatives (undergo Merger, in current terms), while polysyllabic
roots do not.28
Under the derivational architecture in (181), at the
point where Merger applies, information about the phonological make-up of
the root is not yet known.
The paradox is avoided if the conditioning of the application of Merger
does not (despite appearances) make reference to surface phonological characteristics
of roots. A review of the literature suggests that this is in fact
correct. Specifically, I contend that the synchronic application of Merger is
triggered by a diacritic feature on the abstract root (or stem). While there
are some trends and sub-regularities, there are counter-examples to a phonological
account in both directions. Monosyllabic adjectives such as ill and
apt resist -er suffixation, and near minimal pairs exist in disyllabic adjectives
(handsomer versus *irksomer, Clahsen and Temple 2003). Particularly relevant
is the experimental work of Graziano-King (1999) (see also Dalalakis
1994).
There is an over-arching limitation, such that adjectives greater than two
syllables strongly resist the comparative suffix (with well-known exceptions,
including affixation of -y or un-, see below). However, among adjectives
that are not excluded on these grounds, the results are far from neat. Thus,
below the tri-syllabic threshold, Graziano-King shows that the major factor
determining whether comparative formation is via suffixation or periphrasis
241
is frequency, not phonological form. In one of her control tasks, a relative
acceptability judgment with adult native speakers, high-frequency, monosyllabic
adjectives like old, long took the suffixal, rather than periphrastic
comparative in 99% of responses, while low-frequency items like lax, gaunt
took the suffixal comparative in only 15% of responses (Graziano-King 1999,
55). Strikingly, for monosyllabic nonce words on a parallel test, the suffixal
comparative was preferred over the periphrastic construction at exactly 50%
(p.66). Similar results obtained across a range of experimental designs.
The conclusion I draw from the available evidence is that the distribution
of analytic versus synthetic comparatives in English is grammatically regulated
by a diacritic. Certain morphemes are specified as bearing a diacritic
that triggers Merger, say [+M]
, while other morphemes lack the diacritic (or
are marked [−M]
). (In addition to roots, I assume that certain suffixes, such
as the adjectival -y may also bear the diacritic.) The grammar is simple: at
point (226a) in the derivation, if the abstract adj node is marked [+M]
, then
Merger applies, else Merger does not. The interesting variation lies then in
the course of acquisition, namely, how a learner comes to know for a given
root (or affix) whether it has the diacritic, or not (or perhaps, whether the
diacritic is optional, as in polite – politer ∼ more polite).
I presume that the acquisition of diacritic features involves a number of
converging sources of evidence, available to the learner. Positive evidence in
the input (for example, hearing and successfully parsing older, longer) leads
the learner to posit the diacritic for high-frequency forms that do indeed bear
242
the diacritic (including, for example, the suffix -y). Otherwise, statistical regularities
in the input yield partial information about the distribution of the
diacritic. In this manner, I assume, the acquirer can soon deduce that long
adjectives (greater than three syllables) compare analytically, rather than
synthetically, and can use this information in assigning the value [−M]
to the
roots of long adjectives. In the case of experiments involving nonce adjectives,
the task the speaker is faced with is to judge the chance that the new
root “should” bear the diacritic. Evidently, given the phonological make-up
of the items in Graziano-King’s (admittedly small) sample, the results came
out at chance: speakers judged it to be roughly 50% likely that these words
would bear the diacritic [+M]
, based, by hypothesis, on their tacit knowledge
of statistical patterns in their lexicon.29
I would wager that a more extensive
investigation would yield neighbourhood effects (perhaps involving a combination
of phonological and semantic factors) of a familiar kind at play. It is
also worth noting that the distribution of morphological versus periphrastic
comparatives has been relatively unstable in the history of English (Kytö
and Romaine 1997, González-Díaz 2006), again, the kind of behaviour that
one might expect under a diacritic-learning scenario such as that proposed
here.30
If diacritic features such as [+M]
are indeed part of the representation of
the abstract roots (i.e., the ones I have been writing throughout in CAPS),
then there is no ordering paradox. In terms of derivation, Merger does indeed
precede vocabulary insertion, and the operation Merger cannot be, and
243
is not, sensitive to the phonological make up of the root. On the other hand,
phonological information, such as syllable count, can play a role in regulating
the application of merger, but only indirectly, in the sense that the phonological
patterning is one potential source of evidence guiding the learner in
the acquisition of a diacritic feature.
At least superficially, the pattern with morphological comparatives appears
to me to be reminiscent of (the acquisition and distribution of) other
diacritics such as gender assignment, strong versus weak verbs in Germanic,
and the like. There are partial regularities that may generalize (e.g., by redundancy
rule) but no overarching phonological regularity. For example, in
Russian, certain word-forms provide no phonological cue as to gender. In one
study looking at nonce words ending in unstressed [a], which are ambiguous
between feminine and neuter gender, subjects chose feminine over neuter at
roughly a 3:1 ratio, closely tracking the relative frequency of these two genders
in the language (Tarasenkova 2010). The conclusion to be drawn, as
with Graziano-King’s 50% result for monosyllabic nonce adjectives, is that
for gender, in Russian, when phonological or other cues (such as natural gender)
are unavailable for particular items, speakers make an educated guess,
with an appeal to the statistical probability for each gender (feminine:neuter
= 3:1).
We may note in passing that the diacritic approach eliminates the ordering
paradox discussed here, and also avoids the widely-discussed ‘bracketing
paradox’ posed by forms such as unhappier, unfriendlier, unrulier (cf. Peset-
244
sky 1985, among many others). The compositional meaning of these forms is
clearly [ [ un-happy ] - er ] (i.e., ‘more unhappy’ rather than ‘not happier’).
The putative bracketing paradox arises on the assumption that there is a
grammatical prohibition on attaching -er to trisyllabic stems, which would
then be at odds with the semantically-motivated structure. But I have argued
above that there is no grammatical prohibition of this sort. The root (or
base) happy bears the diacritic [+M]
, and it is a general property of English
prefixes that they are transparent for the percolation of diacritics (see, e.g.,
Lieber 1980, 1982). The unhappier examples are simply a special case of this
broader phenomenon: the only grammatical bracketing is the semanticallymotivated
one [ [ un- [ happy ] ][+M]
-er ], in which unhappy inherits the
diacritic from happy.
There is clearly far more to be said on this topic, but it would take the
discussion far further afield from the main thread than it already has gone.
My narrow purpose in this section has been only to show how we might
escape the ordering paradox that seems to arise if (English) morphological
comparative formation is held to be implemented by morphological Merger,
but sensitive to phonological information that is not available to the derivation
until after Merger has (or has not) applied. Treating the alternation as
directly regulated by a diacritic, and only indirectly sensitive to phonological
sub-regularities in the course of acquisition, seems to be correct, and has the
welcome benefit of side-stepping the apparent ordering paradox.31
245
246
Chapter 6
Getting better: Comparison and
deadjectival verbs
6.1 Introduction
In the previous chapters, I have discussed in detail the Comparative Superlative
Generalization in (33):
(33) The Comparative-Superlative Generalization, part 1 (CSG1):
If the comparative degree of an adjective is suppletive, then the superlative
is also suppletive (i.e., with respect to the positive).
This generalization holds that, for a given suppletive alternation between
positive (A) and comparative (B) roots, the corresponding (morphological)
superlative will either inherit the root of the comparative (ABB), or involve
247
a third root (ABC), but will in no case be built on the positive root (*ABA).
Some examples of the licit patterns are give here:
(227) pos cmpr sprl
a. English: good bett-er be-st
b. English: bad worse wor-st
c. Czech : špatn-ý hor-ší nej-hor-ší ‘bad’
d. Estonian: hea pare-m par-im ‘good’
e. Latin: bon-us mel-ior opt-imus ‘good’
f. Welsh: da gwell gor-au ‘good’
In this chapter, we turn to an analogous generalization in the realm of
deadjectival, change-of-state verbs, a subset of the verbs known as ‘degree
achievements’ in the relevant literature. These are verbs which are to a
first (but possibly misleading) approximation, paraphrasable as ‘to become
(more) A’ or ‘to make something (more) A’, for some adjective A. Some
non-suppletive examples are given in (228):
(228) adjective verb
a. English: short short-en
b. English: cool cool
c. German: dick (ver)-dick-en ‘thick-thicken’
d. Russian: dlinn-yj u-dlin-jat’ ‘long-lengthen’
e. Finnish: suuri suure-ntaa ‘large-enlarge’
An influential analysis of these verbs originating in Dowty (1979) holds
that such verbs are derived from adjectives by means of a (possibly covert)
248
sentential operator, become, as in (229a). For transitive senses, Dowty
proposes a further operator cause as in (229b). In much of what follows, I
will simplify the exposition by abstracting away from the distinction between
inchoatives (intransitives) and causatives (transitives), using the symbol V∆
as a shorthand for the deadjectval verbalizing operator(s), and listing only
one verb from for each root in the tables.1
(229) a. [ become [ x cool ]]
b. [ cause [ become [ x cool ]]]
When we turn to adjectives that form suppletive comparatives, we find
with surprising regularity that the verb is formed from the comparative root
allomorph and not from the positive root, as illustrated in (230). A longer
list, with some 50 cognate triples, is given in Table 6.1, below.
(230) pos cmpr verb
a. English: good bett-er (to) bett-er
b. English: bad worse (to) wors-en
c. German : gut bess-er ver-bess-er-n ‘good’
d. Russian: plox-oj xuž-e u-xud-š-at’ ‘bad’
e. Finnish: hyvä pare-mpi para-ntaa ‘good’
f. Georgian: cud-i u-ar-es-i a-u-ar-es-ebs ‘bad’
g. (Late) Latin: bon-us mel-ior mel-i¯or-o ‘good’
What is striking here is not (just) the rather obvious point that words
like worsen contain the comparative form. Rather, what makes this parallel
249
to the superlative cases is that patterns with the positive root are often
impossible: English: *to bad(den), German: *ver-gut-en, Russian *u-plox-at’
etc. (apparent counter-examples will be discussed below). With appropriate
caveats, the following appears to be a contender for a valid generalization,
parallel to (33):
(231) The Comparative-Change of State Generalization (C∆G):
If the comparative degree of an adjective is suppletive, then the corresponding
change-of-state verb is also suppletive (i.e., with respect
to the positive adjective).
The generalization in (231) admits of attested patterns such as bad –
worse – (to) worsen and excludes patterns such as *bad – worse – (to) badden
and analogously for other languages. In light of (229), the C∆G is
surprising, if true. If [ become [ cool ] ] is a possible verb, then why
should [ become [ bad ]] be excluded, in favour of [ become [ worse ]]?
Since the pattern is formally isomorphic to the CSG, I suggest here that the
logic developed in chapter 2 should apply in this case as well. Just as the
representation of superlatives must always contain that of the comparative,
so too must the representation of deadjectival change-of-state verbs always
contain the comparative, even where that relation is not transparent in the
overt morphology. In place of (229), then, we have (232a) and not (232b):
(232) a. [ [ [ adj ] cmpr ] V∆]
b. * [ [ adj ] V∆]
250
Alternatively, the structure may indeed resemble (232b), but the meaning
of the operator is not Dowty’s become but rather a representation in which
V∆ node is internally complex, crucially containing the comparative (cf. Hay
et al.’s 1999 increase operator, or the differential measure function m∆ of
Kennedy and Levin 2008; see the end of this chapter for further dicussion), as
in (233b). Recall from above that in addition to strict nesting, as in (233a)
(=(232a)), two other structures that exclude ABA patterns (by virtue of
triggering elsewhere rule application) were branching affix structures (section
3.2.2) and feature bundles.
(233) a. b.
V
c
a
adj
cmpr
V∆
V
a
adj



V∆
cmpr



If a comparative-containing structure is posited for all cases, then the
account of the C∆G parallels that of the CSG developed above. In that
account, a key assumption was that root suppletion is to be treated as contextual
allomorphy, via rules of exponence whose interaction is governed by
the Elsewhere Principle. The vocabulary item (exponents) competing to realize
the abstract root bad are given in (234). The Elsewhere logic ensures
that the comparative allomorph wins out to express the root whenever it can
251
be inserted. The hypothesis that the superlative contains the comparative
ensures that the superlative is worst rather than *baddest, and the same
reasoning yields (to) worsen rather than *(to) badden, given (232a).2
(The
lack of an overt exponent of the comparative in verbs like (to) cool and (to)
shorten can be modeled by a straightforward extension of the domain of the
zero allomorph of the comparative in (42).)
(234) a. bad → worse / ] cmpr
b. bad → bad
While the proposal in (232a)/(233a) derives the C∆G, it stands directly
at odds with Dowty’s analysis, and those that have built on it (see Abusch
1986, 2005 and Rothstein 2004). One of the goals of this chapter is to show
that the evidence for Dowty’s analysis (essentially (232b)) is not compelling,
as against the alternative in (232a), and that there are existing semantic
proposals we can draw on that explain the same range of facts, starting with
a structure similar to (232a), especially Hay et al. (1999), Kennedy and Levin
(2008) and Winter (2006). In the final section of the chapter, I return to the
empirical basis for the C∆G, investigating a range of prima facie counterexamples.
I show that careful investigation reveals many of them to be in
point of fact consistent with the generalization, though a small residue of
troubling cases remains.
252
6.2 Preliminary remarks
Before delving into the meat of the chapter, though, a few preliminary remarks
are in order regarding some differences between the C∆G and the
CSG.
First, I argued at length above that the Containment Hypothesis for the
superlatives applied to periphrastic as well as morphological superlatives.
For reasons discussed, this is sometimes masked in the (surface) morphology
of periphrastic superlatives, as, for example in Russian and Tikhvin Karelian
and other languages discussed in section 3.4. Nevertheless, positing that
the containment hypothesis holds of the syntactic relationship anchors the
account of four proposed universals: the CSG parts 1 and 2, the RSG, and
the SSG. For change of state verbs, in contrast, it is far from clear that
periphrastic constructions should always contain the comparative. First,
periphrastic expressions (with become, make, get) systematically take both
comparative and positive complements, whether suppletive or not:
(235) a. The weather became worse.
b. The weather became bad.
(236) a. One pill makes you larger.
b. One pill makes you large.
The meanings of these alternations are transparent, up to the vagueness
inherent in gradable adjectives generally. Thus, (236a) clearly has comparative
semantics, entailing that the pill will make you larger (i.e., than you
253
are now), but with no commitment as to whether it will make you large in
any absolute sense. On the other hand, (236b) has exactly that implication,
namely that you will become large by whatever standard is contextually relevant.
It lacks a comparative sense, and because it is a verb of change of
state, it implies a change from not large to large. Even if it may be possible
to derive the containment hypothesis for the superlative from the meaning of
the superlative, and thus have it hold at all relevant levels of representation,
the periphrastic alternation in the verbs thus suggests that (232a) is more
narrowly proscribed; the semantics alone does not preclude (232b), and if
that is to be excluded, this must be a morphological fact.
A second issue worth remarking on briefly concerns the ABC pattern
such as Latin bonus – melior – optimus ‘good – better – best’, and relatedly,
the absence of AAB patterns: *good – gooder – best, which played a role
in the discussion of the CSG. For the deadjectival verbs in (230), I have
presented only ABB patterns, and one could well ask whether the C∆G
patterns with or against the CSG as regards other possibilities.3
For all
practical purposes, it seems to be impossible to provide a clear answer to
this question, for reasons having to do with both the (non)-productivity of
category-chaning morphology and near polysemy in lexical (verbal) roots.
Consider, for example, the question marks in (237).
254
(237) pos cmpr verb
a. many mo-re ???
b. small small-er ???
c. tall tall-er ???
What are the deadjectival change-of-state verb corresponding to many –
more and small? There are no verbs in (standard) English built on these
roots: *(to) (en-)more, *(to) (en-)many, *(to) small(-en); thus if there is
anything in the final column of (237), these must be suppletive. By meaning
alone, perhaps (to) increase or (to) multiply might fit the bill for (237a),
yielding an ABC pattern (but see section 6.4.5 for reasons to doubt that
multiply belongs here). And perhaps something like (to) shrink, (to) reduce,
(to) diminish, (to) lessen belongs in (237b), with (to) grow in (237c) for AAB
patterns. Or not — it is rather doubtful that these should be considered to
constitute suppletive forms, standing in a paradigmatic relation with the
adjectival roots. Far more plausible, it seems to me, is that these are simply
unrelated verbs that are close in meaning and can be pressed into service
to fill a gap.4
In any event, with no clear way to establish that there is
suppletion here, as opposed to a gappy area of lexical derivation, I leave
the possibility of ABC and AAB patterns aside and focus on the contrast
between attested ABB and (largely) unattested *ABA patterns.
Note, relatedly, that positing the nested structure in (232a) suggests the
expectation of an analogue to the SSG: deadjectival degree achievements containing
an adjectival root may only be possible for those roots identified as
255
combinable with a morphological comparative in the first place (see section
3.4).5
Interestingly, this appears to be correct, at least for contemporary English
-en suffixation (shorten) and zero derivation (to cool). All the deadjectival
change of state verbs from gradable adjectives listed in Levin (1993, 245)
embed adjectives that are compatible with -er suffixation for comparatives.
While this seems to provide prima facie support for postulating an analogue
to the Containment Hypothesis for deadjectival degree achievements, there
are other means of forming deadjectival verbs that appear to run against an
extension of the SSG to the verbal domain. For example, verbs in -ize, -ify
may be derived from adjectives that do not permit morphological comparatives:
modern – *moderner – modernize, solid – *solider – solidify. Quite
possibly of relevance here is the distinction between inner (root-attached,
category defining) and outer (category-changing) affixation (Marantz 1997,
2007), to which I return below.
A third remark of note concerns the transparency of embedding. In chapter
3, I noted that systematic embedding of the comparative morphology
(affix) in the superlative is widely attested cross-linguistically, even though
it is synchronically invisible in languages like German and English. When
we turn to the verbal domain, here too we find the comparative affix contained
inside the verb form in a number of examples, as in the majority of
the suppletive roots in (230) and also examples without suppletion in (238):
256
(238) pos cmpr verb
a. English: low low-er (to) low-er
b. German: schön schön-er ver-schön-er-n ‘beautiful’
c. German: groß größ-er ver-größ-er-n ‘big’
d. Late Latin: cert-us cert-ior cert-i¯or-o ‘certain’
Yet the evidence for transparent nesting is weaker for the verbs than it
is for the superlatives. No language (so far as I know) shows systematically
nested morphology in the verbs, the way many languages do in the superlatives.
In German, for example, while some verbs follow the pattern in (238),
others do not (see (228)).6
Moreover, while the comparative morphemes are
frequently identifiable in the superlative examples (230), examples like (238)
are relatively uncommon outside of suppletion. German has a number of
such verbs, but English (to) lower is the only verb derived transparently
from a non-suppletive comparative adjective, and the same may be true of
Late Latin certi¯oro, to judge by a search of Lewis and Short (1879). I will for
the most part abstract away from this difference in the frequency of transparent
nesting between superlatives and verbs in this discussion, thus leaving
an important question unanswered.
I note finally that the discussion in this chapter, and the generalization
presented here, are necessarily more tentative than the preceding chapters.
One significant practical limitation is that the topic of deadjectival verbs
is far less systematically covered in the descriptive grammars, and thus the
empirical domain over which this can readily be tested is more limited. I
257
contend nevertheless that the patterns described here are sufficiently robust
as to encourage further exploration of the C∆G as a candidate linguistic
universal.
6.3 Deadjectival Degree Achievements: Doubting
Dowty
Dowty (1979) famously discussed de-adjectival verbs of change-of-state, noting
that they show apparently ambiguous behaviour with regards to aspect
(telicity) diagnostics. Thus, a verb like (to) cool may appear in both telic
(239a) and atelic (239b) frames, as diagnosed (roughly) by the presence of
time frame (in an hour) and durative (for an hour) adverbials, respectively:7
(239) a. The soup cooled in an hour.
b. The soup cooled for an hour.
This behaviour is puzzling, to the extent that telicity and atelicity should
be mutually exclusive, and thus a given predicate should behave one way or
the other. Compare, in particular, the behaviour of periphrastic expressions
corresponding to (239), in (240). For these, standard telicity diagnostics
yield generally unambiguous results:8
(240) a. The soup got cool in an hour / *for an hour.
b. The soup got cooler for an hour / *in an hour.
258
As Dowty and others have noted, for the deadjectival degree achievements,
there appears to be a prima facie correlation between telicity and
whether or not the positive or comparative degree of the adjective is implied.
Thus, (239a) is most natrually understood as synonymous with (240a) implying
that the soup in fact became cool, whereas (239b) implies only that
the soup became cooler (i.e., than it was), as (240b), though it need not have
actually become cool. In both pairs, only the atelic versions can felicitously
be continued by the phrase ... but it never became cool.
There is a large literature and ongoing debate on this topic. For some
recent treatments, see Hay et al. (1999), Rothstein (2004), Abusch (2005),
Winter (2006), Kennedy and Levin (2008), Kearns (2007), and for a recent
critical overview Kriz (2011). My aim here is not to dive too deeply into this
debate, but to consider only the one aspect of the debate that is of direct
relevance to the C∆G. Specifically, the morphological fact (if it is a fact)
that suppletive allomorphs are always used for a deadjectival verb, when
such an allomorph is available, strongly suggests that the deadjectival verb
always contains the comparative, even when the comparative morpheme is
not overt. The question then is whether the facts discussed in the semantics
literature threaten that conclusion. I will argue, of course, that they do not,
relying in particular on the spirit (though not the details) of proposals in
Hay et al. (1999), Winter (2006) and Kennedy and Levin (2008).
The most potentially problematic view from the semantics literature, for
our perspective, is the original proposal in Dowty (1979), extended in Abusch
259
(1986, 2005), see also Rothstein (2004, 189). Dowty posits that the verb (to)
cool never contains a comparative morpheme, and embeds the positive degree
of the adjective in all readings, as in (229), repeated here:
(229) a. [ become [ x cool ]]
b. [ cause [ become [ x cool ]]]
Dowty proposes to derive the apparent comparative sense of the atelic
expression in (239b) via the pragmatics, exploiting the context-dependent
vagueness of the gradable predicate cool. In brief, his account runs as follows:
Start from the assumption that x cooled always means x became cool, in
other words, that x underwent a change of state from not cool to cool. This
explains the telic case. Now note that cool, as a gradable adjective, does
not correspond to any absolute (temperature) value, but what counts as cool
varies depending on context. What counts as cool for a star is an order of
magnitude hotter than what counts as a cool summer evening, while, on the
same cool summer evening, a beer at the ambient temperature could hardly
be described as a cool refreshment. Dowty proposes that the atelic use of (to)
cool involves an iteration of (telic) changes, from not-cool to cool, with the
contextual standard of cool-ness shifting along the way. As Dowty expresses
it: “The soup cooled for 10 minutes [is true if] ... for each time t within an
interval of 10 minutes duration, there is some resolution of the vagueness of
the predicate cool, by which the soup is cool is true at t but not true at t−1”
(Dowty 1979, 90). He offers the diagram in (241), where P is cool to visualize
260
this:
(241) t−3 t−2 t−1 t
¬ P P
¬ P P
¬ P P
As the soup changes temperature, say, from 100° to 97°, the contextuallydetermined
threshold for cool goes through a progression of changes. At the
first instant, 99° is considered cool and thus the first degree of temperature
loss satisfies the change from not cool to cool; next, the threshold is reset
to 98°, and thus the next change, constitutes again a change from not cool
to cool, and so on. An important assumption here is that while the actual
temperature constituting the boundary between cool and not cool is context
dependent, it is always the case that if some temperature D counts as cool,
then all lower temperatures must also count as cool. This approach ensures
that (to) cool denotes a monotonic decrease in temperature. The internal
semantics of change overrides any default contextual value for cool that may
occur in the phrase the soup cooled in isolation.
In this proposal, the event described by the expression x became cool
involves an iteration of sub-events, each of which is a itself a change from
not-cool to cool (thus also describable as x became cool), and hence, the
overall predicate counts as atelic under standard definitions (Bennett and
Partee 1978, Dowty 1979).9
Dowty’s account stands squarely at odds with the suppletion facts. If
261
the comparative-like readings are always derived from the positive form of
the adjective, then there is no reason why the comparative allomorph of the
adjective root should ever be permitted, let alone required, in the verb. Yet
insofar as I can tell, neither Dowty, nor any other proponent of this general
approach, has argued that the kind of calculus just discussed is the only
means for deriving the atelic meaning. Indeed, Dowty himself commented
that the account in (241) “avoids having to derive The soup cooled from the
morphologically unmotivated BECOME [ The soup is cooler] rather than
simply BECOME [the soup is cool]” (Dowty 1979, 90). This suggests that,
aside from the morphology, there was no compelling reason from the semantics
to take the non-comparative representation as basic in all cases.
Conceivably, then, one could imagine a theory that countenanced two
distinct derivations for deadjectival degree achievements, one with and one
without the comparative, permitting (242b) alongside (242a).
(242) a. [ [ adj ] V∆]
b. [ [ [ adj ] cmpr ] V∆]
This view would permit a committed proponent of a Dowtian account to
maintain that account for (239), while admitting of the derivation in (242b)
for verbs with the suppletive roots (such as (to) worsen, as well as for forms
from other languages in (230)), as well as for the verbs (not typically considered
in the semantic literature) with regular comparative morphology. In
English, there is only one such form, namely (to) lower, but there are more,
262
for example, in German, such as vergrößern ‘to enlarge etc.’ < größer ‘larger’
< groß ‘large’, cf. (238).10
An approach positing two representations along these lines seems inadequate,
however.
In the first place, if both representations in (242) are available, then we
would expect suppletive adjectives to contribute two deadjectival verbs, distinguished
by meaning and telicity. From bad, the derivation in (242a) would
yield a telic verb (to) badden alongside uniquely comparative, thus atelic, to
worsen. But this is not the case. Generally (though there are exceptions),
deadjectival degree achievements from positive roots are unattested when a
suppletive comparative root exists, regardless of telicity. The only verb (pace
remarks in section 6.4.2) corresponding to English bad is (to) worsen, and it
shows the same variable behaviour with respect to telicity as (to) cool does:
(243) a. The weather worsened in an instant.
b. The weather worsened for days.
The same point can be made with respect to verbs that embed regular
comparative morphology. These, too, show the same ambivalent behaviour
with respect to telicity diagnostics that is shown by verbs like (to) cool:
(244) a. The new air conditioner lowered the temperature in an instant.
b. We lowered the rope for hours (but never reached bottom).
Similarly, in German, time-span (in 1 Jahr ‘in 1 year’) and durative (1
Jahr lang ‘1 year long’ = ‘for 1 year’) adverbials correlate with the telicity
263
of the the VP (see Musan 2002; note that as in English, there are numerous
further complexities beyond the basic case, not considered here). Here too,
to the extent I have investigated this, deadjectival degree achievements show
ambivalent behaviour on telicity diagnostics, whether or not they morphologically
contain the comparative. The pair in (245) illustrates this with a
suppletive root, while those in (246)-(247) illustrate with regular, i.e., nonsuppletive
adjectives.11
(245) a. Die
the
Situation
situation
hat
has
sich
refl
1
1
Jahr
year
lang
long
ver-bess-er-t
pref-bett-er-prtcp
‘The situation improved for one year.’ (atelic)
b. Die
the
Situation
situation
hat
has
sich
refl
in
in
1
1
Jahr
year
(um
(by
ein
a
Vielfaches)
severalfold)
ver-bess-er-t
pref-bett-er-prtcp
‘The situation improved (severalfold) in one year.’ (telic)
(246) a. Das
the
Geschwür
abscess
hat
has
sich
refl
1
1
Jahr
year
lang
long
ver-größ-er-t
pref-big-er-prtcp
‘The abscess grew for one year.’ (atelic)
b. Das
the
Geschwür
abscess
hat
has
sich
refl
in
in
1
1
Jahr
year
(um
(by
1
1
cm)
cm)
ver-größ-er-t
pref-big-er-prtcp
‘The abscess grew (1 cm) in a year.’ (telic)
264
(247) a. Der
the
Schneider
tailor
hat
has
die
the
Hose
trousers
eine
1
Stunde
hour
lang
long
ver-läng-er-t.
pref-long-cmpr-ptcp
‘The tailor lengthened the trousers for one hour.’ (atelic)
b. Der
the
Schneider
tailor
hat
has
die
the
Hose
trousers
in
in
einer
1
Stunde
hour
ver-läng-er-t.
pref-long-cmpr-ptcp
‘The tailor lengthened the trousers in one hour.’ (atelic)
What we can conclude from these examples is that variable telicity is not
a function of the presence or absence of comparative morphology. Deadjectival
verbs show ambiguous behaviour on telicity diagnostics, and in particular
are compatible with telic frames, even when they transparently embed comparative
morphology.
In other words, the representation in (242b) is compatible with both telic
and atelic readings. The evidence from suppletion (the C∆G) is that this is
the only representation available for suppleting adjectives, and that (242a)
is unavailable for these adjectives. Why might that be? The strongest view
would be that the representation in (242a) is always unavailable; that the
relevant deadjectival verbs are always derived from comparatives, but that
outside of suppletive contexts, this relationship is often masked on the surface
265
by a null comparative allomorph (42), an entity which was independently
appealed to in the preceding chapters.
Note also in this context that questions arise as to the meaning and combinatorial
properties of V∆ in (242). Roger Schwarzschild points out (personal
communication 2006) that allowing an operator like Dowty’s become to
combine with a comparative as well as the positive form of the adjective may
make incorrect predictions. If become is (as Dowty held it to be) a sentential
operator, then allowing it to embed a comparative as in (242b) would
predict that the arguments of the comparative predicate would be allowed
with the derived verb, just as they are in periphrastic embeddings. For the
standard argument (expressed as a than-clause), this is strikingly false: a
true embedding under become freely allows a than-clause (see (248b)), but
inclusion of a than-clause with the deadjectival verb is simply horrendous
(248c):
(248) a. The soup is cooler (than it was / than the gravy).
b. The soup became cooler (than it was / than the gravy).
c. * The soup cooled {than it was / than the gravy}.
The operator V∆ that embeds the comparative in (242b) thus cannot
be Dowty’s become operator, but must, at the least, contribute the meaning
‘than x was before’ (more precisely, immediately before the event). In
this way, the verbalizing element is similar to the element sprl that embeds
the comparative (discussed at length above). The element sprl means
266
(effectively) ‘than all (others)’ and uses up or binds the argument slot for
standard in the comparative it embeds. Superlatives are thereby incompatible
with a than-clause: Leo is taller than the other children versus Leo is the
tallest (*than the other children) (see also discussion of (135) in Chapter 3).
Now, one might take a different perspective on the facts in (248). Den
Dikken et al. (2010) take issue with the proposal given here and argue that
the impossibility of a than-phrase with the deadjectival verb (248c) shows
instead that there is no comparative morpheme at all within the verb, and
that an alternative semantics should be sought. But the impossibility of an
overt expression of the standard argument with deadjectival verbs such as
to cool cannot be taken as an argument against the presence of a (silent)
comparative element in those verbs, since the same pattern holds when the
comparative element is overt, whether in a suppletive (249) or transparently
compositional (250) derivation:
(249) a. Our chances are better than (they were) yesterday.
b. Our chances became better than (they were) yesterday).
c. * Our chances bettered than (they were) yesterday
267
(250) a. Die
the
U3
U3
ist
is
läng-er
long-cmpr
(als
(than
sie
it
früher
earlier
war
was
/
/
als
than
die
the
U4)
U4)
‘The U3 (subway) is longer (than it was / than the U4)’
b. Sie
they
haben
have
die
the
U3
U3
läng-er
long-cmpr
gemacht
made
(als
(than
sie
it
früher
earlier
war
was
/
/
als
than
die
the
U4).
U4)
‘They have made the U3 longer (than it was / than the U4.’
c. * Sie
they
haben
have
die
the
U3
U3
ver-läng-er-t
pref-long-cmpr-ptcp
(als
(than
sie
it
früher
earlier
war
was
/
/
als
than
die
the
U4).
U4)
‘They have made the U3 longer (than it was / than the U4.)’
Examples such as the above point to the verbal operator as contributing
the meaning ‘than x was before’ in a deadjectival degree achievement. The
presence of this element, like the superlative, uses up the standard-denoting
argument of the comparative, and prevents the expression of a than-phrase.
These considerations remove the major hurdle from Dowty’s analysis for
the general theoretical view put forward here. No compelling reason remains
in the literature surveyed for uniformly deriving degree achievements from
positive adjectives directly. Variable telicity (and incompatibility with a
than phrase) arise even in the presence of overt comparative morphology in
268
the verbs, and are not consequences of embedding a plain adjectival root.
What remains to be shown is how the telic readings might be captured by a
representation that always includes the comparative, even where that element
is not overt.
Various authors, including Hay et al. (1999), Kennedy and Levin (2008)
and Winter (2006) have recently offered proposals that begin to answer this
question. While these proposals do not specifically embed the comparative
morpheme in the verb, the semantics that they offer comes close to doing so.
Such proposals dovetail nicely with the considerations needed to derive the
C∆G. By means of illustration, let us begin with the proposal in Hay et al.
(1999). Simplifying considerably, the leading idea is as follows:
Start from the assumption that x cooled means x became more cool /
became cooler (i.e., than it was before; the Hay et al. proposal combines
‘become’ and ‘more’ into a single operator increase). The atelic cases are
thus straightforward. All else being equal, x became cooler is expected to
behave as an atelic expression for the same reason that (240b) is atelic. In
a standard way of thinking about this, become cooler has the ‘sub-interval
property’ (Bennett and Partee 1978, Dowty 1979) – if become cooler is true
of an event e, then become cooler is also true of the sub-events that constitute
e (up to the limits of granularity). But if x cooled is necessarily atelic, how
are the telic cases to be accommodated?
Hay et al. (1999) note that degree achievements may take an argument
that denotes the extent of the change involved, as in (251). Laying aside
269
iterative and other special contexts, a specific extent argument contributes
a definite endpoint to the event, rendering it telic. Although not discussed,
the extent argument also renders the explicitly comparative periphrastic expression
in (240b) telic (see (251c)):
(251) a. The soup cooled (by) two degrees (in an hour).
b. The soup cooled to 20°C (in an hour).
c. The soup got two degrees cooler (in an hour).
It is relatively clear why the extent argument should have this effect. As
just noted, becom[ing] cooler has the sub-interval property – since temperature
is a continuous scale, any event that is a downward change in temperature
is itself made up of sub-events that are also downward changes in
temperature. But once the extent argument is added, this no longer holds,
becom[ing] 2° cooler lacks the sub-interval property, as it is not made up of
sub-events that are in turn instances of becoming 2° cooler. However this is
to be formalized, the extent argument in these examples thus plays the role
of a ‘measuring out’ expression (Tenny 1987, 1994) or ‘incremental theme’
(Dowty 1991), just as the direct object does in examples such as the oft-cited
(252):
(252) a. Mary drank a glass of beer (in a minute).
b. Mary drank beer (for an hour).
In the pair in (252), the (a) sentence is telic, because no sub-part of the
event of drinking a glass of beer is an event of drinking a glass of beer. The
270
quantized nature of the object glass of beer sets the endpoint for the event.
Within a certain tolerance, the event is not completed until the last drop is
drunk. By contrast, the mass noun beer provides no set endpoint, and an
event of drinking beer consists of multiple proper sub-parts, each of which is
itself an event of drinking beer.
Hay et al. (1999), and with somewhat different technology, Rothstein
(2004) and Kennedy and Levin (2008), argue that otherwise atelic degree
achievements can be rendered telic by the addition of an explicit end-pointdenoting
expression. Hay et al. (1999) and Kennedy and Levin (2008) extend
this approach to the apparent ambiguity in (239) and argue that a silent
analogue of an extent argument may perform the same function even in the
basic cases. For a sentence such as The soup cooled, context and world
knowledge may make accessble an endpoint such as to the point where it is
(too) cool (to eat), yielding the apparently telic behaviour in (239a). To the
extent that the salient endpoint coincides with the meaning of the positive
adjective, the telic use appears to coincide with a positive paraphrase (240a),
although a technically more accurate prose rendering of the semantics would
be something like: The soup has become cool-er to the extent where (we would
now say) it is cool.12
Much work in this vein is devoted to showing the context-sensitivity of
these constructions, and in particular, how different semantic classes of gradable
adjectives vary in the degree to which they make a conventional endpoint
accessible. To see this, consider another telicity diagnostic in English, using
271
the contrast in (252) as a reference point. All else being equal, an imperfective
(progressive) sentence with an atelic VP appears to imply the corresponding
perfective, but the inference fails for a telic predicate:
(253) a. Mary is drinking beer. ⇒ Mary has drunk beer.
b. Mary is drinking a glass of beer. Mary has drunk a glass of
beer.
At first blush, deadjectival degree achievements show inconsistent behaviour
on this test:
(254) The workers are widening the road. ⇒ The workers have widened
the road.
(255) Kim is straightening the rod. Kim straightened the rod.
Yet at least in part this may reduce to the semantics of the adjective in
question, and in particular the associated scale (see especially Rotstein and
Winter 2004). The adjective straight is closed, with a fixed upper bound
beyond which it is impossible to be straighter. Hay et al. (1999) contend
that the closed maximum of the scale (allowing for context-sensitivity in
the granularity of the measurement) provides a naturally salient endpoint
and thus facilitates the telic reading. On the other hand, wide is an openclass
adjective, and thus provides no intrinsic natural endpoint to the event,
leaving the atelic reading more salient. Clearly, there are many additional
factors at play (see Rotstein and Winter 2004, Rothstein 2004, Winter 2006,
272
Kearns 2007, Kennedy and Levin 2008), but the general point remains that
telic-like uses, correlating with the apparent positive paraphrases (become
adjective) may well be accounted for by a complex interplay of factors, even
if the semantic representation is indeed comparative in all instances.
At least to a first approximation, the discussion above jibes with the
observation, raised at various presentations of this material, that aspectual
elements (such as Slavic prefixes and Germanic particles) appear to interact
with perceptions of a meaning distinction between ‘become X’ and ‘become
more X’ readings.13
I suspect that, like English, the appearance of a meaning
‘become X’ (rather than X-er) is nevertheless secondary, a result of the
interaction of a basically comparative meaning for the verb on the one hand,
and the aspectual (in the sense of Aktionsart/telicity, not perfectivity) contribution
of the prefix. If it is at all plausible to think of prefixes as able
to contribute meanings like ‘to the maximal extent’ or ‘a little bit / somewhat’,
then the source of the apparent meaning difference can be located in
the boundedness of the degree/extent, without threatening the account given
here. Indeed, something like this is independently needed, I would contend,
for English particle constructions, for example in the difference between to
dry and to dry up/out. The theoretical discussion above commits me to the
view that to dry is [ become [ dyr-er ]], with the implication of a maximal,
and thus bounded, extent being contributed by the particle up/out. An
analysis of the complex system of Slavic aspectual prefixes is well beyond
the scope of this paper, and thus I leave this as an open challenge, with the
273
hope that further investigation will reveal nothing more sinister in the aspectual
prefixes than what is independently needed to describe their properties
outside of the realm of deadjectival change-of-state verbs.14
A final remark of relevance in this regards builds on an observation by
Klaus Abels (personal communication, 2006). Abels noted that the change
of state verb like to cool on standard approaches should denote a transition
from not cool to cool, but since cool denotes an intermediate range on the
temperature scale, there should be two directions in which such a change
could occur. If we agree for the sake of argument that 20°C is cool, for soup,
then the proposition The soup cooled should be true either by virtue of undergoing
a change from hot to cool, or by undergoing a change in the opposite
direction, for example, by defrosting a block of frozen soup, and bringing it
to 20°C. In both cases, the soup starts out in a state that cannot be felicitously
described with The soup is cool and ends up in such a state. As Abels
notes, The soup cooled is utterly impossible (unquestionably judged false) in
the thawing context.15
A standard reply might appeal to the pragmatics of
scales, holding that frozen soup is indeed cool, but that The soup is cool is
judged inappropriate due to the availability of a stronger proposition, namely
The soup is cold. However, this direction is undermined by the further observation
that the periphrastic expression The soup became (merely) cool is in
fact available (if somewhat awkward) in the thawing context. In other words,
The soup became cool means exactly what it should – a transition from not
cool to cool, with the contextually-sensitive nature of cool held constant. By
274
contrast, The soup cooled cannot mean this, and patterns instead with the
periphrastic expression that embeds the comparative, requiring the endpoint
be cooler than the starting point of the event described.
In sum, the suppletive evidence suggests that the comparative morpheme
is always embedded in the coresponding verbs. This assumption is needed
to force the comparative root allomorph, whenever one is available. The
facts from the behaviour of verbs that overtly embed comparatives, as in
(243)-(246), clearly indicate that the semantics must be able to derive the
relevant range of meanings from this structure, including absolute-like, telic
meanings. The brief review of the relevant literature just presented shows
that there are contenders in the available literature for what such a semantics
might look like. This significantly dilutes the force of a possible objection
based on the success of Dowty (1979) in describing the ambiguity; there is
no compelling reason to believe that the comparative cannot be contained in
the verbs (even where it is not there overtly).
6.4 To good, to badden, and to many
We now turn to the empirical side of things and the investigation of some
apparent counter-examples to the C∆G. Note first that the C∆G is narrowly
formulated to apply to change-of-state verbs derived from gradable
adjectives. The claim, by parity of reasoning to the account of the CSG, is
that there exists an abstract morpheme V∆, and that this morpheme may
275
only combine with comparative stems (see (242)). Other morphemes may
exist which derive verbs from adjectives, which are not restricted, or even
expected, to embed the comparative. Russian, for example, has deadjectival
verbs which are ambiguous between stative and change-of-state readings,
such as that in (256):16
(256) bel-yj ‘white’ → bel-et’ ‘(to) whiten’ or ‘(to) be white’
Another verb type that is not subject to the C∆G is what might be
called ‘ascriptive’ verbs, with meanings such as belittle and exaggerate. Like
the stative verbs, their meanings do not denote a change of state (such as
coming to have a property). In current usage at least, to belittle someone is
not to cause them to become little, but rather to (attempt to) cause them to
appear little, or to decry their stature. Verbs with ascriptive meanings are
not expected to show suppletion, even when built on otherwise suppletive
adjectival roots. Russian again provides a nice minimal pair in this regards.
The adjective meaning ‘small’ shows suppletion in the comparative (257),
and there are two corresponding deadjectival verbs, one from each of the
roots. The verb with the basic change-of-state meaning is, as predicted by
the C∆G, formed on the comparative root (257a). On the other hand, the
verb from the positive root is an ascriptive, rather than a change-of-state
verb (257b):
276
(257) positive comparative verb
a. mal-yj men’-še u-men’-š-at’ ‘to shrink’
b. mal-yj men’-še u-mal-jat’ ‘to belittle’
Yet other verbs have only a tenuous synchronic connection to the adjective,
in semantic terms. Once again, Russian provides an example. The
adjective meaning ‘good’ is suppletive in the comparative: xoroš-ij – lučš-e,
and again there are verbs from each of the two roots. The verb from the
comparative root u-lučš-at’ has the regular change-of-state semantics, i.e.,
‘to improve’ (to make better). The verb from the positive root xoroš-et’
is also a change of state verb, but rather than meaning ‘to make/become
good’ it means ‘to become prettier’, and is thus in terms of meaning related
to a distinct adjectival root krasiv-yj ‘pretty, beautiful’. There is a historical
connection between xoroš-ij and xoroš-et’, but semantic drift has pulled
apart their meanings, with root xoroš- taking over the basic sense of ‘good’
(expressed by dobr- in the other Slavic languages).
The root for ‘good’ in a range of languages appears to be particularly
susceptible to semantic drift in deadjectival verb formation. For example, Old
English, Basque and Russian have verbs built on a basic (non-comparative)
root for ‘good’ which mean ‘to fertilize’ (presumably from ‘to make the land
good for farming’), while Czech and Bulgarian have verbs from the ‘good’
root meaning ‘to conciliate’ (cf. the English fixed expression ‘to make good’).
Some of these verbs have additional meanings, but with the exception of
Old English, these verbs have a drastically limited range of meaning when
277
compared to the basic adjectives. While there are thus some fairly clear
cases where the relation between the verb and the adjective is diachronic
and not synchronic, there remains a rather sizeable grey area. Thus English
(to) worsen and especially (to) better do not have the full range of meanings
available to the corresponding adjectives, yet seem (to me) nevertheless to
be close enough in meaning to the adjectives to count as the corresponding
change-of-state verbs, thus relevant to evaluating the C∆G. I acknowledge,
though, that this opens up something of a slippery slope, and thus that the
accuracy of the C∆G will depend on a more precise resolution of some of
these considerations, something I leave to future research.
Verbs built on the root for ‘good’ illustrate a further point that is relevant
here. In some languages, the root meaning ‘good’ has both adjectival and
nominal senses and corresponding morphosyntax. In English (and Dutch
and German), for example, there is a noun good(s) meaning something like
‘property or possessions’ (the goods will be delivered. . . ’). Verbs with the
root good may, with appropriate semantics, be derived from this noun, rather
than (directly) from the adjective root good. Thus we find Dutch ver-goed-en
‘compensate’ (i.e. ‘to pay back for something in money or goods’) seems
to be derived from the noun goed ‘good(s)’ (and similarly, one sense of Old
English gódian ‘(to) good’ reported in the OED). There is certainly no reason
to expect comparative morphology (and thus no reason to expect comparative
suppletion) in these denominal constructions.
278
With these points in mind, Table 6.1 provides a listing of cases that satisfy
the C∆G. As with the CSG data, I have counted cognate triples, listing for
each triple only one example (and where there are multiple, related verbs, for
example, with different derivational morphology, I have listed only one verb).
There is of course overlap still, in particular where positive roots with a given
meaning differ among related languages, but there is a shared comparative
and verb. Note that many of the examples listed participate in doublet
patterns, existing alongside a regular AAA pattern from the same root. Note
also that I have not attempted to distinguish the range of meanings associated
with a given verb. Thus, English (to) better corresponds to only a very
restricted range of the semantics of make/get good/better, with improve being
the far more general term. This contrasts with German verbessern which has
a much wider distribution than the English (partial) cognate.
279
Table 6.1: Deadjectival verbs from suppletive adjectives
gloss adj cmpr verb Language; Cognates
good good bett-er bett-er English; oth. Germanic
good dobry lepsz-y po-lepsz-yć Polish; Czech
good xoroš-ij luč-š-e u-luč-š-at’ Russian
good dobar bol-ji po-bolj-š-ati Serbo-Croatian; Slovenian
good



dobr-yj
harn-yj



krashch-yj po-krashch-yty Ukranian
good bonus mel-ior mel-ior-are Latin; Modern Romance
good da gwell gwell-a Welsh
good agath-ós belt-í¯on belt-i-ó¯o Anc. Greek
good agath-ós kreíss-¯on kreiss-ó¯o Anc. Greek
good on hobe hobe-tu Basque
good hyvä pare-mpi para-ntaa Finnish
good k’argi-i u-mˇob-es-i a-u-mˇob-es-ebs Georgian
280
Table 6.1: Deadjectival verbs from suppletive adjectives
gloss adj cmpr verb Language; Cognates
bad bad worse worse-n English
bad



dårlig
ille
ond
vond



ver-re for-ver-re(s) (seg) Norwegian; oth. Scandinavian
bad dålig säm-re för-säm-ra(-s) Swedish
bad šlext erg-er var-erg-er-n Yiddish
bad



špatn-ý
zl-ý



hor-ší z-hor-š-ít (se) Czech
bad plox-oj xuž-e u-xud-š-at’ Russian
bad pohan-yj hir-š-yj po-hir-š-yty Ukranian
bad mal-us p¯ej-or pej-or-are Latin; Modern Romance
bad dolent pitjor em-pitj-or-ar Catalan
281
Table 6.1: Deadjectival verbs from suppletive adjectives
gloss adj cmpr verb Language; Cognates
bad cattiv-o peggiore peggior-are Italian
bad drwg gwaeth gwaeth-ygu Welsh
bad kak-ós cheiró-ter-os cheiro-ter-eýo Mod. Greek
bad kak-ós héss-¯on héss-áomai Anc. Greek
bad cud-i u-ar-esi a-u-ar-es-ebs Georgian
big velk-ý vět-ší z-vět-š-ít Czech; Polish, Serbo-Croatian
big velyk-yj bil’-sh-yj z-bil’-sh-yty Ukrainian
large, big mawr mwy mwy-hau Welsh
much, many viel mehr vermehr-en German; oth. Germanic
many,much asko gehi-ago gehi-ago-tu Basque
much paljo-n ene-mmä-n ene-tä Finnish
many bevri met’-i amet’ebs Georgian
many küp artıq arttı-rırˇga Tatar
282
Table 6.1: Deadjectival verbs from suppletive adjectives
gloss adj cmpr verb Language; Cognates
small mal-i, malen man-ji uman-ji-ti Serbo-Croatian; oth. Slavic
small, few



lille,lidt
små



mind-re (for-)mind-ske(-s) Danish; oth. Scandinavian
small, little parv-us; paul-um min-or min-or-are Latin; Modern Romance
few weinig mind-er vermind-er-en Dutch
few, (a) little



few
(a) little



less less-en English
small bach llai llei-hau Welsh17
small mikr-ós meí-¯on mei-ó¯o Anc. Greek
small mikr-ós eláss-¯on elass-ó¯o Anc. Greek
little, few olíg-os héss-¯on héss-áomai Anc. Greek
little, few olíg-os meí-¯on mei-ó¯o Anc. Greek
few cot’a na-k’l-ebi -k’l-18
Georgian
283
Table 6.1: Deadjectival verbs from suppletive adjectives
gloss adj cmpr verb Language; Cognates
old gammel æld-re (for-)æld-es Danish; oth. Scandinavian
near agos nes nes-u Welsh
strong cryf trech trech-u Welsh
284
We may now move on to a consideration of some apparent challenges
to the C∆G. These challenges take the form of verbs that appear to derive
from the positive allomorph of roots that supplete in the comparative. I
aim to show here that a large number of apparent counter examples are
in fact not problems for the C∆G, once other relevant considerations are
taken into account. Even though accounts for all the apparent problems are
not available, it is worth noting that verbs apparently built on a positive
adjective are in a distinct minority, a surprising fact on a Dowtian approach
to deadjectival degree achievements, as noted above. Although a good deal of
space is devoted here to seeking alternate explanations for apparent counterexamples,
it should be borne in mind that the number of apparent problems
is quite small in comparison to the 50 or so cases just presented that are all
consistent with the generalization (but which for that very reason require no
further discussion).
6.4.1 Ancient Greek goods
We may begin with a consideration of Ancient Greek, as it provides a relatively
straightforward case.19
The Ancient Greek adjective agath-ós ‘good’ is
a rich example of suppletion, entering into a one:many relation with a variety
of comparative allomorphs. The following is a standard listing, from Seiler
(1950), Chantraine (1967), repeated from (141):
285
(258) pos cmpr sprl
Anc. Greek: agathós ameín¯on ‘good’
(also): beltí¯on béltistos
(also): kreíss¯on krátistos
(also): lóïon lõistos
(also): phéristos
Additional suppletive comparative forms in -teros, such as bélteros, l¯o´ïteros,
phérteros, with corresponding superlatives in -tatos, are also attested. Verbal
derivatives from the suppletive comparatives are well attesteed, as expected
under the C∆G. However, verbs derived from the root agath- are also attested,
with (among other senses) the meaning ‘make good’ (thus: agath-ýn¯o,
agath-ó¯o (Liddell and Scott 1996), in apparent violation of the C∆G.
On closer inspection, with careful attention to sources, the counter-example
evaporates. The term ‘Ancient Greek’ spans a long history of the language
and an assortment of varieties. The problematic verbs are first attested in the
post-classical Koine Greek of the Septuagint (Old Testament), dating from
about the 3rd century BCE. And beginning in the Septuagint, regular (nonsuppletive)
comparative agath´¯oteros and superlative agath´¯otatos appear as
well (Liddell and Scott 1996). The appearance of a counter-example here
is thus an artefact of the standard presentation, which omits the regular
comparatives in tables such as (258), yet includes the offending verbs (they
have separate entries in Liddell and Scott 1996). But in fact, there are two
periods of Ancient Greek to consider, with no evidence of an ABA pattern
286
at any one point in the history of Greek. Through the Classical period, the
comparative and superlative were uniquely suppletive, and verbs were built
only on the suppletive allomorphs, while in the post-classical Koine period,
a regular (non-suppletive) comparative doublet arises, and with it, a regular
deadjectival verb.
6.4.2 Baddening (up)
A somewhat similar case comes from English. The (standard) English paradigm
for bad is suppletive, with comparative worse, and the corresponding verb is
(to) worsen. Yet colloquially, at least, the verb (to) badden (up) occurs sporadically.
A Google search yields a handful of examples such as the following
(there are no occurrences in the COCA corpus).20
(259) I’m trying to badden up my image, like that Sabrina the Teenage
Witch is trying to do. (http://www.melissajoanhart.net/sightings.
shtml, attributed to Ray Romano, speaking to Regis Philbin, April
2000)
(260) Or is it, as we suspect, part of a new programme to badden up the
image of the perennial good guy? (http://www.paulcarr.com/tft/
people/001223.php)
The context of these examples makes the meaning extremely clear. The
sense of bad is a positive, or at least a prestige, characteristic, corresponding
to senses IV.12 and 13 in the OED:21
287
(261) IV. (orig. U.S.) Formidable, good. 12. As a general term of approbation:
good, excellent, impressive; esp. stylish or attractive....
13. Originally in African-American usage. Of a person: (originally)
dangerous or menacing to a degree which inspires awe or admiration;
impressively tough, uncompromising, or combative; (in later use also)
possessing other desirable attributes to an impressive degree; esp.
formidably skilled.
And as noted in chapter 2, in exactly this sense, the adjective bad compares
regularly, not suppletively: bad – badder – baddest. Examples from the
OED are given here (repeated from (140)).
(262) a. We stayed in our Beemer for a bit, givin them a chance to check
out our badder alloy wheels, . . .
b. Sheep Eye’s here, and I’m the baddest sonofabitch that ever
moved.
What we have, then, is distinct senses of bad correlated with distinct morphological
patterning. The garden-variety bad compares suppletively (worse)
and that suppletion is carried over to the verb worsen. The hip bad by contrast
compares regularly (badder), and thus the verb corresponding to this
sense takes the plain root badden, an ABB and an AAA pattern, distinguished
by meaning. The examples in (259)-(260) turn out on inspection to
be exactly as predicted by the C∆G.
288
6.4.3 Moral goodness and ambiguity in Serbian
Another set of examples in which distinct senses of a single root play a role
is drawn from colloquial forms in some varieties of Serbian and Croatian.22
The adjectives meaning ‘good’ and ‘bad’ compare suppletively, following the
general pattern in Slavic, as in (263). As also shown, there are corresponding
verbs, with the expected meanings (‘to improve, to worsen’) derived from the
comparative root allomorphs. (The clitic se is the mediopassive which marks
the inchoative forms; I lay aside here questions of the distribution of transitive
and intransitive alternants.)
(263) pos cmpr verb
a. dobar bol-ji po-bolj-š-ati (se) ‘good’
b. zao (zl-) gor-i po-gor-š-ati (se) ‘bad’
c. loš gor-i po-gor-š-ati (se) ‘bad’
So far, so good. But alongside the standard forms in (263), the forms prodobr-iti
(se) and pro-zl-iti (se) are acceptable to some speakers, and occur in
various contexts. Not all speakers accept the verbs based on the positive root,
but those who do report differences in meaning.23
Unlike English badden, the
solution is not as simple as a doublet with a regularly comparing adjective.
Speakers who accept prodobriti se nevertheless do not accept a comparative
other than bolji.
Despić (2008) and Despić and Sharvit (To appear) explore the issues and
the meaning difference here in some detail. Despić’s findings are striking (the
289
following examples are all from his work). As in many languages, including
English, the adjective dobar ‘good’ has a variety of senses, including an intersective
reading and a non-intersective reading. Thus (264) is ambiguous,
and can either refer to a person who is ‘good’ i.e., skillful at being a thief
(the non-intersective reading), or can refer to a person who is both ‘good’ in
some absolute, typically moral, sense, and also a thief (for example, Robin
Hood).
(264) On
He
je
is
dobar
good
lopov.
thief
‘He is a good thief.’
For speakers that permit prodobriti se alongside standard poboljšati se,
this ambiguity appears to be resolved in the corresponding verbs. If a thief
gets better at being a thief, only the verb built on the comparative root is
available (265a). If the sentence in (265b), with the verb from the positive
root, is acceptable, then it means only that the thief became good in the
moral sense.
(265) a. Ovaj
This
lopov
thief
se
refl
po-boljš-ao
prf-better-tv.msg
‘This thief got better (more adept).’
b. Ovaj
This
lopov
thief
se
refl
pro-dobr-io
prf-better-tv.msg
‘This thief became (morally) good.’
290
Of particular relevance here is the further observation that comparative
morphology also disambiguates the senses of ‘good’. More specifically, the
comparative bolji, whether attributive or predicative, lacks the moral reading
available in the positive.24
The paradigm in (266) illustrates:
(266) a. On
He
je
is
dobar
good
fudbaler.
soccer.player
‘He is a good soccer player.’ (ambiguous: moral or skillful)
b. On
He
je
is
bolji
better
fudbaler.
soccer.player
‘He is a better soccer player.’ (unambiguous: more skillful)
c. On
He
je
is
dobar
good
fudbaler,
soccer.player,
ali
but
ovaj
this
je
is
još
even
bolji.
better
‘He is a good soccer player, but this (one) is even better.’ (unambiguous:
more skillful)
Despić reports that parallel considerations apply to the special form prozliti
(se) ‘(to) become (morally) bad’, alongside expected and normal pogoršati
(se) ‘(to) worsen’. Despić’s conclusion is that dobar ‘good’ is ambiguous, between
an intersective, moral reading (‘x is a thief, and x is good’) and a
non-intersective reading (‘x is good (at being a) thief’), but that only the
latter is subject to comparison. For unclear reasons, moral goodness is evidently
not a gradable property in Serbian. As such, only the non-intersective
reading gives rise to morphological comparatives, which, as expected, serve
291
as the basis for the deadjectival verb. The verb prodobrio se is derived from
a non-gradable adjective, and thus falls outside the scope of the C∆G.25
6.4.4 Russian: bad feelings
The Russian adjective plox-oj ‘bad’ is suppletive, with comparative xuž-e <
xud-. The regular verb meaning ‘(to) worsen’ is from this comparative allomorph:
u-xud-š-at’.26
Alongside this verb are, however, various verbs built
on the positive root plox-. One of these verbs, s-plox-ovat’ ‘(to) make a blunder’,
falls semantically outside the range of verbs relevant to the C∆G, but a
second verb, the colloquial po-plox-et’ seems more problematic.27
This verb
is used impersonally, as illustrated in (267b) corresponding to the impersonal
construction in (267a). In this usage, the adverb appears to compares suppletively:
mne xuže ‘I feel worse’, so the considerations regarding previous
apparent counter-examples do not apply.28
(267) a. mne
me.dat
plox-o
bad-adv
‘I’m sick’ / ‘I don’t feel good.’
b. mne
me.dat
po-plox-el-o
pref-bad-pst-neu
‘I suddenly felt sick, felt worse.’
The Russian speakers I have discussed the example with consistently draw
attention to a sense of ‘sudden onset’ in (267b). This may suggest a derivation
292
of the apparently offending verb that follows an alternate route to verbhood,
avoiding V∆ and the associated prediction about suppletion. Recall from the
beginning of this section that Russian does have de-adjectival stative verbs
(meaning ‘be X’ rather than ‘become X’). In addition, the complex system
of aspectual prefixes includes inceptive prefixes, indicating the start of an
event. While not the most frequent exponent of inceptive aspect, the prefix
po- does have this use, as in bežat’ ‘(to) run’ → po-bežat’ ‘(to) start to run’.
In theory, then, the morphology of Russian provides at least two routes to
a meaning that is very close to ‘become adj’. The structure in (268a) is
the one that has been the focus of this chapter, and involves V∆ and thus
the comparative, deriving u-xud-š-at’ with the comparative allomorph. The
structure in (268b) is the combination of the inceptive po- with the (abstract)
stative-forming morpheme VSTATE, mentioned in conjunction with (256).
(268) a. [ [ [ adj ] cmpr ] V∆]
b. [ incept [ [ adj ] VSTATE ] ]
Transparent use of inceptive morphology to form inchoative verbs from
adjectives is evident in Ingush (Nakh-Dagestanian, Nichols 2011, Chh. 15,
21). The elements -lu and -d.oal form inceptives with verbs (as in (269b))
but intransitive change-of-state verbs from adjectives (in (269a)).29
293
(269) a. adjective verb
d.waaixa ‘hot’ d.woax-lu ‘get hot’
niisa ‘straight’ niis-lu ‘become straight, level’
kura ‘arrogant’ kur-d.oal ‘become arrogant
b. verb inceptive
qier- ‘fear’ qiera-lu
d.iesh- ‘read’ diesha-lu
laz ‘hurt’ laza-d.oal ’
The Ingush examples establish the independent plausibility of using inceptive
morphology with stative (adjectival) predicates to derive inchoative
meanings. To the extent an inceptive analysis can be sustained as a possible
alternative route to verb formation, bypassing V∆ and the comparative, it is
somewhat of a two-edged sword. On the one hand, it avoids the otherwise
problematic character of (267b), removing yet one more apparent counterexample.
On the other, it points to a subtlety in the predictions of the C∆G
— the prediction is about the results of a particular derivation, a derivation
that may be circumvented if the morphological resources of the language allow
it. The combination of inceptive and stative morphology yields a meaning
(‘start to be X’) that is extremely difficult to tease apart from the meaning
relevant to the C∆G (‘become X’). The existence of such an alternate derivation
opens a back door in the analytical realm which, if not circumscribed,
threatens to bleed the predictive force of the generalization as it permits
294
surface violations of the C∆G. For Russian, we may at least point to the independent
evidence for the constituent pieces of this alternative derivation,
but the general concern remains.
6.4.5 Multiplicity
Alternate routes to verb-hood may be at work in another set of challenges
coming from expressions meaning ‘to multiply’. If (the translation equivalent
of) this verb is taken to be the corresponding change-of-state to many/much,
then there are many challenging examples in this domain. Some are given
in (270), with the corresponding suppletive comparatives (see section 4.3 for
many more examples of suppletive comparatives of ‘many/much’ terms):
(270) pos cmpr verb
a. German viel meh-r ver-viel-fach-en
b. Hungarian sok tö-bb sok-as-odik
c. Hungarian sok tö-bb sok-szorosít
d. Finnish monet usea-mmat monin-kertaistaa
e. Russian mnogo bol’š-e u-množ-at’(-sja)
It is not clear that ‘(to) multiply’ really stands in the right semantic relation
to ‘many’ to consider these problematic in the first place. At a minimum,
it appears that the element meaning ‘many’ in various languages shows some
measure of an ambiguity between a quantificational element (subject to comparison)
and a numeral-like entity (cardinal many). In some languages, it
295
can also be used as a noun, with a meaning such as ‘a crowd’ or ‘a large
amount’. The numeral-like usage of ‘many’ is illustrated by Hungarian sok,
which may take morphology that is otherwise restricted to numerals, for example,
ordinal-forming -dik, as in (271) (distinct from the intransitive verbal
affix in (270b)). With sok ‘many’, the ordinal corresponds roughly to casual
English umpteenth.
(271) a. öt = 5 ötö-dik = 5th öt-szöröz = ‘quintuple’
b. sok = ‘many’ soka-dik = ‘many-th’ sok-szorosít = ‘multiply’
The verb-forming patterns in (270) follow the pattern for forming verbs
meaning ‘to double, triple, quadruple, etc.’ from numerals. Since cardinal
numerals do not denote gradable concepts, the morphology that derives these
verbs must be of a form that does not require a comparative complement. The
categorial ambiguity of many ‘many’-words thus provides an alternative route
to verb-hood, bypassing the morphology (by hypothesis, V∆) that derives
degree achievements from adjectival roots.
In addition, some examples, including the Hungarian and German examples
in (270a-b) involve additional derivational morphology between the
quantifier root and the verbalizer (or a hypothetical hidden comparative morpheme,
selected by the verbalizer). Given the adjacency condition, the extra
derivational morphology shields the root from the triggering context for suppletion.
We see this effect, for example, in the English pair good – better
versus good-ly – good-li-er, where the intervening derivational morphology
blocks root suppletion in the latter. German vervielfachen in (270a) is de-
296
rived, not from viel itself, but rather from the derived adjective viel-fach
‘multiple’ (cf. English mani-fold), with -fach predictably blocking suppletion.
In a similar manner, Hungarian sokasodik appears to have a transparent
derivation from sok in its nominal, rather than adjectival, guise, via the
suffix -as, which forms adjectives from nouns, and then -odik, which forms
verbs from adjectives. The derivation thus parallels: ház ‘house’ > ház-as
‘married’ > ház-as-odik ‘to marry’.
With these considerations in place, it is not clear that the meaning ‘to
multiply’ is related in the right way to the gradable quantificational element
‘many’, which undergoes suppletion, to be subject to the C∆G. A closer
meaning might be ‘to increase (in quantity)’. And indeed, German shows a
contrast between the verb ver-viel-fach-en ‘(to) multiply’, in (270) on the one
hand, and verbs meaning to increase in quantity on the other, including mehren
and ver-mehr-en, both built, as expected, on the comparative allomorph
of the quantificational root.
6.4.6 Compounds and roots
It is worth noting that some of the forms in (270) involve compounds, rather
than affixal structures. Thus in Hungarian, the verb szoroz on its own means
‘(to) multiply’; the forms in (271) appearing to be compounds of this verb
with a numeral. In discussing Armenian superlatives above (sections 3.4.1
and 4.3.1), I noted that compounds appear to involve a distinct internal
structure. For present purposes, if that is the case, then the roots for many
297
in compound forms like (270c) fall beyond the reach of the trigger for comparative
suppletion.
This consideration appears to be relevant for other derived verbs as
well. The Modern Romance languages have verbs such as Italian bonificare,
French bonifier ‘improve’, apparently derived from the positive root b(u)on‘good’,
alongside miglioràre, ameliorer from the suppletive comparative root.
The historical sources for both are clearly compatible with the generalizations
as presented here: the latter verbs involve(d) morphological derivation
from the adjective and transparently embed the comparative (as required by
the C∆G). On the other hand, the verbs containing -ifi- ultimately derive
from periphrastic / compound constructions in Latin bonu(m) facere ‘good
make/do’ (with subsequent morphophonological adjustments, A. Calabrese,
personal communication, 2009). The theory of locality presented in Chapter
3 ensures that syntactic (periphrastic) complementation structures will not
be suppletive (see 235). What remains to be shown is how the synchronic
analysis of the Italian and French verbs relates to the Latin periphrastic
construction. One (uninteresting, but not implausible) option is that the
modern verbs are descended independently of the adjectival roots they (historically)
contain. An alternative would be to posit a richer structure in the
complement of -ifi-.
As noted in the previous subsection, the adjacency condition appealed
to elsewhere requires that the trigger for comparative suppletion must be
adjacent to the root, in order for that root to undergo suppletion. Now,
298
throughout this work, I have not distinguished a root from its category.
Much work in Distributed Morphology does draw this distinction however,
and would represent a positive adjective, such as wide, in the syntax, as in
(272), with a root (symbolized
√
) dominated by a category-defining phrase
(indicated by a small letter, here a for adjective; not to be confused with the
mnemonic node-designators I have used in the morphological representations
elsewhere in this book). See Pesetsky (1995), Marantz (1996, 2007), Harley
and Noyer (1999), Embick (2010), among many others; see also Borer (2005,
2009) for a dissenting view.
(272) aP
√
P
√
wide
a
From this perspective, Marantz (1997, 2007) draws a distinction between
‘inner’ and ‘outer’ morphology. Inner morphology attaches to the root, or at
least within the domain of the first category-defining head. Outer morphology
attaches outside a category-defining root. For a given root that normally
occurs as an adjective (i.e., in the syntactic configuration in (272), there are
thus (at least) two distinct derivations that will yield a verb, as in (273):
299
(273) a. b.
vP
√
P
√
wide
v
vP
aP
√
P
√
wide
a
v
Of these, only (273b) is truly category-changing, in the sense that it actually
embeds one category-defining node inside another. In principle, either of
these could be the abstract representation of a verb like widen. However, with
the adjacency condition on suppletion (applying to the complex heads derived
from (273) by Merger or head movement), a key distinction potentially
emerges. The root in (273a) is close enough to the head v to be conditioned
by it, while the root in (273b) is not. (This reasoning assumes that the comparative
element is near or within the v node in (273); cf. (233)b)). If the
-ifi- derivations involve outer morphology, as in (273b), consistent, perhaps,
with their historical source as phrasal complementation structures, then the
lack of suppletion in forms like bonifier follows from the locality conditions
on suppletion already posited. This conclusion holds even if structural adjacency
is replaced by the more complex combination of structural and linear
locality conditions proposed in Embick (2010), in which (in effect) an outer
category-denoting head may not condition allomorphy across an intervening
category-denoting head.
300
As noted earlier in this chapter, the inner/outer distinction could provide
an insight into the difference between two types of verbalizing morphology,
as regards a potential SSG-like generalization. Degree achievements derived
from adjectives by means of zero-derivation (a null affix) or the -en suffix
appear to inherit the lexical restrictions of morphological comparatives, and
thus only form degree achievements from roots that take morphological comparatives.
On the other hand, -ify and -ity derive change of state verbs from a
wide range of stems, including adjectives that do not permit of morphological
comparative formation (modernize, solidify versus *moderner, *solider). If
these represent different abstract morphemes, it is plausible that an account
of this difference could be found in terms of the differing types of meaning
available to inner and outer morphology differing relations (both semantic
and morphological) that inner and outer category-defining nodes may bear
to a given root (see Marantz 2007 and especially Ramchand 2008 for relevant
proposals).
Like the inceptive morphology discussed for Russian, then, the theory
allows for multiple routes to the verbal structures, with the C∆G predicted
to hold only of certain derivations, and thus with different routes (derivations)
correlating with different roots (allomorphs). As with the Russian inceptive
derivations, recognizing this possibility makes the analytical task that much
harder, in determining which derivation applies in any given case, in noncircular
fashion. I have not investigated this for bonifier and leave this as
a tentative and programmatic indication of a direction that seems to me to
301
have merit in accounting for these forms.
6.4.7 Outstanding examples
To this point, we have considered a variety of prima facie counter-examples to
the C∆G, and I have argued that, with varying degrees of confidence, these
may be safely explained away. There remain, however, a few outstanding
examples for which no obvious solution presents itself. For the record, I note
these here.
First, while some uses of Old English gódian, and Middle English goden,
(to) good (attested into the 17th Century) fall outside the realm of relevant
meanings (as noted above), a residue of problematic uses remain. The OED
gives examples supporting the meaning ‘to become better, improve’ and ‘to
make good’. A 13th century example (for which no gloss is provided) juxtaposes
igodet ‘gooded’ and iwurset ‘worsened’, strongly suggesting a meaning
relevant to the current study. Note that Old English gódian forms a doublet,
alongside bétan with the suppletive root as expected.
Another apparent counter-example is from Old Church Slavonic, in which
the adjective velik˘u ‘big’ undergoes suppletion in the comparative (with two
distinct roots: bol- and vęšt-), but where the change-of-state verb is built
on the positive root: v˘uz-velič-iti. It is perhaps worthy of note that in the
modern Slavic languages, the situation has resolved itself for the most part.
Thus in Russian, the basic adjective for ‘big’ has become bol’š-ój (comparative:
ból’š-e) with velik-ij relegated to the narrowed meaning ‘great’. The
302
verb u-velič-iva-t’ retains the general meaning ‘(to) increase, magnify’ but
the replacement of the positive root means this is no longer an ABA pattern,
but if anything, an AAB pattern (on which see above). On the other hand,
Ukranian retains the suppletive pattern from OCS (velyk-yj – bil’š-yj ‘big –
bigger’) but has undergone changes in the verbal domain, with voz-velyč-yty
taking on the specialized meaning of ‘to glorify’ and the general verb meaning
to increase built on the comparative: z-bil’š-yty. Serbo-Croatian retains
the suppletive adjectives, and a doublet in the verbs (with the problematic
u-velič-ati limited to the causative/transitive usage).30
Also in Slavic, Polish presents an apparently more serious challenge, as
discussed by P. Jabłońska (personal communication, cf. Jabłońska 2007,
141). In Polish, transitive and anticausative deadjectival verbs conform to
the C∆G, but simple inchoative verbs formed from adjectival roots with
no derivational morphology seem to violate it. For example, corresponding
to the suppletive pair dobr-y – lepsz-y ‘good – better’, one finds the
verbs u-lepsz-yć ‘(to) make sth. better’ (with causative prefix u-) and anticausative
po-lep-sz-yć się ‘(to) get better’ but also an inchoative verb dobrzeć
‘to get better’, and also mal-eć alongside z-mniej-sz-yć się (both) ‘(to)
get smaller’ (W. Browne, personal communication). This looks like a real
counter-example, though note that there appears to be some systematicity
here in that only the true inchoatives fail to embed the comparative stem.31
303
6.5 Summary: What’s the Difference?
This chapter has explored the possibility that a generalization parallel to
the CSG(1) exists in the domain of deadjectival verbs, stated as the C∆G.
I considered some of the theoretical ramifications of that conclusion. On
the empirical side, this chapter has been somewhat more tentative than the
preceding chapters — the C∆G is surprisingly robust, though problematic
cases remain. Although I have devoted much more discussion to the problematic
cases, note that they are firmly in the minority. Against the three
truly problematic-looking cases in section 6.4.7, note that the tally of C∆Gconsistent
ABB examples collected in Table 6.1 runs to some 50 or so cognate
sets. If the remaining counter-examples can be explained away, then the logic
of an account presents itself by parity of reasoning to Chapter 2, implying
a parallel Containment Hypothesis. Yet for the verbal domain, an appeal
to semantic complexity seems a less obvious underpinning to this hypothesis
than it might for the superlatives (see Chapter 7).
A somewhat more radical possibility (though one that is not free of problems),
is to build the comparative meaning into V∆ itself, in a way that allows
— or rather requires — it to trigger comparative root suppletion. Hay et al.
(1999) and Kennedy and Levin (2008) pursue approaches very much of this
sort. Thus, Kennedy and Levin (2008) propose that (a component of) the
meaning of the grammatical element that characterizes degree achievements
(regardless of their derivation) is a measure-of-change function m∆. The
304
function, applied to an individual x and an event e, measures the difference
in which x changes in m-ness over the course of e, where m is some property,
denoted by the adjective in the case of deadjectival degree achievements.
Kennedy and Levin (2008) suggest that m∆ is a special case of what they
call in related work (see Kennedy and McNally (2005)) a differential measure
function md.32
The core of this proposal holds that a simple adjective denotes
a function relating an individual to a (position on a) scale of measurement
(for example, the height scale for the adjective tall), the comparative degree
of an adjective denotes a function that measures the difference between an
object’s associated value on the scale and a second value, provided by the
standard of comparison. Seeing the measure of change as difference over
time, the key meaning component of V∆ effectively incorporates a special
case of the comparative meaning. Like the superlatives, the degree achievements
are thus a proper subset of ‘comparative’ environments, but unlike the
superlatives, this need not be encoded by a nested structure. Put differently,
this direction would amount to positing the structure [[ adj ] V∆], but with
the understanding that the ∆ component is in all relevant respects the same
linguistic entity as cmpr, though here rolled into the verbalizing affix, as in
(233b).
An advantage of this direction might be that it would force comparative
suppletion of the root in the verbal environment, without recourse to the
comparative morpheme itself. No analogue of the SSG is expected on this
account. Since the comparative morpheme per se is not implicated, there is
305
no expected transitivity effect regarding the operation M, and it should be
possible to derive degree achievements from adjectives that lack a morphological
comparative, or for that matter, from adjectives that are not gradable
(a simple change of state would thus be a (positive) change along a two-point
scale: the difference from 0 (false) to 1 (true)).
I acknowledge a certain intuitive appeal to this direction, but it appears
to stumble over the cases in which overt comparative morphology is contained
in the deadjectival verbs, as in the verbs in (238) from regular comparatives,
and all of the suppletive forms in Table 6.1 for which the comparative has
regular comparative morphology (like bett-er as opposed to worse). What
appears to be unattested is a form in which a comparative root allomorph
occurs without comparative morphology as the base of a degree achievement,
unless that root allomorph is a portmanteau in its regular comparative sense.
We do not find *good – bett-er – (to) bett-en, German *gut – besser – ver-bessen
or Russian *plox-oj – xuž-e (< xud-) – u-xud-et’33
. Such forms appear to
be what a theory treating the ∆ of m∆ (or V∆) as the comparative might
predict.34
Acknowledging that important questions thus remain, I leave further exploration
of this direction for future work.
306
Chapter 7
Complexity, Bundling, and
Lesslessness
7.1 Introduction
I have argued at length above that UG admits of no ‘superlative’ morpheme
understood in the familiar sense as a morpheme that combines with an adjective
X to yield the meaning ‘more X than all (others).’ In order to make sense
of a wide range of data, I have posited that this meaning is always expressed
bi-morphemically, as in (274a) (and/or in the branching-affix structure in
(68)), with the comparative part (‘more’) represented separately from the
‘than all others’ component (which latter I have glossed sprl). Despite surface
appearances in languages such as English, the structure in (274b) must
be impossible.
307
(274) a. superlative b. impossible
s
c
a
adj
cmpr
sprl
* s
a
adj
sprl
The underlying bi-morphemic structure can be masked on the surface; for
example, English -est appears to attach directly to adjectival bases, but is
in fact mediated by an intervening, but phonologically null, allomorph of the
comparative (42). The decomposition of the superlative in this manner, even
where not morphologically transparent, is the cornerstone of the account of
(both parts of) the CSG and the SSG. In Chapter 6 I extended the account,
arguing that a parallel generalization may hold of the derivation of deadjectival
change-of-state verbs, suggesting a parallel containment hypothesis:
[[[ adj ] cmpr ] V∆], and not [[ adj ] V∆].
The preceding chapters have explored in detail the consequences of these
hypotheses, and the evidence for them, but I have for the most part refrained
from speculating on their nature. How should the Containment Hypothesis
be expressed formally, and why should it hold? It may be that this is an
irreducible property of language — that UG provides a universal vocabulary
in the sense that the set of (functional) morphemes is a closed class, with the
(abstract) morphemes of any individual language drawn from this set (see,
308
for example, Beard 1995). The discovery of what morphemes are in this set
would then be an empirical matter; the present study constitutes an argument
that the morpheme sprl, meaning ‘than all others’ and necessarily combining
with the comparative, is in the universal vocabulary, while a morpheme with
the more complex meaning ‘more than all others’, typically assigned to -est,
is not.
But a more interesting question, of course, would be whether we could
derive this result from some prior or deeper consideration. Rather than
stipulating a universal morpheme inventory, one would like to understand
in more general terms what the constraints are on possible and impossible
morphemes, and answer to the question of why the more complex meaning is
not in the universal functional vocabulary. Ultimately, it seems this will only
push the question back to some degree: a component of this endeavour will
have to include a list of some sort, but perhaps rather than morphemes as
such, the list will be a list of features or other primitives that may combine to
constitute morphemes, along with constraints on permissible combinations
thereof.
This seems to be a promising line in the characterization of person features,
for example. An argument originally due to Silverstein (1976) and
Ingram (1978) and debated in much subsequent work, holds that the inventory
of person categories (first, second, third; inclusive/exclusive) is explained
by assuming that UG constructs the familiar categories (such as “first person
exclusive”) out of exactly two binary (or privative) features: [±I,±you] (or
309
some close variant thereof). The argument runs as follows (see Bobaljik 2008
for a review of the literature):1
Admitting only the three classical values of person (first, second, third)
undergenerates, in failing to describe robustly attested exclusive versus inclusive
contrasts in person marking (pronouns, agreement, clitics). The existence
of monomorphemic ‘inclusive’ markers (meaning ‘we, including you’)
suggests that feature composition is needed in addition: the inclusive category
is [1+2]. Admitting of composition, along with the three values, yields
the maximal seven-way contrast in (275a) (when cross-classifying features
such as number, gender, etc. are factored out):
(275)
a. Possible b. attested c. binary
1+2+3
inclusive [+I, +you]
1+2
1+3
exclusive [+I, -you]
1
2+3
second [-I, +you]
2
3 third [-I, -you]
However, the seven-way contrast is famously unattested. The maximal
attested contrast is the four-way contrast in (275b). Logically possible contrasts,
such as a [2+3] versus [2] contrast in the plural (hearer and others,
versus multiple hearers) are never attested as monomorphemic distinctions
310
(though they are made compositionally). There are at least five such universally
unattested contrasts:
(276) U1 No language distinguishes [1+1] from [1+3].
(No “choral" we)
U2 No language distinguishes [2+2] from [2+3].
(No “exclusive" 2, mentioned in main text)
U3 No language distinguishes [1+2] from [1+2+3].
(No “complete" person)
U4 No language has minimal [1+3]. (Apparent examples are [+I,you]
dual)
U5 No language has minimal [2+3]. (Apparent examples are [I,+you]
dual)
Instead of the classical three-value system, the two value binary system in
(275c) — plus the general constraint that no morpheme may be specified for
conflicting values of the same feature — provides for the description of all and
only the attested person marking inventories. Inventories smaller than the
maximum involve syncretism, with ‘natural’ syncretic patterns (characterized
by neutralization of one or more possible contrasts) outnumbering ‘unnatural’
patterns (presumably historical accidents) by a factor of roughly 9:1.2
The conclusion to be drawn from this data is that UG includes a set of
primitives from which morpheme meanings are constructed. In the realm of
person features, this set includes [(+)1] and [(+)2] (and perhaps a grouping
311
feature, [ participant ] defined as the union of these), but no others. To
date, UG remains the most plausible source of these feature restrictions.
No language-external, functionalist account has succeeded in providing any
deeper understanding of why language marks only a small, and specific subset
of the relevant conceptual distinctions in this domain (see Levinson 1988
and discussion in Cysouw 2003, Bobaljik 2008; but see Wechsler 2010 for
suggestions couched in terms of deixis and Theory of Mind.)
Returning to the realm of comparatives and superlatives, we should ask
whether a plausible account of the primitives of meaning might be given
that would force the Containment Hypothesis (as in (274)) from some prior
and more general consideration, given a suitable inventory of primitives for
morpheme meanings. Although I can do little more than speculate at this
point, an obvious candidate is a limit on the maximal internal complexity of
(functional) morphemes. A (syntactic) proposal of this sort was put forward
in Kayne (2005, 212), as the Principle of Decompositionality, and similar
ideas are entertained in related Cartographic approaches with rich sub-lexical
decomposition:3
(277) UG imposes a maximum of one interpretable syntactic feature per
lexical or functional element.
I suspect the stricture in (277) is too severe (even allowing for some leeway
in what an ‘interpretable syntactic feature is, as Kayne recognizes), but that
nevertheless, some version of a Complexity Condition in this general line may
312
be correct. Most plausibly, it seems to me, limits on morpheme complexity
will prove to be best stated in semantic, rather than syntactic terms — certain
types of meanings are complex in ways that the resources of UG cannot pack
into a single morpheme.4
Hackl (2009) argues this point for most, contending
that it must be analyzed compositionally as the superlative of many. In so
doing, he argues that the resources of Generalized Quantifier Theory, which
provides a monomorphemic meaning for most (Barwise and Cooper 1981),
are too rich; the primitives which UG builds quantifier meanings out of should
not have the power to construct a semantics of the proportional quantifier
most, other than via (syntactic) composition.
In light of the evidence presented in this book, inasmuch as he invokes an
element -est in forming the superlative from the adjective (or determiner),
Hackl did not decompose enough: the combination of the comparative degree
operator and sprl (containing at the least a universal quantifier and/or a
definite determiner) is also more than can be combined into a single functional
morpheme.
If this is correct, then the Containment Hypothesis, as a condition specific
to superlatives, need not be stated in UG at all. The impossibility of (274b)
follows on the one hand, since there cannot be a single, indivisible morpheme
with the complex meaning ‘more than all (others)’, and on the other, because
an attempt to combine a sprl morpheme (meaning just ‘than all (others)’)
with the adjective root in the absence of a comparative element will simply
fail to yield a valid interpretation. The Containment Hypothesis is thus a
313
specific consequence of the Complexity Condition.
7.2 Lesslessness
A general complexity condition of this sort should manifest itself in a variety
of detectable contexts. If the C∆G is correct, and deadjectival change-ofstate
verbs systematically embed comparative adjectives (Chapter 6), then
these provide another clue as to the range of meanings that may be expressed
monomorphemically. Outside of comparatives, there is at least a point of
contact with the long-discussed domain of missing (i.e., never lexicalized)
logical operators, famously discussed in Neo-Gricean terms in Horn (1972,
1989) and reexamined by many since (for a sample, see Jaspers 2005, Penka
2007, Katzir and Singh 2009). Of the four corners of the traditional Square of
Opposition, three are found as lexical items across languages: an existential,
a universal, and a(n often marked) negative, cf. English some, all, none,
but the fourth member of the set, a quantifier meaning not all, is never
lexicalized (*nall). The pattern repeats itself through an array of quantifiers
and connectives, a deeply systematic gap. A component of (at least) some
approaches to this distribution rests in part on the internal complexity of
the logical operators, given a set of primitives — the missing meanings are
necessarily complex, and blocked by an economy condition favouring the use
of simpler inventories.
Returning to the domain of comparative morphology, there is one further
314
generalization that may support the general idea of constraints on morpheme
meanings, namely, the generalization I called Lesslessness in the Introduc-
tion.
(278) Lesslessness
No language has a synthetic comparative of inferiority.
Comparison of superiority (‘more X’) is affixal in many languages, as in
long – long-er, but comparison of inferiority (‘less X’) never is. In the schema
in (279), the lower right hand cell is universally empty.
(279) analytic synthetic
a. superiority more adj adj-er
b. inferiority less adj *
This generalization is empirically the strongest of all the generalizations
considered in this book. In none of the more than 300 languages examined
for this study did anything remotely resembling a counter-example appear.5
The Complexity Condition may get us some way (though not all the
way) towards an understanding of (278). Assume that elements meaning
‘less (than)’ are necessarily composed of the comparative operator > in tandem
with an operator that reverses the polarity of a scale of measurement,
which I will indicate as ⇄. Such an operator has been invoked, for example,
to relate polar antonyms, yielding the marked (i.e., negative) member in pairs
such as short – tall: thus short is a portmanteau expressing [[ tall ] ⇄ ]
(see, e.g., Rullmann 1995, Heim 2006, Büring 2007 who call the operator
315
little, and also the operator ◦ in Katzir and Singh 2009; on the semantics
of such an operation see especially Sassoon 2007, 2010). If there is indeed
a Complexity Condition along the lines of (277), it seems a priori plausible
that the combination of these operators would be subject to this condition,
and thus impossible at the sub-morphemic level. The underlying representation
of a comparison of inferiority would necessarily be morpho-syntactically
decomposed, along the lines in (280):
(280) CmprP
cmpr ⇄P
⇄ AdjP
adj
A periphrastic comparative of inferiority, as in English less tall, is modeled
straightforwardly, if cmpr and ⇄ combine, but the two do not undergo
merger with the adjectival root. The derivation is given here, and is entirely
parallel to the derivation of a periphrastic superlative in (115) in Chapter 3.
316
(281) CmprP
cmpr ⇄P
⇄ AdjP
adj
This analysis treats less as a portmanteau spelling out cmpr + ⇄ (and is
why Rulmann and Heim treat ⇄ as expressing little, in its quantificational
sense, opposite of much). The key morphological pieces are clearly visible
in languages like German (wenig-er ‘less’ = wenig ‘little, few’ -er cmpr)
and (Modern) Greek (ligo-ter-os ‘less’ < lig-os ‘little, few’). The following
examples (from Jason Merchant, personal communication 2011) show the
Greek element in its quantificational sense (comparative of ‘few’) and in its
use forming periphrastic comparatives of inferiority.6
(282) a. Exo
I.have
lig-a
few-n.pl
vivlia.
book.n.pl
‘I have few books.’
b. Exo
I.have
ligo-ter-a
few-cmpr-n.pl
vivlia
book.n.pl
apo
from
sena.
you
‘I have fewer books than you.’
317
(283) Imo
I.am
ligo-ter-o
few-cmpr-n.sg
nevriasmen-os
annoyed-m.sg
apo
from
sena.
you
‘I am less annoyed than you.’
As Merchant observes, the Greek data illustrate another generalization,
which provides additional grounds for deriving comparatives of inferiority
from comparatives of superiority, as proposed here. Specifically, in languages
with morphological or periphrastic comparatives, the gross syntax including
the standard marker is the same for inferiority and superiority (English than,
German als, Greek apo in (283) etc.).7
There is no general reason why this
formal identity should hold; for other degree constructions, such as those
of equality, a different marker of the standard, and sometimes a different
morphosyntax altogether, is used. So, if more and less were simply distinct
degree heads, then like equative as, we might expect to find them combining
with differing markers for the standard phrase. However, while the Englishlike
pattern in (284a) is replicated in language after language, the pattern
in (284b), with a distinct marker for the standard of inferiority appears to
be unattested. This follows, of course, if the major functional component of
a comparison of inferiority is the same cmpr element as in comparatives of
superiority, as schematized in (281).
318
(284) a. attested b. unattested
equality as . . . as as . . . as
superiority more . . . than more . . . than
inferiority less . . . than less . . . X
The polarity-reversing operator ⇄ clearly has affinities with negation
(though it is not semantically identical to simple negation as such, see Büring
2007, Sassoon 2010 for comments). If we allow that it may in fact be expressed
as a species of negation in some languages (short is not-tall) then we
find transparent reflections of (281) in some languages. In Komi, the comparative
suffix -Ãïk attaches to adjectives to form regular comparatives ((285a),
from Lytkin 1966, 288), while less-comparatives are formed by adding the
comparative suffix to the negative particle abu, as in (285b) (Hausenberg
1998, 314, see Coates 1982 for further discussion). The comparative in (285b)
transparently corresponds to the output of Merger in (281), with abu as the
exponent of ⇄.
(285) a. bur bur-Ãïk
‘good’ ‘good-cmpr’ i.e., ‘better’
b. mičj
a abu–Ãïk mičj
a
‘beautiful’ ‘neg-cmpr beautiful’ = ‘less beautiful’
Now consider the derivation that would be needed to represent a synthetic
comparison of inferiority. Just as (280) and (281) derive periphrastic
constructions, the corresponding synthetic construction requires Merger (or
319
Head Movement, or Lowering) to combine all three heads in (280), as shown
in (286a), generating the complex morphological object in (286b):
(286) a. CmprP
cmpr ⇄P
⇄ AdjP
adj
b. c
b
a
adj
⇄
cmpr
Recall that the initial motivation for the operator ⇄ was to represent
polar antonyms: short is [ [ tall ] ⇄ ]. With this in mind, we see that the
putative synthetic comparative of inferiority in (286b) for some adjective, is
in fact indistinguishable from the regular comparative (i.e., of superiority) of
the polar antonym. In other words, if [ [ tall ] ⇄ ] is pronounced short, then
[ [ [ tall ] ⇄ ] cmpr ] = shorter. In essence, if the Complexity Condition can
be invoked to force (286) to be the representation of comparatives of inferiority,
the apparent universal absence of synthetic expressions of this construct
becomes a blocking effect — there are synthetic less-comparatives after all,
and in fact they abound, but they are pronounced as (-er)-comparatives of
the polar antonym.
The assumption, mentioned at various points above, that vocabulary insertion
proceeds cyclically from the most deeply embedded node outwards
320
(see Bobaljik 2000b, Embick 2010) will ensure that spell out of the node b
in (286b) will take precedence over alternatives, ensuring the blocking result
just mentioned.8
The competition effect invoked her appears to be directly visible in Sonora
Yaqui, as described in Dedrick and Casad (1999). Comparatives in this languages
are formed periphrastically, with the (sometimes optional) particle
čé'a. For adjectives that have a lexical polar antonym, a comparison of inferiority
is expressed by the (simple) comparative of the antonymous adjective
(i.e., ‘more small’ for ‘less big’). But for adjectives that lack a lexical polar
antonym, negation is used to derive one, as in (287) among other examples
(Dedrick and Casad 1999, 111). The two alternations flesh out exactly the
alternation just posited, with all the pieces of (280) overt when no portmanteau
form is listed to express the polar antonym.
(287) a. čé'a
more
húni'i
even
tú'ii
good
‘it is even better’
b. čé'a
more
húni'i
even
kaa-tú'ii
not-good
‘it is even worse’
Decomposing antonymous adjective pairs in this way will interact with
the adjacency condition, invoked (with some caveats) in Chapter 5. Under
the view just considered, the representation of bad is now held to be the
321
polar antonym of good, and thus internally complex, as in (288a), with the
corresponding comparative in (288b):
(288) a. b
a
good
⇄
b. c
b
a
good
⇄
cmpr
English worse spells out the structure in (288b) and patently involves (at
the least) a special form of the root in the environment of the comparative.
Maintaining the adjacency condition would add a further reason to treat (at
least some) portmanteaus as the spell-out of a complex X0
, either directly
(as in Radkevich 2010) or after fusion (as in Chung 2007b and others), rather
than as contextual allomorphy of terminals, with concomitant zero exponence
of higher nodes. If bad is a portmanteau for the b node in (288), then no
adjacency violation arises with worse, which latter is either a portmanteau
for the whole structure in (288b), or a contextual allomorph for the b node,
conditioned by the adjacent cmpr head.
Of course, not all gradable adjectives denote properties for which there
are polar antonyms (see Sassoon 2007, 2010). For example, colour terms
such as blue are gradable: blue – bluer – bluest — ‘Leo’s eyes are bluer than
mine.’ They also enter into comparisons of inferiority: ‘My eyes are less
blue than Leo’s’. And yet, there is no polar antonym that would spell-out
322
the morphological structure: [ [ blue ] ⇄ ]. The blocking account just
sketched therefore does not extend to gradable adjectives that lack a polar
antonym. A solution may be had by stipulating that adjectives which lack
a polar antonym may not undergo merger with the head ⇄. That is, the
combination [ [ blue ] ⇄ ] is simply undefined. If this is sufficient to block
the lower step of Merger in (286a), then the logic would directly parallel the
derivation of the SSG in section 3.4 — the addition of a higher head in the
structure cannot ‘rescue’ an otherwise impossible lower step of merger, and
only the periphrastic comparative less blue < [[ cmpr ⇄ ] blue ] is possible,
in accord with the facts.
The appeal to the Complexity Condition in explaining the universal impossibility
of synthetic comparatives of inferiority (278) thus finds some initial
plausibility. Fleshing it out in further detail may nevertheless prove tricky for
various reasons. One area of concern is alternative structures to (280). The
complexity condition only ensures that cmpr and ⇄ enter the derivation as
distinct heads, but it does not specifically require the hierarchical nesting in
(280). If it is possible to entertain the opposite scope order (⇄ > cmpr), or
a branching affix structure (as considered for superlatives in section 3.2.2),
the results from (280)-(286) could be undermined, and an alternative derivation
for the universally unattested morphological comparison of inferiority
may be opened up. I see no obvious way to exclude this at this point, and it
looks like a serious lacuna in the direction suggested here.
There are also some well-known concerns about the plausibility of using
323
⇄ as a means to derive antonymous adjective pairs. Authors who propose
such a derivation often express some reservations (see Rullmann 1995, 96
and Heim 2006, 54-58). Thus, as Heim notes, a proposal such as (286) assigns
the same structure, and thus the same logical meaning, to expressions
such as less fast and slower, but it is not clear that they are entirely logically
equivalent. Some of the differences have to do with evaluative (i.e.,
pragmatic (un)markendess) aspects of antonymous pairs (on which see Rett
2008), which may be orthogonal to the structural question. In addition, if
antonymous pairings truly are bi-morphemic, we should surely expect to see
this in the productive morphology in some languages.
In English, basic antonymous pairings (good-bad, long-short etc.) are
rarely expressed in this way, and the domain of polarity-reversing affixes
such as un- is quite limited. There are a number of languages in which fairly
basic negative terms such as ‘bad’ are transparently derived from ‘good’ via
negative morphology. An example from Sonora Yaqui was given in (287b);
an analogous pattern occurs in Kham (Sino-Tibetan): ca-o ‘good’ – ma-ca-o
‘neg-good’ = ‘bad’ (Watters 2002, 127), though suppletive antonymous pairings
exist in these languages as well. A language in which the derived nature
of the negative member of antonymous pairing is routinely transparent is
Hixkaryana. Thus Derbyshire (1985, 14) writes:
In Hixkaryana, there is a highly productive negative process,
and antonymy is almost exclusively restricted to pairs where one
item is a negative form derived from the same root as the posi-
324
tive form: kawo ‘long’ and kawohra ‘short’ ... t1yoke ‘sharp’ and
1yohra ‘blunt’ ... atahurmakaxaho ‘one that is open’ and atahurmakahn1
‘one that is shut’...
Hixkaryana shows that a language may make transparent use of compositional
means to express polar antonyms with a negative-like, polarityreversing
operator. While perhaps uncommon — or more accurately: frequently
masked by an inventory of portmanteau terms for basic pairs —
I take it that the possibility of this derivation is established. Building on
that observation, the complexity condition envisaged here could underlie an
account of the Lesslessness generalization in (278). I have sketched the beginnings
of what such an account might look like, tying it to some existing
proposals in the literature, but leave further exploration of this direction to
future work.
7.3 Conservative decomposition: Adjacency and
Bundling
I suggested earlier in this chapter that the motivation for lexical decomposition
is a complexity condition, which I speculated is ultimately semantic
in nature. The appeal to this condition as the basis for the Containment
Hypothesis, and especially as a syntactic reflection of the markedness hierarchy
positive < comparative < superlative (Greenberg 1966, Canger 1966,
325
Ultan 1972), raises the question of whether a much more radical lexical decomposition
should be entertained. In the Nanosyntax framework version of
the Cartographic framework, put forward in recent work by M. Starke and
colleagues, it is suggested that all markedness hierarchies should be cast as
nested structures in the syntax (see e.g. Caha 2009, and for related ideas,
Williams 2003). Recall also from the beginning of this chapter that Kayne
(2005) entertains (277):
(277) UG imposes a maximum of one interpretable syntactic feature per
lexical or functional element.
On both theory-internal and empirical grounds, I am led not to adopt
this view, in favour of the more conservative perspective that the limit will
be some (yet-to-be specified) measure of semantic complexity, rather than a
narrow limit on syntactic features (interpretable or otherwise). Whether or
not there is a substantive debate here depends on many definitional questions
(is “comparative” a feature? what features are “interpretable”?), and deriving
testable predictions depends on many other assumptions. In this section, I
offer only a few remarks on the issues involved, with no pretense that the
arguments are in any way conclusive.
Consider first the adjacency condition on allomorphy, adopted (albeit
with reservations) in Chapter 5. The condition banned C from serving as a
context for allomorphy of a root A in the configuration: [ [ [ A ] B ] C ].
The main work done by this condition in the current study lay in excluding
326
the unattested AAB pattern in superlatives. In this regard, superlatives contrast
minimally with the distribution of German verbal ablaut patterns, as
described by Wiese (2004, 2005), discussed in section 5.4, where AAB patterns
are attested. In both empirical domains, the account makes reference
to a markedness hierarchy (see (289)):
(289) a. present < participle < preterite
b. positive < comparative < superlative
And in both domains, the hierarchies are represented as relations of
proper containment. The feature structure of a less marked element in the
hierarchy is properly contained in the feature structure of a more marked
element on the hierarchy. This containment relation in each case triggers the
Elsewhere effect, and leads to the exclusion of the unattested *ABA patterns.
But the nature of the containment differs slightly, as in (224), repeated here
as (290):
(290) a. b.
V
V
verb



past
(finite)



s
c
a
adj
cmpr
sprl
327
The structure of superlatives assigns each meaningful element to its own
terminal node in the structure in (290b), and here, I hold that this should
follow from the proper version of the Complexity Condition. With the adjacency
condition, the superlative head is simply too far away from the root
to condition allomorphy. By contrast, admitting a feature bundle in (290a)
allows for AAB patterns in this domain — correctly, as discussed in section
5.4. In other words, both [past] and the combination [past, finite] may
trigger stem readjustment, hence both must be sufficiently local to the root.
If adjacency is the correct locality condition, and if these are both interpretable
features, subject to (277), then (277) is overly restrictive in terms
of the bundling it permits. In other words, if adjacency is the correct locality
condition, we have a partial argument in favour of admitting feature
bundling, as in (290a).
The argument for bundling can perhaps be made more general. Continuing
to assume an adjacency condition on allomorphy, any root allomorphy
conditioned by a combination of (interpretable) features will require that the
features be bundled into a single node (otherwise, they could not be simultaneously
adjacent to the root). German provides a handy illustration of this
point.9
In addition to ablaut, some German strong verbs are subject to umlaut
(a → ä, also e → ie) in the second and third person singular of the present
tense. The basic pattern is illustrated here, with fahren ‘to drive’:
328
(291) present
sg pl
1 fahre fahren
2 fährst fahrt
3 fährt fahren
As the paradigms illustrate, the conditioning environments for umlaut are
defined morphosyntactically: [2 sg] and [3 sg]. Both person and number
features must be mentioned in the context of the readjustment rule. Hence,
under the adjacency condition, both must be adjacent to the root, requiring
a feature bundle for person and number. But beyond person and number, the
accounts of ablaut and umlaut, invoking adjacency, come together in a potentially
interesting way. Assume that the general structure of German verbs
is [[[ root ] (tense) ] agreement ], a decomposition that is transparent
in the ‘weak’ and ‘mixed’ conjugations: (du) sag-te-st ‘(you) say-past-2sg’
(see Bobaljik and Thráinsson 1998, Bobaljik 2002b for the proposal that tense
and agreement may be bundled in some languages, but separate in others,
along with a discussion of the syntactic consequences of such a distinction;
see Thráinsson 2010 for a current assessment of apparent challenges). Positing
an adjacency condition for readjustment rules requires that the present
tense (which is morphologically unmarked, and treated as lacking features
in (222b)) is entirely absent at the point where adjacency is computed (cf.
Embick 2003, 2010 on ‘pruning’). By the same token, the ability of the past
tense to trigger ablaut is not only consistent with the structure just given
329
(tense, when present in the structure, is adjacent to the root), but it also
implies that the tense marker will block the application of person-number
conditioned readjustment (umlaut). That is, the considerations just given
correctly predict the restriction of umlaut to the present (i.e. absence of)
tense in German (see Bobaljik 2000a):10
(292) present preterite
sg pl sg pl
1 fahre fahren fuhr fuhren
2 fährst fahrt fuhrst fuhrt
3 fährt fahren fuhr fuhren
In sum, I have argued above that a locality condition on contextual allomorphy,
stated over the pre-exponence feature structures, provides a piece
of the explanation of the CSG2 (the absence of AAB patterns). The hypothesis
that structural adjacency is this condition leads to the conclusion that
some bundling of features into complex terminal nodes is inescapable, as, at
least in some cases, multiple features may condition allomorphy at a single
root. Of course, adjacency may not be the correct formulation of the locality
condition, which would require a reevaluation of this conclusion.
7.4 Concluding Remarks
I began this study with the observation that there are recurring cross-linguistic
regularities in the domain of comparative suppletion. English good, bad com-
330
pare via suppletion: better, worse, and both the superlatives, best, worst, and
corresponding degree achievement verbs, (to) better, (to) worsen, are derived
from the suppletive comparatives, and not from the positive roots. In undertaking
a large, cross-linguistic survey, I have argued that these isolated facts
of English form part of a broader network of intersecting generalizations, with
sporadic apparent counter-examples plausibly susceptible to alternative explanations.
The generalizations, to somewhat varying degrees of confidence,
appear to constitute strong candidates for morphological universals.
Approaching the empirical phenomena from the perspective of a formal
Generative Typology (Baker and McCloskey 2007, Baker 2009), I have accordingly
offered theoretical explanations for these generalizations. The core
of the account relies on the Elsewhere Condition as a factor regulating the
competition among suppletive root allomorphs. In tandem with hypotheses
about the structure of the relevant representations, most notably the claim
that superlatives always contain comparatives, even where the morphology
may be partly hidden, provides for an account not only of the observed
forms, but crucially of the absence of patterns that appear to be systematically
unattested. Moreover, each additional theoretical assumption invoked
yielded new predictions, which in turn revealed new generalizations (see in
particular Chapters 3 and 5).
I have argued that Distributed Morphology provides the right framework
of core assumptions in which to couch the theory presented here, and have
offered some arguments against conceivable alternatives. In so doing, I have
331
also made suggestions regarding choices among competing assumptions internal
to DM (such as the specific form a theory of locality might take).
Two aspects of DM loom large in the results presented here. First is the
realizational nature of the theory, which allows for competition among root
allomorphs, regulated by the Elsewhere Condition. Coupled with the Containment
Hypothesis, the Elsewhere Condition does the majority of the work
in explaining the absence of ABA patterns in comparative suppletion. While
many theories of inflection are realizational in nature, DM is distinguished
from the majority of its competitors in assuming that the morphosyntactic
representations that serve as input to rules of exponence are hierarchically
arranged, and thus the output of the syntax. Other frameworks deny this;
for example, A-morphous Morphology (Anderson 1992) holds that order and
structure in morphology are introduced by, and not prior to, the rules of exponence.
In an A-morphous framework (or its Word-and-Paradigm cousins),
locality conditions cannot be stated over hierarchical arrangements of features,
prior to their exponents being inserted. Thus, as a structural locality
condition, the adjacency condition invoked in Chapters 2 and 5 to explain the
absence of AAB patterns (the CSG2) is simply unstatable in an A-morphous
framework, as are related conditions framed in terms, for example, of cyclic
or phasal nodes (as in Embick 2010).
Leaving the details aside, by identifying apparently significant regularities
over a large sample in precisely that part of morphology which is by
definition the most irregular, and by showing how a relatively small set of
332
generally current theoretical assumptions derive this regularity, the core message
I hope to have offered here is a challenge to the view that morphology
is in some way fundamentally unconstrained, that ‘anything goes,’ and that
the best that can be hoped for in morphology is description. By contrast,
I contend that there are universals of morphological structure to be discovered
— and explained — and that current theoretical frameworks may be
differentiated by the degree to which the demarkation they define between
the possible and the impossible accords with the division between attested
and unattested. The success of morphological theories generally, and theories
of suppletion in particular, should be judged not only on what attested
forms they may describe, but also on how accurately they may predict the
absence of unattested patterns: an adequate theory, it seems, must exclude
unattested *good – better – goodest; simply getting good – better – best is not
good enough.
333
334
Appendices
335
Appendix A
The Broad Sample
This appendix reports the results of the broad sample, as described in section
1.3. There are 148 languages in this sample, which have been grouped into
5 broad geographic regions, with genetic affiliations as indicated. For each
language, four properties are indicated to the extent data was available. Descriptions
of additional languages were consulted, but were not included here
if there was insufficient information about comparatives to make even a reasonably
informed entry in the table. This was particularly true of indigenous
languages of the Americas, where many descriptions fail to comment at all
on the expression of comparison. This contributed to an underrepresentation
of many of these languages in the overall dataset.1
The column headed cmpr typ gives the type of comparative construction
indicated in the descriptions consulted. The three main values in this column
indicate the three types of comparative construction set out (with additional
337
examples) in section 1.4 (cf. Stassen 1985, 2008):
EX ‘Exceed’ comparative; comparison is expressed by a verb meaning ‘exceed,
surpass’ etc. in construction with the property-denoting predicate.
An example from Tamashek is given here:
(293) Ø-j`ær-\a-hi
3m.sg.s-surpass-\O-1sg
t-à-y@tte
fem-sg-intelligence
‘He is smarter than I’ lit: ‘He surpasses me in smartness.’
(Tamashek, Heath 2005, 244)
CN Conjoined comparative; comparison is expressed via the conjunction
of two clauses with, for example, antonymous adjectives, or adjectives
representing differing degrees. An Itelmen example illustrates:
(294) tinuPn
these
LeNu-Pn-č
berry-pl-dim
č’eBuz-laX-aPn,
sweet-adj-pl,
a
but
xaNnaPn
those
qaPm
not
‘These berries are sweeter than those.’ (Itelmen, field notes
[SB14A])
ST Standard comparative; the comparative is monoclausal, characterized by
means of a particular morphosyntax for the standard of comparison,
for example a than-phrase, as in English, or an oblique case marking
(with or without marking of the property-denoting predicate, see
below). Most Indo-European languages are of this sort:
(295) Leo is tall-er than Mika.
338
Languages with multiple strategies are indicated with multiple values in
this column. Since there are often many ways to paraphrase a comparative
meaning, it is not clear how clear-cut the difference is between, say, ST and
ST; CN. As little hinges on these labels, I have followed the descriptions that
I relied on. Questionable classifications are marked with a question mark.
The few languages marked OTH as the strategies for forming comparatives
do not obviously fit into the three main types; see section 1.4 for additional
discussion.
The column headed cmpr indicates the type of morphological expression
of comparison associated with the adjective (or other property-denoting predicate,
in languages which lack adjectives). The main values in this column
are:
M Morphological (synthetic) expression of comparison. There is an affix
(or morphological process) on the adjective that is associated with the
expression of comparison. A language is classified as M if this strategy
is available for at least some adjectives. Thus English would have an
M in this table, for the -er suffix, even though not all adjectives may
be so marked.
PER There are no regular synthetic comparatives; comparison is expressed
analytically, with an adverb (or other free element) that modifies the
phrase headed by the adjective. Romanian is a language of this sort,
with all comparatives expressed on the pattern: bun ‘good’ – mai bun
339
‘more good’ = ‘better’.2
ZERO There is no marking of a comparative construction associated with
the adjective (though there may be a mark on the standard). Most,
if not all, ‘exceed’-type and conjoined-type comparatives are of this
sort (see (293) and (294) above). This type is also common among
standard-comparative languages. Japanese is a language with no overt
marking of comparison:
(296) Sally-wa
Sally-top
Bill-yori
Bill-from
kasikoi
smart
‘Sally is smarter than Bill.’ (Beck et al. 2004, 327)
The values (M) and (PER) are used for languages in which an affixal
or periphrastic expression is available, but described as optional. Udmurt
illustrates an optional affix in (297) and Armenian an optional periphrastic
marker (adverb) in (298).
(297) kilometr
kilometer
i´škem-le´š
mile-abl
vak´či-(ges)
short-cmpr
‘A kilometer is shorter than a mile.’ (Winkler 2001, 40-41)
(298) Artak-@
Artak-def
Bagrat-e-n
Bagrat-abl-def
(aveli)
more
partsrahasag
tall
e
be.3sg.pres
‘Artak is taller than Bagrat.’
Note that there is a blurry border between ZERO and (PER). The difference
between Armenian, with an optional comparative adverb, and Japanese,
340
with unmarked comparatives, but an adverb/particle that may be used for
reinforcement, seems likely to turn out to be an artefact of different descriptive
traditions. Fortunately, nothing of substance here hinges on this
classification.
Comparative-forming elements, especially adverbs, have been borrowed
in many languages. Thus one finds Spanish mas or Portuguese mais forming
periphrastic comparatives in South and Central America, and similarly with
Russian bolee among indigenous languages of the former Soviet Union. I have
not included these constructions where the comparative is a clear calque, in
particular, where it exists alongside a form or construction that is not an
obvious borrowing. I have not, however, attempted to exclude borrowings
generally. Thus, Mari (Finno-Ugric) (in the focussed sample in Appendix B)
has an optional comparative suffix -rak (Kangasmaa-Minn 1998, 237), which
seems likely to be a Turkic loan, as is the (periphrastic) superlative in the
language.
The column headed sprl indicates the type of morphosyntactic expression
of the relative superlative. Information was not available for many languages,
and an n/i in this column indicates “no information (available)”. M
and PER are as in the previous column. The other values are:
ABS The only superlatives indicated in the source are absolute superlatives.
ALL Superlatives are expressed by the comparative construction, with a
universal quantifier serving as the standard of comparison.3
Russian
341
illustrates (see section 3.2.3):
(299) Vanja
Vanja
vyš-e
tall-er
vse-x.
all-gen.pl
‘Vanja is the tallest.’ (lit: ‘Vanja is taller than everyone.’)
CPR There is no clear morphosyntactic distinction between comparatives
and superlatives.
DEF The superlative is derived from the comparative (or if that is zero, from
the positive) by a marker of definiteness (or, in one case, a nominalizer
with this function).
VERY The superlative is derived from the comparative by means of an
intensifier. (It is not always clear that these are relative, rather than
absolute superlatives, though they are presented as such in the three
instances cited in this table).
POS The positive form of the adjective is used, with the superlative sense
derived from context.
NONE Descriptions report that there is no fixed superlative construction.
OTH See notes at entry.
Many languages have multiple means of expressing superlatives, and I
have not aimed for exhaustivity in this list. The primary aim has been to
identify morphological superlatives, as these are most relevant to the concerns
342
of this book. There is at least one other construction described as a means
of forming superlatives that I have not included in this table. Namely, quite
a few languages, in particular across southern and central Asia, are reported
to express a superlative meaning by a construction roughly equivalent to
English whiter than white, i.e., where the property predicate is compared to
itself. Of the languages in this appendix, such a construction is indicated
for Udmurt, Kashmiri, Burushaski, Tyvan, and Khakas (as well as in other
Turkic languages and the Mongolic languages in the next appendix). Other
than the Mongolic languages Dagur and Shira Yughur, this is never the only
means of expressing superlatives indicated in the relevant sources.
The final column indicates whether there is evidence of suppletion in
comparative (and/or superlative) formation. Values in this column are:
YES There is suppletion.
NO There is no suppletion.
MORE The only pair for which suppletion is indicated is many/much –
more. See section 4.3.
V There may be suppletive verbs replacing an ‘exceed’ comparative; see
discussion at (27) in Chapter 1.
n/i No suppletion is indicated in the available description. Unfortunately,
this last category is quite common; sources vary in how much confidence
we may have that suppletion would have been noted had it been
343
present. See discussion of this point in the next Appendix.
344
language lineage cmpr typ cmpr sprl suppl
Africa
Tamashek Afro-Asiatic, Berber ST; EX ZERO DEF V
Hausa Afro-Asiatic, Chadic (West) EX ZERO ALL n/i
Lele Afro-Asiatic, Chadic (East) EX ZERO n/i n/i
Mina Afro-Asiatic, Chadic (Daba) ST; EX ZERO n/i n/i
Qimant Afro-Asiatic, Cushitic ST ZERO ALL n/i
Somali Afro-Asiatic, Cushitic ST PER PER n/i
Arabic (Std.) Afro-Asiatic, Semitic (Central) ST M DEF n/i
Neo-Aramaic Afro-Asiatic, Semitic (Aramaic) ST (PER) DEF n/i
Hebrew Afro-Asiatic, Semitic ST (PER) PER MORE
Khoekhoe (Nama) Khoe (North Khoekhoe) ST ZERO ALL n/i
Khwe Khoe (Kalahari) ST; EX ZERO POS4
n/i
Koromfe Niger Congo, Gur EX ZERO NONE NO
I.jo. Niger-Congo, Ijoid EX ZERO n/i n/i
Kana Niger-Congo, Benue-Congo EX ZERO n/i n/i
345
language lineage cmpr typ cmpr sprl suppl
Ogbronuagum Niger-Congo, Benue-Congo EX ZERO ALL n/i
Banda Niger-Congo, Volta-Congo ST; EX ZERO ALL n/i
Lingala Niger-Congo, Bantu, NW Narrow EX ZERO ALL n/i
Nandi Niger-Congo, Bantu (Central J) EX ZERO ALL NO
Sotho (Northern) Niger-Congo, Bantu (Central S) EX ZERO n/i n/i
Zulu Niger-Congo, Bantu (Central B) EX ZERO ALL, M? n/i
Lango Nilo-Saharan, E. Sudanic EX ZERO ALL n/i
Mbay Nilo-Saharan, C Sudanic EX ZERO PER n/i
Koroboro Seno Nilo-Saharan, Songhai ST; EX ZERO ALL n/i
Kanuri Nilo-Saharan, W Saharan ST; EX (M) ALL n/i
Eurasia
Irish IE Celtic ST M CPR Y
Modern Greek IE, Hellenic ST M DEF Y
Kashmiri IE, Indo-Aryan (Dardic) ST ZERO ALL n/i5
Persian (Farsi) IE, Iranian (SW) ST M M Y
346
language lineage cmpr typ cmpr sprl suppl
Czech IE, Slavic ST M M Y
Finnish Uralic, Finno-Ugric ST M M YES
Udmurt Uralic, Permian ST (M) ALL n/i
Nenets Uralic, Samoyed ST M M ?
Ket Yeniseyan ST ZERO PER; ALL n/i
Even Tungusic ST ZERO ALL n/i
Evenki Tungusic ST M ? M; ALL n/i
Khalkha Mongolian ST ZERO ALL n/i
Lezgian Nakh-Dagestanian, Lezgic ST ZERO ALL MORE ?
Hunzib Nakh-Dagestanian, Tsezic ST ZERO ALL n/i
Abkhaz Northwest Caucasian ST ZERO ALL n/i
Bashkir Turkic ST (M) PER n/i
Khakas Turkic ST ZERO PER; n/i
Tyvan Turkic ST (PER) PER, n/i
Itelmen Chukotko-Kamchatkan CN ZERO NONE NO
347
language lineage cmpr typ cmpr sprl suppl
Chukchi Chukotko-Kamchatkan ST M M n/i
Malayalam Dravidian, Tamil ST ZERO6
VERY; ALL n/i
Ainu NE Asia, isolate ST; EX (PER) (PER) n/i
Nivkh NE Asia, isolate ST (M) ABS/VERY?7
n/i
Burushaski S Asia, isolate ST ZERO ALL n/i
Yukagir N Asia, isolate ST ZERO ALL n/i
Basque Europe, isolate ST M M Y
ASL Sign (American) CN; ST M M NO
E. Asia
Cantonese Sino-Tibetan, Chinese EX ZERO PER NO
Kunming Sino-Tibetan, Chinese ST; EX (M) (M) n/i
Mandarin Sino-Tibetan, Chinese EX ZERO PER NO
Dzongkha Sin.-Tib., Tib.-Burman (Tibetan) ST ZERO PER ? n/i
Kham Sin.-Tib., Tib.-Burman (Kham) ST ZERO n/i n/i
Limbu Sin.-Tib., Tib.-Burm. (E. Kiranti) ST ZERO ALL n/i
348
language lineage cmpr typ cmpr sprl suppl
Newar (Dolokha) Sin.-Tib., Tib.-Burm. (Newari) ST ZERO ALL n/i
Bisu Sin.-Tib., Tib.-Burman (Ngwi) ST ZERO n/i NO
Mongsen Ao Sin.-Tib., Tib.-Burm. (Naga) CN; ST M8
VERY n/i
Mizo Sin.-Tib., Tib.-Burm. (Kuki-Chin) ST PER PER NO
Dumi Sin.-Tib., Tib.-Burm. (W. Kiranti) ST ZERO ALL n/i
Lao Tai-Kadai ST; EX ZERO PER; ALL n/i
Thai Tai-Kadai ST PER PER n/i
Japanese East Asia ST ZERO (PER) NO
Korean East Asia ST (PER) (PER) NO
Pacific
Mayali Australian, Gunwingguan CN; EX ZERO CPR n/i
Nunggubuyu Australian, Gunwingguan CN ZERO NONE NO?
Arrernte Austr., Pama-Nyungan (Arandic) ST M ALL NO
Dyirbal Austr., Pama-Nyungan (Dyribalic) CN M n/i n/i
Gumbaynggir Austr., Pam-Nyu (Gumbainggiric) CN; ST M9
M n/i
349
language lineage cmpr typ cmpr sprl suppl
Walpiri Austr., Pam-Nyu (South-West) CN ZERO n/i NO
Yindjibarndi Austr., Pam-Nyu (South-West) CN; ST ZERO NONE n/i
Kayardild Austr., Pam-Nyu (Tangkic) CN; ST ZERO n/i n/i
Jingulu Australian, West Barkly CN ZERO NONE NO
Amis Austronesian, E. Formosan CN; EX ZERO PER n/i
Paiwan Austronesian, Paiwan ST M M n/i
Puyuma Austronesian, Puyuma ST ? M M? n/i
Rukai Austronesian, Rukai ST ? M M n/i
Seediq Austronesian, Atayalic EX ZERO CPR n/i
Rapanui Austronesian, Malayo-Polynesian
(Oceanic, E. Nuclear Polynesian)
ST; EX PER ALL n/i
Maori Austronesian, Malayo-Polynesian
(Oceanic, E. Nuclear Polynesian)
ST PER VERY n/i
Muna Austronesian, Malayo-Polynesian
(Sulawesi - Munic)
ST ZERO n/i n/i
350
language lineage cmpr typ cmpr sprl suppl
Pendau Austronesian, Malayo-Polynesian
(Sulawesi - Tomini)
ST M n/i n/i
Yapese Austronesian, Malayo-Polynesian
(Oceanic, Yapese)
ST ZERO PER n/i
Palauan Austronesian, Malayo-Polynesian
(Palauan)
ST ZERO n/i n/i
Fijian (Boumaa) Austronesian, Malayo-Polynesian
(Oceanic, W. Fijian-Rotuman)
CN; ST (PER) n/i n/i
Tuvaluan Austronesian, Malayo-Polynesian
(Oceanic, Samoic)
CN; ST; EX PER POS n/i
Manam Austronesian, Malayo-Polynesian
(Oceanic, W. Oceanic)
CN ZERO n/i n/i
Sinaugoro Austronesian, Malayo-Polynesian
(Oceanic, W. Oceanic)
ST ZERO ALL n/i
Amele Trans-New Guinea, Madang EX; ST ? PER ? ALL n/i
351
language lineage cmpr typ cmpr sprl suppl
Klon Tr-New Guin., Timor-Alor-Pantor ST M CPR n/i
Teiwa Tr-New Guin., Timor-Alor-Pantor CN PER POS10
n/i
Mian Tr-New Guin., Central-W. (Ok) CN ZERO PER ? NO
Ekagi Tr-New Guin., Central-W. (W.
Bomberai)
CN ZERO ALL11
n/i
Koiari Tr-New Guin., Eastern (Koiarian) CN; EX ZERO NONE n/i
Mauwake Tr-New Guin., Adelbert Range CN; EX ZERO ALL n/i
Arapesh PNG Torricelli ST ? PER NONE n/i
Skou Papuan CN ZERO ZERO/NONE n/i
Mundari Austro-Asiatic, Munda (North) ST ZERO ALL n/i
Santali Austro-Asiatic, Munda (North) ST (PER) ALL n/i
Remo Austro-Asiatic, Munda (South) ST ZERO ALL n/i
Semalai Aust-Asia, Mon Khmer (Aslian) CN; ST M n/i NO
Palaung Aust-Asia, Mon Khmer (N.) ST ZERO ? ALL n/i
352
language lineage cmpr typ cmpr sprl suppl
Vietnamese Aust-Asia, Mon Khm. (Viet-
Muong)
EX ZERO OTH12
n/i
Khmer Aust-Asia, Mon Khm. (E., Khmer) ST PER PER n/i
Sapuan Aust-Asia, Mon Khm. (E., Bah-
naric)
ST ZERO PER n/i
Americas
Aleut Eskimo-Aleut ST ZERO n/i n/i
West Greenlandic Eskimo-Aleut ST M CPR n/i
Straits Salish Salishan CN ZERO NONE n/i
Navajo Na-Dene, Athapaskan ST M M + PER n/i
Nishnaabemwin Algic, Algonquian ST PER NONE n/i
Cherokee Iroquoian ST M M n/i
Kiowa Kiowa Tanoan ST ZERO CPR; ALL NO
Chimariko isolate (Hokan?) ST? M M n/i13
Kiliwa Yuman (Hokan) CN? M14
n/i n/i
353
language lineage cmpr typ cmpr sprl suppl
Choctaw Muskogean EX ZERO n/i n/i
Crow Siouan OTH15
M ? n/i n/i
Wappo Yukian ST ZERO PER n/i
Nahuatl Uto-Aztecan, Nawan ST PER NONE n/i
Comanche Uto-Aztecan, Numic ST ZERO ? NONE n/i
Tumpisa Soshone Uto-Aztecan, Numic ST ZERO ALL NO
Cupeño Uto-Aztecan, Takic EX; ST (PER) (ALL) n/i16
Sonora Yaqui Uto-Aztecan, Sonoran ST PER ALL n/i
Tohono O’odham Uto-Aztecan, Sonoran ST PER VERY n/i
Tarascan N. Amer, isolate OTH M ? n/i n/i
Totonac N. Amer, isolate ST PER CPR; ALL n/i
Miskitu Misumalpan CN (PER) ABS n/i
Mixtec Oto-Manguean CN; ST M DEF17
NO ?
Wari’ Chapacuran CN ZERO n/i n/i
Yagua Peba-Yaguan CN; EX ZERO n/i n/i
354
language lineage cmpr typ cmpr sprl suppl
Bare Arawakan, N. (Central Upper Ama-
zonian)
ST (PER) n/i n/i
Tariana Arawakan, N. (Eastern Nawiki) ST; EX M PER n/i
Warekena Arawakan, N. (Western Nawiki) CN ZERO n/i n/i
Paumarí Arawan CN; ZERO18
n/i n/i
Apalai Carib CN; ST ZERO n/i n/i
Hixkaryana Carib CN; OTH19
ZERO n/i n/i
Bororo Bororoan (Macro-Ge ?) ST ZERO n/i n/i
Kayapó Jêan (Macro-Ge ?) CN ZERO ABS n/i
Pirahã Muran CN ZERO n/i n/i
Mbyá (Guaraní) Tupi-Guarani (Guaraní) ST M ALL NO
Urubu-Kaapor Tupi-Guarani (Tupi) CN; EX ZERO n/i n/i
Andoke S. Amer, isolate ST ZERO ABS (PER) n/i
Aymara S. Amer, isolate ST; EX PER PER, ALL n/i
Mapuche S. Amer, isolate ST PER ALL ?
355
language lineage cmpr typ cmpr sprl suppl
Mosetén S. Amer, isolate CN ZERO ABS n/i
Urarina S. Amer, isolate EX ZERO ALL n/i
Kwaza S. Amer, isolate CN ? ZERO NONE n/i
356
Appendix B
The Focussed Survey
The following reports on the second stage of the database construction. The
columns and headings are as described in the previous appendix. This survey
telescoped in on languages identified in the broad sample as having comparative
morphology and/or comparative suppletion, then expanded outwards to
more and less closely related languages, stopping where a point of diminishing
returns seemed to have been reached. For example, morphological marking
of comparison shows up in the Nakh languages, and thus I examined the
Nakh-Dagestanian family; however, after 15 languages of the Dagestanian
branches were all described as lacking a morphological comparison, I did not
probe those language groups any deeper (and listed only half of those surveyed
here). Similar remarks hold of Indo-Aryan, where there appears to be
strikingly little variation across the modern languages in how comparatives
and superlatives are formed, and thus only a representative set of languages
357
is given here. On the other hand, there are languages in this table (such as
Vlach Romani) in which there is a single table entry, but substantial variation
among dialects (see Chapter 3). Despite a certain arbitrariness in divisions,
then, in all, this survey has entries for 174 languages, of which 20 duplicate
entries from Appendix A (20 languages are in both the broad sample and
focussed survey). Between the two surveys, a total of 302 languages are
reported on here.
The aim of this telescoping was to construct a listing of languages with
the potential for suppletive superlatives that is as comprehensive as possible.
Nevertheless, coverage remains uneven in various ways. As the references
indicate, for some families more than for others, I relied on short sketches
(such as those in Vinogradov et al. 1966-1968, and in the LINCOM EUROPA
series), but consulted more extensive descriptions wherever I could,
in particular where the sketches pointed to potentially relevant phenomena
(as for example in Finno-Ugric). Relying on grammatical descriptions is the
only feasible way to undertake a study of this breadth, but I acknowledge
that my conclusions are not only at the mercy of the (highly variable) quality
of the descriptions I relied on, but involve in addition a layer of interpretation
on my part in comparing across differing traditions and terminologies, a
potential source of error. Note in particular that some grammars do explicitly
comment on the absence of suppletion, or of morphological comparisons
of inferiority, but many do not (indicated as n/i for ‘no information’ in the
table). It seems a reasonable guess that many, if not most, of these languages
358
lack suppletion, but the evidence is entirely circumstantial, namely: (i) suppletion/irregularity
is generally noted elsewhere in these descriptions, hence
should have been noted if it were attested in comparison; (ii) for many of
these, the examples given include ‘good’ and ‘many’, by far and away the
most commonly suppletive roots, hence if these are regular, chances are all
other roots are as well; and (iii) where the facts are clear, languages with
exclusively zero comparatives lack suppletion (with some possible exceptions
meaning ‘more’, see section 4.3); I find it implausible to suppose that the
phenomenon exists in languages of this type, but by coincidence only where
it happens to have escaped notice.
359
language lineage cmpr sprl suppl
Indo-European
Albanian Albanian PER PER NO
Hittite Anatolian ZERO n/i n/i1
Tocharian Tocharian ZERO? n/i n/i
Armenian (Classical) Armenian (M) ALL, CPR Y
Armenian Middle Armenian PER DEF n/i
Armenian Western Armenian (PER) PER; M; ALL Y
Armenian Eastern Armenian PER PER, M n/i
Latvian Baltic M M Y
Lithuanian Baltic M M NO
Old Church Slavonic Slavic M M Y
Belorussian Slavic M M Y
Bulgarian Slavic M2
M Y
Czech Slavic M M Y
Macedonian Slavic M3
M Y
360
language lineage cmpr sprl suppl
Polish Slavic M M Y
Russian Slavic M M Y
Serbian/Croatian Slavic M M Y
Slovak Slavic M M Y
Slovenian Slavic M M Y
Sorbian Slavic M M Y
Ukrainian Slavic M M Y
Old Irish Celtic M M Y
Breton Celtic M M Y
Irish Celtic M CPR Y
Manx Celtic M CPR Y
Scottish Gaelic Celtic M CPR Y
Welsh Celtic M M Y
Latin Romance M M Y
Catalan Romance PER PER Y
361
language lineage cmpr sprl suppl
French (Old) Romance M PER Y
French (Modern) Romance PER PER Y
Italian Romance PER PER Y
Portuguese Romance PER PER Y
Romanian Romance PER PER NO
Spanish Romance PER PER Y
Gothic Germanic M M Y
English (Old) Germanic M M Y
Icelandic (Old) Germanic M M Y
Afrikaans Germanic M M Y
Danish Germanic M M Y
Dutch Germanic M M Y
English Germanic M M Y
Faroese Germanic M M Y
Frisian Germanic M M Y
362
language lineage cmpr sprl suppl
German Germanic M M Y
German (Cimbrian) Germanic M M Y
Icelandic Germanic M M Y
Norwegian Germanic M M Y
Swedish Germanic M M Y
Yiddish Germanic M M Y
Greek (Ancient) Hellenic M M Y
Greek (Modern) Hellenic M M Y
Sanskrit Indo-Aryan M M Y
Bagri Indo-Aryan ZERO ALL n/i
Hindi Indo-Aryan ZERO ALL Y4
Kashmiri Indo-Aryan ZERO ALL ?5
Marathi Indo-Aryan ZERO ALL n/i6
Nepali Indo-Aryan ZERO ALL n/i
Sinhalese Indo-Aryan PER PER n/i
363
language lineage cmpr sprl suppl
Romani Indo-Aryan M PER n/i
Romani (Vlach) Indo-Aryan PER DEF7
Y8
Persian (Old) Iranian M M Y9
Persian (Middle) Iranian M M Y
Sogdian Iranian M PER n/i
Khotanese (Gy¯aźdian) Iranian M M Y
Balochi Iranian M M,ALL10
n/i
Gilyan Iranian M ALL Y
Kurdish, Kurmanji Iranian M CPR, PER Y
Kurdish, Sorani Iranian M M n/i
Ossetian Iranian M ALL, PER Y
Pashto Iranian ZERO ALL, VERY n/i
Persian (Farsi) Iranian M M Y
Shughni-Rushan Iranian M ALL n/i
Tajik Iranian (M) M; ALL Y
364
language lineage cmpr sprl suppl
Talysh Iranian ZERO PER n/i
Tat Iranian M M n/i
Wakhi Iranian (M) ALL n/i
Yagnobi Iranian ZERO11
ALL n/i
Zazaki Iranian ZERO ZERO n/i
Uralic
Estonian Fennic M M, ALL YES
Finnish Fennic M M YES
Ingrian Fennic ALL n/i
Karelian Fennic M M, ALL12
YES
Livonian Fennic M ALL, PER, DEF n/i
Lude Fennic M ALL n/i
Veps Fennic M PER, ALL YES
Votic Fennic M PER, ALL YES
Saami (North) Saamic M M n/i
365
language lineage cmpr sprl suppl
Saami (South) Saamic M M n/i
Saami, Kildin Saamic M M YES
Hungarian Hungarian M M YES
Khanty (Ostyak) Khanty ZERO n/i n/i
Mansi (Vogul) Mansi (M) PER n/i
Mari (Cheremis) Mari (M) PER n/i
Mordva Mordvin ZERO PER, ALL n/i
Komi Permian M PER n/i
Udmurt Permian (M) ALL n/i
Nenets (Yurak) Samoyed M M; ALL n/i
Nganasan (Tavgy) Samoyed M?13
n/i ?14
Selkup Samoyed ZERO PER n/i
Altaic
Even Tungusic ZERO ALL n/i
Evenki Tungusic M (?) M; ALL n/i
366
language lineage cmpr sprl suppl
Manchu Tungusic ZERO n/i n/i
Nanai Tungusic ZERO ALL ? n/i
Negidal Tungusic (M) (M) n/i
Orochi Tungusic ZERO PER n/i
Orok Tungusic (M) ALL; VERY n/i
Ulch Tungusic ZERO PER n/i
Azeri Turkic (PER) PER n/i
Bashkir Turkic (M) PER n/i
Chuvash Turkic (M) PER; ALL n/i
Gagauz Turkic PER PER n/i
Karachay-Balkar Turkic (M) PER n/i
Karaim Turkic M PER n/i
Kazakh Turkic M PER n/i
Khakas Turkic ZERO PER n/i15
Kirghiz Turkic (M) PER n/i
367
language lineage cmpr sprl suppl
Kumyk Turkic PER PER n/i
Nogai Turkic ZERO PER n/i
Shor Turkic ZERO PER n/i
Tatar Turkic (M) PER Y
Turkish Turkic (PER) N N
Turkic, Old Turkic M PER Y?16
Turkmen Turkic ZERO ? ABS n/i
Tyvan Turkic ZERO PER n/i
Uyghur Turkic M PER n/i
Uzbek Turkic (M) PER n/i
Bonan Mongolian ZERO n/i n/i
Buryat Mongolian ZERO VERY n/i
Dagur Mongolian ZERO 17
n/i
Kalmyk Mongolian ZERO PER n/i
Khalkha Mongolian ZERO ALL n/i
368
language lineage cmpr sprl suppl
Moghol Mongolian M18
M n/i
Mongghul Mongolian ZERO n/i n/i
Oirat Mongolian ZERO n/i n/i
Ordos Mongolian ZERO n/i n/i
Santa Mongolian ZERO n/i n/i
Shira Yughur Mongolian ZERO 19
n/i
Yeniseian
Ket Yeniseian ZERO ALL n/i
Kott Yeniseian ZERO CPR; ALL Y?20
Yugh Yeniseian (M) ALL n/i 21
Chukotko-Kamchatkan
Itelmen Chukotko-Kamchatkan ZERO n/i NONE22
Alutor Chukotko-Kamchatkan ZERO M n/i
Chukchi Chukotko-Kamchatkan M M n/i
Kerek Chukotko-Kamchatkan M M n/i
369
language lineage cmpr sprl suppl
Koryak Chukotko-Kamchatkan M M n/i
Kartvelian
Georgian Kartvelian M M Y
Mingrelian Kartvelian M PEIPRH Y
Laz (Chan) Kartvelian ZERO ALL n/i23
Svan Kartvelian M M Y
NW Caucasian
Abaza NW Caucasian PER n/i Y
Abkhaz NW Caucasian (PER) ALL Y
Adyghe NW Caucasian M?24
PER n/i
Kabardian NW Caucasian M? PER Y?25
Ubykh NW Caucasian M M n/i
Nakh-Dagestanian
Chechen Nakh (M) PER n/i
Ingush Nakh M PER n/i
370
language lineage cmpr sprl suppl
Batsbi Nakh M M n/i
Avar Dagestanian ZERO PER n/i
Andi Dagestanian ZERO n/i n/i
Botlkh Dagestanian ZERO n/i n/i
Bagvalal Dagestanian ZERO n/i n/i
Tsez Dagestanian ZERO n/i n/i
Hunzib Dagestanian ZERO ALL n/i
Lezgian Dagestanian ZERO ALL Y?26
Tabasaran Dagestanian ZERO ALL n/i
Semitic
Amharic Afro-Asiatic, Semitic ZERO ALL n/i
Arabic, Standard Afro-Asiatic, Semitic M DEF n/i
Arabic Gulf Afro-Asiatic, Semitic M DEF n/i
Arabic, Moroccan Afro-Asiatic, Semitic M POS/DEF27
n/i
Assyrian Neo-Aramaic Afro-Asiatic, Semitic (PER) DEF n/i
371
language lineage cmpr sprl suppl
Hebrew, Biblical Afro-Asiatic, Semitic ZERO ALL; DEF n/i
Hebrew, Modern Afro-Asiatic, Semitic (PER) PER Y
Maltese Afro-Asiatic, Semitic M DEF n/i
372
Appendix C
Principal Sources
In addition to the published sources noted below, I am grateful to the large
number of scholars and speakers who answered queries regarding specific
languages, including: A. Calabrese, W. U. Dressler, M. Peters, M. Weiss
(Ancient Greek, Latin, Indo-European), D. Kallulli (Albanian), H. Khanjian,
B. Vaux (Armenian), L. Kamandulyt˙e (Lithuanian), R. Pancheva (Bulgarian),
P. Caha, I. Kučerová, H. Skoumalova (Czech), S. Dyła, P. Jabłonska
(Polish), Zh. Glushan, N. Fitzgibbons, N. Radkevich, O. Tarasenkova
(Russian), Ž. Bošković, M. Despić, M. Marelj (Serbo-Croatian), A. Harrison
(Slovenian), N. Kariaeva, H. Koulidobrova (Ukranian), R. Feldstein (Old
Church Slavonic), A. Carnie (Irish), J. Phillips, E. Pyatt (Welsh), D. Adger
(Scottish Gaelic), A. Calabrese (Italian), P. Schlenker (French), S. Herdan
(Romanian), C. Buesa-García (Spanish), I. Oltra Massuet (Catalan), M. DeGraff
(Haitian Creole), H. Broekhuis, H. van der Hulst (Dutch), S. Dyk
373
(Frisian), S. Wurmbrand, W. U. Dressler, W. Öller, M. Prinzhorn, M. Hackl,
U. Sauerland (German), J. G. Jónsson (Icelandic), R. Bhatt (Hindi), B. Hettich
(Ossetian), A. Kahnemuyipour, I. Cagri (Persian) I. Laka, G. Rebuschi
(Basque), L. Nash (Georgian), M. Kelepir, N. Şener, S. Şener (Turkish), M.
Erdal (Old Turkish), Y. Sharvit (Hebrew), P. Kiparsky (Finnish), G. M.
Alekseev, N. Antropova, N. Bukina, E. Filipova, M. V. Matveena, P. M.
Zajkov, Zh. Glushan (Karelian), A. Szabolcsi, K. Szendrői (Hungarian), F.
Mc Laughlin (Wolof), C. Creider, J. T. Creider (Nandi, Swahili), J. Rennison
(Koromfe), C. Armer (Cherokee), D. Harbour (Kiowa), A. R. Deal (Nez
Perce), M. Macauley (Mixtec), G. Thomas (Mbyá, Tupi-Guaraní), I.-T. C.
Hsieh (Mandarin), I. Chung (Korean), D. Lillo-Martin (ASL), D. Nash, J.
Simpson (Warlpiri), and R. Pensalfini (Jingulu).
Africa
Afro-Asiatic
Amharic Cohen (1936)
Arabic (Gulf) Holes (1990)
Arabic (Moroccan) Ennaji et al. (2004), Harrell (2004)
Arabic (Standard) Schulz et al. (2000), Ryding (2005)
Assyrian Neo-Aramaic Arsanis (1968)
Hausa Schön (1862), Migeod (1914), Robinson (1897)
Hebrew (Biblical) Pratico and Van Pelt (2009), Joüon and Muraoka
(2005)
Hebrew (Modern) Glinert (1989)
374
Lele Frajzyngier (2001)
Maltese Aquilina (1959), Borg and Azzopardi-Alexander
(1997)
Mina Frajzyngier and Edwards (2005)
Qimant Rossini (1912)
Somali Saeed (1999)
Tamashek (Berber) Hanoteau (1896), Heath (2005)
Niger-Congo
Banda Tisserant (1930)
Fulfulde Taylor (1921), Labouret (1952)
I.jo. Williamson (1965)
Kana Ikoro (1996)
Koromfe Rennison (1997)
Lingala Odhner (1981), Meeuwis (1998)
Nandi Creider and Creider (1989)
Ogbronuagum Kari (2000)
Sotho Louwrens et al. (1995)
Swahili Wilson (1970), Ashton (1947), Steere (1884)
Tswana Cole (1955)
Wolof Mc Laughlin (2004)
Zulu Poulos and Msimang (1998), Doke (1963)
Nilo-Saharan
Koroboro Seno Heath (1999)
375
Kanuri Lukas (1937), Hutchison (1981)
Lango Noonan (1992)
Mbay Keegan (1997)
Nuer Crazzolara (1933)
Khoe
Khoekhoe (Nama) Meinhof (1909), Hagman (1977)
Khwe Kilian-Hatz (2008)
Europe and North Asia
IndoEuropean
Baltic
Latvian University of Latvia (1999)
Lithuanian Ambrazas (1997), Vasiliauskiene and Slocum
(2005-2007)
Celtic
Celtic (general) Lewis and Pedersen (1937)
Breton Hemon (1975, 1995), Ternes (1970)
Irish (Old) Thurneysen (1909)
Irish (Modern) O’Siadhail (1989), Doyle (2001)
Manx Kelly (1859), Phillips (2004)
Scottish Gaelic Calder (1990), Lamb (2003)
Welsh Rhys Jones (1977)
Germanic
376
Gothic Bopp (1856), Wright (1910)
English (Old) Sievers (1882), Bosworth and Toller (1898)
Icelandic (Old) Cleasby and Vigfusson (1874), Noreen (1923)
Afrikaans Donaldson (1993)
Danish Krymova et al. (1960)
Dutch Broekhuis (1999)
English Huddleston and Pullum (2001), Oxford English
Dictionary (OED) (online) http://www.oed.com
Faroese Thráinsson et al. (2004)
Frisian Tiersma (1985)
German Grimm and Grimm (1854-1961), Das Digitale
Wörterbuch der deutschen Sprache des 20. Jh.
(DWDS) and Wörterbuch der deutschen Gegenwartssprache
(WDG) both at: http://www.
dwds.de
German (Cimbiran) Schweizer (2008)
Icelandic Einarsson (1945), Hólmarsson et al. (1989)
Norwegian Kirkeby (1989), Strandskogen and Strandskogen
(1986)
Swedish Holmes and Hinchcliffe (2003), Prisma (1995),
Svenska Akademiens ordbok (SAOD) http://g3.
spraakdata.gu.se/saob/
Yiddish Weinreich (1984), Jacobs (2005)
377
Indo-Iranian
Sanskrit Bopp (1856), Debrunner and Wackernagel (1929)
Bagri Gusain (2000)
Hindi McGregor (1972), Kachru (2006)
Kashmiri Wali and Koul (1997)
Marathi Pandharipande (1997)
Nepali Acharya (1991)
Sinhalese Garusinghe (1962)
Romani Weigand (1888), Pobożniak (1964), Gołąb (1984),
Katsánis and Dínas (1990), Hancock (1995), Matras
(2002)
Persian (Old) Kent (1953), Branenstein and Mayrhofer (1964),
Skjærvø (2009b)
Persian (Middle) Skjærvø (2009a)
Sogdian Yoshida (2009)
Khotanese (Gy¯aźdian) Emmerick (2009)
Balochi Rastorgueva (1966), Jahani and Korn (2009)
Gilyan Rastorgueva et al. (1971)
Kurdish Bakaev (1966), Thackston (2006b,a)
Ossetian Abaev (1964), Isaev (1966), Hettich (2010)
Pashto Tegey and Robson (1996), Robson and Tegey
(2009)
378
Persian (Farsi) Hayyim (1934-1936), Mace (2003), Windfuhr and
Perry (2009)
Shughni-rushnan Sokolova (1966), Edelman and Dodykhudoeva
(2009)
Tajik Rastorgueva (1963), Windfuhr and Perry (2009)
Talysh Pirejko (1966), Schulze (2000)
Tat Grjunberg (1966)
Wakhi Pakhalina (1966), Bashir (2009)
Yagnobi Khromov (1972)
Zazaki Paul (2009)
Romance
Latin Lewis and Short (1879), Kühner and Holzweissig
(1912), Weiss (2009)
Catalan Badia Margarit (1962)
French (Old) Bauer and Slocum (2006)
French (Modern) Dietiker (1983), Judge and Healey (1983)
Italian Grandgent and Wilkins (1915)
Portuguese Perini (2002)
Spanish Bruyne (1995)
Romanian Nandris (1986)
Slavic
Old Church Slavonic Leskien (1962), Lunt (1959), Diels (1963), Vondrák
(1900)
379
Belorussian Mayo (1993)
Bulgarian Hauge (1999), Alexander (2000)
Czech Janda and Townsend (2000)
Macedonian Friedman (2002), Lunt (1952)
Polish Rothstein (1993)
Russian Kuznecova and Efremova (1986), Katzner (1994),
Garde (1998)
Serbo-Croatian Stanojčić and Popović (1992), Browne (1993),
Barić et al. (1997), Klajn (2005)
Slovak Short (1993)
Slovenian Herrity (2000)
Sorbian Seiler (1830), Schuster-Šewc (1996)
Ukrainian Shevelov (1993)
Other Indo-European
Proto Indo-European Bopp (1856), Kuryłowicz (1964), Weiss (2009)
Albanian Kacori (1979)
Armenian (Classical,
Middle)
Karst (1901), Krause and Slocum (2004)
Armenian (Middle) Karst (1901)
Armenian (Western) Riggs (1856), Bardakjian and Vaux (2001)
Armenian (Eastern) Kozintseva (1995), Dum-Tragut (2009)
Greek (Ancient) Seiler (1950), Frisk (1960-1970), Chantraine
(1967), Liddell and Scott (1996)
380
Greek (Modern) Sofroniou (1962), Holton et al. (1997)
Hittite Lehmann and Slocum (2005)
Tocharian (A,B) Krause and Slocum (2007-2010)
Uralic (Finno-Ugric, Samoyedic)
general: Fuchs (1949), Nau (1992)
Estonian Tauli (1973, 1983), Viitso (1998)
Finnish Hakulinen (1957), Karlsson (2008)
Hungarian Tompa (1968), Országh (1988)
Kamass Castrén (1854), Künnap (1999)
Karelian Rjagoev (1977), Nau (1992), Zajkov (1999)
Khanty (Ostyak) Redei (1965), Gulya (1966), Abondolo (1998a)
Komi Lytkin (1966), Coates (1982), Hausenberg (1998)
Livonian Raun (1949), Nau (1992)
Lude Nau (1992)
Mansi (Vogul) Rombandeeva (1966), Keresztes (1998)
Mari (Cheremis) Kovedjaeva (1966a,b), Kangasmaa-Minn (1998)
Mordva Zaicz (1998), Ermushkin (2004)
Nenets (Yurak) Castrén (1854), Décsy (1966), Salminen (1998)
Nganasan (Tavgy) Castrén (1854), Tereshchenko (1966, 1979), Helimski
(1998a)
Saamic Sammallahti (1998)
381
Saami (Kildin) Kert (1971)
Saami (North) Nickel (1990)
Saami (South) Bergsland (1982)
Selkup Prokofjeva (1966), Helimski (1998b)
Udmurt (Votyak) Tepljashina (1966), Csúcs (1998), Winkler (2001)
Veps Zajtseva (1981), Nau (1992)
Votic Ariste (1968), Nau (1992)
Tungusic
Even Benzing (1955), Malchukov (1995)
Evenki Nedjlakov (1997), Bulatova and Grenoble (1999),
Bulatova (1999)
Manchu von Möllendorff (1892), Avrorin (2000)
Nanai Avrorin (1982)
Negidal Kolesnikova and Konstantinova (1968), Tsintsius
(1982)
Oroči Avrorin and Lebedeva (1968)
Orok Petrova (1967)
Udihe Sunik (1968a)
Ulch Sunik (1968b)
Yeniseian
Ket Castrén (1858), Vall and Kanakin (1990), Werner
(1997c), Vajda (2004)
Kott Castrén (1858), Werner (1997a)
382
Yugh Castrén (1858), Werner (1997b)
Chukotko-Kamchatkan
Itelmen Bogoras (1922), Volodin (1976), and field notes
Alutor Zhukova (1968a), Nagayama (2003)
Chukchi Bogoras (1922), Skorik (1961, 1977)
Kerek Skorik (1968)
Koryak Zhukova (1968b, 1972)
Turkic
Azeri Gadzhieva (1966)
Bashkir Poppe (1964), Juldashev (1966), Usmanova
(2007)
Chuvash Krueger (1961), Andreev (1966)
Gagauz Pokrovskaja (1966)
Karachay-Balkar Khabichev (1966)
Karaim Musaev (1966)
Kazakh Kenesbaev and Karasheva (1966), Krippes (1996)
Khakas Karpov (1966)
Kirghiz Hebert and Poppe (1963), Junusaliev (1966)
Kumyk Magomedov (1966)
Nogai Baskakov (1966)
Shor Babushkin and Donidze (1966)
Tatar Dmitrieva (1966), Zakiev (1966), Burganova et al.
(1969), Zakiev et al. (1997)
383
Turkish Lewis (1967)
Turkic, Old Erdal (2004)
Turkmen Azimov et al. (1966)
Tyvan Sat (1966), Anderson and Harrison (1999), Harrison
(2000)
Uyghur Kajdarov (1966)
Uzbek Sjoberg (1962), Reshetov (1966)
Mongolic
Bonan Hugjiltu (2003)
Buryat Bertagaev (1968), Skribnik (2003)
Dagur Tsumagari (2003)
Kalmyk Todaeva (1968), Bläsing (2003)
Khalkha Street (1963), Kullmann and Tserenpil (1996),
Svantesson (2003)
Moghol Weiers (2003)
Mongghul Georg (2003a)
Oirat Birtalan (2003)
Ordos Georg (2003b)
Santa Kim (2003)
Shira Yughur Nugteren (2003)
Kartvelian
Georgian Aronson (1989), Hewitt (1995, 1996)
Mingrelian Kipshidze (1914), Kiziria (1967), Harris (1991b)
384
Laz (Chan) Marr (1910), Dirr (1928), Holisky (1991)
Svan Topuria (1967), Gudjedjiani and Palmaitis (1986),
Schmidt (1991)
NW Caucasian
Abaza Lomtatidze (1967a), Tabulova (1976), O’Herin
(1995)
Abkhaz Aristava et al. (1968), Chirikba (2003), Hewitt
and Khiba (1979), Hewitt (1991), Lomtatidze
(1967b)
Adyghe Jakovlev and Ashxamaf (1941), Rogava and Kerasheva
(1966), Kumakhov (1967)
Kabardian Shagirov (1967), Colarusso (1992)
Ubykh Dumézil (1931)
Nakh-Dagestanian
Chechen Desheriev (1967b), Nichols (n.d.)
Ingush Dolakova (1967), Nichols (2011)
Batsbi Desheriev (1967a)
Avar Madieva (1967)
Andi Tsertsvadze (1967)
Botlkh Gudava (1967b)
Godoberi Gudava (1967c)
Karatin Magomedbekova (1967a)
Akhvakh Magomedbekova (1967a)
385
Bagvalal Gudava (1967a)
Tindi Gudava (1967d)
Chamalin Magomedbekova (1967b)
Tsez Bokarev (1967)
Hunzib van der Berg (1995)
Lak van der Berg (1995)
Dargi Abdullaev (1967)
Lezgian Haspelmath (1993)
Tabasaran Magometov (1965)
Dravidian
Dravidian (general) Steever (1998a), Krishnamurti (2003)
Malayalam Asher and Kumari (1997)
Eurasia: Isolates
Ainu Batchelor (1905), Refsing (1986), Tamura (2000)
Basque Trask (1997, 2003), de Rijk (2008)
Burushaski Berger (1999)
Nivkh Krejnovich (1968), Gruzdeva (1998)
Yukagir Maslova (2003)
Southeast Asia
Sino-Tibetan
Cantonese Matthews and Yip (1994)
386
Kunming Gui (2000)
Mandarin Li and Thompson (1981)
Dzongkha van Driem (1998)
Kham Watters (2002)
Lahu Matisoff (1973)
Limbu van Driem (1987)
Newar (Dolokha) Genetti (2007)
Bisu Shixuan (2001)
Mongsen Ao Coupe (2007)
Mizo (Lúshái) Shaha (1884)
Dumi van Driem (1993)
Tai-Kadai
Lao Morev et al. (1972), Enfield (2007)
Thai Iwasaki and Ingkaphirom (2005)
Other
Japanese Martin (2004), Beck et al. (2004), Oda (2008)
Korean Lukoff (1993), Underwood (1914)
Australia and Pacific
Australian
Mayali (Bininj Gun-
wok)
Evans (2003)
387
Nunggubuyu Heath (1984)
Arrernte Strehlow (1942)
Dyirbal Dixon (1972)
Gumbaynggir Smythe (1948), Eades (1979)
Walpiri Nash (1980), Simpson (1983)
Yindjibarndi Wordick (1982)
Kayardild Evans (1995)
Jingulu Pensalfini (2003)
Austronesian
Amis He et al. (1986a)
Paiwan Egli (1990)
Puyuma Teng (2008)
Rukai Zeitoun (2007)
Seediq Holmer (1996)
Rapanui Du Feu (1996)
Maori Bauer (1983)
Muna van den Berg (1989)
Pendau Quick (2007)
Yapese Jensen (1977)
Palauan Josephs (1975)
Fijian (Boumaa) Dixon (1988)
Tuvaluan Besnier (2000)
388
Manam Lichtenberk (1983)
Sinaugoro Tauberschmidt (1999)
Papua New Guniea
Amele Roberts (1987)
Klon Baird (2008)
Teiwa Klamer (2010)
Mian Fedden (2007)
Ekagi Drabbe (1952)
Koiari Dutton (2003)
Mauwake Berghäll (2010)
Arapesh Fortune (1942)
Skou Donohue (2004)
Austro-Asiatic
Semalai Kruspe (2004)
Vietnamese Thompson (1987)
Khmer Huffman (1970)
Sapuan Jacq and Sidwell (1999)
Palaung Milne (1921)
Santali Neukom (2001)
Remo Fernandez (1967)
389
America
North America
Aleut Bergsland (1997)
West Greenlandic Fortescue (1984)
Straits Salish Jelinek and Demers (1994)
Navajo Young and Morgan (1987), Bogal-Allbritten
(2008)
Nishnaabemwin Valentine (2001)
Cherokee Pulte and Feeling (1975)
Kiowa Watkins and MacKenzie (1984)
Chimariko Jany (2009)
Kiliwa Mixco (2000)
Choctaw Nicklas (1979), Broadwell (2006)
Crow Graczyk (2007)
Wappo Thompson et al. (2006)
Nahuatl (Classical) Andrews (1975)
Comanche Charney (1993)
Tumpisa Soshone Dayley (1989)
Cupeño Hill (2005)
Sonora Yaqui Dedrick and Casad (1999)
Tohono O’odham (Pa-
pago)
Zepeda (1983)
390
Purépecha (Tarascan) de S. Juan Crisóstomo Nájera (1879), Foster
(1969), Swadesh (1969), Chamoreau (2003)
Totonac (Misantla) MacKay (1999)
Mixtec Macaulay (1996)
Miskitu Conzemius (1929)
South America
Wari’ Everett and Kern (1997)
Yagua Payne and Payne (1990)
Bare Aikhenvald (1995)
Tariana Aikhenvald (2003)
Warekena Aikhenvald (1998)
Paumarí Chapman and Derbyshire (1991)
Apalai Koehn and Koehn (1986)
Hixkaryana Derbyshire (1985)
Bororo Crowell (1979)
Kayapó Sala (1914)
Pirahã Everett (1986)
Mbyá (Guaraní) Pederson (1977)
Urubu-Kaapor Kakumasu (1986)
Andoke Landaburu (1979)
Aymara Inojosa (1966)
Mapudungun Zuñiga (2006), Smeets (2007)
391
Mosetén Sakel (2004)
Urarina Olawsky (2006)
Kwaza van der Voort (2004)
392
Notes
1
Not all languages draw a clear distinction between the two types of superlative,
and grammatical descriptions are sometimes vague on this point.
As a rule of thumb, I have tried to note all questionable cases below wherever
an analytical decision bears on the status of the empirical generalizations
under discussion.
2
As is often the case, there is some degree of inconsistency in the use of
these terms in the literature. Some (for example, Hackl 2009) use the terms
‘relative’ and ‘absolute’ superlative not in the sense just indicated, but rather
to refer to two different readings of the (relative) superlative, namely those
termed the ‘comparative’ and ‘absolute’ readings elsewhere.
3
This generalization is mentioned tentatively in Cuzzolin and Lehmann
(2004, 1213), with no indication of the sample size they drew on. I have
found no other mention of this in the literature (other than in the descriptions
of individual languages), nor any systematic survey, though I imagine it is
widely suspected to hold.
4
For differing views on this point within DM, see Embick (2010, 193, n1)
and Marantz (1996, 2007).
5
Compare Halle’s 1997 formulation of the Subset Principle which combines
the Elsewhere Condition as a condition regulating rule interaction, with
a general condition on rule application that a rule may apply only when its
structural description (including contextual restrictions) constitutes a subset
393
of the context being considered. Thus, a rule may not apply if its description
includes features that are not part of the node to which insertion applies.
6
See also Anderson (1992), Stump (2001) on disjunctive rule application.
In contrast to these theories, DM holds that the locus of disjunctivity is the
terminal node, thus ‘rule blocks’ are in the general case featurally coherent.
However, DM provides mechanisms for manipulating the structure after
syntax but prior to vocabulary insertion, which allows for the description of
departures from a 1:1 mapping of terminal nodes to exponents. See Halle
and Marantz (1993), Noyer (1997) for discussion and comparison with the
framework of Anderson (1992) in particular.
7
Marantz (1989, 261) defines Morphological Merger as follows:
(i) At any level of syntactic analysis (d-structure, s-structure, phonological
structure), a relation between X and Y may be replaced by
(expressed by) the affixation of the lexical head of X to the lexical
head of Y
The core cases of Merger encompass rebracketing under adjacency as an
account of certain cliticization effects: [ X [ Y ... ]] → [ X+Y ... ], and the
affixation of a head X to the head of its complement, as in affix-hopping:
[ X [Y P Y ]] → [Y P X+Y]. In practice, Merger of the affix hopping type
tends to effect a lowering operation, though this is not strictly a part of the
definition. See Bobaljik (2002a) for more discussion.
8
This is a matter of ongoing debate, see Bonet and Harbour (To appear),
394
Embick (2010) for current discussion.
9
Embick (2010) proposes a condition of linear adjacency, in addition to a
domain-based structural condition. In essence, Embick’s theory incorporates
strict adjacency, but allows for the deletion or ‘pruning’ of X when X is
non-overt.
10
A large number of grammatical descriptions, in particular older descriptions
of languages of the Americas, provide no information on how comparison
is expressed. Many of these languages may well have no clear comparative
construction and/or may be classified in the conjoined type (see section 1.4).
In choosing representatives for a given family, stock or area, some priority was
given to languages discussed in previous typological surveys of comparison
(Ultan 1972, Stassen 1985, 2008) to facilitate comparability across studies,
although this proved impractical in many cases (either due to difficulty in
obtaining data, or since many of these studies include an overabundance of
languages from some families).
11
The conjoined comparative strategy may be endangered. Of the 109
languages in Stassen (1985), 42 are spoken by more than 1 million people
according to the figures in Ethnologue Gordon (2005). Taking population to
be a rough proxy for endangered status, we might consider these 42 to be
relatively safe. In the 109-language sample, 20 (18%) are conjoined comparative
languages, but only one of these (Nahuatl, with 1.4 million speakers)
is among the 42 relatively large languages. The relative proportion of the
other comparative types remains essentially constant between the full 109
395
languages, and the 42 large languages, with standard comparatives outnumbering
exceed comparatives by roughly 3½:1.
12
For some languages, there is somewhat of a fuzzy boundary between
‘exceed’ type comparatives and standard comparatives. In Mandarin and
Thai, for example, the element that marks comparison is historically derived
from a verb meaning ‘exceed, surpass’ but is no longer obviously a verb
synchronically, and may instead be analyzed as a standard marker. Deciding
the issue one way or the other for these languages does not affect the points
of interest in this book.
13
An apparent additional type (not generally noted in the previous literature)
is represented in my sample in the (distantly related) Formosan
languages Amis and Rukai, and possibly in Purepecha (Tarascan), an isolate
in Mexico.
In Amis and Rukai, there is no distinct adjectival category; propertydenoting
predicates are verbal. In both languages one way of expressing
comparison is to express the standard of comparison as the direct object of
the property denoting-predicate, with partial reduplication in the verb in
Rukai (Zeitoun 2007, 182), but no further marking in Amis, illustrated here.
(i) maSu@Su-(aj)
fat-prog
kuni
this
a
(gen)
wawa
kid
‘This kid is fat.’
396
(ii) taPaNajaj
big
ku
subj
waÙu
dog
tu
obj
puSi
cat
‘The dog is bigger than the cat.’ (He et al. 1986b, 61-63)
I suspect this is best seen as a special case of standard-marking where
there is no morphological distinction between direct objects (patients) and
(certain) obliques, a property shared with other Formosan languages. In
Paiwan, the standard in a comparative construction is marked by the particle
tua/tjai which also marks direct objects, but which is described as a sort of
elsewhere marker for phrases in a variety of syntactic functions (Egli 1990,
189-199). Kuo (2008) shows that Amis has a variety of means for expressing
comparison, including conjoined comparatives, exceed constructions (with
main predicate ‘more’, as in other Austronesian languages) and standardlike
constructions with oblique marking of the standard NP; one of these
constructions involves a prefix on the predicate. (I thank I.-T. C. Hsieh for
providing English translations of examples in He et al. 1986b.)
In Purepecha (Tarascan), as described in Foster (1969, 126-128) a transitivizing
morpheme -ku- (which has a variety of functions) may be added to
a verbalized adjective to yield a comparative, with the standard expressed as
a the direct object. This construction is somehow ephemeral—older descriptions
(de S. Juan Crisóstomo Nájera 1879, Swadesh 1969) give an exceed
comparative, while a more recent description (Chamoreau 2003) shows a
periphrastic construction with the borrowed Spanish adverb mas.
14
But note the brief discussion of comparison in the Hokan language Kiliwa
397
in Mixco (2000, 41-42). This language appears to use conjoined clauses, with
loose literal paraphrases such as ‘he’s not like me; he’s tall-er’; though there
is no discussion of whether the affix glossed more can be used outside of
this construction. Another relevant example may be Dyirbal. Dixon (1972)
give a comparative affix -baóa, used in examples lacking an overt standard,
such as (i). (Dixon 1972, 227) notes that when “an explicit comparison is
needed”, a conjoined construction is used, as in (ii)-(iii), which may contain
the comparative on one or both clauses.
(i) Ninda
you
midi–baóa
small-cmpr
‘You are smaller.’
(ii) Nad%a
I
bulgan-baóa
big-cmpr
bayi
he
midi–baóa
small-cmpr
‘I’m bigger [than him]; he’s smaller [than me]’
(iii) bayi
he
bulgan
big
Nad%a
I
bulgan-baóa
big-cmpr
‘He’s big [but] I’m bigger.’
Dixon (1977, 245) reports a similar pattern for the affix -wad¸an in Yidiñ,
though notes that it is also used as a marker of emphasis, glossable as ‘very’.
15
This was first drawn to my attention by F. Mc Laughlin.
16
Intrinsically comparative verbs for some adjectival meanings are also
reported for Mayali (Evans 2003, 569-570).
398
1
For familiarity, examples will generally be given in the standard orthography
or standard transliteration of the languages in question, and in citation
forms wherever practical. Thus, even though our interest is in the roots or
stems and the comparative and superlative morphology, many examples will
contain additional (irrelevant) inflectional material, such as the masculine,
nominative, singular ending -us in Latin (28b) and elsewhere.
2
Ultan suggested that the generalization represents a strong tendency and
not an absolute. In part, this is because the ABC pattern is a violation of
the pattern he saw, which recognizes only the ABB pattern, in contrast to
the CSG below, which also admits ABC. In addition, Ultan believed ABA
patterns to occur, however, the one apparent ABA pattern he noted seems
to be misanalyzed (see n. 23 below.)
3
Here and throughout, diagrams represent stems to which inflectional
morphology may be added. The comparative embeds the positive stem (or
base), minus inflectional morphology.
4
The Nesting Hypothesis is also reminiscent of proposals for an implicational
hierarchy among a series of functional heads in particular domains,
where the inventory of functional heads is subject to cross-linguistics variation,
but the presence of a higher head entails the presence of lower heads. See
Wurmbrand (2008, 2011c) for a proposal regarding variation in the structure
of infinitival complements, and Radkevich (2010) on PP structure.
5
Note also that I must reject a pos morpheme in the positive, a morpheme
that, although not clearly manifest in overt morphology in any known lan-
399
guage, is an important component of some semantic treatments; see Kennedy
2007b. I return to this question in section 4.2, below.
6
Here and below, the rule fragments presented are only those required for
the point under discussion. The Czech vocabulary includes additional rules
introducing other allomorphs of the comparative suffix, notably -ější/-ejší,
along with contextual restrictions regulating their distribution, and the same
holds for comparative and superlative exponents in other languages below.
These complexities do not affect the structure of the argument and are left
aside for ease of exposition.
7
Note further that the rule introducing the zero comparative in (42) is formulated
as a rule of exponence. On the assumption that vocabulary insertion
proceeds cyclically from the root outwards, (42) thus automatically follows
vocabulary insertion at the root. Rule (42) must be prevented from deleting
or pruning the comparative node prior to root insertion, else it would render
the root and the superlative adjacent, incorrectly feeding rules like (45a)
and bleeding those like (39d). The approach to impoverishment in Trommer
(1999) treats all impoverishment rules as special cases of zero-vocabulary insertion,
which would ensure the correct ordering in the cases at hand. See
also Embick (2010) for discussion of pruning and adjacency.
8
My thanks to Jessica Rett for this observation.
9
I thank Ivona Kučerová and Pavel Caha (personal communication 2010)
for discussion of the Czech forms. Both note that pře-dobr-ý is somewhat
stilted or ‘bookish’, but is clearly the form that fills the slot indicated in
400
(50a), as opposed to a form built on the root lep-.
The point here is that suppletion in the comparative automatically extends
to the superlative, unless there is further suppletion (an ABC pattern),
but the suppletion in the comparative does not automatically extend to the
equative. Nothing prevents the comparative allomorph being specified to occur
in the equative context as well. Thus, Welsh bach ‘small’ has suppletive
allomorph llai ‘smaller’, which occurs also in both the superlative and equative;
Welsh da ‘good’ has four distinct roots, one for each grade (including
the positive).
10
In chapter 7, I will offer a very tentative speculation on why this might be
a property of UG; the leading idea being that UG may impose a complexity
threshold on certain classes of morphemes. The meaning of the superlative,
involving comparison and universal quantification over the standard, would
exceed that threshold, and would thus necessarily be bi-morphemic.
11
A rather naive application of basic statistical tests here yields robust
significance, even with the apparent problematic patterns included. But
what is really driving this result is the overwhelming number of ABB patterns
as against anything else. Within a given language, it is plausible that the
ABB pattern for one root influences the abstract pattern for others, and
thus without knowing the weight of this effect, applying statistical measures
seems at best premature. Relatedly, an assumption of equal proportionality
as the null hypothesis seems highly implausible, but I see no clear way of
establishing a baseline chance distribution of patterns.
401
12
That is not to say that there is no borrowing in this domain. The particles
that make periphrastic comparatives (and superlatives) are borrowed,
creating in some cases new comparative constructions in the target languages.
Some languages of South and Central America have borrowed Spanish mas or
Portuguese mais to form periphrastic comparatives (Tarascan, Chamoreau
2003; Paumarí, Chapman and Derbyshire 1991), while indigenous languages
of Russia have borrowed Russian bolee and/or samyj. In a few cases, morphological
comparatives and superlatives are borrowed, but in borrowings from
Sanskrit and Persian into Hindi, (McGregor 1972), and from Tajik into Yagnobi
(Khromov 1972), for example, the borrowed morphology is only used
with borrowed adjectives. More common is borrowing of forms that constitute
suppletive comparatives in the donor language, but lose their comparative
morphosyntax in the recipient language, serving as intensifiers, in some
cases subject to further comparison or superlative formation. Examples of
this sort include Persian beh-tar ‘better’ into Hindi and Turkish, and (ultimately)
Latin optimus into English optimal, an adjective with a meaning like
‘best’ but which is not morphosyntactically a superlative form. See section
4.1.
13
The status of the superlative here is somewhat unclear, but orthogonal
to the immediate point. Most sources note that the superlative is formed
by adding naj- to the comparative (e.g., Vondrák 1900, 161) and none note
any irregularities in the formation of naj- derivatives. On the other hand,
Diels (1963, 201, n.3) claims that OCS had no means of forming a superlative,
402
stating that the naj- prefix is very rarely found, and when it is, has in the first
instance essentially no superlative meaning. Evidently, the prefix became the
standard means of forming morphological superlatives in all the daughter
languages, so this quibble is moot here.
14
Superlatives for (55c,f) were not given in the sources from which these
were taken, but the sources indicates superlatives formed by naj- prefixation
to the comparative.
15
That the Greek forms belong in this paradigm, with the comparative
cognate to those in min- but the positive root unrelated, is asserted based
on Frisk (1960-1970). The Slavic root mal- is related to Germanic small, and
not to the comparative, according to Vasmer (1953-1958).
16
The most extensive cases of one:many relations in suppletive comparatives
are in Old Church Slavonic and Ancient Greek, both written languages
representing an area with multiple dialects. There is thus a specter of artificiality
about this relationship, though one that does not affect the point
made here, namely, that there is variation within Slavic.
1
A reviewer points out that the examples in (60e-h) support the Containment
Hypothesis if their structure is [ sprl [ X cmpr ] ] but that they are
also superficially consistent with an alternative parse: [ [ sprl X ] cmpr ].
Note that for none of these is there any compelling evidence (that I am
aware of) for the alternative parse. For example, in all cases, the smaller
constituent [ X cmpr ] surfaces as the comparative, but a putative intermediate
[ sprl X ] is ill-formed (notwithstanding the remark on Hungarian
403
locative adverbs in n. 4).
2
Note that Ultan’s classificatory criteria were different from those used
here in ways which mean that his claim, though ultimately correct, needed
to be reevaluated. For example, Ultan does not distinguish, in making this
claim, among morphological and periphrastic marking of comparative and
superlative, and includes ‘exceed’ type comparatives, which are always formally
identical with superlatives, in his pool of data. Note also that some
of Ultan’s illustrative examples turn out, on closer inspection, to be absolute,
not relative superlatives. His first example of a superlative formed of a
comparative plus an additional element is the Tswana example in (i), where
Ultan indicates that reduplication of the intensifying adverb bogolo, used in
comparatives, constitutes the formal marking of the superlative.
(i) tl“ou
elephant
ethata
strong
bogolo-bogolo
more-more
m“odiph“ol“og“ol“ong
loc.animals
‘The elephant is the strongest of the animals.’ (Ultan 1972, 140)
However, the description of comparative and superlative formation in
Cole (1955, 423-5), from which Ultan’s example is taken, makes it clear that
the adverb here is an optional intensifier in the comparative construction
(and not the marker of comparison per se), and that there is no formal distinction
between comparative and superlative, although the latter is formed
“sometimes with the additional use of the reduplicated and therefore more
intensive adverbs bogolo-bogolo and thata-thata” (Cole 1955, 424).
404
3
Various qualifications are in order regarding the older languages, and I
thank Andrea Calabrese, W. U. Dressler, and Michael Weiss for their help
at various points with this material. Proto-Indo-European had at least two
formatives that yielded comparatives, only one of which, *-i
“
os-, with zerograde
*-is-, was contained in the corresponding superlative. Comparatives
in *-tero- have corresponding superlatives in *-tm
˚
mo, with no overt containment.
Note also that the Latin -issimus was not the only means of
forming superlatives. See Kühner and Holzweissig (1912), Cowgill (1970),
Weiss (2009) for discussion of Latin comparatives and superlatives in their
historical context.
4
The prefix leg- attaches to spatial adverbs without a comparative suffix,
to form pairs such as alsó – leg-alsó ‘down’ – ‘bottommost’. The superlative
prefix can also be doubled to form an ‘excessive’ grade, thus: nagy ‘big’ –
nagy-obb ‘bigger’ – leg-nagy-obb ‘biggest’ , and leg-es-leg-nagy-obb ‘the very
biggest’ (Kiefer 2001, 277). On the etymological source of the Hungarian superlative,
and evidence that it is a relatively late innovation (16th Century),
see Fuchs (1949). Fuchs (p.226) also gives examples from Hungarian dialects
of a borrowed Slavic prefix for superlatives, which adheres to the pattern in
(60), thus compare: náj-nagy-obb ‘biggest’ to the standard Hungarian forms
just cited.
5
The use of a definiteness marker (or demonstrative) to form the superlative
also occurs in languages in which a comparative construction involves
no overt marking on the adjective, or property-denoting predicate, as in
405
Tamashek (Berber) (Hanoteau 1896). If the definite article is indicative of
an NP (or DP) structure, then languages that use a nominalizer to form the
superlative would also belong to this type. Mixtec is of this sort (the ‘additive’
suffix in these examples appears to be a type of contrastive or emphatic
marker, and not a comparative suffix as such):
(i) rùPù
I
s´¯uk´¯u=ka=rí
tall=add=1
asù
than
roPo
you
‘I am taller than you.’
(ii) roPo
you
kúu
cop
xa-náxini=ka
nom-drunk=add
‘You are the drunkest.’ (Macaulay 1996, 162-164)
6
I thank John Phillips for drawing my attention to the distinct descriptive
traditions here, and to the question of whether this is a distinction without
a difference.
7
The superlative in Moroccan Arabic can be formed either by placing the
comparative form (for adjectives that have one) in the construct state with
the noun it modifies, or by using the simple adjective in conjunction with
the definite article (Harrell 2004, 205).
8
In this variety, the definite article cliticizes to the adjective: bunu
“
– cama
bunu
“
– cama bun-lu ‘good – more good – more good-the’; my thanks to Jason
Merchant for the pointer to this variety.
9
Original: “Die Setzung oder Nichtsetzung des Artikels dürfte nicht zur
406
Unterscheidung der beiden Steigerungsstufen (wie teilweise in romanischen
Sprachen) dienen...”
10
This pattern also holds for some adjectives in Standard Estonian (Tauli
1973, Viitso 1998), although the synthetic superlative in Estonian is characterized
as occurring primarily in the written language, where some authors
claim it is a Finnish calque (see Nau 1992, 16 and Laakso 2001, 198). Other
Fennic languages do not have a synthetic superlative.
11
The comparative suffix is reconstructed to Proto-Uralic (see, e.g., Fuchs
1949, Sammallahti 1998, 81), but the superlative element in -i- appears to be
a Fennic innovation. Saami shows comparatives in -bu, cognate to Finnish
-mpi (and Hungarian -(V)bb), but the Saami superlative suffix is -mus, for
which Sammallahti (1998, 81) gives a distinct etymology, not containing the
comparative. Rießler (2007) suggests that Kildin Saami has innovated a
transparently nested structure, with superlatives in -a-mus containing the
comparative -a, from older -amp together with the Saami superlative forma-
tive.
Although the suffix in question is reconstructed for Proto-Uralic, it only
has a comparative sense in Saamic-Fennic and Hungarian, whereas in the
Samoyedic languages the reflex of this suffix survives with a tempering or
relativizing meaning, for example as in Nenets -mboj ‘rather X’ (Décsy 1966,
59, see also Fuchs 1949). A blurry boundary between a relativizing and a
comparative function is also seen in Turkic -raX. It is worth noting in light of
section 2.3 that the group of languages within Uralic for which the modern
407
suffix is a comparative do not form a single branch, but are those which are
spoken in areas historically adjacent to Indo-European languages.
12
These languages are well known for a rich and complex system of stem
alternations, and the final -i- is characteristic of (a class of) plural stems. In
Estonian, for example, Tauli (1973, 86) notes that the stem of the superlative
for some adjective classes is identical to the so called i-stem of the plural. At
least in theory, then, it is possible that the -i- is part of a stem alternation
and not an exponent of the superlative, as such. It is perhaps worth noting
that in related Udumrt, the comparative suffix follows the plural suffix on
adjectives (Winkler 2001, 40).
13
Such structures could be base generated or derived from (67) by morphological
operations such as rebracketing under adjacency (Kiparsky 1983,
Sproat 1985, Radkevich 2010) or a variety of merger (Marantz 1989, Embick
and Noyer 1999).
14
I treat the alternation pare- ∼ parha- as irregular, and not suppletive.
15
The forms for ‘good’ are also suppletive in Basque, with suppletion carrying
over to the -xe- comparative: on – hobe – hobe-xe-ago ‘good – better – a
little better’ (de Rijk 2008, 711). There are two further complications regarding
the ‘good’ paradigm, namely, the absence of -ago in the true comparative,
versus its presence in the -xe- grade, and the interaction of adjacency with
the corresponding adverbial forms; I return to these in Chapter 5.
16
An apparently similar construction exists in Lithuanian, with the element
-˙el- interposed between the adjective root and the comparative suffix, as in
408
maž-˙el-èsnis ‘small-el-cmpr’ = ‘a little smaller’ (Vasiliauskiene and Slocum
2005-2007). However, Lithuanian shows no suppletion in adjectival gradation
and thus the prediction discussed immediately above cannot be tested in that
language.
17
The morpheme order adj-sprl-cmpr is also given for Epena Pedee in
Harms (1994, 81) but from the description and the one example given (glossed
as ‘the other smallest one’) this appears to be an intensified comparative
(‘even more’) and/or absolute superlative. If the Paiwan superlative tjala...-an,
mentioned briefly below (60) does indeed contain the comparative tja,
then the structure of this superlative is also more complex than a simple
nesting structure. The element -la- in the prefix tjala- may be segmentable
as an emphatic element, on the strength of ‘weak intensifier’ ka-...-an versus
superlative kala-...-an (Egli 1990, 148-149). The element that derives
superlatives in Mongsen Ao likewise occurs inside the comparative-forming
suffix, but it is not clear that these are genuine comparatives as opposed to
relative clauses; see note 8.
18
Though note that for Lechner (2004), the comparative affix is not the
interpreted degree head but in essence a type of agreement between the
adjective and the null comparative head.
19
The tripartitie pattern in (74b) occurs only with the long form of ‘better’,
itself one of the very small set of suppletive inflected comparatives; otherwise,
the superlative embeds the positive grade. The form sam-yj xoroš-ij is
possible alongside (74b).
409
20
A superlative suffix -tku is also possible in place of -tmer. Bulatova and
Grenoble (1999, 21) also give a form with just the bare adjective. Bulatova
and Grenoble state that the -tku form is the superlative, while the comparative
consisting of the positive form with the universal quantifier forms “an
even higher degree of the adjectival quality.” This may be what Dressler and
Kiefer (1990) discuss for Hungarian and Viennese German as regards a degree
beyond superlative (the best of the best). In any event, compare (75d)
to the example from related Even in (76a), in which the suffix (evidently
cognate to the Evenki compartive marker) is glossed as an intensifier, rather
than a comparative. It is not clear to me whether there is a real grammatical
distinction between these languages in this respect.
21
Searched July 2010.
22
This may be related to a puzzle in plural superlatives identified by Stateva
(2005). Stateva noted that theories of the superlative predict a contradictory
reading for (i), much as (ii) is contradictory without some special
context (such as: ‘in their respective ranges’):
(i) Everest and K2 are the highest mountains.
(ii) # Everest is the highest mountain and K2 is too.
Stateva argued, in effect, for an accommodation mechanism that is similar
to that needed for the ‘more than all’ examples (though she did not phrase
it in these terms). See Fitzgibbons et al. (2009) for criticism of Stateva’s proposal,
and an alternative account (which, so far as I can see, would leave the
410
plural superlatives unrelated to the ‘more than all’ construction discussed in
the main text).
23
Compare also Lao, which forms a superlative with a word glossed ‘peer,
(member of) group’ in the comparative construction, which is etymologically,
at least, an ‘exceed’-type (Enfield 2007, 257).
24
I put aside here the many interesting questions of the syntax of comparatives
(including the question of whether cmpr resides in a degP for
example), on which there is a vast literature, including Bresnan (1973), von
Stechow (1984), Kennedy (1997), Lechner (2004), among many others. Important
in these discussions is the position of the than-phrase, taken to be
the first sister of the cmpr head in some accounts. Note that I also ignore
throughout the difference between clausal (Leo is taller than [IP Talia is ])
and phrasal (Leo is taller than [DP Talia ]) comparatives. My own sample
confirms the observation in Bhatt and Takahashi (2008) that no language
uses distinct degree markers (-er, more) for the two types of comparative;
where languages do draw a morphosyntactic distinction between clausal and
phrasal comparison, the distinction is always internal to the marking of the
standard (than and its complement); see Bhatt and Takahashi (2008), Merchant
(2009) on Hindi-Urdu and Japanese and Greek, respectively.
25
These forms are from Aristava et al. (1968, 49-50), see also Chirikba
(2003, 30) (the superlative is from Hewitt 1991, 47); the morphological alternations
on the adjective reflect the use of predicative forms in the comparative
construction, as compared to the basic form given for the positive. I have
411
regularized differences in transcription across sources.
26
The Late Latin examples are from Cuzzolin and Lehmann (2004, 1217)
and the Greek from González-Díaz (2006, 726) and Holton et al. (1997, 87).
27
For French, there may be variation, as plus meilleure apparently occurs,
at least inasmuch as the Office québécois de la langue française devotes a page
to it, with examples given, describing it as a pleonasm, rather than unacceptable,
see http://66.46.185.79/bdl/gabarit_bdl.asp?id=2114 (September
2010), yet some speakers report it as entirely unacceptable (P. Schlenker,
personal communication, 2010).
28
An apparently similar pattern arises in the Northwest Caucasian languages
Abaza and Abkhaz, where regular comparatives are periphrastic but
‘good’ and ‘bad’ have suppletive forms (in Abkhaz as doublets to regular
periphrastic comparatives), generally occurring without the comparative adverb
(see Tabulova 1976, 71 and Aristava et al. 1968, 49-50).
29
We are comparing across frameworks with differing underlying assumptions
here, as Vincent and Börjars (1996) in particular adopt a lexicalist
approach in which morphology does not realize syntactic structure, but is
rather associated with such structure by rules. I have attempted to abstract
away from this difference among frameworks in representing the core idea of
their approach in the main text.
30
Compare also the non-supletive periphrastic form for ‘most’ in Ludian
Karelian in (173c).
31
I have replaced Votic with related Veps in (107) as the sources consulted
412
do not provide a complete paradigm for both a regular and a suppletive adjective
in the nominative singular in each language. The textual descriptions
indicate that they are the same as regards the points of interest. Rjagoev
(1977) does not give a comparative for hoikku, but the superlative with suamo
clearly lacks comparative morphology and there is no reason to suspect an
irregularity here.
32
In theory, this should play out internal to a single language, if (i) the
language has different strategies for forming superlatives, some of which embed
the comparative and some do not, and (ii) there are roots that undergo
comparative suppletion. Ossetian is such a language, see (164) and
n. 37. However, the only root that undergoes suppletion in the comparative
(biræ – fil-dær ‘many – more’) forms superlatives with elements that
embed the comparative, as in æppætu fildær, and does not combine with
(positive-embedding) iuuIl. My thanks to B. Hettich for the Ossetian data
and discussion thereof.
33
As noted elsewhere, Greenberg (1966), Canger (1966), Ultan (1972) propose
the markedness hierarchy positive < comparative < superlative, which
may imply the inventory universal in (89), although the authors cited do not
note or explore this implication.
34
Bobaljik (1995) argues for an antilocality constraint prohibiting syntactic
movement of a head to the next head up (that is, prohibiting movement of Y
to X in the structure [XP X [Y P Y ]], but permitting that movement as a step
in a longer chain of movement. The proposal is used to account for patterns
413
of verb movement in the Germanic languages (Bobaljik and Thráinsson 1998,
Bobaljik 2002b). If that proposal is correct, then affixation in comparative
structures such as (110) must be achieved by Morphological Merger, rather
than by syntactic head movement. See also Embick and Noyer (1999) for
related discussion.
35
The particle -ma appears to be a clitic, sometimes written together with
its host and sometimes not. Superlatives may also be formed without vád.¯a,
in which case the particle follows the adjective; the particle may also follow
the noun in some contexts.
36
Hewitt (1995, 49) gives this pattern exists in Georgian as well, alongside
a periphrastic superlative, while L. Nash (personal communication, 2010) and
Aronson (1989) suggest that u-. . . -es only form comparatives for a handful of
suppletive forms, with ulamazasi having only a elative / absolute superlative
sense. These examples are only relevant if these constructions involve a
superlative head or projection; see section 3.2.3 for discussion.
37
The Ossetian example is from Abaev (1964, 20); Isaev (1966, 243) notes
that among superlative-forming elements, only iuuIl embeds the positive,
others embed the comparative. The st∼šč alternation in the Russian root is
phonological, not suppletive. The Arrente example may not belong in this
list. It is not clear from the description (i) whether indora marks an absolute
or a relative superlative, and (ii) whether the affix versus free-standing
distinction made by Strehlow is a genuinely sharp grammatical distinction.
38
The locality condition in (90) precludes in particular positing a zero
414
allomorph of the comparative as an affix on the adjectives, extending the
analysis in (42) in chapter 2. Since the superlative is periphrastic, it would
not be local enough to govern such allomorphy.
39
An issue here (raised by G. Corbett, personal communication) is why
the Russian adjective meaning ‘good’ alone permits both the positive and
the long form (declinable) comparative in the periphrastic superlative, as
shown in (i):
i. pos cmpr sprl
a. xoroš-ij luč-še samyj xoroš-ij ‘good’
b. luč-š-ij samyj luč-š-ij
I suspect that the reason is that luč-š-ij alone among the four declinable
comparatives can also be used on its own as a superlative (Garde 1998, 238)
(discounting fixed expressions). If correct, then (ia) is the true periphrastic
superlative, parallel to (123d) and all regular adjectives, while (ib) represents
a vestigial morphological superlative, with samyj here in a reinforcing
function, as discussed for Romance comparatives in section 3.3.2.
40
For extensive help with the Armenian data, I thank H. Khanjian, and
for his assistance in collecting and analyzing this with respect to the theory
in Beck et al. (2009), my thanks to I.-T. C. Hsieh.
41
Oda (2008), developing a refinement of the proposal in Beck et al. (2004),
posits no comparative head in the syntax for Japanese. However, what Oda
proposes is that the comparative element is included in the representation of
all adjectives in Japanese in the lexicon. What Japanese lacks, in her view, is
415
positive forms of gradable adjectives — they are always comparative. If this
analysis is correct then a language of the sort she envisages has ‘morphological
comparatives’ by definition and will thus always satisfy the SSG (although
note that structure-building in the lexicon is at face value incompatible with
the general theoretical framework adopted here). For a proposal rejecting
a null comparative in Japanese, see the response to Beck et al. (2004) in
Hayashishita (2009)
42
Although Armenian shows unambiguous evidence of degree semantics,
in Beck et al.’s terms, it should also be noted that Armenian shows a mix
of properties relative to their other two parameters, which on their theory
should not be allowed. For what Beck et al. term Degree Abstraction, Armenian
appears to permit scope interactions with a degree operator (see
note 43), but also appears to lack negative island effects in comparatives,
two properties which should not combine. Similarly problematic, for their
Degree Phrase parameter, Armenian appears to allow subcomparatives, but
lacks measure phrases with positive, non-comparative, adjectives. Exploring
Armenian and its theoretical implications in greater detail are left for future
research.
43
A rather complex, but much-discussed example of this sort involves apparent
scope interactions between the comparative operator and modals, and
is given in (i), from Heim (2000).
(i) This draft is 10 pages long. The (final) paper is required to be exactly 5
pages longer than that.
416
The second sentence is ambiguous as regards the length requirement.
One reading holds that the final paper must be exactly 15 pages long; neither
longer nor shorter papers will be accepted. In this reading, the requirement
is that the paper be [exactly 5 pages longer than 10 pages].
The other reading is somewhat more subtle, but holds that the requirement
is a minimum of 15 pages, but is consistent with longer papers being
accepted. For this reading, the logical paraphrases given by Heim and others
take the exactly and the comparative part to scope above the (operator in)
the word required. The (minimum) requirement is 15 pages, and 15 is exactly
5 pages longer than 10.
Preliminary investigation suggests that even in such cases, Armenian
patterns with English in the readings allowed, to the extent that (ii.b) is also
ambiguous, even without aveli, in the same way as English (i) is.
(ii) a. as
this
hotvadz-@
article-def
das-@
10-def
etš
page
e/uni
be/have.3sg
‘This draft is 10 pages (long).’
b. vertšnagan
final
hotvadz-@
article-def
bedk-e
must-be
džišt
exactly
hing
five
etš-e-n
page-abl-def
aveli
more
@lla
be.3.sg
‘The final paper must be exactly 5 pages longer than that.’
44
One further examples in (one variety of) Vlach Romani, is discussed in
n. 29 in Chapter 4.
417
45
This does leave a residue of recalcitrant cases, such as as winningest
versus *winninger (than), which does appear to have a relative superlative
(but no comparative) usage, in particular in sporting contexts, as in: “The
winningest coach in Division I history is hearing Geno Auriemma’s footsteps
loud and clear” (Hartford Courant, 9 April 2003).
46
I am indebted to Jon Gajewski and Chris Hsieh for helpful discussions
of the semantics discussed in this section.
47
Although note that there are theories of the comparative that formally
assign it a meaning like (131), consisting of a conjunction plus a negation,
such as Lewis (1970), Seuren (1973); see von Stechow (1984), Klein (1991),
Schwarzschild (2008) for discussion. The two approaches to the formal semantics
of the comparative in a rough sense paraphrase two of the major
types of overt comparison in the worlds’ languages (Stassen 1985, 2008),
namely the ‘exceed’ (i) and conjoined (ii) types:
(i) Uqa
3sg
cecela.
long
Uqa
3sg
ija
1sg
wol-te-na.
surpass-1sg-3sg-pres
(Amele, Tr-N-Guinea)
‘He is taller than me.’ (Roberts 1987, 91)
(ii) a’e
he
puku,
long
ne
you
an¯ı
no
(Urubu-Kaapor, Tupi-Guaraní)
‘He is taller than you.’ (Kakumasu 1986, 346)
48
Whether Czech pře- includes an ‘excessive’ meaning is perhaps debatable,
but the same behavior characterizes the Slovenian cognate pre- which
418
is explicitly excessive (Herrity 2000).
49
For the reader familiar with Hackl (2009), note that his use of the terms
‘relative’ and ‘absolute’ superlative is distinct from how those terms are used
here; see n. 2 of Chapter 1.
50
Very few languages have an affixal equative degree, hence an analogue
of the SSG for the equative may be surface-true but one cannot exclude the
possibility that this is accidental.
51
Plausible meanings could be assigned to the superlative examples. Thus,
(136) could mean something like: ‘To the extent the father is the tallest
(e.g., by 5 cm), the child is the shortest by the same extent’, or for (137):
‘Whatever he repeated the most, he became the most convinced of.’ But
the constructions are uniformly out, eliciting reactions such as ‘impossible’
or ‘word salad’). A second Hungarian comparative correlative construction
given in Den Dikken (2005) shows the same comparative/superlative asymmetry
(i), but the superlative in this latter construction may be excluded on
independent grounds.
(i) Hungarian (A. Szabolcsi, personal communication, 2010)
a. Minél
what.adess
több-et
more-acc
olvasol,
you.read
annál
that.adess
több-et
more-acc
{meg-értesz
vm-you.understand
/ értesz
you.understand
meg}
vm
‘The more you read, the more you understand.’ (Den Dikken
(2005))
419
b. * Minél
what.adess
leg-több-et
sprl-more-acc
olvasol,
you.read
annál
that-adess
leg-több-et
sprl-more-acc
{meg-értesz
vm-you.understand
/ értesz
you.understand
meg}
vm
1
The source does not provide the corresponding superlative, but Sorbian
superlatives are formed on the general Slavic pattern with naj- prefixed to
the comparative.
2
The suppletive comparatives of Latin are retained in the Modern Romance
languages, sometimes with a non-cognate positive root (especially in
the forms meaning ‘bad’), but none of the Modern Romance languages retains
a synthetic superlative.
3
Comparative suppletion for this and some of the other patterns below
is retained in modern Irish, Scottish Gaelic and Manx, but these languages
lack a distinct superlative.
4
See note 14 below.
5
The adjective k’argi-i formed a regular (non-suppletive) comparative in
Old Georgian (Fähnrich 1991).
6
Gudjedjiani and Palmaitis (1986) list four suppletive comparatives in
Svan; but note also that the comparative forms in xo- -a for these adjectives
are used with a positive sense, and subject to further comparative formation
in xo- - el. It may thus be synchrnoically inappropriate to include these
forms here.
7
This suppletive comparative pattern is retained for ‘bad’ in Modern
420
Greek, but the superlative is periphrastic.
8
Additional doublets are given in some sources.
9
This form is possibly irregular, rather than suppletive.
10
For Old Persian, Kent (1953) and Skjærvø (2009b) give superlative maT-išta
as a suppletive form corresponding to vaza
˙
rka ‘great’. No source gives a
corresponding comparative for Old Persian; to judge by Branenstein and
Mayrhofer (1964, 64-65), the surviving Old Persian materials provide only
limited evidence of adjectival gradation, and the comparative for ‘great’ is
not among the attested forms.
11
Små is a suppletive plural for singular lille.
12
The forms for small are given in Skjærvø (2009a, 206) in a list of suppletive
forms, separately from a list of irregular forms; the superlative is not
given in Skjærvø (2009a), the form in the table is the corresponding Parthian
form from Boyce (1977).
13
I thank C. and J. Creider for discussion of these examples.
14
Even if superlative phér-istos constitutes an archaic remnant, not paradigmatically
related to ameín-¯on, Ancient Greek would contribute an ABC triple
to the extent that a superlative built on ameín- is lacking, and the corresponding
superlative is drawn from one of the other roots. Alternatively, the
agath-ós – ameín-¯on pattern may simply be defective, as presented in (141),
in which case there is no ABC pattern here.
15
Although the standard grammars reject a regular comparative (*on-ago),
Itziar Laka (personal communication 2006) and a Basque audience member
421
at a presentation of this material (Rutgers, 2006) note that the form is indeed
occasionally met with. However, from what I can glean, this is unlikely to
constitute the general situation, and the pattern in (143) likely represents
the norm.
16
Thanks to C.E. Ariel Diertani for drawing my attention to this case.
17
Ancient Greek eu ‘well’, which is apparently the suppletive adverb corresponding
to agathos ‘good’, may be another instance of the pattern shown by
English ‘well’, being compared by the suppletive forms of the corresponding
adjective, though I was unable to extract this information definitively from
Liddell and Scott (1996).
18
Hewitt (1995, 49) provides more than a dozen examples of regular morphological
comparatives, but at least some speakers report that, with the
exception of the suppletive forms (as in (154)) the comparative reading is
unavailable, and the forms reported by Hewitt have only an intensified or
absolute superlative sense (L. Nash, personal communication 2010, see also
Aronson 1989, 246)
19
Rjagoev (1977, 163) gives hüve-mm-in as an alternative alongside paremm-in
for the Tikhvin dialect of Karelian.
20
Nivkh also shows the pattern in (158a) for adverbial comparatives, derived
from property-denoting roots, which are verbal in the language (which
lacks adejctives): eG-gur – eG-jo-gur ‘be.in.hurry-(cmpr)-conv’ = ‘quickly
– more quickly’ (Gruzdeva 1998, 37). From the scanty description, the -joformative
appears to be a diminutive marker for property-denoting verbs;
422
the one example of a comparative in Gruzdeva (1998) is unmarked.
21
A relevant observation in this context, perhaps, is that there is independently
a range of variation in the parts of speech that comparative morphemes
may attach to. Basque is on the liberal side in allowing comparative affixes
to attach to nouns as well as to adjectives, thus from gizon, gau ‘man, night’:
gizon-ago, gau-ago ‘more of a man, more night’ (De Rijk 2008, 710). The
comparative affix also attaches to nouns in Macedonian (and its close relative
Bulgarian), as in prijatel ‘friend’, po-prijatel ‘cmpr-friend’ = ‘more of
a friend’ (Friedman 2002, 22). There is indirect evidence that Macedonian
(and Bulgarian) pattern with Basque in the affix-ordering in (158). This
is hard to determine directly, since the comparative is a prefix, and there
is no productive de-adjectival adverbial affix. However, we can note that
the productive means of forming a deadjectival adverb in Macedonian (as
in Slavic generally) is by means of the neuter inflectional suffix -o, whereas
comparative and superlative markers appear to be able to attach peripheral
to inflection, or inflection-like elements such as clitics and functional heads,
as in na jug ‘to south’, po-na-jug ‘cmpr-to-south’ = ‘more southerly’, and
naj ne go saka ‘sprl neg 3sg.obj like’ = ‘he dislikes him the most’ (Friedman
2002, 22). Clearly, there is a topic for further investigation here. Lunt
(1952, 36) suggests the combination of comparative (or superlative) marker
with nouns is rare in Macedonian, and applies to a group of words “mostly
of Turkish origin”. In addition, Lunt suggests that the phrasal examples argue
for an adverbial, rather than affixal, status for these elements, despite
423
the orthographic convention of writing them as a single word. In Bulgarian,
the comparative and superlative constitute an independent prosodic domain,
having their own stress (Alexander 2000, 198). There are fewer languages
where clearly affixal comparatives attach to categories other than adjectives
(or property denoting predicates), though W. U. Dressler, personal communication
2010, notes that the comparative affix attaches to verbs in Sanskrit.
22
More accurately: “the adverb txarki [i.e., the regularly derived from expected
from adjective txar ‘bad’ -JDB] is little used... Its place is taken by
gaizki, derived from the adjective gaitz, which, however, has changed its former
meaning ‘bad’ to ‘difficult’ (and also ‘tremendous’), so that we are faced
with a typical case of suppletion: txar ‘bad’, gaizki ‘badly”’ (de Rijk 2008,
236).
23
Ultan (1972, 144) offers this adverb as an ABA pattern: bald – eher
– (am) baldigsten. Yet on both semantic and distributional grounds, this
appears to be in error. The form (am) baldigsten does not appear to be
the (relative) superlative of bald; that is, it does not mean “sooner than
all others”, rather, it is restricted in meaning to something like “as soon as
possible”, an absolute superlative (compare also the positive adjective baldig
‘early’). The superlative corresponding to comparative eher is the regular
(am) ehesten, see, e.g., WDG, which notes also dialectal regular gradation:
bald – bälder/balder – (am) bäldesten/baldesten. I thank M. Wagner and S.
Wurmbrand for discussion of these examples.
24
The cognate triples for the many and much roots do not pattern the
424
same across Germanic; for example, in Danish, Norwegian and Swedish, the
cognate to English much compares, like English, with a cognate to more, but
the cognates of many compare via the root fle- (see below).
25
A lot of or lots of has replaced much as the positive form of the mass
quantifier in many contexts, but is of course still compared exclusively with
more. Including this item does raise a question of boundaries — I have not
included, for example, other roughly similar measure expressions, such as the
count expression a bunch (of).
26
Recall from section 4.1 that it is unclear if the Basque suffix -en is a true
superlative; if it is not, then the Basque triple is not relevant.
27
The word for ‘more’ lacks comparative morphology, while the word for
‘most’ has comparative, rather than the expected superlative morphology.
28
The form zyaadaa appears to be used as a suppletive comparative ‘more’
to kai ‘many’, although in some (but not all) contexts, zyaadaa may have
a non-comparative sense. I thank Rajesh Bhatt, personal communication
2010, for discussion of this element.
29
The superlative-forming particle nái- is borrowed into the Arumanian
variety of Vlach Romani spoken in Macedonia, yielding apparent SSG violations:
bunu
– k’ama bunu
– náibűnu
‘good – better – best. Yet in describing
these forms, Gołąb (1984, 85) notes that, as in the source language Macedonian,
“the morpheme nái can be treated as a particle . . . since it clearly bears
a secondary stress, and besides that it can also be used with compound adverbs
or adverbial expressions” similar to (166). Thanks to Jason Merchant
425
for providing these examples and the reference.
30
I wish to thank Zhanna Glushan for conducting the survey in Karelia, and
G. M. Alekseev, N. Antropova, N. Bukina, E. Filipova, M. V. Matveena, and
two other speakers, for sharing their intuitions. The majority of questionnaire
respondents characterized themselves as speakers of the Livvi dialect; Zajkov
(1999, 9) notes that this dialect has final y where ‘Karelian proper’ has final
ä; suggesting that the dictionary as well is based on a Livvi variety. Likewise,
Zajkov (1999, 10) notes that Livvi mb corresponds to mp in the variety he
descries.
31
The one speaker of the Ludian variety offered äijäon to render comparative
bol’še ‘more’ in isolation, but gave äjnä-mbi (plausibly a mis-transcription
of enämbi) in the context in (172).
32
‘Most’ was also translated as kaikis suurin, ‘biggest of all’ in one other
context by one respondent. This may have been influenced by the Russian
expression for ‘most’ in this context: samoe bol’šoe količestvo ‘most big quantity’.
Another isolated option offered for (171c) was (i), which seems, like
its translation, to be a reasonable paraphrase in the context given, but not
a grammatical superlative construction.
(i) Lauri-l
Lauri-adess
on
is
vie
still
enä-mbi.
more-cmpr
‘Lauri has still/even more.’
Fennic dialectology is apparently quite complex. Zajkov (1999, p.7) identifies
33 dialects of Karelian, in 3 major dialect groups (narečija), including
426
Ludian, and Nau (1992) notes specifically that there is substantial variation
in superlative formation (as seen in (173)) within Karelian. Of interest, perhaps,
is that the only speaker to offer the form in (173c) was the one speaker
identifying themselves as a speaker of Ludian Karelian; this contrasts with
the remark in Nau (1992) that gives superlatives with ylen only for nonLudian
varieties.
33
But see note 16 in section 5.3.
1
In Bobaljik (2000a,b) I argued that rules of exponence are rewrite rules,
rather than information-preserving rules as in Halle and Marantz (1993).
This has implications for what information may be available to condition
allomorphy at the point where non-root nodes are subject to vocabulary
insertion. As our focus here is insertion at roots, these additional questions
are not directly relevant. For a variety of contrasting views on this topic, see
Carstairs-McCarthy (2001, 2003), Adger et al. (2003), Chung (2007b), Bonet
and Harbour (To appear) and Embick (2010).
2
See Seiler (1950, 25-26) and Kinchin Smith and Melluish (1968, 187);
a similar rule applies in Latin, thus: mag-nus ‘big’, comparative mai¯or =
/majj¯or/ < *mag-j¯os (Weiss 2009, 359). The theory pursued here treats
the established reconstructions as constituting derivations from underlying
to surface representations in the synchrnonic phonology of Latin and Ancient
Greek.
3
It is clear how the architecture in (181) excludes root allomorphy conditioned
by morphophonological contexts. A trickier case is to exclude reference
427
to the word-boundary symbol, as in (i), suggested by Danny Fox (or various
other formulations referring to material that is unique to the comparative,
such as inflectional morphology following the comparative affix). This would
allow reference to the comparative, to the exclusion of the superlative.
(i) / ] cmpr ] #
It may be that the word-edge symbol # is a part of the morphophonological
symbol inventory, not present at the stage of the derivation where
root exponence occurs. In addition, the considerations of locality in the next
section may also be relevant for excluding this configuration.
4
Wurzel (1985) treats suppletion as a cline, and includes German hoch –
höher – am höchsten [hox] – [hœ-er] – am [hœç-sten] ‘high – higher – highest’
as a canonical instance of ‘weak’ suppletion, in virtue of the irregularity of
fricative deletion in the comparative; compare schwach – schwächer – am
schwächsten ‘weak – weaker – weakest’. Setting aside vowel quality (umlaut),
and the fully predictable [x] ∼ [ç] alternation, if this hoch pattern were treated
as suppletive, then this would appear to be an ABA pattern. In the theory
considered here, the hoch pattern is of course (merely) irregular, with the
unique root allomorph /hox/, subject to an idiosyncratic rule of intervocalic
velar-deletion.
5
Although it is related to the Romance varieties for which a phonological
solution is not supported, Anderson argues that stress, rather than morphological
patterning, is the determining factor for stem alternations in Surmi-
428
ran, on the basis, for example, of differences in stem form in the infinitive
depending on whether the root is stressed (infinitives in -@r) or unstressed
(infinitives in -Vr, with non-schwa V). At worst, phonologically-governed
suppletion would require either a revision to the relationship between rules
of exponence and other phonological rules in (181), or a re-consideration
of the limits of the power of readjustment rules. Since the theory developed
here relies on the Elsewhere Condition, the considerations for phonologicallyconditioned
suppletion, if it exists, are analogous to the discussion of irregulars
— the CSG is predicted to hold wherever the relevant subset-superset
relations hold of conditioning environments.
6
As a further consequence, the adjacency condition precludes comparative
allomorphy of roots where the adjectival stem is internally complex. For
example, if an adjective like happy is internally complex with a suffix -y:
[ [ happ ] y ], then root allomorphy in the comparative [ [ [ happ ] i ] er ] is
precluded. Compare, for example, good – better – best, where comparative
-er is root-attached, to good-ly – good-li-er – good-li-est, with an intervening
derivational suffix and a failure of suppletion. I have not investigated this
prediction in detail, but know of no problematic cases. Also relevant here is
that comparatives in some Indo-European languages (including Proto-IndoEuropean)
are typically root-attaching, with stem augments in the positive
lost in the comparative, thus Russian (typical of Slavic) adjectives with a
positive in -(o)k- lose that augment in the comparative: vys-ok-ij ‘high-okm.sg’
– vyš-e ‘high-er’.
429
7
The VIP prevents, for example, the exponent cmpr → -er from being
inserted at node c of (195), which would have strange consequences, such as
yielding -er (or perhaps more) as the comparative of all regular adjectives:
big – more rather than big – bigg-er.
8
A node α is the minimal node dominating features F1...Fn iff α dominates
F1...Fn and there is no node β such that α dominates β and β dominates
F1...Fn.
9
In principle, Fusion and VIP could be made to differ in their interaction
with cyclic, bottom-up vocabulary insertion, as a reviewer notes. See Chung
(2007a) and Radkevich (2010, 62-65) for discussion. Radkevich suggests that
insertion applies at lower nodes, and is overwritten on a later cycle by insertion
of portmanteaus at a higher node. Another option is to assume a
form of delayed spell-out, where application of a rule of exponence targeting
a node on an inner cycle is suspended if the Vocabulary includes a rule that
targets the same node but involves outer material as well, as in (194). See
section 5.3.2 for such a proposal.
10
With perhaps a question about Ancient Greek, see n. 14 to Chapter 4.
11
An initial investigation shows that the same conclusion appears to hold
for the examples in the Surrey Suppletion Database http://www.smg.surrey.
ac.uk/Suppletion/, to the extent that the relevant information can be extracted
from examples presented there.
12
The prediction from adjacency is distinct from the prediction regarding
the relative frequency of occurrence of stem alternations with the different
430
types of trigger, as discussed in Bybee (1985) with respect to this hierarchy.
Bybee argues that root suppletion governed by more peripheral elements
should be less frequent than suppletion governed by less peripheral ones, but
her framework does nt (so far as I can see) preclude the unattested cases of
on-contiguous syncretism excluded by (199).
13
This excludes the treatment of certain types of “defective paradigms”
(adjectives having comparative and/or superlative forms but no positive) as
simply lexical gaps. For example, English rather, which is a comparative,
but lacks a corresponding positive, could not be treated simply in a manner
parallel to (203a), lacking a default.
14
This is not to say that the “word” better cannot be aquired before the
adjective good. Rather, it is a narrower claim, namely, that better can only
be understood as the suppletive comparative of good after good has been
learned. There is anecdotal evidence at least that English children might
learn forms like better first as intensified adjectives, see Ohtaki (2010) and
references there.
15
The parallel can be seen more sharply, if (204b) is restated via the equivalent,
but slightly non-standard, notation in (i.b), yielding the pair (i.a-b),
with cmpr preserved in the output so as to allow insertion of -ior):
(i) a. good, cmpr → opt- / ] sprl ]
b. good, cmpr → mel-cmpr
16
Incidentally, (202) turns out to exclude the hypothetical derivation in
(179) considered, but not adopted, in section 4.3.
431
17
Key ingredients of this alternative were suggested by Susi Wurmbrand
and Harry van der Hulst.
18
The cyclic status of the node a is irrelevant to this illustration: since
the deepest node has no complement (within the complex X0
), no rules of
exponence would apply even if it were cyclic.
19
Radkevich understood adjacency as overt string adjacency, or contiguity
in the case of portmanteaus expressing multiple features, for the purposes of
her study. Additional work would be needed to see whether the more restrictive
requirements of the combination of adjacency and (202) are supported.
20
One point against treating the Korean facts as a lexical-semantic or pragmatic
alternation is that the honorific suppletion is lost in passive, even
though the agent may still be honorified (and marked as such with an honorific
suffix); I thank Inkie Chung (personal communication 2011) for discussion
of this point.
21
A reviewer notes another possible adjacency-violating example in Tamil
pronouns (Asher 1982), some of which show root allomorphy (suppletion)
for nominative versus non-nominative case, but have the general structure:
root - number - case. Yoshiyuki Shibata notes also instances of numbergoverned
root suppletion in Ainu verbs that appear to violate the adjacency
condition.
22
Although note that if the Adjacency condition proves untenable, we lose
only the account of the CSG2 — the account of the CSG1 would not be
affected.
432
23
I thank Klaus Abels, Gereon Müller, and Curt Rice for drawing my
attention to this general domain and Gereon Müller for making me aware of
Wiese’s work which this section reports on.
24
Umlaut in the 2nd and 3rd person singular of the present, and in the
subjunctive past, are predictable from the principal parts.
25
A possible counter-example is shear, past sheared, participle sheared or
shorn. Note also variant past shore, given in the OED.
26
In particular from a semantic perspective, Wiese’s labels for the features
can undoubtedly be improved upon; what is central here is the containment
relationship. I also set aside the ge- prefix, characteristic of participles. As
a place holder for a fuller analysis, I assume that the German participle has
further structure as in: [ ge- [ [ sproch ] -en] ], and that the features of
present concern are in the lower, i.e., suffixal portion of this structure. If
the syntax (and/or semantics) requires an additional participle-defining (or
perfect) feature, then that feature is associated with the prefixal node ge-.
Wurmbrand (2010, 2011b) argues on syntactic grounds (unrelated to present
concerns) that ge- is associated with the syntactically relevant participlecharacterizing
feature. Part of her argument comes from a cross-linguistic
comparison of subtle differences in the internal order of verb clusters among
Germanic dialects with and without (descriptively) circumfixal participles.
27
Wiese notes precedents for this treatment, motivated by the patterns of
stem syncretism discussed here, citing Adelung (1782) and Johnston (1997).
28
I thank Andrew Nevins for raising this point (personal communication
433
2006). My thanks to Nilüfer Şener for introducing me to Graziano-King’s
work, discussed below.
29
Graziano-King herself suggests that the experiments yield the conclusion
that acquisition of the analytic-syntheitc alternation in comparatives involves
lexical listing, rather than rule-governed behaviour. My interpretation in
the text is a hybrid — a rule, triggered by a lexically listed (i.e., diacritic)
property. Under a literal interpretation of her proposal, where only listed
morphological comparatives, learned by positive exposure, are grammatical,
it seems that the expectation for novel forms should be 0% morphological
comparatives, rather than 50%, i.e., because morphological comparatives of
nonce adjectives do not occur in the input and thus cannot be listed. The historical
variation also seems to argue against a strict lexical-learning analysis,
which would seem to imply that the set of adjectives undergoing morphological
comparison could only decrease, and never increase, over time.
30
A relevant observation in this regards is the study of English children’s
spontaneous production in Ohtaki (2010). Examining four children in the
CHILDES corpus, Ohtaki found that the children never over-extend the use
of -er to adjectives that do not take morphological comparatives in the adult
grammar, but moreover, that they also do not over-use the periphrastic comparative,
e.g., in contexts where the adult grammar requires a synthetic form
(*more big). In fact, the children in his study did not produce simple periphrastic
comparatives at all, even where this is the only form admitted by
the target (adult) grammar. Ohtaki suggests that this requires an interpre-
434
tation whereby in spontaneous speech, as opposed to experimental settings,
children do not guess and have no default, requiring evidence of one sort or
another to establish the value of [±M]
for any given adjective. A confounding
factor, noted by Ohtaki, and earlier in Gathercole (1983), is that children’s
use of morphological comparative forms, both regular (tighter) and suppletive
(better) often appear to lack comparative semantics, being used as a variant
of the corresponding positive, perhaps with some degree of intensification.
If this is the proper interpretation of all the morphologically comparative
forms in the corpus that Ohtaki examined (which is not certain), then these
forms do not directly bear on the representation and acquisition of the [±M]
diacritic.
31
At various presentations of this material, audience members have suggested
acquisition or artificial learning experiments, involving nonce suppletive
adjectives. One experiment might present the basic and comparative:
grock – santer, eliciting a superlative. All else being equal, the theory developed
throughout this book suggests learners would produce and/or prefer
santest rather than *grockest. A more subtle twist would present the basic
and superlative, seeking the missing comparative: grock – – santest, with the
predicted form being santer. Another experiment type might expose learners
to an apparent ABA paradigm (grock – santer – grockest to see if it poses
special problems in contrast to ABB or ABC paradigms. I remain somewhat
skeptical of the utility of this latter experiment type, as the result would
have to be extremely subtle. As noted in chapter 2, given the possibility of
435
ABC patterns, UG must allow direct learning of three distinct roots in the
three grades, and thus a putative ABA pattern in the input could, as a worst
case, be analyzed as an ABC pattern with accidental homophony of the A
and C allomorphs (see also remarks on apparent ABA patterns in section 4).
Another source of an apparent ABA distribution in the input could be the
conflation of two distinct, nearly synonymous paradigms (either an AAA and
an ABB pattern, or two distinct regular patterns), with different statistical
preferences in the different grades (see the remarks on English small, little
at example (142)). An additional concern with artificial learning in this context
is the possibility of explicit rules for individual lexical items (or classes
thereof) overriding grammatical rules, allowing for the learning and use of
unnatural patterns. Thus one can learn to write the data are in English, despite
the overwhelming grammatical basis for treating data as synchronically
a mass noun, necessitating singular agreement. Thus even in the extremely
conservative OED, which treats the mass noun use as ‘erroneous’, definitions
in sub-entires show the mass noun use: data structure...the way data is
organized....
1
I thus abstract away from the derivation of morphological anti-causatives,
in which the intransitive verb appears to be derived from the causative as
opposed to the other way around; see Koontz-Garboden (2009) for recent
discussion. In a few instances, the causative and inchoative verbs behave
differently in a way that is potentially significant for matters discussed here,
such as in the problematic Polish examples in section 6.4.7. Where I know
436
of such issues, they are duly noted below.
2
A structure like (233b) would provide the same account of why (234)
yields (to) worsen rather than *(to) badden in English. It is less obvious that
it would generate the right forms where the verb embeds an overt comparative
morpheme (as in to better, and examples in (238) below). For the bulk of
this chapter, I will mostly set aside (233b), returning to it occasionally, as
the discussion warrants.
3
Deciding the issue could be relevant for evaluating the viability of (233b);
compare the discussion of bundling and allomorphy in section 5.4.
4
Deadjectival derivation of change of state verbs is quite unproductive,
for ill-understood reasons, although there are sub-regularities, such as the
near absence of inchoative and causative derivatives from human propensity
adjectives (see Dixon 1982, 21-24). Note in this regard the recency (and
humorous overtones) of the coinage embiggen filling a prior gap.
For some adjectives, the change-of-state verb with the expected meaning
is apparently derived morphologically from (or via) a noun: old – to age;
long – to lengthen. In some cases, such as lengthen, the denominal derivation
is used where a deadjectival derivation in -en is unavailable for phonological
reasons (Jespersen 1909-1949).
If one sees this derivational area as abstractly productive, rather than
gappy, with suppletive relationships populating lists such as (237) (see Apresjan
1992 for this view regarding analogous Russian examples) then it must
be a domain of uncommonly rich suppletive relationships.
437
5
The issues in this paragraph were raised by Edwin Williams and Mark
Baker at presentations of this material.
6
Kriz (2011) notes that the degree achievements that include comparative
morphology in German (ad English) are all built on what he calls ‘stubbornly
relative’ adjectives — these are open scale adjectives that resist modification
with adverbs such as completely or almost.
7
There are numerous complexities in the extremely large literature on this
topic. See Rothstein (2004) for a relatively recent treatment with review of
the literature.
8
The expression ‘for an hour’ is licit in (240a) on a different reading,
namely, in which it modifies the extent not of the change event, but rather
the duration of the result state. This situation is implausible with cooling
soup, but compare: The weather got cool for an hour, and then it warmed
up again.
9
The approach in Abusch (1986) differs in important details, but also embeds
only the positive form of the adjective, appealing to contextual resolution
of the vagueness in the adjective meaning to derive the comparative-like
sense.
10
See Kriz (2011) for an account along these lines, although the ambiguity
that Kriz posits includes an additional pos operator in (242a); Kriz argues
moreover that not all adjective classes allow for both representations, and
that ‘stubbornly relational’ adjectives might allow only (242b).
11
The atelic examples imply that the salient endpoint is not reached, and
438
beg a continuation (as in (244b)). There may be alternative, more natural,
means of describing the relevant situations, but for present concerns, it
suffices to establish that the presence of comparative morphology does not affect
the ambiguous telicity of degree achievements. I thank Susi Wurmbrand,
Martin Prinzhorn, and Wilfried Öller for discussion of these examples.
12
Winter (2006) makes a similar proposal, but relates change along a scale
to movement along a path, thus drawing parallels to telicity – reaching an
endpoint – with motion verbs; see also Jackendoff (1996, 331).
13
I thank Mirjam Fried for first raising this point for Czech, and Zhanna
Glushan for discussion of Russian. Csirmaz (2009) notes for Hungarian that
aspectual particles serve to disambiguate these readings in deadjectival degree
achievements in that are ambiguous without such particles; the perfective
verbal marker meg and other telicity-inducing elements yield the reading
where the contextually salient endpoint of the scale is reached, with Hungarian
thus overtly marking an aspectual distinction that is largely unmarked
in English.
14
Note also in this regard the deadjectival verbs of Karo Batak as presented
in Woollams (1996, 62), some of which are given in (i)-(ii). The glosses
suggest a morphological distinction between ‘make X’ and ‘make more X’,
but the textual discussion seems to make clear that the sense of ‘more’ here is
one of intensification of the verb. The affixes pe- and -ken are both causativeforming
affixes, and the combination of the two yields an intensification. A
similar process happens with comparatives, which can be doubly marked
439
(prefix ter- and suffix -en yielding an “intensive comparative degree ‘even
more (adjective)”’ (p.55).
(i) ganjang pe-ganjang pe-ganjang-ken
high to put up high to put up even higher
(ii) uli pe-huli pe-huli-ken
good to make good, repair to make better, improve
15
The following contrast illustrates:
(i) I put the frozen soup in the microwave for 10 min, but. . .
a. it only became cool (not even warm).
b. # it only cooled.
16
The semantics of the stative sense are more complex than simply ‘be
white’, as illustrated by the notoriously difficult to translate poem ‘The Sail’
(Lermontov 1832), which begins: Beleet parus odinokij // v tumane morja
golubom., rendered rather clumsily as ‘A lonely sail gleams white [?] in the
(light) blue fog of the sea.’
17
Compare also possible cognate triples in other Celtic languages, such as
Irish beag, tearc – lu – laghdaíonn and Manx beg, tiark – s’loo – leodhaghay.
18
The corresponding verb here is -k’l- ‘to take away’ which shares a root
with the comparative, but in this case the derivation goes the other way —
the comparative is a participial form of the verb (L. Nash, personal communication
2006).
19
I thank W. U. Dressler for raising the example, and M. Peters for help
in finding the solution.
440
20
It should all but go without saying that Google hits are of limited reliability,
and will turn up examples from non-native speakers, errors of various
kinds, intentionally humorous coinages (New Study Finds IQ of American
Journalists Baddening), and the like. Nevertheless, the following examples
seem perfectly plausible to my ear, and were accepted as colloquial, but
grammatical, by other speakers of English I have informally queried.
21
The example in (260) refers to an incident involving an actor known
for playing ‘good guys’ trying to reinvent himself as a rap star. It continues:
“After all, rap is a tough game and if the Hoff is to make it as the new Eminem,
he is going to have to start playing the role of the mean motherfucker, which
is nowhere near as easy as it sounds.” This is precisely sense 13 in (261).
22
This section relies heavily on observations by M. Marelj and especially
M. Despić, whose assistance I acknowledge with gratitude.
23
Despić’s report of the meaning differences, from speakers representing
a variety of areas, is presented below. M. Marelj, personal communication
2006, also reports that some speakers who accept prodobriti (se) understand
it only in a facetious or humorous sense.
24
Despić reports variation among English speakers as to whether (and to
what extent) English shows the same contrast. A reviewer reports losing the
moral sense for better in Mary is a better politician than John, while finding
Mary is a good politician ambiguous. I find the English data quite murky
here, but there is a potential problem lurking here if the English comparative
adjective better lacks a moral reading (out of context), while the English verb
441
to better can retain that moral sense.
25
Other languages in which the adjective meaning ‘good’ shows special
grammatical properties in its ‘moral’ sense include French: Bouchard (1998,
142) provides the contrast between un bon chef ‘a good (at cooking) chef’
versus un chef bon ‘a chef [who is] good as a human being’, and Dutch, in
which the attributive adjective goed ‘good’ lacks the otherwise systematic
schwa inflection in prenominal position when used in its moral sense (Marcel
den Dikken, personal communication, 2011).
26
I take it that the -š- in u-xud-š-at’ ‘(to) worsen’ is the comparative
morpheme, Zaliznyak (1977); compare the long form comparative xud-š-ij.
27
Thanks to G. Corbett and A. Krasovitsky for the example; N. Radkevich
notes s-plox-et’ for some speakers with the same meaning; I thank also O.
Tarasenkova, N. Fitzgibbons and N. Kariaeva for discussion. M. Despić and
others have noted that Serbian and Croatian have analogous forms, to which
the analysis offered here should extend.
28
This assumes that mne xuže is indeed the comparative of (267a); note
also positive mne xud-o ‘me.dat bad-adv’ alongside (267a) with similar
meaning, which provides a regular (non-suppletive) source for comparative
xuž-e. The puzzle posed by (267b) is the absence of the suppletive root in
the derived verb, while adjective/adverb plox- lacks a non-suppletive com-
parative.
29
In Nichols’s list of deadjectival forms of this sort (p. 334), some are
glossed with a positive meaning and others with a comparative meaning,
442
such as d.weax-lu ‘get longer’ from d.weaxa ‘long’. Transitive counterparts
of the deadjectival verbs (‘make X’) are formed from the adjectival root with
causative morphology.
30
Data and discussion reported in this paragraph from L. Veselinova, R.
Feldstein, N. Kariaeva, M. Marelj, and Ž. Bošković.
31
M. Krämer (personal communication, 2006) observes that the inchoative
versus anticausative contrast may have a reflex in German as well, where
(among zero-derived verbs) inchoatives appear to be built directly on the
root (without umlaut) but transitives and anti-causitives undergo umlaut if
the comparative does:
(i) a. alt – älter – veralten ‘old – older – to age’
arm – ärmer – verarmen ‘poor – poorer – to become poor’
b. hart – härter – sich verhärten ‘hard – harder – to harden (intr.)’
jung – jünger – sich verjüngen ‘young – younger – to become
younger’
32
For differential measure functions, including m∆, they stipulate that
the function returns a positive value if the difference is positive, and zero
otherwise; this applies both to comparatives, which are always of superiority,
never of inferiority (see Chapter 7) and to deadjectival degree achievements,
which likewise always denote a positive change. I have not found deadjectival
443
verbs meaning, for example, ‘to become less X’ without an explicit marker of
negation. The adjectival domain thus contrasts with denominal verbs such
as to skin (an animal), to scale (a fish) etc.
33
Russian xud-et’ exists as a verb, but means ‘to lose weight, become thin’,
a regular derivation from xud-oj ‘thin.’
34
Another approach that shares with the proposal under discussion in this
paragraph the idea that the comparative semantics in degree achievements
does not arise from a hidden instance of the (regular) comparative morpheme
is that of Den Dikken et al. (2010). Unlike the V∆ proposal, Den Dikken et
al. build a comparative feature into all gradable adjectival roots. In their account,
this comparative feature is normally uninterpretable, hence eliminated
and uninterpreted, when the adjective is in a positive context, but the feature
is interpreted in the verbal context, yielding the uniformly comparative
semantics I have argued for above. Their proposal holds that the derivation
offered in the main text, with a comparative morpheme (head) in the
deadjectival verb [ [ [ adj ] -cmpr ] -v] is universally excluded, and avoids
the narrow problem just mentioned, by treating the suppletive comparatives
such as English better (and by implication the other suppletive comparatives
in Table 6.1) as synchronically underived. This strong view would appear to
stumble on German verbs that transparently embed regular (non-suppletive)
comparatives, such as ver-schön-er-n ‘to beautify’ in (238). A major focus of
their aper is understanding why a simple adjective in Mandarin is necessarily
understood as comparative without additional syntactic context: ta gao ‘he
444
tall’ can only mean ‘he is taller’; the positive reading (‘he is tall’) requires a
pleonastic element hen.
1
Key contributions in this large literature include Sokolovskaja (1980),
Cysouw (2003), Simon (2005); for related alternatives invoking a mix of privative
and binary features, see Harbour (2006), Nevins (2007).
2
The figure is calculated from the data in the extensive survey reported in
an appendix to Cysouw (2001). See also Baerman et al. (2005) and Pertsova
(2007, 2011).
3
The context of Kayne’s proposal is for a decomposition of many and few
into an adjective plus a silent noun number.
4
I suspect that the Complexity Condition regulates particular types of
meaning, call them ‘grammatical meaning’, for example the primitives that
underlie the logical syntax of semantic computations. I do not expect the
kind of lexical-conceptual complexity in, say, underived noun meanings to be
within the scope of this constraint. Note also that the complexity condition,
or Kayne’s (277), is a condition on abstract representations, but is not taken
here to be a condition on exponents/vocabulary items, which may indeed define
correspondences between a single exponent and multiple syntactic nodes;
see the discussion of portmanteaus / cumulative exponence in Chapter 5.
5
Beyond mention for specific languages, the only prior mention of this
generalization that I am aware of is in Cuzzolin and Lehmann (2004, 1213),
for which no indication of the sample size is given. I believe this generalization
has been widely suspected, but never (to my knowledge) systematically
445
investigated in prior work. Note that many languages have an approximative
or relativizing affix, such as English -ish. In an appropriate context, these
can be pressed into service to yield an implicature of a lesser degree: Yao
Ming is tall, but alongside Yao, Emeka Okafor is merely tall-ish, but these
affixes are distinct in meaning from a comparative of inferiority, and cannot,
for example be used with a comparative syntax: *Emeka Okafor is tallish
than Yao. versus . . . less tall than Yao.
6
There may be even more structure here. The point that Greek and
German illustrate is that the element forming a periphrastic comparative of
inferiority is the comparative of the polar antonym of the element that forms
comparatives of superiority. This is true of English less and more, but Greek
and German differ in having the comparative affix transparently contained
in both poles.
7
One qualification regarding Greek is that only synthetic comparatives
may take a genitive standard, rather than a standard introduced by a particle;
since there are no synthetic comparatives of inferiority, the standard in a
comparison of inferiority can only be marked by a particle (apo in (283)) and
not by genitive; see Merchant (to appear).
There are languages of the conjoined comparative type that mark the
distinction in other ways. Urubu-Kaapor has a strategy for expressing comparison,
using a nominalized form of the adjective, and a main locative predicate;
superiority is marked by the adverbial meaning ‘above’ and inferiority
by one meaning ‘below’, as in the following, from Kakumasu (1986, 344-345):
446
(i) a'e
he
ne
you
'ar
above
koty
towards
we
somewhat
tam˜ui-ha
old.man-nml
‘He is older than you.’
(ii) a'e
he
ne
you
wyr
below
koty
towards
we
somewhat
tam˜ui-ha
old.man-nml
‘He is younger than you.’
8
It is necessary to prevent operations that would subvert this result. For
example, if rebracketing under adjacency could perturb the structure prior
to root insertion, the bleeding effect could be circumvented.
9
If agreement features are not “interpretable”, then the German facts in
(292) are not relevant to (277), though they may be relevant to stronger
claims of decomposition, as in Nanosyntax.
10
Recall from Chapter 5 that the adjacency condition holds of rules making
reference to morphosyntactic conditioning environments, but is not expected
to be a condition on phonological rules. Icelandic provides a contrast that
seems relevant, but also a potential problem. Like German, Icelandic has
a morphosyntactically (person-number) conditioned i-umlaut, which is restricted
to the present (unmarked) tense, but Icelandic also has an u-umlaut
rule, (a → ö) which is (morpho-)phonologically conditioned, triggered by an
u in the next syllable (and subject to morphological restrictions). The uumlaut
rule appears to occur across an overt tense marker: ég tal-d-i ‘I
count-past-sg’ versus við töl-d-um ‘we count-past-1.pl’. On the other
hand, Icelandic strong verbs are characterized by a number-conditioned vowel
447
alternation in the preterite, which would appear to violate the adjacency condition.
Perhaps relevant here is the observation that the form of the preterite
suffix itself is conditioned by number: weak -ið- sg ∼ -uð- pl, and strong
-Ø- sg ∼ -u- pl, suggesting an adjacency-by-transitivtiy effect. I leave the
detailed exploration of Icelandic from this perspective for a future project.
1
Australian languages are also perhaps underrepresented; see Schweiger
(1984) for an overview. Note that there are comparative affixes in some
Australian languages, though I have as yet found no report of suppletion in
adjectival gradation; although see n. 16 to Chapter 1 on verbs.
2
For the purposes of classification, Modern Romance languages such as
French and Spanish are classified as periph, as the only synthetic comparatives
are suppletive, with all others periphrastic. See the text, especially
Chapter 3.3.2, for discussion.
3
I have included Nilo-Saharan languages Lango and Kanuri as being of
this type; they are like the ALL cases except that in place of a universal
quantifier, a kind noun serves as the standard. Thus in Lango, ‘elephants are
the biggest animals’ is rendered ‘elephants are big exceeding (the) animals’
(Noonan 1992, 229). In Nuer, another Eastern Sudanic (Nilotic) language,
the superlative is rendered as a comparative against the third person plural
pronoun: roughly ‘surpasses (among) them’ (Crazzolara 1933, 55-56).
4
There are also periphrastic expressions meaning very roughly ‘above all’
or ‘most’.
5
Kashmiri, like Hindi, has some borrowed adjectives (from Persian) which
448
occur only in comparative/superlative meanings.
6
The lack of morphological marking for comparative and superlative is reportedly
characteristic of Dravidian (Caldwell 1875, 211-212, Steever 1998b,
19, Krishnamurti 2003, 29), but Elfenbein (1998, 398) reports the frequent,
though optional use of a comparative suffix (borrowed from Persian via
Balochi) in Brahui, a North Dravidian language geographically remote from
the rest of the family.
7
Gruzdeva (1998, 37) notes reduplication as a means for forming superlatives
of adverbs. Cross-linguistically, this is a frequent strategy for forming
absolute superlatives, or other intensified degrees, but the one example given
by Gruzdeva is glossed as an apparent relative superlative: ‘most quickly [of
all]’.
8
The element that forms comparatives in this language is a nominalizing
affix that also forms nominalized relative clauses; thus a single form could be
rendered as ‘(the) nearer (one)’ or ‘(the one) that is near’ (Coupe 2007, 61),
making it less clear that it is correct to see the affix involved as representing
a comparative structure per se.
9
Both Smythe (1948) and Eades (1979) give a standard comparative construction
alongside the conjoined comparative, but only the former gives a
comparative suffix.
10
In addition to the positive adjective with an existential verb, Klamer
(2010) gives an apparently idiomatic serial verb construction with the verbs
glossed ‘not know’ and ‘come’. The comparative is formed with an intensi-
449
fier meaning ‘truly’, forming an conjoined contruction when the standard is
explicitly mentioned.
11
The Ekagi superlative is not exactly parallel to the conjoined comparative,
but close enough for the purposes of this table.
12
Property predicates are verbal, the superlative meaning is rendered with
aspectual morphology.
13
The available information is sketchy and includes no discussion of the
syntax of comparatives, such as the expression of the standard. The language
is of particular relevance here, as the superlative is derived from the
comparative (Jany 2009, 97).
14
See n. 14 of Chapter 1.
15
Property-denoting predicates are verbal; comparison is marked via aspectual
morphology.
16
Hill (2005) notes that borrowing of the comparative adverb and particle
from Spanish has rendered an older comparative syntax somewhat trickier
to discern.
17
The superlative is formed from the comparative by the addition of a
nominalizing prefix, “yielding, in effect, ‘the one that is more X’, hence ‘the
X-est”’ (Macaulay 1996, 164).
18
The Portuguese loanword mais ‘more’ can be used in the expression of
comparison.
19
One comparative strategy involves a relational element marking the standard,
glossed as ‘bigger than’, and related to the word meaning ‘above’; in
450
the examples given, the phrase containing this element and the standard
appears to act as the main predicate.
1
See Benveniste (1948, 126).
2
Although described as a prefix, the comparative and superlative markers
in Bulgarian and Macedonian are likely not (morphological) affixes; see
section 4.3.
3
See n. 2.
4
In Hindi, Kashmiri and Marathi, morphological comparative and superlative
marking, borrowed from Persian and Sanskrit, can occur on adjectives
borrowed from those langauges. Some borrowings are reported as suppletive
(or lexical) comparatives or superlatives; it is not clear whether these constitute
grammatical superlatives, or instead patern with English optimal. See
also note 28 in Chapter 4
5
See previous note.
6
A form bara, glossed ‘better’ in Pandharipande (1997, 223), appears unrelated
to the form given for ‘good’ in that source.
7
There is variation among Romani varieties; some are described as marking
no distinction between comparative and superlative; see Weigand (1888)
and Katsánis and Dínas (1990) on varieties spoken in Greece, brought to my
attention by Jason Merchant. See Chapter 3.
8
Pobożniak (1964) gives a suppletive form for better, but Hancock (1995)
states that this is a relic, occuring in fixed expressions.
9
Kent (1953, 201) gives maT-išta- ‘greatest’ as a superlative, with no cor-
451
responding positive or comparative; Skjærvø (2009b, 80) lists this form as a
suppletive superlative to vaza
˙
rka- ‘great’. A corresponding (non-suppletive)
positive maz-, mazant- is given in Bartholomae (1904, 1158), but not in Kent
(1953) or Skjærvø (2009b).
10
Rastorgueva (1966) states that no superlative is attested in (Soviet)
Balochi. Jahani and Korn (2009, 656) give a morphological form that derives
superlatives from comparatives in some environments, noting that an ALL
strategy is used in predicative position.
11
Morphological comparatives exist for Tajik loans.
12
There is substantial dialect variation in Karelian, and not all varieties
have a morphological superlative. See Rjagoev (1977) and Nau (1992).
13
Helimski (1998a, 497) gives morphological comparatives for “a small
group of qualitative adjectives”, but Tereshchenko (1979, 130-131) argues
that these are augmentative and not comparative in meaning.
14
Castrén (1854, 189) gives four words for Nganasan that occur only as
comparatives, namely, words for ‘better’, ‘more’, ‘less/fewer’, ‘worse’. Although
this is suggestive of suppletion, the available data is insufficient to
draw firm conclusions.
15
The difference between Khakas, which allows for the diminutive particle
arax following the adjective in the comparative, and the Turkic languages
with the optional comparative suffix -raX may be an artefact of the descrip-
tions.
16
According to Lewis (1967, 55), Turkish had a comparative suffix -rek,
452
also used as a diminutive, but this is now obsolete. Turkish also has some
borrowed comparatives from Persian and Arabic. M. Erdal (personal communication
2006) suggests a possible suppletive pair in Old Turkish: edgü
‘good’ – yeğ ‘better’; though Lewis gives yeğ ‘good’, yeğ-rek ‘better’ as a relic
surviving in proverbs.
17
The only strategy indicated for this language is the whiter than white
construction mentioned above, also available in other Mongolic languages,
and attested in Middle Mongol (Rybatzki 2003, 68).
18
Mongolic languages generally lack morphological comparatives; the morphology
in Moghol is comparative -tar and superlative -tar-iin, borrowed
from Persian.
19
See note 17.
20
Werner (1997a, n. 10) notes an apparent suppletive ‘better’ in Castrén
(1858), but suggests this may be typo.
21
Werner (1997b, 88) reports one speaker’s use of an apparent comparative
suffix, not confirmed with other speakers. In general, regular comparatives
in Yugh are of the ZERO type, with an ablative-mrked standard.
22
Early descriptions, starting with Bogoras (1922, 719), note a comparative
suffix -čeje in Itelmen. Volodin (1976, 326) notes that his consultants could
produce elicited examples with this suffix, but that not a single example
was attested in recorded texts or spontaneous speech. In my own work, a
generation later, speakers did not recognize forms cited by earlier authors,
and produced only conjoined comparatives. Volodin also gives a second suffix,
453
-Ns, but noting that it too was not part of the living language by the 1970s
and not recognized by all speakers. For the superlative, Volodin gives analytic
and synthetic constructions, but these are absolute superlatives.
23
Marr (1910, 17) notes a single morphological comparative in Laz, but
states that this is a borrowing from Mingrelian.
24
In Adyghe and Kabardian, the comparative marker nax sometimes occurs
as a separate word and sometimes as a prefix (and sometimes both). It
is not clear whether this should be considered a morphological comparative.
The pair for ‘many’ – ‘more’ appears to be suppletive, though this is not
entirely clear from the available sources.
25
See previous note on the comparative element nax. In Kabardian, the
pair for ‘many’ – ‘more’ appears to be suppletive, though this is not entirely
clear from the available sources.
26
On a possible suppletive doublet for ‘more’, see Haspelmath’s remark
cited in section 4.3.
27
In this language, superlatives are formed with the definite article without
embedding the comparative.
454
Bibliography
Abaev, V. I. (1964) A grammatical sketch of Ossetic [English translation].
The Hague: Mouton and Co.
Abdullaev, Z. G. (1967) Darginskij jazyk. In Vinogradov et al. (1967), 508–
523.
Abondolo, Daniel (1998a) Khanty. In Abondolo (1998b), 358–386.
Abondolo, Daniel, ed. (1998b) The Uralic Languages. London: Routledge.
Abusch, Dorit (1986) Verbs of change, causation, and time. Technical Report
CSLI-86-50.
Abusch, Dorit (2005) Causatives and mixed aspectual type. In Reference and
quantification: The Partee effect, Greg Carlson and Jeffrey Pelletier, eds.,
Stanford: CSLI.
Acharya, Jayaraj (1991) A Descriptive Grammar of Nepali and an Analyzed
Corpus. Washington, D.C.: Georgetown University Press.
455
Adelung, Johann Chr. (1782) Umständliches Lehrgebäude der Deutschen
Sprache, zur Erläuterung der Deutschen Sprachlehre für Schulen. Leipzig:
Breitkopf.
Adger, David, Susana Béjar, and Daniel Harbour (2003) Directionality in Allomorphy:
A reply to Carstairs-McCarthy. Transactions of the Philological
Society 101(1): 109–115.
Aikhenvald, Alexandra (1995) Bare. München: Lincom Europa.
Aikhenvald, Alexandra Y. (1998) Warekena. In Handbook of Amazonian Languages
vol. 4, vol. 4, Desmond C. Derbyshire and Geoffrey K. Pullum, eds.,
Berlin: Mouton de Gruyter, 225–440.
Aikhenvald, Alexandra Y. (2003) A Grammar of Tariana, from Northwest
Amazonia. Cambridge: Cambridge University Press.
Alexander, Ronelle (2000) Intensive Bulgarian, vol. 1: A textbook and reference
grammar. Madison: University of Wisconsin Press.
Allen, Margaret (1979) Morphological Invesitgations. Ph.D. thesis, University
of Connecticut, Storrs.
Ambrazas, Vytautas (1997) Lithuanian Grammar. Vilnius: Baltos Lankos.
Andersen, Paul Kent (1983) Word Order Typology and Comparative Constructions.
Amsterdam: John Benjamins.
456
Anderson, Gregory David and K. David Harrison (1999) Tyvan. München:
Lincom Europa.
Anderson, Stephen R. (1992) A-Morphous Morphology, Cambridge Studies
in Linguistics, vol. 62. Cambridge University Press.
Anderson, Stephen R. (2008) Phonologically Conditioned Allomorphy in Surmiran
(Rumantsch). Word Structure 1: 109–134.
Andreev, I. A. (1966) Čuvašskij jazyk. In Vinogradov et al. (1966b), 43–65.
Andrews, J. Richard (1975) Introduction to Classical Nahuatl. Norman: University
of Oklahoma Press, 2 ed.
Apresjan, Yuri D. (1992) Lexical Semantics: User’s guide to contemporary
Russian vocabulary. Ann Arbor: Karoma.
Aquilina, Joseph (1959) The Structure of Maltese: A Study in Mixed Grammar
and Vocabulary. The Royal University of Malta.
Aristava, Š. K., X. S. Bgažba, M. M. Tsikolija, L. P. Čkadua, and K. S. Šakryl
(1968) Grammatika abkhazskogo jazyka: fonetika i morfologija. Suxumi:
Alašara.
Ariste, Paul (1968) A Grammar of the Votic Language. Indiana University
Press.
Aronoff, Mark (1994) Morphology by itself: Stems and inflectional classes.
Cambridge: MIT Press.
457
Aronson, Howard I. (1989) Georgian: A reading grammar. Bloomington, IN:
Slavica.
Arsanis, G. V. (1968) Sovermennyj assirijskij jazyk. In Vinogradov et al.
(1968), 489–507.
Asher, R. E. (1982) Tamil. Amsterdam: North Holland.
Asher, R. E. and T. C. Kumari (1997) Malayalam. NY, USA: Routledge.
Ashton, E. O. (1947) Swahili Grammar. London: Longmans, 2 ed.
Avrorin, V. A. (1982) Sintaksičeskie issledovanija po nanajskomu jazyku.
Leningrad: Nauka.
Avrorin, V. A. (2000) Grammatika man’čžurskogo pis’mennogo jazyka. SanktPeterburg:
Nauka.
Avrorin, V. A. and E. P. Lebedeva (1968) Oročskij jazyk. In Vinogradov et al.
(1968), 191–209.
Azimov, P., J. Amansaryev, and K. Saryev (1966) Turkmenskij jazyk. In
Vinogradov et al. (1966b), 91–111.
Babushkin, G. F. and G. I. Donidze (1966) Šorskij jazyk. In Vinogradov et al.
(1966b), 467–481.
Badia Margarit, Antonio M. (1962) Gramatica catalana. Madrid: Editorial
Gredos.
458
Baerman, Matthew, Dunstan Brown, and Greville G. Corbett (2005) The
syntax-morphology interface: a study of syncretism. Cambridge: Cambridge
University Press.
Baird, Louise (2008) A Grammar of Klon: A Non-Austronesian Language of
Alor, Indonesia. Canberra: Pacific Linguistics.
Bakaev, Ch. Kh. (1966) Kurdskij jazyk. In Vinogradov et al. (1966a), 257–
280.
Baker, Mark (1988) Incorporation: A Theory of Grammatical Function
Changing. Chicago: The University of Chicago Press.
Baker, Mark (2009) Formal generative typology. In The Oxford Handbook of
Linguistic Analysis, Bernd Heine and Heiko Narrog, eds., Oxford: Oxford
University Press, 285–312.
Baker, Mark C. and James McCloskey (2007) On the rlationship of typology
to theoretical syntax. Linguistic Typology 11: 285–296.
Bardakjian, Kevork B. and Bert Vaux (2001) A textbook of Modern Western
Armenian. New York: Caravan Books.
Barić, Eugenija, Mijo Lončarić, Dragica Malić, Slavko Pavešić, Mirko Peti,
Vesna Zečević, and Marija Znika (1997) Hrvatska gramatika. Zagreb: Školska
Knjiga.
459
Bartholomae, Christian (1904) Altiranisches Wörterbuch. Strassburg: Karl
J. Trübner.
Barwise, Jon and Robin Cooper (1981) Generalized Quantifiers and Natural
Language. Linguistics and Philosophy 4: 159–219.
Bashir, Elena (2009) Wakhi. In Windfuhr (2009), 825–858.
Baskakov, Nikolai Aleksandrovich (1966) Nogajskij jazyk. In Vinogradov
et al. (1966b), 280–300.
Batchelor, John (1905) An Ainu-English-Japanese dictionary (Including a
grammar of the Ainu language). Tokyo: Methodist Publishing House.
Bauer, Brigitte L. M. and Jonathan Slocum (2006) Old French Online.
http://www.utexas.edu/cola/centers/lrc/eieol/ofrol-0-X.html.
Bauer, Winfred (1983) Maori: Descriptive Grammars. London; New York:
Routledge.
Beard, Robert (1995) Lexeme-morpheme base morphology: a general theory
of inflection and word formation. SUNY Series in Linguistics, Albany:
SUNY Press.
Beck, Sigrid (1997) On the semantics of comparative conditionals. Linguistics
and Philosophy 20: 229–271.
Beck, Sigrid, Sveta Krasikova, Daniel Fleischer, Remus Gergel, Stefan Hofstetter,
Christiane Savelsberg, John Vanderelst, and Elisabeth Villalta
460
(2009) Crosslinguistic Variation in Comparison Constructions. Linguistic
Variation Yearbook 9: 1–66.
Beck, Sigrid, Toshiko Oda, and Koji Sugisaki (2004) Parametric variation in
the semantics of comparison: Japanese vs. English. Journal of East Asian
Linguistics 13: 289–344.
Bennett, Michael and Barbara Partee (1978) Toward the logic of tense and
aspect in English. Indiana University Liguistics Club, Bloomington.
Benveniste, Émile (1948) Noms d’agent et noms d’action en indo-européen.
Paris: Adrien-Maisonneuve.
Benzing, Johannes (1955) Lamutische grammatik. Wiesbaden: Franz Steiner
Verlag.
Berg, Helma van der (1995) A Grammar of Hunzib (with Texts and Lexicon).
München: Lincom Europa.
Berg, René van den (1989) A Grammar of the Muna Language. Dodrecht:
Foris Publication.
Berger, Hermann (1999) Die Burushaski-Sprache von Hunza und Nager, Teil
I: Grammatik. Wiesbaden: Harassowitz Verlag.
Berghäll, Liisa (2010) Mauwake reference grammar. Ph.D. thesis, University
of Helsinki.
Bergsland, Knut (1982) Sydsamisk Grammatikk. Oslo: Universitetsforlaget.
461
Bergsland, Knut (1997) Aleut Grammar: Unangam Tunuganaan Achixaasiˆx.
Fairbanks: Alaska Native Language Center.
Bertagaev, T. A. (1968) Burjatskij jazyk. In Vinogradov et al. (1968), 13–33.
Besnier, Niko (2000) Tuvaluan: a Polynesian language of the Central Pacific.
London: Routledge.
Bhatt, Rajesh and Shoichi Takahashi (2008) Reduced and Unreduced
Phrasal Comparatives, ms. University of Massachussetts at Amherst
and University of Tokyo [http://people.umass.edu/bhatt/papers/
glow-phrasal.pdf].
Bickel, Balthasar (2008) A refined sampling procedure for genealogical control.
Sprachtypologie und Universalienforschung 61: 221–233.
Bickel, Balthasar and Johanna Nichols (1996ff) The AUTOTYP database.
Birtalan, Ágnes (2003) Oirat. In Janhunen (2003), 210–228.
Blake, Barry J. (2001) Case. Cambrdige: Cambridge University Press, 2 ed.
Bläsing, Uwe (2003) Kalmuck. In Janhunen (2003), 229–247.
Bobaljik, Jonathan David (1995) Morphosyntax: the syntax of verbal inflection.
Ph.D. thesis, MIT, Cambridge.
Bobaljik, Jonathan David (1997) If the head fits...: on the morphological
determination of Germanic syntax. Linguistics 35: 1029–1055.
462
Bobaljik, Jonathan David (2000a) Implications of Itelmen Agreement Asymmetries.
In Proceedings of the Berkeley Linguistics Society Annual Meeting
(1999), vol. 25, Steven S. Chang, Lily Liaw, and Josef Ruppenhofer, eds.,
Berkeley: Berkeley Linguistics Society, 299–310.
Bobaljik, Jonathan David (2000b) The ins and outs of contextual allomorphy.
In University of Maryland Working Papers in Linguistics, vol. 10, Kleanthes
K. Grohmann and Caro Struijke, eds., College Park, MD: University
of Maryland, 35–71.
Bobaljik, Jonathan David (2002a) A-chains at the PF-interface: Copies and
‘Covert’ Movement. Natural Language and Linguistic Theory 20(2): 197–
267.
Bobaljik, Jonathan David (2002b) Realizing Germanic inflection: why morphology
does not drive syntax. Journal of Comparative Germanic Linguistics
6(2-3): 129–167.
Bobaljik, Jonathan David (2008) Missing Persons: A case study in morphological
universals. The Linguistic Review 25(1-2): 203–230.
Bobaljik, Jonathan David (In preparation) Distributed Morphology. In Handbook
of Morphology, Andrew Spencer, ed., London: Wiley-Blackwell.
Bobaljik, Jonathan David and Höskuldur Thráinsson (1998) Two heads aren’t
always better than one. Syntax 1(1): 37–71.
463
Bobaljik, Jonathan David and Susi Wurmbrand (2005) The domain of agreement.
Natural Language and Linguistic Theory 23: 809–865.
Bogal-Allbritten, Elizabeth (2008) Gradability and Degree Constructions in
Navajo. Honors Thesis, Swarthmore College.
Bogoras, Waldemar (1922) Chukchee. Washington: Government Printing Of-
fice.
Bokarev, E. A. (1967) Tsezskij jazyk. In Vinogradov et al. (1967), 404–420.
Bonet, Eulàlia and Daniel Harbour (To appear) Contextual Allomorphy. In
The Handbook of Exponence, Jochen Trommer, ed.
Bonet, Eulàlia, Maria-Rosa Lloret, and Joan Mascaro (2007) Allomorph selection
and lexical preference: two case studies. Lingua 117(9): 903–927.
Bopp, Franz (1856) A comparative grammar of the Sanskr.it, Zend, Greek,
Latin, Lithuanian, Gothic, German, and Sclavonic Languages, vol. 1.
Williams and Norgate.
Borer, Hagit (2005) Structuring Sense (2 vols). Oxford: Oxford University
Press.
Borer, Hagit (2009) Categorizing roots. Handout of a talk given at CASTL
Tromsø.
Boretzky, Norbert (1994) Romani: Grammatik des Kalderaš-Dialekts mit
Texten und Glossar. Wiesbaden: Harassowitz Verlag.
464
Borg, Albert and Marie Azzopardi-Alexander (1997) Maltese. London: Rout-
ledge.
Bosworth, Joseph and T. Northcote Toller (1898) An Anglo-Saxon Dictionary.
Oxford: Clarendon Press.
Bouchard, Denis (1998) The distribution and interpretation of adjectives in
French: A consequence of Bare Phrase Structure. Probus 10: 139–184.
Bošković, Željko (2010) Phases beyond clauses, ms. University of Connecti-
cut.
Boyce, Mary (1977) A word-list of Manichaean Middle Persian and Parthian.
With a reverse index by Ronald Zwanziger. Tehran: Bibliotheque Pahlavi.
Branenstein, Wilhelm and Manfred Mayrhofer (1964) Handbuch des Altpersischen.
Wiesbaden: Otto Harrassowitz.
Bresnan, Joan (1973) Syntax of the comparative clause construction in English.
Linguistic Inquiry 4: 275–343.
Broadwell, George Aaron (2006) A Choctaw Reference Grammar. Lincoln,
NE, USA: University of Nebraska Press.
Broekhuis, Hans (1999) Adjectives and adjective phrases. University of
Tilburg Modern Grammar of Dutch Occasional Papers 2.
Browne, Wayles (1993) Serbo-Croat. In Comrie and Corbett (1993), 306–387.
465
Bruyne, Jacques De (1995) A comprehensive Spanish grammar. Oxford:
Blackwell.
Bulatova, N. Ja. (1999) Jazyk saxalinskix evenkov. Sankt-Peterburg: Beskonfliktnyj
Sever.
Bulatova, Nadezhda and Lenore Grenoble (1999) Evenki. München: Lincom
Europa.
Burganova, N. B., M. Z. Zakiev, and Kh. R. Kurbatov (1969) Sovremennyj
tatarskij literaturnyj jazyk. Moscow: Nauka.
Büring, Daniel (2007) More or Less. Proceedings from the Annual Meeting of
the Chicago Linguistic Society 43(2): 3–17.
Bybee, Joan L. (1985) Morphology: A Study of the Relation between Meaning
and Form. Amsterdam: John Benjamins.
Caha, Pavel (2009) The Nanosyntax of case. Ph.D. thesis, CASTL Tromsø.
Calabrese, Andrea (2005) Markedness and economy in a derivational model
of phonology. Berlin: Mouton de Gruyter.
Calder, George (1990) A Gaelic Grammar [1923]. Glasgow: Gairm Publica-
tions.
Caldwell, Robert (1875) A comparative grammar of the Dravidian or SouthIndian
family of languages. London: Trübner, 2nd ed.
466
Canger, U. (1966) The category of comparison. Bulletin de la societé polonaise
de linguistique 24: 145–161.
Carstairs, Andrew (1987) Allomorphy in Inflexion. Croom Helm Linguistics
Series, London: Croom Helm.
Carstairs, Andrew (1988) Some implications of phonologially conditioned
suppletion. In Yearbook of Morphology 1998, Geert Booij and Jaap van
Marle, eds., Dordrecht: Foris, 67–94.
Carstairs-McCarthy, Andrew (1992) Current Morphology. London: Rout-
ledge.
Carstairs-McCarthy, Andrew (2001) Grammatically conditioned allomorphy,
paradigmatic structure, and the Ancestry Constraint. Transactions of the
Philological Society 99(2): 223–245.
Carstairs-McCarthy, Andrew (2003) Directionality and Locality in Allomorpy:
A Response to Adger, Béjar, and Harbour. Transactions of the
Philological Society 101(1): 117–124.
Castrén, M. Alexander (1854) Grammatik der samojedischen Sprachen
[reprint]. Bloomington: Indiana University, 1966 ed.
Castrén, M. Alexander (1858) Versuch einer jenissei-ostjakischen und kottischen
Sprachlehre. St. Petersburg: Kaiserlich Akademie der Wissenschaften.
467
Chamoreau, Claudine (2003) Purépecha de Jarácuaro, Michoacán. México:
El Colegio de México.
Chantraine, Pierre (1967) Morphologie historique du grec. Paris: Klincksieck.
Chapman, Shirley and Desmond Derbyshire (1991) Paumarí. In Handbook
of Amazonian Languages vol.3, vol. 3, Desmond C. Derbyshire and Geoffrey
K. Pullum, eds., Berlin: Mouton de Gruyter, 161–352.
Charney, Jean Ormsbee (1993) A Grammar of Comanche. University of Nebraska
Press.
Chirikba, Viacheslav A. (2003) Abkhaz. München: Lincom Europa.
Chomsky, Noam (1957) Syntactic Structures. The Hague: Mouton.
Chomsky, Noam (2001) Derivation by phase. In Ken Hale: A life in language,
Michael Kenstowicz, ed., MIT Press, 1–52.
Chung, Inkie (2007a) Ecology of PF: A study of Korean phonology and morphology
in a derivational appraoch. Ph.D. thesis, University of Connecticut.
Chung, Inkie (2007b) Suppletive negation in Korean and Distributed Morphology.
Lingua 117: 95–148.
Chung, Inkie (2009) Suppletive verbal morphology in Korean and the mechanism
of vocabulary insertion. Lingua 45: 533–567.
468
Cinque, Guglielmo (1999) Adverbs and functional heads: A cross-linguistic
perspective. Oxford: Oxford University Press.
Clahsen, Harald and Christine Temple (2003) Words and rules in Williams
Syndrome. In Language Competence across Populations, Yonata Levy and
Jeannette Schaeffer, eds., Erlbaum, 323–352.
Cleasby, Richard and Gudbrand Vigfusson (1874) An Icelandic-English Dictionary.
London: Clarendon Press.
Coates, J. G. (1982) The -dž1k comparative suffix in contemporary Komi
usage. Transactions of the Philological Society 80(1): 119–129.
Cohen, Marcel (1936) Traité de langue amharique (Abyssinie). Paris: Institut
d’ethnologie.
Colarusso, John (1992) A grammar of the Kabardian language. Calgary: University
of Calgary Press.
Cole, Desmond T. (1955) An Introduction to Tswana Grammar. London:
Longmans, Green and Co.
Comrie, Bernard and Greville G. Corbett, eds. (1993) The Slavonic Languages.
London: Routledge.
Conzemius, Eduard (1929) Notes on the Miskito and Sumu languages of
eastern Nicaragua and Honduras. International Journal of American Linguistics
5(1): 57–115.
469
Corbett, Grenville G. (2007) Canoncial typology, suppletion and possible
words. Language 83(1): 8–42.
Coupe, Alexander Robertson (2007) A Grammar of Mongsen Ao. Berlin;
New York: Mouton de Gruyter.
Cowgill, Warren (1970) Italic and Celtic Superlatives and the Dialects of
Indo-European. In Indo-European and the Indo-Europeans, George Cardona,
Henry M. Hoenigswald, and Alfred Senn, eds., Philadelphia: University
of Pennsylvania, 113–153.
Crazzolara, J. P. (1933) Outlines of a Nuer grammar. Wien: Anthropos.
Creider, Chet A. and Jane Tapsubei Creider (1989) A Grammar of Nandi.
Hamburg: Helmut Buske Verlag.
Crowell, Thomas Harris (1979) A Grammar of Bororo. Ph.D. thesis, Cornell
University, Ithaca.
Csirmaz, Aniko (2009) Scales, verbs and verbal modifiers, ms. University of
Utah.
Csúcs, Sándor (1998) Udmurt. In Abondolo (1998b), 276–304.
Cuzzolin, Pierluigi and Christian Lehmann (2004) Comparison and gradation.
In Morphologie: ein internationales Handbuch zur Flexion und Wortbildung,
Geert Booij, Christian Lehmann, and Joachim Mugdan, eds.,
Berlin: de Gruyter, 1212–1220.
470
Cysouw, Michael (2001) The paradigmatic structure of person marking. Ph.D.
thesis, Katholieke Universiteit Nijmegen.
Cysouw, Michael (2003) The Paragdigmatic Structure of Person Marking.
Oxford: Oxford University Press.
Dalalakis, Jenny (1994) English adjectival comparatives and Familial Language
Impairment. In Linguistic Aspects of Familial Language Impairment.
McGill Working Papers in Linguistics, vol. 10, John Matthews, ed., Montréal:
McGill University, 50–66.
Dayley, Jon P. (1989) Tümpisa (Panamint) Shoshone grammar. Berkeley:
University of California Press.
de S. Juan Crisóstomo Nájera, Fr. Manuel (1879) Gramática del Tarasco.
Morelia: Octaviano Ortiz.
Debrunner, Albert and Jacob Wackernagel (1929) Altindische Grammatik
III. Deklination der Nomina, Zahlwörtet – Pronomina. Göttingen: Vandenhoeck
and Ruprecht.
Décsy, Gyula (1966) Yurak Chrestomathy. Bloomington: Indiana University.
Dedrick, John M. and Eugene H. Casad (1999) Sonora Yaqui language structures.
Tucson: University of Arizona Press.
Derbyshire, Desmond C. (1985) Hixkaryana and Linguistic Typology. Arlington,
Texas: SIL and University of Texas.
471
Derbyshire, Desmond C. and Geoffrey K. Pullum, eds. (1986) Handbook of
Amazonian Languages vol. 1, vol. 1. Berlin: Mouton de Gruyter.
Desheriev, Ju. D. (1967a) Batsbijskij jazyk. In Vinogradov et al. (1967),
228–246.
Desheriev, Ju. D. (1967b) Čečenskij jazyk. In Vinogradov et al. (1967), 190–
209.
Despić, Miloje (2008) Explaining prodobriti se, manuscript, University of
Connecticut.
Despić, Miloje (2011) Syntax in the absence of Determiner Phrase. Ph.D.
thesis, University of Connecticut, Storrs.
Despić, Miloje and Yael Sharvit (To appear) Some Non-Intersective Adjectives
are Genuinely Noun-Taking. In Proceedings of NELS 39, Ithaca: Cornell
Linguistics Club.
Diels, Paul (1963) Altkirchenslavische Grammatik I: Grammatik, vol. 1. Heidelberg:
Carl Winter Universitätsverlag, 2nd ed.
Dietiker, Simone Renaud (1983) En Bonne Forme. Lexington, Mass.: D.C.
Heath and Co., 3rd ed.
Dikken, Marcel den (2005) Comparative correlatives comparatively. Linguistic
Inquiry 36(4): 497–532.
472
Dikken, Marcel den (2007) Phase Extension: Contours of a theory of the role
of head movement in phrasal extraction. Theoretical Linguistics 33: 1–41.
Dikken, Marcel den, Yang Gu, and Jie Guo (2010) Positively comparative,
ms. CUNY and The Chinese University of Hong Kong.
Dirr, Adolf (1928) Einführung in das Studium der kaukasischen Sprachen.
Leipzig: Verlag der Asia Major.
Dixon, R. M. W. (1972) The Dyirbal Language of North Queensland. Cambrdige:
Cambridge University Press.
Dixon, R. M. W. (1988) A Grammar of Boumaa Fijian. Chicago: University
of Chicago Press.
Dixon, R.M.W. (1982) Where have all the adjectives gone? In Where have
all the adjectives gone?, R.M.W. Dixon, ed., berlin: Mouton, 1–62.
Dixon, Robert M. W. (1977) A grammar of Yidiñ. Cambridge, UK; New
York, USA: Cambridge University Press.
Dmitrieva, L. V. (1966) Jazyk barabinskix tatar. In Vinogradov et al.
(1966b), 155–172.
Doke, Clement M. (1963) Textbook of Zulu Grammar. Cape Town: Longmans,
6th ed.
Dolakova, P. I. (1967) Ingušskij jazyk. In Vinogradov et al. (1967), 210–227.
473
Donaldson, Bruce R. (1993) Afrikaans. Berlin: Mouton.
Donohue, Mark (2004) A grammar of the Skou language of New
Guinea. Ms. Nationl University of Singapore [available at: http://ldh.
livingsources.org/archive/].
Dowty, David (1979) Word Meaning and Montague Grammar. Dordrecht:
Reidel.
Dowty, David R. (1991) Thematic Proto-Roles and Argument Selection. Language
67: 547–619.
Doyle, Aidan (2001) Irish. München: Lincom Europa.
Drabbe, P. (1952) Spraakkunst van het Ekagi. ’s-Gravenhage: Martinus Ni-
jhoff.
Dressler, Wolfgang U. and Ferenc Kiefer (1990) Austro-hungarian morphopragmatics.
In Contemporary Morphology, Wolfgang U. Dressler, Hans C.
Luschützky, Oskar E. Pfeiffer, and John R. Rennison, eds., Walter de
Gruyter, 69–77.
Driem, George van (1987) Grammar of Limbu. Berlin: Mouton de Gruyter.
Driem, George van (1993) A Grammar of Dumi. Berlin: Mouton de Gruyter.
Driem, George van (1998) Dzongkha (Languages of the Greater Himalayan
Region). Leiden, the Netherlands: Research School CNWS, School of
Asian, African, and American Studies.
474
Du Feu, Veronica (1996) Rapanui. London: Routledge.
Dum-Tragut, Jasmine (2009) Armenian: Modern Eastern Armenian. Amsterdam:
John Benjamins.
Dumézil, George (1931) La langue des oubykhs. Paris: Honoré Champion.
Dutton, Thomas Edward (2003) A dictionary of Koiari, Papua New Guinea,
with grammar notes. Canberra, Australia: Pacific Linguistics, Researxch
School of Pacific and Asian Studies, The Australian National University.
Eades, Diana (1979) Gumbaynggir. In Handbook of Australian Languages,
R. M. W. Dixon and Barry J. Blake, eds., Amsterdam: John Benjamins,
245–361.
Edelman, D. (Joy) I. and Leila R. Dodykhudoeva (2009) Shughni. In Windfuhr
(2009), 787–824.
Egli, Hans (1990) Paiwangrammatik. Wiesbaden: Harassowitz Verlag.
Einarsson, Stefán (1945) Icelandic. Baltimore: Johns Hopkins Press.
Elfenbein, Josef (1998) Brahui. In Steever (1998a), 388–.
Embick, David (2003) Locality, Listedness, and Morphological Identity. Studia
Linguistica 57(5): 143–169.
Embick, David (2007) Blocking effects and analytic/synthetic alternations.
Natural Language and Linguistic Theory 25(1): 1–37.
475
Embick, David (2010) Localism versus Globalism in Morphology and Phonology.
Cambridge: MIT Press.
Embick, David and Alec Marantz (2008) Architecture and Blocking. Linguistic
Inquiry 39(1): 1–53.
Embick, David and Rolf Noyer (1999) Locality in post-syntactic operations.
Emmerick, Ronald E. (2009) Khotanese and Tumshuqese. In Windfuhr
(2009), 377–415.
Enfield, N. J. (2007) A Grammar of Lao. Berlin: Mouton de Gruyter.
Ennaji, Moha, Ahmed Makhoukh, Hassan Es-saiydy, Mohamed Moubtassime,
and Souad Slaoui (2004) A Grammar of Moroccan Arabic. Mohammedia:
Imprimerie de Fédala.
Erdal, Marcel (2004) A grammar of Old Turkic. Leiden: Brill.
Ermushkin, Grigorij I. (2004) Srednetjoshskij dialekt èrzja-mordovskogo
jazyka. Mitteilungen der Societas Uralo-Altaica 24.
Evans, Nicholas (1995) A grammar of Kayardild: with historical-comparative
notes on Tangkic. Berlin; New York: Mouton de Gruyter.
Evans, Nicholas (2003) Bininj Gun-Wok: A Pan-Dialectal Grammar of Mayali,
Kunwinjku, and Kune. Canberra, Australia: Pacific Linguistics, Researxch
School of Pacific and Asian Studies, The Australian National Uni-
versity.
476
Everett, Daniel L. (1986) Pirahã. In Derbyshire and Pullum (1986), 200–325.
Everett, Daniel L. and Barbara Kern (1997) Wari’. London: Routledge.
Fähnrich, Heinz (1991) Old Georgian. In Harris (1991a), 129–218.
Fedden, Olcher Sebastian (2007) A Grammar of Mian, a Papuan Language
of New Guinea. Ph.D. thesis, University of Melbourne, Melbourne.
Fernandez, Frank (1967) A Grammatical Sketch of Remo: A Munda Language.
Chapel Hill, North Carolina, USA: PhD thesis, University of North
Carolina at Chapel Hill.
von Fintel, Kai (1999) NPI licensing, Strawson entailment, and context dependency.
Journal of Semantics 16: 97–148.
Fitzgibbons, Natalia, Yael Sharvit, and Jon Gajewski (2009) Plural superlatives
and distributivity. In Proceedings of SALT XVIII, 302–318.
Fortescue, Michael (1984) West Greenlandic. London: Croom Helm.
Fortune, R. F. (1942) Arapesh. New York: J. J. Augustin.
Foster, Mary LeCron (1969) The Tarascan Language. Berkeley: University
of California Press.
Frajzyngier, Zygmunt, Eric Johnston and Andrian Edwards (2005) Grammar
of Mina. Berlin; New York: Mouton de Gruyter.
477
Frajzyngier, Zygmunt (2001) A Grammar of Lele. Standford, California:
Center for the Study of Language and Information.
Friedman, Victor A. (2002) Macedonian. München: Lincom Europa.
Frisk, Hjalmar (1960-1970) Griechisches etymologisches Wörterbuch. Heidelberg:
Carl Winter.
Fuchs, David R. (1949) Der Komparativ und Superlativ in den finnischugrischen
Sprachen. Finnsich-ugrischen Forschungen 30: 147–230.
Gadzhieva, N. Z. (1966) Azerbajdžanskij jazyk. In Vinogradov et al. (1966b),
66–90.
Gajewski, Jon (2010) Superlatives, NPIs and most. Journal of Semantics
27(1): 125–137.
Garde, Paul (1998) Grammaire russe: phonologie et morphologie. Paris: Institute
détudes slaves, 2 ed.
Garusinghe, Dayaratne (1962) Sinhalese: The Spoken Idiom. München: Max
Hueber Verlag.
Gathercole, Virginia C. (1983) Haphazard examples, prototype theory, and
the acquisition of comparatives. First Language 4: 169–196.
Genetti, Carol (2007) A Grammar of Dolakha Newar. Berlin; New York:
Mouton de Gruyter.
478
Georg, Stefan (2003a) Mongghul. In Janhunen (2003), 286–306.
Georg, Stefan (2003b) Ordos. In Janhunen (2003), 193–209.
Glinert, Lewis (1989) The grammar of modern Hebrew. Cambridge: Cambridge
University Press.
Gołąb, Zbigniew (1984) The Arumanian dialect of Kruševo in SR Macedonia,
SFR Yugoslavia. Skopje: Macedonian Academy of Sciences and Arts.
González-Díaz, Victorina (2006) The origin of English periphrastic comparatives.
English Studies 87(6): 707–739.
Gordon, Jr., Raymond, ed. (2005) Ethnologue: Languages of the world. Dallas:
SIL International, 15 ed.
Graczyk, Randolph (2007) A Grammar of Crow. Lincoln, NE, USA: University
of Nebraska Press.
Grandgent, Charles Hall and E. H. Wilkins (1915) Italian Grammar. Boston:
D. C. Heath.
Graziano-King, Janine (1999) Acquisition of comparative forms in English.
Ph.D. thesis, CUNY.
Greenberg, Joseph H. (1966) Language Universals. Mouton.
Grimm, Jacob and Wilhelm Grimm (1854-1961) Deutsches Wörterbuch von
Jacob Grimm und Wilhelm Grimm. Online edition http://germazope.
uni-trier.de:8080/Projekte/DWB.
479
Grjunberg, A. L. (1966) Tatskij jazyk. In Vinogradov et al. (1966a), 281–301.
Gruzdeva, Ekaterina (1998) Nivkh. München: Lincom Europa.
Gudava, T. E. (1967a) Bagvalinskij jazyk. In Vinogradov et al. (1967), 351–
367.
Gudava, T. E. (1967b) Botikhskij jazyk. In Vinogradov et al. (1967), 293–306.
Gudava, T. E. (1967c) Godoberinskij jazyk. In Vinogradov et al. (1967),
307–321.
Gudava, T. E. (1967d) Tindinskij jazyk. In Vinogradov et al. (1967), 368–383.
Gudjedjiani, Chato and Mykolas L. Palmaitis (1986) Upper Svan: Grammar
and Texts. Viliuns: Mokslas.
Gui, Ming Chao (2000) Kunming Chinese. München: Lincom Europa.
Gulya, Janos (1966) Eastern Ostyak Chrestomathy. Indiana University Pub-
lications.
Gusain, Lakhan (2000) Bagri. München: Lincom Europa.
Hackl, Martin (2000) Comparative Quantifiers. Ph.D. thesis, M.I.T.
Hackl, Martin (2009) On the grammar and processing of proportional quantifiers:
most versus more than half. Natural Language Semantics 17: 63–98.
Hagman, Roy S. (1977) Nama Hottentot Grammar. Bloomington, IN, USA:
Indiana University.
480
Hakulinen, Lauri (1957) Handbuch der finnischen Sprache [1. Band]. Wiesbaden:
Otto Harrassowitz.
Halle, Morris (1990) An approach to morphology. In Proceedings of NELS xx,
vol. 20, J. Carter, R.-M. Dechaine, B. Philip, and T. Sherer, eds., Amherst:
GLSA, 150–184.
Halle, Morris (1997) Some Consequences of the representation of words in
memory. Lingua 100: 91–100.
Halle, Morris and Alec Marantz (1993) Distributed Morphology and the
pieces of inflection. In The view from Building 20: Essays in Linguistics
in honour of Sylvain Bromberger, Ken Hale and Samuel Jay Keyser, eds.,
Cambridge: MIT Press, 111–176.
Halle, Morris and Alec Marantz (1994) Some Key Features of Distributed
Morphology. In Papers on Phonology and Morphology, vol. 21, Andrew
Carnie, Heidi Harley, and Tony Bures, eds., Cambridge: MIT Working
Papers in Linguistics, 275–288.
Hancock, Ian (1995) A handbook of Vlax Romani. Ann Arbor: Slavica.
Hankamer, Jorge and Line Mikkelsen (2005) When Movement Must Be
Blocked: A Reply to Embick and Noyer. Linguistic Inquiry 36(1): 85–
125.
Hanoteau, A. (1896) Essai de grammaire de la langue tamachek’. Alger:
Adolphe Jourdan, 2 ed.
481
Harbour, Daniel (2006) Person hierarchies and geometries without hierarchies
or geometries. QM Occasioanl Paper Advancing Linguistics 6.
Harley, Heidi and Rolf Noyer (1999) Distributed Morphology. Glot International
4(4): 3–9.
Harms, Phillip Lee (1994) Epena Pedee Syntax. Arlington: Summer Institute
of Linguistics.
Harrell, Richard S. (2004) A Short Reference Grammar of Moroccan Arabic
with Audio CD. Washington, D.C.: Georgetown University Press.
Harris, Alice C., ed. (1991a) The Indigenous Languages of the Caucasus vol.
1, vol. 1. Delmar, N.Y.: Caravan Books.
Harris, Alice C. (1991b) Mingerlian. In Harris (1991a), 313–394.
Harrison, K. David (2000) Topics in the phonology and morphology of Tuvan.
Ph.D. thesis, Yale University.
Haspelmath, Martin (1993) A Grammar of Lezgian. Berlin; New York: Mouton
de Gruyter.
Hauge, Kjetil Rå(1999) A short grammar of contemporary Bulgarian. Bloomington:
Slavica.
Hausenberg, Anu-Reet (1998) Komi. In Abondolo (1998b), 305–326.
482
Hay, Jennifer (2000) Morphological adjacency constraints: a synthesis. In
Proceedings of SCIL 9, Cambridge: MITWPL, 17–29.
Hay, Jennifer, Christopher Kennedy, and Beth Levin (1999) Scalar structure
underlies telicity in "degree acheivements". In Proceedings of SALT IX,
T. Mathews and D. Strolovitch, eds., CLC Publications, 127–144.
Hayashishita, J.-R. (2009) Yori-comparatives: a reply to Beck et al. (2004).
Journal of East Asian Linguistics 18: 65–100.
Hayyim, Sulayman (1934-1936) New Persian-English dictionary. Teheran:
Librairie-imprimerie Béroukhim, online edition http://dsal.uchicago.
edu/dictionaries/hayyim/ ed.
He, Rufen, Siqi Zeng, Shanzhong Tian, and Dengxian Lin (1986a) Zhongguo
shao shu min zu yu yan jian zhi cong shu: Gaoshan zu yuyan zhi-Amis.
Beijing: Minzu.
He, Rufen, Siqi Zeng, Shanzhong Tian, and Dengxian Lin (1986b) Zhongguo
shao shu min zu yu yan jian zhi cong shu: Gaoshan zu yuyan zhi-Amis.
Beijing: Minzu.
Heath, Jeffrey (1984) Functional grammar of Nunggubuyu. Canberra: Australian
Institute of Aboriginal Studies.
Heath, Jeffrey (1999) A Grammar of Koyraboro (Koroboro) Senni: The Songhay
of Gao, Mali. Cologne, Germany.
483
Heath, Jeffrey (2005) A Grammar of Tamashek (Tuareg of Mali). Berlin;
New York: Mouton de Gruyter.
Hebert, Raymond J. and Nicholas Poppe (1963) Kirghiz Manual. Bloomington
and the Hague: Indiana University and Mouton &Co.
Heim, Irene (1985) Notes on comparatives and related matters, unpublished
manuscript, University of Texas, Austin.
Heim, Irene (2000) Degree operators and scope. In Proceedings of Semantics
and Linguistic Theory 10, Brandon Jackson and Tanya Matthews, eds.,
Ithaca: CLC Publications, 40–64.
Heim, Irene (2006) Little. In Proceedings of SALT, M. Gibson and J. Howell,
eds., Cornell University, 35–58.
Heine, Bernd (1997) Cognitive Foundations of Grammar. Oxford: Oxford
University Press.
Helimski, Eugene (1998a) Nganasan. In Abondolo (1998b), 480–515.
Helimski, Eugene (1998b) Selkup. In Abondolo (1998b), 548–579.
Hemon, Roparz (1975) A Historical Morphology and Syntax of Breton. Dublin
Institute for Advanced Studies.
Hemon, Roparz (1995) Breton Grammar [10th ed, transl. M. Everson]. Everson
Gunn Teoranta.
484
Herdan, Simona and Yael Sharvit (2006) Definite and non-definite superlatives
and NPI licensing. Syntax 9(1): 1–31.
Herrity, Peter (2000) Slovene: A comprehensive grammar. London: Rout-
ledge.
Hettich, Bela (2010) Ossetian. München: Lincom.
Hewitt, B. George (1991) Abkhaz. In The Indigenous Languages of the Caucasus
vol, vol. 2, B. George Hewitt, ed., Delmar, N.Y.: Caravan Books,
38–88.
Hewitt, B. George (1995) Georgian: A structural reference grammar. Amsterdam:
John Benjamins.
Hewitt, B. George and Z. K. Khiba (1979) Abkhaz. Amsterdam: North Hol-
land.
Hewitt, George (1996) Georgian: A Learner’s Grammar. London: Routledge.
Hill, Jane H. (2005) A grammar of Cupeño. Berkeley: University of California
Press.
Holes, Clive (1990) Gulf Arabic. London: Routledge.
Holisky, Dee Ann (1991) Laz. In Harris (1991a), 395–472.
Hólmarsson, Sverrir, Christopher Sanders, and John Tucker (1989) Íslenskensk
orðabók / Concise Icelandic-English Dictionary. Reykjavík: Iðunn.
485
Holmer, Arthur J. (1996) A Parametric Grammar of Seediq. Sweden: Lund
University Press.
Holmes, Philip and Ian Hinchcliffe (2003) Swedish: A Comprehensive Grammar.
London: Routledge, 2nd ed.
Holton, David, Peter Mackridge, and Irene Philippaki-Warburton (1997)
Greek: A comprehensive grammar of the modern language. London: Rout-
ledge.
Horn, Laurence R. (1972) On the Semantic Properties of Logical Operators
in English. Ph.D. thesis, UCLA.
Horn, Laurence R. (1989) A Natural History of Negation. Chicago: University
of Chicago Press.
Huddleston, Rodney and Geoffery K. Pullum (2001) The Cambridge Grammar
of the English Language. cambridge: Cambridge University Press.
Huffman, Franklin E. (1970) Modern Spoken Cambodian. New Haven and
London: Yale University press.
Hugjiltu, Wu (2003) Bonan. In Janhunen (2003), 325–346.
Hutchison, John P. (1981) The Kanuri Language: a reference Grammar.
Madison, Wisconsin: African Studies Program, University of Wisconsin,
Madison.
486
Ikoro, Suanu M. (1996) The Kana Language. Leiden, the Netherlands: Research
School CNWS.
Ingram, David (1978) Typology and universals of personal pronouns. In Universals
of Human Language, vol. III: Word Structure, Joseph H. Greenberg,
ed., Stanford: Stanford University Press, 213–248.
Inojosa, Mario Franco, ed. (1966) Arte de la lengua aymara de Diego de
Torres Rubio. Lima: Lyrsa.
Isaev, M. I. (1966) Osetinskij jazyk. In Vinogradov et al. (1966a), 237–256.
Ishii, Yasuo (1991) Operators and Empty Categories in Japanese. Ph.D. thesis,
University of Connecticut, Storrs.
Iwasaki, Shoichi and Preeya Ingkaphirom (2005) A Reference Grammar of
Thai. New York: Cambridge University Press.
Jabłońska, Patrycja (2007) Radical Decomposition and Argument Structure.
Ph.D. thesis, University of Tromsø.
Jackendoff, Ray (1996) The proper treatment of measuring out, telicity, and
perhaps even quantification in English. Natural Language and Linguistic
Theory 14: 305–354.
Jacobs, Neil G. (2005) Yiddish: A linguistic introduction. Cambridge: Cambridge
University Press.
487
Jacq, Pascale and Paul Sidwell (1999) Sapuan (S@puar). München: Lincom
Europa.
Jahani, Carina and Agnes Korn (2009) Balochi. In Windfuhr (2009), 634–
692.
Jakobson, Roman (1941) Kindersprache, Aphasie, und allgemeine Lautgesetzt.
Uppsala: Universitets Aarskrift.
Jakovlev, N. and D. Ashxamaf (1941) Grammatika adygejskogo literaturnogo
jazyka. Moscow: Izdatel’stvo akademii nauk SSSR.
Janda, Laura A. and Charles Townsend (2000) Czech. Berlin: Lincom Eu-
ropa.
Janhunen, Juha, ed. (2003) The Mongolic Languages. London: Routledge.
Jany, Carmen (2009) Chimariko grammar: areal and typological perspective.
Berkeley: University of California Press.
Jaspers, Dany (2005) Operators in the Lexicon: On the Negative Logic of
Natural Language. Ph.d. dissertation, Leiden University.
Jelinek, Eloise and Richard A. Demers (1994) Predicates and Pronominal
Arguments in Straits Salish. Language 70(4): 697–736.
Jensen, John Thayer (1977) Yapese reference grammar. Honolulu: University
of Hawai‘i Press.
488
Jespersen, Otto (1909-1949) A modern English grammar on historical principles.
Heidelberg: Carl Winter.
Johnston, Jason (1997) Systematic Homonymy and the Structure of Morphological
Categories: Some Lessons from Paradigm Geometry. Ph.D. thesis,
University of Sydney.
Josephs, Lewis S. (1975) Palauan reference grammar. Honolulu: University
of HawaiÊżi Press.
Joüon, Paul and T. Muraoka (2005) A grammar of biblical Hebrew. Rome:
Editrice Pontificio Istituto Biblico.
Judge, Anne and F. G. Healey (1983) A reference grammar of Modern French.
London: E. Arnold.
Juldashev, A. A. (1966) Baškirskij jazyk. In Vinogradov et al. (1966b), 173–
193.
Julien, Marit (2002) Syntactic Heads and Word Formation. Oxford: Oxford
University Press.
Junusaliev, B. M. (1966) Kirgizskij jazyk. In Vinogradov et al. (1966b), 482–
505.
Kachru, Yamuna (2006) Hindi. Amsterdam: John Benjamins.
Kacori, Thoma (1979) A handbook of Albanian. Sofia: Sofia University "Kliment
Ohridski".
489
Kajdarov, A. T. (1966) Ujgurskij (novoujgurskij) jazyk. In Vinogradov et al.
(1966b), 363–386.
Kakumasu, James Y. (1986) Urubu-Kaapor. In Derbyshire and Pullum
(1986), 326–403.
Kangasmaa-Minn, Eeva (1998) Mari. In Abondolo (1998b), 219–248.
Kari, Ethelbert E. (2000) Ogbronuagum (the Bukuma language). München:
Lincom Europa.
Karlsson, Fred (2008) Finnish: an essential grammar. London: Routledge, 2
ed.
Karpov, B. G. (1966) Khakasskij jazyk. In Vinogradov et al. (1966b), 428–
445.
Karst, Josef (1901) Historische grammatik des kilikisch-armenischen. Strassburg:
Karl J. Trübner.
Katsánis, N. and K. Dínas (1990) Grammatiki tis Koinis Koutsovlachikis.
Thessaloniki: Archeio Koutsovlachikon Meleton.
Katzir, Roni and Raj Singh (2009) On the Absence of XOR in Natural Language.
In Presuppositions and Implicatures: Proceedings of the MIT-Paris
Workshop, Paul Égré and Giorgio Magri, eds., Cambridge: MITWPL.
Katzner, Kenneth (1994) English-Russian Russian-English Dictionary (Revised
and Expanded ed.). New York: John Wiley and Sons.
490
Kayne, Richard (2005) Movement and silence. Oxford: Oxford University
Press.
Kearns, Kate (2007) Telic senses of deadjectival verbs. Lingua 117: 26–66.
Keegan, John (1997) A reference grammar of Mbay. München: Lincom Eu-
ropa.
Kelly, John (1859) A Practical Grammar of the Antient Gaelic, or Language
of the Isle of Man, Usually Called Manks. Douglas, Isle of Man: Manx
Society.
Kenesbaev, S. K. and N. B. Karasheva (1966) Kazakhskij jazyk. In Vinogradov
et al. (1966b), 320–339.
Kennedy, Christopher (1997) Projecting the adjective: The syntax and semantics
of gradability and comparison. Ph.D. thesis, University of California
at Santa Cruz.
Kennedy, Christopher (2007a) Modes of comparison. In Proceedings of CLS
43, Malcolm Elliott, James Kirby, Osamu Sawada, Eleni Staraki, and Suwon
Yoon, eds., Chicago: Chicago Linguistic Society.
Kennedy, Christopher (2007b) Vagueness and grammar: the semantics of relative
and absolute gradable adjectives. Linguistics and Philosophy 30(1):
1–45.
491
Kennedy, Christopher (Forthcoming) Vagueness and comparison. In Vagueness
and language use, Paul Egré and Nathan Klinedinst, eds., Palgrave
Macmillan.
Kennedy, Christopher and Beth Levin (2008) Measure of change: The adjectival
core of degree achievements. In Adjectives and adverbs: Syntax,
semantics and discourse, Louise McNally and Christopher Kennedy, eds.,
Oxford: Oxford University Press, 156–182.
Kennedy, Christopher and Louise McNally (2005) Scale structure and the
semantic typology of gradable predicates. Language 81: 345–381.
Kent, Roland G. (1953) Old Persian [Second Edition]. New Haven: American
Oriental Society.
Keresztes, László (1998) Mansi. In Abondolo (1998b), 387–427.
Kert, Georgij Martynovič (1971) Saamskij jazyk (kil’dinskij dialekt). Nauka.
Khabichev, M. A. (1966) Karačaevo-balkarskij jazyk. In Vinogradov et al.
(1966b), 213–233.
Khromov, A. L. (1972) Jagnobskij jazyk. Moscow: Nauka.
Kiefer, Ferenc (2001) Morphology and pragmatics. In The Handbook on Morphology,
Andrew Spencer and Arnold M. Zwicky, eds., Oxford: Blackwell,
272–279.
492
Kilian-Hatz, Christa (2008) A grammar of modern Khwe. Köln: Rüdiger
Köppe Verlag.
Kim, Stephen S. (2003) Santa. In Janhunen (2003), 346–363.
Kinchin Smith, F. and T.W. Melluish (1968) Greek. London: Hodder and
Stoughton.
Kiparsky, Paul (1973) “Elsewhere” in phonology. In A Festschrift for Morris
Halle, New York: Holt, Rinehart and Winston, 93–106.
Kiparsky, Paul (1983) Word formation in the lexicon. In Proceedings of the
1982 Mid-America Linguistics Conference, F. Ingemann, ed.
Kiparsky, Paul (2008) Universals constrain change, change results in typological
generalizations. In Linguistic universals and language change, Jeff
Good, ed., Oxford: Oxford University Press, xx.
Kipshidze, I. (1914) Grammatika mingrel’skago (iverskago) jazyka s"
xrestomatieju i slovarem". Sankt-Peterburg: Imperial Academy of Sci-
ences.
Kirkeby, Willy A. (1989) English-Norwegian Dictionary. Oslo: Norwegian
University Press.
Kiziria, A. I. (1967) Zanskij jazyk. In Vinogradov et al. (1967), 62–76.
Klajn, Ivan (2005) Gramatika srpskog jezika. Beograd: Zavod za udžzbenike
i nastavna sredstva.
493
Klamer, Marian (2010) A grammar of Teiwa. Berlin: Mouton de Gruyter.
Klein, Ewan (1991) Comparatives. Berlin: de Gruyter, 673–691.
Koehn, Edward H. and Sally S. Koehn (1986) Apalaí. In Derbyshire and
Pullum (1986), 33–127.
Kolesnikova, V. D. and O. A. Konstantinova (1968) Negidal’skij jazyk. In
Vinogradov et al. (1968), 109–128.
Koontz-Garboden, Andrew (2009) Anticausativization. Natural Language
and Linguistic Theory 27: 77–138.
Kovedjaeva, E. I. (1966a) Gornomarijskij jazyk. In Vinogradov et al. (1966c),
241–260.
Kovedjaeva, E. I. (1966b) Lugovo-vostochnyj marijskij jazyk. In Vinogradov
et al. (1966c), 221–240.
Kozintseva, Natalia (1995) Modern Eastern Armenian. München: Lincom
Europa.
Krause, Todd B. and Jonathan Slocum (2002-2004) Old Church Slavonic
Online. http://www.utexas.edu/cola/centers/lrc/eieol/ocsol-0-X.html.
Krause, Todd B. and Jonathan Slocum (2004) Classical Armenian Online.
http://www.utexas.edu/cola/centers/lrc/eieol/armol-0-X.html.
494
Krause, Todd B. and Jonathan Slocum (2007-2010) Tocharian Online.
http://www.utexas.edu/cola/centers/lrc/eieol/tokol-0.html.
Krejnovich, E. A. (1968) Nivkhskij jazyk. In Vinogradov et al. (1968), 408–
434.
Krippes, Karl A. (1996) Kazakh Grammar with Affix List. Kensington, MD:
Dunwoody Press.
Krishnamurti, Bhadriraju (2003) The Dravidian languages. Cambridge:
Cambridge University Press.
Kriz, Manuel (2011) Issues in the Semantics of Degree Achievements. Mag.
diplomarbeit, Universität Wien, Vienna.
Krueger, John R. (1961) Chuvash Manual: Introduction, Grammar, Reader
and Vocabulary. Bloomington, IN: Indiana University Publications.
Kruspe, Nicole (2004) A Grammar of Semelai. Cambridge, UK: Cambridge
University Press.
Krymova, N.I., A. Emsina, and G. Moltke (1960) Datsko-russkij slovar’.
Moscow: Gosudarstvennoe izdatel’stvo inostrannyx i nacional’nyx slovarej.
Kühner, Raphael and Friedrich Holzweissig (1912) Ausführliche Grammatik
der Lateinischen Sprache, Teil I. Hahnsche Buchhandlung, 2 ed.
Kullmann, Rita and D. Tserenpil (1996) Mongolian Grammar. Hong Kong:
Jensco.
495
Kumakhov, M. A. (1967) Adygejskij jazyk. In Vinogradov et al. (1967), 145–
164.
Künnap, Ago (1999) Kamass. München: Lincom.
Kuo, Cheng-chuen (2008) Comparative constructions in Amis. Master’s thesis,
National Taiwan University.
Kuryłowicz, Jerzy (1964) The inflectional categories of Indo-European. Heidelberg:
Carl Winter.
Kuznecova, A.I. and T.F. Efremova (1986) Slovar’ morfem ruskogo jazyka.
Moscow: Russkij jazyk.
Kytö, Merja and Suzanne Romaine (1997) Competing forms of adjective
comparison in Modern English: what could be more quicker, easier and
more effective? In To Explain the Present: Studies in the Changin English
Language in Honour of Matti Rissanen, T. Nevalainen and L. KahlasTarkka,
eds., Helsinki: Société Néophilologique, 329–352.
Laakso, Johanna (2001) The Finnic languages. In The Circum-Baltic languages:
Typology and Contact, vol. 1, Östen Dahl and Maria KoptjevskajaTamm,
eds., John Benjamins, 179–212.
Labouret, H. (1952) La langue des peuls ou foulbé. Dakar: IFAN.
Lamb, William (2003) Scottish Gaelic. München: Lincom Europa.
496
Landaburu, Jon (1979) La langue des andoke (Amazonie colombienne):
Grammaire. Paris: CNRS.
Lechner, Winfried (2004) Ellipsis in Comparatives. Berlin: Mouton de
Gruyter.
Lehmann, Winfred P. and Jonathan Slocum (2005) Hittite Online.
http://www.utexas.edu/cola/centers/lrc/eieol/hitol-0-X.html.
Lermontov, Michail Yuryevich (1832) Parus [The Sail].
Leskien, A. (1962) Handbuch der altbulgarischen (altkirchenslavischen)
Sprache. Heidelberg: Carl Winter Universitätsverlag.
Levin, Beth (1993) English Verb Classes and Alternations: A preliminary
investigation. Chicago: University of Chicago Press.
Levinson, Steven C. (1988) Putting linguistics on a proper footing: Explorations
in Goffman’s concepts of participation. In Erving Goffman: Exploring
the interaction order, Paul Drew and Anthony Wootton, eds., Oxford:
Polity Press, 161–227.
Lewis, Charlton T. and Charles Short (1879) A Latin Dictionary. Oxford:
Clarendon Press.
Lewis, David (1970) General Semantics. Synthese: An international journal
for epistemology, methodology and philosophy of science 22: 18–67.
Lewis, G. L. (1967) Turkish Grammar. Oxford: Oxford University Press.
497
Lewis, Henry and Holger Pedersen (1937) A concise comparative Celtic grammar.
Göttingen: Vandenhoeck and Ruprecht.
Li, Charles N. and Sandra A. Thompson (1981) Mandarin Chinese: A functional
reference grammar. Berkeley: University of California Press.
Lichtenberk, Frantisek (1983) A Grammar of Manam. Honolulu, HI, HI:
University of Hawaii Press.
Liddell, Henry George and Robert Scott (1996) Greek-English Lexicon. Oxford:
Clarendon Press, 9 ed.
Lieber, Rochelle (1980) On the organization of the lexicon. Ph.D. thesis,
Massachusetts Institute of Technology.
Lieber, Rochelle (1982) Allomorphy. Linguistic Analysis 10(1): 27–52.
Lomtatidze, Ketevan V. (1967a) Abazinskij jazyk. In Vinogradov et al.
(1967), 123–144.
Lomtatidze, Ketevan V. (1967b) Abkhazskij jazyk. In Vinogradov et al.
(1967), 101–122.
Louwrens, Louis J., Ingeborg M. Kosch, and Albert E. Kotzé (1995) Northern
Sotho. München: Lincom Europa.
Lukas, Johannes (1937) A Study of the Kanuri Language. Oxford: Oxford
University Press.
498
Lukoff, Fred (1993) An introductory course in Korean. Seoul: Yonsei University
Press.
Lunt, Horace G. (1952) A grammar of the Macedonian Literary Language.
Skopje: Državno knigoizdatelstvo na N.R. Makedonija.
Lunt, Horace G. (1959) Old Church Slavonic Grammar. ’s-Gravenhage: Mouton
and Co.
Lytkin, V. I. (1966) Komi-zyrjanskij jazyk. In Vinogradov et al. (1966c),
281–299.
Macaulay, Monica Ann (1996) A Grammar of Chalcatongo Mixtec. Berkeley;
Los Angeles: University of California Press.
Mace, John (2003) Persian Grammar. London: Routledge Curzon.
MacKay, Carolyn J. (1999) A Grammar of Misantla Totonac. Salt Lake City:
University of Utah Press.
Madieva, G. I. (1967) Avarskij jazyk. In Vinogradov et al. (1967), 255–271.
Magomedbekova, E. M. (1967a) Karatinskij jazyk. In Vinogradov et al.
(1967), 322–335.
Magomedbekova, E. M. (1967b) Čamalinskij jazyk. In Vinogradov et al.
(1967), 384–399.
499
Magomedov, A. G. (1966) Kumykskij jazyk. In Vinogradov et al. (1966b),
194–212.
Magometov, A. A. (1965) Tabasaranskij jazyk. Tbilisi: Metsniereba.
Maiden, Martin (2005) Morphological autonomy and diachrony. In Yearbook
of Morphology 2004, Geert Booij and Jaap van Marle, eds., Dordrecht:
Kluwer, 137–175.
Malchukov, Andrej L. (1995) Even. München: Lincom Europa.
Marantz, Alec (1989) Clitics and phrase structure. In Alternative Conceptions
of Phrase Structure, Mark Baltin and Anthony Kroch, eds., Chicago: The
University of Chicago Press, 99–116.
Marantz, Alec (1996) ‘Cat’ as a phrasal idiom, ms. MIT.
Marantz, Alec (1997) No Escape from Syntax: Don’t Try Morphological
Analysis in the Privacy of Your Own Lexicon. In Proceedings of the 21st
Penn Linguistics Colloquium, vol. 4, Alexis Dimitriadis, Laura Siegel,
Clarissa Surek-Clark, and Alexander Williams, eds., Philadelphia: U Penn
Working Papers in Linguistics, 201–225.
Marantz, Alec (2007) Phases and words. In Phases in the theory of grammar,
Sook-Hee Choe, ed., Seoul: Dong In, 199–222.
Marr, N. Ja. (1910) Grammatika chanskago (lazskago) jazyka s" xrestomatieju
i slovarem". Sankt-Peterburg: Imperatorskaja Akademija Nauk.
500
Martin, Samuel Elmo (2004) A reference grammar of Japanese. Honolulu:
University of HawaiÊżi Press.
Maslova, Elena (2003) A Grammar of Kolyma Yukaghir. Berlin; New York:
Mouton de Gruyter.
Matisoff, James A. (1973) The grammar of Lahu. Berkeley: University of
California Press.
Matras, Yaron (2002) Romani: A Linguistic Introduction. Cambridge: Cambridge
University Press.
Matthews, P.H. (1972) Inflectional Morphology: A theoretical study based
on aspects of Latin verb conjugation. Cambrdige: Cambridge University
Press.
Matthews, P.H. (1991) Morphology. Cambridge: Cambridge University Press,
2 ed.
Matthews, Stephen and Virginia Yip (1994) Cantonese: A Comprehensive
Grammar. London: Routledge.
Mayo, Peter (1993) Belorussian. In Comrie and Corbett (1993), 887–946.
Mc Laughlin, Fiona (2004) Is there an adjective class in Wolof. In Adjective
classes: A cross-linguistic typology, R.M.W. Dixon and Alexandra Y.
Aikhenvald, eds., Oxford: Oxford University Press, 242–262.
McGregor, R. S. (1972) Outline of Hindi Grammar. Oxford: Clarendon Press.
501
Meeuwis, Michael (1998) Lingala. München: Lincom Europa.
Meinhof, Karl (1909) Lehrbuch der Nama-sprache. Berlin: Georg Reimer.
Mel’čuk, Igor A. (1994) Suppletion: Toward a Logical Analysis of the Concept.
Studies in Language 18(2): 339–410.
Mel’čuk, Igor A. (2006) Aspects of the Theory of Morphology. 2, Berlin: Mouton
de Gruyter.
Merchant, Jason (2009) Phrasal and clausal comparatives in Greek and the
abstractness of syntax. Journal of Greek Linguistics 9: 134–164.
Merchant, Jason (to appear) Genitives of comparison in Greek. In Proceedings
of ICGL9, Katerina Chatzopoulou, Alexandra Ioannidou, and Suwon
Yoon, eds.
Migeod, Frederick William Hugh (1914) A grammar of the Hausa language.
London: Kegan Paul, Trench, Trubner & Co.
Milne, Leslie (1921) An Elementary Palaung Grammar. Oxford: Clarendon
Press.
Mitxelena, Koldo (1988) Euskal idazlan guztiak. Donostia: Euskal Editoreen
Elkartea [9 vols.].
Mixco, Mauricio J. (2000) Kiliwa. München: Lincom Europa.
502
von Möllendorff, P. G. (1892) A Manchu Grammar. Shanghai: American
Presbyterian Mission Press.
Morev, L. N., A. A. Moskalev, and Ju. Ja. Plam (1972) Laosckij jazyk.
Moscow: Nauka.
Munro, Pamela and Dieynaba Gaye (1991) Ay baati Wolof: a Wolof dictionary.
Los Angeles: UCLA Occasional Papers in Linguistics.
Musaev, K. M. (1966) Karaimskij jazyk. In Vinogradov et al. (1966b), 260–
279.
Musan, Renate (2002) The German Perfect: Its semantic composition and
its interaction with temporal adverbials. Dordrecht: Kluwer.
Nagayama, Yukari (2003) Očerk grammatiki aljutorskogo jazyka [Endangered
Languages of the Pacific Rim A2-038]. Kyoto: Nakanishi Printing Co.
Nandris, Grigoe (1986) Colloquial Rumanian. London: Routledge.
Nash, David (1980) Topics in Warlpiri Grammar. Ph.D. thesis, M.I.T., Cam-
bridge.
Nau, Nicole (1992) Der ostseefinnische Superlativ im arealen Kontext.
Finnisch-ugrische Mitteilungen 16/17: 13–23.
Nedjlakov, Igor (1997) Evenki. London: Routledge.
503
Neeleman, Ad and Kriszta Szendrői (2007) Radical Pro Drop and the Morphology
of Pronouns. Linguistic Inquiry 38(4): 671–714.
Neukom, Lukas (2001) Santali. München: Lincom Europa.
Nevins, Andrew (2007) The representation of third person and its consequences
for the person-case constraint. Natural Language and Linguistic
Theory 25(2): 273–313.
Nichols, Johanna (2011) Ingush Grammar. Berkeley: University of California
Press.
Nichols, Johanna (n.d.) Chechen. Unpublished manuscript at http://
ingush.narod.ru/chech/book/.
Nickel, Klaus Peter (1990) Samisk grammatikk. Universitetsforlaget.
Nicklas, T. Dale (1979) Reference grammar to the Choctaw Language. Durant,
Okla.: Choctaw Bilingual Education Program, Southeastern Oklahoma
State University.
Noonan, Michael P. (1992) A Grammar of Lango. Berlin; New York: Mouton
de Gruyter.
Noreen, Adolf (1923) Altisländische und altnorwegische Grammatik (Lautund
Flexionslehre) unter Berücksichtigung des Urnordischen. Halle: M.
Niemeyer.
504
Noyer, Rolf (1997) Features, positions and affixes in autonomous morphological
structure. New York: Garland.
Noyer, Rolf (1998) Impoverishment theory and morphosyntactic markedness.
In Morphology and its Relation to Phonology and Syntax,, Steven G. Lapointe,
Diane Brentari, and Patrick Farrell, eds., Palo Alto: CSLI, 264–
285.
Nugteren, Hans (2003) Shira Yughur. In Janhunen (2003), 265–285.
Oda, Toshiko (2008) Degree constructions in Japanese. Ph.D. thesis, University
of Connecticut.
Odhner, John D. (1981) English-Lingala manual. Landham, MD, USA: University
Press of America.
O’Herin, Brian (1995) Case and agreement in Abaza. Ph.D. thesis, University
of California at Santa Cruz.
Ohtaki, Koichi (2010) The Syntax-Lexicon Interface in the Acquisition of English
Comparative Morphology. In Proceedings of the 2009 Mind/Context
Divide Workshop, Michael Iverson, Ivan Ivanov, Tiffany Judy, Jason Rothman,
Roumyana Slabakova, and Marta Tryzna, eds., Somerville: Cascadilla
Press, 96–103.
Olawsky, Knut J. (2006) A Grammar of Urarina. Berlin; New York: Mouton
de Gruyter.
505
Országh, László (1988) Angol-Magyar / Magyar-Angol Szótár [16th ed.]. Budapest:
Akadémiai Kiadó.
O’Siadhail, Michael (1989) Modern Irish: Grammatical structure and dialect
variation. Cambridge: Cambridge University Press.
Osthoff, Hermann (1888) Etymologica I. Beiträge zur Geschichte der
deutschen Sprache und Literatur 13: 395–463.
Osthoff, Hermann (1899) Vom Suppletivwesen der indogermanischen
Sprache. Heidelberg: J. Hörning.
Pakhalina, T. N. (1966) Vakhanskij jazyk. In Vinogradov et al. (1966a), 398–
418.
Pandharipande, Rajeshwari (1997) Marathi. London; York: Routledge.
Paul, Ludwig (2009) Zazaki. In Windfuhr (2009), 545–586.
Payne, Doris L. and Thomas Payne (1990) Yagua. In Handbook of Amazonian
Languages vol. 2, vol. 3, Desmond C. Derbyshire and Geoffrey K. Pullum,
eds., Berlin: Mouton de Gruyter, 249–474.
Pederson, Donald Leslie (1977) Grammatical structures of Guaraní. Master’s
thesis, University of Southern California, Los Angeles.
Penka, Doris (2007) Negative indefinites. Ph.D. thesis, Universität Tübingen.
506
Pensalfini, Robert (2003) A Grammar of Jingulu. Canberra: Pacific Linguis-
tics.
Perini, Mário A. (2002) Modern Portuguese: A reference grammar. New
Haven: Yale University Press.
Pertsova, Katya (2007) Learning form-meaning mappings in presence of
homonymy. Ph.D. thesis, UCLA, Los Angeles.
Pertsova, Katya (2011) Grounding systematic syncretism in learning. Linguistic
Inquiry 42(2): 225–266.
Pesetsky, David (1985) Morphology and Logical Form. Linguistic Inquiry
16(2): 193–246.
Pesetsky, David (1995) Zero Syntax: Experiencers and Cascades. Cambridge:
MIT Press.
Petrova, T. I. (1967) Jazyk orokov (Ul’ta). Leningrad: Nauka.
Phillips, John D. (2004) Manx. München: Lincom Europa.
Pinker, Steven (1984) Language Learnability and Language Development.
Cambridge: MIT Press.
Pirejko, L. A. (1966) Talyshskij jazyk. In Vinogradov et al. (1966a), 302–322.
Pobożniak, Tadeusz (1964) Grammar of the Lovari dialect. Kraków: Państwowe
wydawnictwo naukowe.
507
Pokrovskaja, L. A. (1966) Gagauzskij jazyk. In Vinogradov et al. (1966b),
112–138.
Poppe, Nicholas (1964) Bashkir Manual: Descriptive Grammar and Texts
with A Bashkir-English Glossary. Bloomington, IN, USA, and the Netherlands:
Indian University and Mouton &Co., the Hague.
Poser, William J. (1992) Blocking of phrasal constructions by lexical items.
In Lexical Matters, Ivan Sag and Anna Szabolcsi, eds., Stanford: CSLI,
111–130.
Poulos, George and Christian T. Msimang (1998) A linguistic analysis of
Zulu. Cape Town: Via Afrika.
Pratico, Gary D. and Miles V. Van Pelt (2009) Basics of Biblical Hebrew
Grammar. Grand Rapids, Mich.: Zondervan.
Prisma (1995) Prisma’s Swedish-English and English-Swedish Dictionary
(Third Ed.). Minneapolis: University of Minnesota Press.
Prokofjeva, E. D. (1966) Seljkupskij jazyk. In Vinogradov et al. (1966c),
396–415.
Pulte, William and Durbin Feeling (1975) Outline of Cherokee Grammar.
Tahlequah: Cherokee Nation of Oklahoma.
Quick, Phil (2007) A Grammar of the Pendau Language of central Sulawesi,
508
Indonesian. Canberra, Australia: Pacific Linguistics, Researxch School of
Pacific and Asian Studies, The Australian National University.
Radkevich, Nina (2010) On Location: The structure of case and adpositions.
Ph.D. thesis, University of Connecticut.
Ramchand, Gillian Catriona (2008) Verb Meaning and the Lexicon: A First
Phase Syntax. Cambridge: Cambridge University Press.
Rastorgueva, V. S. (1963) A short sketch of Tajik grammar. Bloomington
and the Hague: Indiana University and Mouton and Co.
Rastorgueva, V. S. (1966) Beludžskij jazyk. In Vinogradov et al. (1966a),
323–341.
Rastorgueva, V. S., A. A. Kerimova, A. K. Mamedzade, L. A. Pirejko, and
D. I. Edel’man (1971) Giljanskij jazyk. Moscow: Nauka.
Raun, Alo (1949) Zum Komparativ und Superlativ im Livischen. Finnsichugrischen
Forschungen 30: 231–241.
Redei, Karoly (1965) Northern Ostyak Chrestomathy. Indiana University
Publications.
Refsing, Kirsten (1986) The Ainu language. Aarhus: Aarhus University Press.
Reiter, Norbert (1979) Komparative. Wiesbaden: Harrassowitz.
Rennison, John R. (1997) Koromfe. London: Routledge.
509
Reshetov, V. V. (1966) Uzbekskij jazyk. In Vinogradov et al. (1966b), 340–
362.
Rett, Jessica (2008) Degree modification in natural language. Ph.D. thesis,
Rutgers University.
Rhys Jones, T. J. (1977) Teach Yourself Living Welsh. Sevenoaks, Kent:
Hodder and Stoughton.
Rießler, Michael (2007) Grammatical borrowing in Kildin Saami. In Grammatical
borrowing in cross-linguistic perspective, Yaron Matras and
Jeanette Sakel, eds., Berlin: Walter de Gruyter, 229–244.
Riggs, Elias (1856) A grammar of the Modern Armenian language as spoken
in Constantinople and Asia Minor. Constantinople: A. B. Churchill, 2 ed.
Rijk, Rudolf P. G. de (2008) Standard Basque. Cambrdige: MIT Press.
Rjagoev, Vladimir Dmitrievich (1977) Tixvincskij govor karel’skogo jazyka.
Leningrad: Nauka.
Roberts, John (1987) Amele. Beckenham: Croom Helm.
Robinson, Charles Henry (1897) Hausa grammar with exercises, readings and
vocabulary. London.
Robson, Barbara and Habibullah Tegey (2009) Pashto. In Windfuhr (2009),
721–772.
510
Rogava, G. V. and Z. I. Kerasheva (1966) Grammatika adygejskogo jazyka.
Krasnodar: Krasnodarskoe knizhnoe izdatel’stvo.
Rombandeeva, E. I. (1966) Mansijskij jazyk. In Vinogradov et al. (1966c),
343–360.
Rossini, Carlo Conti (1912) La langue des kemant en Abyssine. Wien: Alfred
Hölder.
Rothstein, Robert A. (1993) Polish. In Comrie and Corbett (1993), 686–758.
Rothstein, Susan (2004) Structuring Events. Malden, Mass.: Blackwell.
Rotstein, Carmen and Yoad Winter (2004) Total adjectives vs. partial adjectives:
Scale structure and higher-order modifiers. Natural Language Semantics
12: 259–288.
Rullmann, Hotze (1995) Maximality in the semantics of wh-constructions.
Ph.D. thesis, University of Massachusetts, Amherst.
Rybatzki, Volker (2003) Middle Mongol. In Janhunen (2003), 57–82.
Ryding, Karin C. (2005) A reference grammar of modern standard Arabic.
Cambridge: Cambridge University Press.
Saeed, John I. (1999) Somali. Amsterdam; Philadelphia: John Benjamins
Publishing Company.
Sakel, Jeanette (2004) Mosetén. Berlin: Mouton de Gruyter.
511
Sala, Antonio Maria (1914) Essai de Grammaire kaiapó, langue des Indiens
kaiapó, Brésil. Anthropos 9: 233–240.
Salminen, Tapani (1998) Nenets. In Abondolo (1998b), 516–547.
Sammallahti, Pekka (1998) Saamic. In Abondolo (1998b), 43–95.
Sapir, Edward (1944) Grading: A Study in Semantics. Philosophy of Science
11: 93–116.
Sassoon, Galit Weidman (2007) Vagueness, gradability and typicality, a comprehensive
semantic analysis. Ph.d. dissertation, Tel Aviv University.
Sassoon, Galit Weidman (2010) The degree functions of negative adjectives.
Natural Language Semantics 18: 141–181.
Sat, Sh. Ch. (1966) Tuvinskij jazyk. In Vinogradov et al. (1966b), 387–402.
Schmidt, Karl Horst (1991) Svan. In Harris (1991a), 473–556.
Schön, James Frederick (1862) Grammar of the Hausa language. London:
Church Missionary House.
Schulz, Eckehard, Günther Krahl, and Wolfgang Reuschel (2000) Standard
Arabic: An elementary - intermediate course. Cambridge: Cambridge University
Press.
Schulze, Wolfgang (2000) Northern Talysh. München: Lincom Europa.
512
Schuster-Šewc, Hinc (1996) Grammar of the Upper Sorbian Language.
München: Lincom Europa.
Schwarzschild, Roger (2008) The Semantics of Comparatives and Other Degree
Constructions. Language and Linguistics Compass 2(2): 308–331.
Schweiger, Fritz (1984) Comparative: a neglected category in Australian
linguistics. Working papers in language and linguistics 18: 28–38.
Schweizer, Bruno (2008) Zimbrische Gesamtgrammatik. Stuttgart: Franz
Steiner Verlag.
Seiler, Andreas (1830) Kurzgefasste Grammatik der sorben-wendischen
Sprache nach dem Budissiner Dialekte. Budissin: K. A. F. Weller.
Seiler, Hansjakob (1950) Die primären griechischen Steigerungsformen. Hamburg:
Hansischer Gildenverlag.
Seuren, Pieter (1973) The comparative. In Generative Grammar in Europe,
Ferenc Kiefer and Nicolas Ruwet, eds., Doredrecht: Riedel, 528–564.
Seuren, Pieter (1984) The comparative revisited. Journal of Semantics 3:
109–141.
Shagirov, A. K. (1967) Kabardinskij jazyk. In Vinogradov et al. (1967), 165–
193.
Shaha, Brojo Nath (1884) A grammar of the Lúshái language. Calcutta:
Bengal Secretariat Press.
513
Shevelov, George Y. (1993) Ukrainian. In Comrie and Corbett (1993), 947–
998.
Shixuan, Xu (2001) The Bisu language. München: Lincom Europa.
Short, David (1993) Slovak. In Comrie and Corbett (1993), 533–592.
Siegel, Dorothy (1978) The Adjacency Constraint and the theory of Morphology.
In Proceedings of NELS 8, M. Stein, ed., Amherst: GLSA, 189–197.
Sievers, Eduard (1882) Angelsächsische Grammatik. Halle: Max Niemeyer.
Silverstein, Michael (1976) Feature hierarchies and Ergativity. In Grammatical
categories in Australian languages, Richard M.W. Dixon, ed., Canberra:
Australian Institute of Aboriginal Studies, 112–171.
Simon, Horst (2005) Only you? Philological investigations into the alleged
inclusive-exclusive distinction in the secon person plural. In Clusivity,
Elena Filimonova, ed., Amsterdam: John Benjamins, 113–150.
Simpson, Jane (1983) Aspects of Warlpiri morphology and syntax. Ph.D. thesis,
M.I.T., Cambridge.
Simpson, Jane and Meg Withgott (1986) Pronominal Clitic Clusters and
Templates. In The syntax of pronominal clitics, Syntax and Semantics,
vol. 19, Hagit Borer, ed., New York: Academic Press, 149–174.
Sjoberg, Andrée F. (1962) Uzbek structural grammar. Bloomington and the
Hague: Indiana University and Mouton &Co.
514
Skjærvø, Prods Oktor (2009a) Middle West Iranian. In Windfuhr (2009),
196–278.
Skjærvø, Prods Oktor (2009b) Old Iranian. In Windfuhr (2009), 43–195.
Skorik, Piotr Ja. (1961) Grammatika čukotskogo jazyka, čast’ I: fonetika i
morfologija imennyx častej reči, vol. 1. Moscow: Izdatel’stvo akademii nauk
SSSR.
Skorik, Piotr Ja. (1968) Kerekskij jazyk. In Vinogradov et al. (1968), 310–
333.
Skorik, Piotr Ja. (1977) Grammatika chukotskogo jazyka, vol. 2. Leningrad:
Nauka.
Skribnik, Elena (2003) Buryat. In Janhunen (2003), 102–128.
Smeets, Ineke (2007) A Grammar of Mapuche. Berlin; New York: Mouton
de Gruyter.
Smythe, W.E. (1948) Elementary Grammar of the GumbáiNgar Language
(North Coast, N.S.W.). Oceania 19(2): 130–191.
Sofroniou, S.A. (1962) Teach Yourself Modern Greek. London: Hodder and
Stoughton.
Sokolova, V. S. (1966) Šugnano-rušanskaja jazykovaja gruppa. In Vinogradov
et al. (1966a), 362–397.
515
Sokolovskaja, N. K. (1980) Nekotorye semantičeskie universalii v sisteme
ličnyx mestoimenij. In Teorija i tipologija mesoimenij, I. F. Vardul’, ed.,
Moscow: Nauka, 84–102.
Sproat, Richard William (1985) On Deriving the Lexicon. Ph.D. thesis,
M.I.T.
Stanojčić, Živojin and Ljubomir Popović (1992) Gramatika srpskog jezika.
Beograd: Zavod za udžzbenike i nastavna sredstva.
Stassen, Leon (1985) Comparison and Universal Grammar. Oxford: Basil
Blackwell.
Stassen, Leon (2008) Comparative constructions. In World Atlas of Language
Structures Online, Martin Haspelmath, Matthew S. Dryer, David
Gil, and Bernard Comrie, eds., chap. 121, Max Planck Digital Library,
http://wals.info/feature/121.
Stateva, Penka (2002) How different are different degree constructions?
Ph.D. thesis, University of Connecticut.
Stateva, Penka (2005) Presuppositions in superlatives, ms. ZAS, Berlin.
von Stechow, Arnim (1984) Comparing semantic theories of the comparative.
Journal of Semantics 3: 1–77.
Steele, Susan (1987) Review of Stassen (1985). Language 63(3): 629–632.
516
Steere, Edward LL.D. (1884) A Handbook of the Swahili Language as spoken
at Zanzibar. London: Society for Promoting Christian Knowledge.
Steever, Sanford B., ed. (1998a) The Dravidian languages. London: Rout-
ledge.
Steever, Sanford B. (1998b) Introduction to the Dravidian languages. In
Steever (1998a), 1–39.
Strandskogen, Åse-Berit and Rolf Strandskogen (1986) Norwegian: An essential
grammar. London: Routledge.
Street, John C. (1963) Khalkha structure. Bloomington and the Hague: Indiana
University and Mouton &Co.
Strehlow, T. G. H. (1942) Aranda grammar. Oceania 13(1): 71–103.
Stump, Gregory T. (2001) Inflectional Morphology: A theory of paradigm
structure. Cambridge: Cambridge University Press.
Sunik, O. P. (1968a) Udègejskij jazyk. In Vinogradov et al. (1968), 210–232.
Sunik, O. P. (1968b) Ul’čskij jazyk. In Vinogradov et al. (1968), 149–171.
Svantesson, Jan-Olof (2003) Khalkha. In Janhunen (2003), 154–176.
Swadesh, Mauricio (1969) Elementos del tarasco antiguo. México: Universidad
nacional autónoma de méxico.
517
Szabolcsi, Anna (1986) Comparative superlatives. In MIT Working Papers
in Linguistics, vol. 8, Naoki Fukui, Tova R Rapoport, and Elizabeth Sagey,
eds., 245–265.
Tabulova, N.T. (1976) Grammatika abazinskogo jazyka: fonetika i morfologija.
Cherkessk: Karachaevo-Cherkesskoe otdelenie Stavropol’skogo
knizhnogo izdatel’stva.
Takahashi, Masahiko (2011) Some theoretical consequences of Case-marking
in Japanese. Ph.D. thesis, University of Connecticut, Storrs.
Tamura, Suzuko (2000) The Ainu language. Tokyo: Sanseido.
Tarasenkova, Oksana (2010) Acquisition of Agreement and Morphological
Features Within the Noun Phrase in Russian. Ph.D. thesis, University of
Connecticut, Storrs.
Tasseva-Kurktchieva, Mila (2006) The Categorial Status of Quantifiers
in Bulgarian: Evidence for DP over QP. In Proceedings of FASL 14:
The Princeton Meeting, James Lavine, Steven Franks, Mila TassevaKurktchieva,
and Hana Filip, eds., Ann Arbor: Michigan Slavic Publications,
378–393.
Tauberschmidt, Gerhard (1999) A Grammar of Sinaugoro. Canberra, Australia:
Pacific Linguistics, Researxch School of Pacific and Asian Studies,
The Australian National University.
518
Tauli, Valter (1973) Standard Estonian Grammar, Part I: Phonology, morphology,
word-formation. Uppsala: Almqvist and Wiksell.
Tauli, Valter (1983) Standard Estonian Grammar, Part II: Syntax. Uppsala:
Uppsala University.
Taylor, F. W. (1921) A first grammar of the Adamawa dialect of the Fulani
language. Oxford: Clarendon Press.
Tegey, Habibullah and Barbara Robson (1996) A reference grammar of
Pashto. Washington: Center for Applied Linguistics.
Teng, Stacy Fang-Ching (2008) A reference grammar of Puyuma. Canberra:
Pacific Linguistics.
Tenny, Carol (1987) Grammaticalizing Aspect and Affectedness. Ph.D. thesis,
MIT, Cambridge.
Tenny, Carol (1994) Aspectual roles and the syntax-semantics interface. Dordrecht:
Kluwer Academic Publishers.
Tepljashina, T. I. (1966) Udmurtskij jazyk. In Jazyki narodov S.S.S.R. t.
III: Finno-ugorskie i samodijskie jazyki, V. V. Vinogradov, ed., Moscow:
Nauka, 261–280.
Tereshchenko, N. M. (1966) Nganasanskij jazyk. In Vinogradov et al. (1966c),
416–437.
Tereshchenko, N. M. (1979) Nganasanskij jazyk. Moscow: Nauka.
519
Ternes, Elmar (1970) Grammaire structurale du breton de l’ile de Groix. Carl
Winter Universitätsverlag.
Thackston, Wheeler M. (2006a) Kurmanji Kurdish: A Reference Grammar
with Selected Readings, ms. Harvard.
Thackston, Wheeler M. (2006b) Sorani Kurdish: A Reference Grammar with
Selected Readings, ms. Harvard.
Thompson, Laurence C (1987) A Vietnamese Reference Grammar. Honolulu.:
University of Hawaii Press.
Thompson, Sandra A., Joseph Sung-Yul Park, and Charles N. Li (2006) A
reference grammar of Wappo. Berkeley: University of California Press.
Thráinsson, Höskuldur (2010) Predictable and unpredictable sources of variable
verb and adverb placement in Scandinavian. Lingua 120(5): 1062–
1088.
Thráinsson, Höskuldur, Hjalmar P. Petersen, Jógvan í Lon Jacobsen, and Zakaris
Svabo Hansen (2004) Faroese: An overview and reference grammar.
Tórshavn: Føroya Fróðskaparfelag.
Thurneysen, Rudolf (1909) Handbuch des Alt-Irischen. Heidelberg: Carl
Winter.
Tiersma, Pieter Meijes (1985) Frisian reference grammar. Dordrecht: Foris.
520
Tisserant, Charles (1930) Essai sur la grammaire Banda. Paris: Institut
d’ethnologie.
Todaeva, B. Kh. (1968) Kalmykskij jazyk. In Vinogradov et al. (1968), 34–52.
Tompa, József (1968) Ungarische Grammatik. Berlin: Mouton.
Topuria, B. T. (1967) Svanskij jazyk. In Vinogradov et al. (1967), 77–94.
Trask, R. L. (1997) The history of Basque. London: Routledge.
Trask, R.L. (2003) The Noun Phrase: nouns, determiners and modifiers;
pronouns and names. In A grammar of Basque, José Ignacio Hualde and
Jon Ortiz de Urbina, eds., Berlin: Mouton, 113–170.
Travis, Lisa deMena (1984) Parameters and Effects of Word Order Variation.
Ph.D. thesis, MIT, Cambrdige.
Trommer, Jochen (1999) Morphology consuming syntax’s resources: Generation
and parsing in a Minimalist version of Distributed Morphology.
Tsertsvadze, I. I. (1967) Andijskij jazyk. In Vinogradov et al. (1967), 276–292.
Tsintsius, V. I. (1982) Negidal’skij jazyk. Leningrad: Nauka.
Tsumagari, Toshiro (2003) Dagur. In Janhunen (2003), 129–153.
Ultan, Russell (1972) Some features of basic comparative constructions.
Working papers on language universals 9: 117–162.
521
Underwood, Horace Grant (1914) An introduction to the Korean spoken language.
New York: MacMillan, 2 ed.
University of Latvia (1999) Latvian Grammar (online).
http://www.ailab.lv/valoda/v1.htm.
Usmanova, M. G. (2007) Grammatika baškirskogo jazyka. Ufa: Kitap.
Vajda, Edward J. (2004) Ket. München: Lincom Europa.
Valentine, Randy (2001) Nishnaabemwin reference grammar. Toronto: University
of Toronto Press.
Vall, M. N. and I. A. Kanakin (1990) Očerk fonologii i grammatiki ketskogo
jazyka. Novisibirsk: Nauka.
Vasiliauskiene, Virginija and Jonathan Slocum (2005-2007) Baltic Online.
http://www.utexas.edu/cola/centers/lrc/eieol/armol-0-X.html.
Vasmer, Max (1953-1958) Russisches etymologisches Wörterbuch. Heidelberg:
Carl Winter.
Veselinova, Ljuba N. (2006) Suppletion in verb paradigms. Amsterdam: John
Benjamins.
Viitso, Tiit-Rein (1998) Estonian. In Abondolo (1998b), 115–148.
Vincent, Nigel and Kersti Börjars (1996) Suppletion and syntactic
theory. In The Proceedings of the LFG ’96 conference,
522
Miriam Butt and Tracy Holloway King, eds., Stanford: CSLI,
http://cslipublications.stanford.edu/LFG/1/VincentBorjars.html.
Vinogradov, V. V., B. A. Serebrennikov, N. A. Baskakov, Ju. D. Desheriev,
P. Ja. Skorik, and V. F. Beljaev, eds. (1966a) Jazyki narodov S.S.S.R. t.
I: indo-evropejskie jazyki, vol. 1. Moscow: Nauka.
Vinogradov, V. V., B. A. Serebrennikov, N. A. Baskakov, Ju. D. Desheriev,
P. Ja. Skorik, and V. F. Beljaev, eds. (1966b) Jazyki narodov S.S.S.R. t.
II: tjurkskie jazyki, vol. 2. Moscow: Nauka.
Vinogradov, V. V., B. A. Serebrennikov, N. A. Baskakov, Ju. D. Desheriev,
P. Ja. Skorik, and V. F. Beljaev, eds. (1966c) Jazyki narodov S.S.S.R. t.
III: Finno-ugorskie i samodijskie jazyki, vol. 3. Moscow: Nauka.
Vinogradov, V. V., B. A. Serebrennikov, N. A. Baskakov, Ju. D. Desheriev,
P. Ja. Skorik, and V. F. Beljaev, eds. (1967) Jazyki narodov S.S.S.R. t. IV:
iberijsko-kavkazskie jazyki, vol. 4. Moscow: Nauka.
Vinogradov, V. V., B. A. Serebrennikov, N. A. Baskakov, Ju. D. Desheriev,
P. Ja. Skorik, and V. F. Beljaev, eds. (1968) Jazyki narodov S.S.S.R. t. V:
Mongol’skie, tunguso-manˇvžurskie i paleoaziatskie jazyki, vol. 5. Moscow:
Nauka.
Volodin, Aleksandr P. (1976) Itel’menskij jazyk. Leningrad: Nauka.
Vondrák, Václav (1900) Altkirchenslavische Grammatik. Berlin: Weidmannsche
Buchhandlung.
523
van der Voort, Hein (2004) A Grammar of Kwaza. Berlin: Mouton de
Gruyter.
Wali, Kashi and Omker N. Koul (1997) Kashmiri: A Cognitive-Descriptive
Grammar. London: Routledge.
Watkins, Laurel J. and Parker MacKenzie (1984) A grammar of Kiowa. Lincoln,
Nebr.: University of Nebraska Press.
Watters, David E. (2002) A Grammar of Kham. Cambridge, UK: Cambridge
University Press.
Wechsler, Stephen (2010) What ’you’ and ’I’ mean to each other: Person
indexicals, self-ascription, and theory of mind. Language 86(2): 332–365.
Weiers, Michael (2003) Moghol. In Janhunen (2003), 248–264.
Weigand, Gustav (1888) Die Sprache der Olympo-Walachen. Leipzig: Johann
Ambrosius Barth.
Weinreich, Uriel (1984) Modern English-Yiddish, Yiddish-English Dictionary.
New York: Random House.
Weiss, Michael (2009) Outline of the historical and comparative grammar of
Latin. Ann Arbor: Beech Stave Press.
Werner, Heinrich (1997a) Abriß der kottischen Grammatik. Wiesbaden: Harassowitz
Verlag.
524
Werner, Heinrich (1997b) Das Jugische (Sym-Ketische). Wiesbaden: Harassowitz
Verlag.
Werner, Heinrich (1997c) Die ketische Sprache. Wiesbaden: Harassowitz Ver-
lag.
Wiese, Bernd (2004) Unterspezifizierte Stammparadigmen: Zur Systematik
des Verbablauts im Gegenwartsdeutschen, unpublished ms. IDS
Mannheim.
Wiese, Bernd (2005) Form and function of verb ablaut in contemporary standard
German. In Studies in Integrational Linguistics, Robin Sackmann,
ed., Amsterdam: John Benjamins, To appear.
Williams, Edwin (2003) Representation Theory. Cambridge: MIT Press.
Williamson, Kay (1965) A grammar of the Kolokuma dialect of İj˙o. Cambrdige:
Cambridge University Press.
Wilson, P. M. (1970) Simplified Swahili. Nairobi: East African Literature
Bureau.
Windfuhr, Gernot, ed. (2009) The Iranian Languages. London: Routledge.
Windfuhr, Gernot and John R. Perry (2009) Persian and Tajik. In Windfuhr
(2009), 416–544.
Winkler, Eberhard (2001) Udmurt. München: Lincom Europa.
525
Winter, Yoad (2006) Closure and telicity across categories. In Proceedings
of Semantics and Linguistic Theory 16, Christopher Tancredi, Makoto
Kanazawa, Ikumi Imani, and Kiyomi Kusumoto, eds., Ithaca: Cornell
Linguistics Club.
Wold, Dag (1991) A few properties of the ... the ... comparative constructions.
In Proceedings of the Second Annual Formal Linguistic Society of MidAmerica
(FLSM2), 272–281.
Woollams, Geoff (1996) A grammar of Karo Batak, Sumatra. Canberra: Department
of Linguistics, the Australian National University.
Wordick, F. J. F. (1982) The Yindjibarndi language. Canberra: Pacific Lin-
guistics.
Wright, Joseph (1910) Grammar of the Gothic Language. Oxford: Clarendon
Press.
Wurmbrand, Susi (2008) WollP: Where syntax and semantics meet. In Proceedings
of NELS 37, Emily Elfner and Martin Walkow, eds., Amherst,
Mass: Graduate Linguistic Student Association.
Wurmbrand, Susi (2010) Parasitic morphology in Germanic: Consequences
for the theory of syntactic licensing, ms. University of Connecticut [Detailed
handout available at: http://wurmbrand.uconn.edu/Susi/Agree.
html].
526
Wurmbrand, Susi (2011a) On Agree and Merge, ms., Univeristy of Connecticut,
Storrs. [Lecture notes available at: http://wurmbrand.uconn.edu/
Susi/Agree.html].
Wurmbrand, Susi (2011b) Parasitic participles in Germanic: Evidence for
the theory of verb clusters., ms. University of Connecticut, Storrs.
Wurmbrand, Susi (2011c) Tense and aspect in English infinitives., ms. University
of Connecticut, Storrs. [Available at: http://wurmbrand.uconn.
edu/Susi/Infinitives.html].
Wurzel, Wolfgang (1985) Die Suppletion bei den Dimensionsadjektiven. Studien
zur Morphologie und Phonologie A126: 114–143.
Yoshida, Yutaka (2009) Sogdian. In Windfuhr (2009), 279–335.
Young, Roberts W. and William Morgan (1987) The Navajo Language: A
Grammar and Colloquial Dictionary. Albuquerque: University of New
Mexico Press.
Zaicz, Gábor (1998) Mordva. In Abondolo (1998b), 184–218.
Zajkov, P. M. (1999) Grammatika karel’skogo jazyka (fonetika i morfologija).
Petrozavodsk: Periodika.
Zajtseva, Maria Ivanovna (1981) Grammatika vepsskogo jazyka (fonetika i
morfologia). Leningrad: Nauka.
527
Zakiev, Mirfatyx Zakievich (1966) Tatarskij jazyk. In Vinogradov et al.
(1966b), 139–154.
Zakiev, Mirfatyx Zakievich, Fuat Ashrafovich Ganiev, and Klara Zievna Zinnatullina,
eds. (1997) Tatarskaja grammatika, tom II: Morfologija. Kazan:
Tatar Academy of Sciences.
Zaliznyak, Andrey Anatolyevich (1977) Grammaticheskij slovar’ russkogo
jazyka. Moscow: Nauka.
Zeitoun, Elizabeth (2007) A Grammar of Mantauran (Rukai). Taipei: Acadmica
Sinica.
Zepeda, Ofelia (1983) A Papago Grammar. Tuscon, AZ, USA: The University
of Arizona Press.
Zhukova, Alevtina N. (1968a) Aljutorskij jazyk. In Vinogradov et al. (1968),
294–309.
Zhukova, Alevtina N. (1968b) Korjakskij jazyk. In Vinogradov et al. (1968),
271–293.
Zhukova, Alevtina N. (1972) Grammatika korjakskogo jazyka: fonetika, morfologija.
Nauka.
Zuñiga, Fernando (2006) Mapudungun: El habla mapuche. Santiago de Chile:
Centro de esturios públicos.
528