Decomposing derived collectives in West Slavic:
Experimental evidence from Czech and Polish
Introduction. Though the heterogeneous semantic nature of collective nouns has been
known for a long time and keeps posing a challenge for a proper treatment, it was commonly
assumed that collectives constitute a uniform category (e.g., Landman 1989, Barker 1992,
Schwarzschild 1996). However, recent findings suggest that there are different types of
such expressions (Pearson 2011, Henderson 2017). In this paper, we examine 3 classes of
derived collectives in Czech and Polish: i) group nouns, e.g., rytířcz / rycerzpl (‘knight’)
→ rytířstvocz / rycerstwopl (‘group/totality of knights’), ii) bunch numerals, e.g., třicz
/ trzypl (‘three’) → trojicecz / trójkapl (‘group of three’), and iii) aggregate nouns,
e.g., listcz / liśćpl (‘leaf’) → listí cz / listowiepl (‘foliage’). Though all 3 classes involve
collective inferences, they differ in a number of other properties, e.g., bunches are count
whereas groups and aggregates are not. Unlike other classes, groups seem to have a generic
flavor since they can combine with kind-level predicates. On the other hand, aggregates
constitute clusters, i.e., spatial groupings involving topology (Grimm 2012) whereas groups
and bunches do not seem to assert spatial configurations. Our aim is to investigate to
what degree different modes of group-formation relate to decomposability of particular
collective nouns, i.e., to what extent one can access atomic members of a group plurality
to distribute a certain property. In this regard, we focus on the interaction between
collectives and so-called A-different expressions such as jinýcz / innypl (‘different’). Such
expressions can be bound within a clause to express covariation with a plural argument
but unlike, e.g., English same or German verschieden (‘different’), they do not have a
built-in distributive operator and do not express covariation in the absence of a distributive
universal quantifier in subject position (Beck 2000, Dotlačil 2010), see (1).
(1) a. Each man is from a different town. → covariation (sentence-internal reading)
b. All the men are from a different town. → lack of covariation
Experiments. In order to test the interaction we designed analogous experiments on
Czech and Polish. Both were based on a Latin Square design involving 9 items in each of
the 3 classes, i.e., 27 items in total plus 27 fillers. The participants were asked to judge
whether a sentence is adequate in a context supporting a strong reciprocal scenario (a
truth value judgment task using a 5-point Likert scale: 1=worst, 5=best). We defined 3
conditions on each item: i) col : the target involved group, bunch, or aggregate,
ii) bp : a bare plural NP corresponding to col, and iii) qua : a universal distributive
quantifier with a singular bare NP corresponding to col. See (2) for an example of an
itemcz; itemspl were analogous. Given that qua was set as the reference level (nearly
total acceptability), we expected col and bp to be judged as significantly worse than
qua. The crucial questions concerned potential differences between col and bp as well
as between particular classes of collectives, i.e., group, bunch, and aggregate.
(2) Context: All the knights of the kingdom gathered to fight the final battle against
the musketeers. A historian who observed the battlefield realized that none of the
knights has the same armor as the others. He recorded the fact in a chronicle:
a. Rytířstvo
knightcoll
má
has
jinou
different
zbroj.
armor
‘(A group/totality of) Knights have different armor.’ col
b. Rytíři
knights
mají
have
jinou
different
zbroj.
armor
‘Knights have different armor.’ bp
c. Každý
every
rytíř
knight
má
has
jinou
different
zbroj.
armor
‘Every knight has different armor.’ qua
Both experiments were designed in Ibex and run online. There were 51cz and 48pl
participants all of whom were successful with the fillers and, thus, included in the analysis.
Results. Figure 1 represents the acceptability ratingscz of the 3 classes and their
conditions. Responses were modeled by the mixed-effects ordered probit regression
in the R package ordinal. The model had 1 predictor, namely the condition qua, and
included the subject and item slope+intercept random effects. The statistical outcome
is as follows. In the class bunch, bare plural NPs (βcz = −1.6289, zcz = −7.626, pcz <
0.001; βpl = −1.8408, zpl = −11.56, ppl < 0.001) and collectives (βcz = −1.6819, zcz =
−7.783, pcz < 0.001; βpl = −1.7601, zpl = −11.03, ppl < 0.001) were judged as worse than
quantifiers. There was no statistical difference between collectives and bare NPs in this
class (βcz = 0.05298, zcz = 0.267, pcz = 0.789; βpl = −0.0807, zpl = −0.578, ppl = 0.563).
In group, bare plurals (βcz = −2.1113, zcz = −10.70, pcz < 0.001; βpl = −2.2537, zpl =
−11.15, ppl < 0.001) and collectives (βcz = −2.8955, zcz = −13.56, pcz < 0.001; βpl =
−2.6726, zpl = −12.45, ppl < 0.001) were also judged as worse than Qs, however, bare plural
NPs (βcz = 0.7842, zcz = 4.831, pcz < 0.001; βpl = 0.4189, zpl = 2.486, ppl < 0.05) were
judged better than collectives. Similar, in aggregate bare plurals (βcz = −1.9916, zcz =
−8.166, pcz < 0.001; βpl = −2.3726, zpl = −13.58, ppl < 0.001) and collectives (βcz =
−2.7251, zcz = −10.731, pcz < 0.001; βpl = −2.7958, zpl = −15.02, ppl < 0.001) got
significantly worse results than Qs and again bare plurals ranked higher than collectives
on the scale (βcz = 0.7335, zcz = 3.45, pcz < 0.001; βpl = 0.4232, zpl = 3.054, ppl < 0.01).
Discussion. The results show that Czech and Polish bunch numeral phrases are easier
to decompose than group and aggregate collectives, i.e., despite shared collective
inferences there is an asymmetry in the accessibility to the members of a denoted plurality.
We attribute the difference to the fact that group and aggregate collectives are
semantically more complex, i.e., group-formation involves an additional mode such as
kind inference or mereotopology whereas bunch numerals simply establish a membership
relation between particular entities and a group they constitute. Though the ppl values
tend to be higher than pcz, the effect is almost identical in both languages. Therefore, our
findings suggest the existence of a scale of decomposability which holds in West Slavic:
(3) group ≈ aggregate < bunch ≈ bare plural NP < distributive QP
Selected References. Barker
(1992) Group terms in English: Representing
groups as atoms, JoS 9 •
Beck (2000) The semantics of different:
Comparison operator and relational
adjective, L&P 23 • Dotlačil
(2010) Anaphora and Distributivity •
Grimm (2012) Number and Individuation
• Henderson (2017) Swarms: Spatiotemporal
grouping across domains,
NLLT 35 • Landman (1989) Groups,
L&P 12 • Pearson (2011) A new semantics
for group nouns, WCCFL 28
• Schwarzschild (1996) Pluralities