Introducing Linguistics This outstanding series is an indispensable resource for students and teachers - a concise and engaging introduction to the central subjects of contemporary linguistics. Presupposing no prior knowledge on the part of the reader, each volume sets out the fundamental skills and knowledge of the field, and so provides the ideal educational platform for further study in linguistics. 1 Andrew Spencer 2 John I. Saeed 3 Barbara Johnstone 4 Andrew Cainie Phonology Semantics, Third Edition Discourse Analysis, Second Edition Syntax, Second Edition Semantics Third Edition John I. Saeed OoJ WILEY-BLAK3CWELL AJohn Wiley SiSons.l.td'., Publication •_____i 116 Semantic Description 11 Since this relation is clearly similar to the biconditional connective described earlier, we could give a logical definition of synonymy as in: p and q are synonymous when the expression p a 5.12 The amendment remains in force. H 5.13 Jenny loves to ski. We will say that adjectives and stative verbs are inherently static, i.e. that it is part of their lexical semantics to portray a static situation type. We have already briefly mentioned the dimension of tense. As we will describe in section 5.2.4 many languages have grammatical forms, such as verb endings, which allow a speaker to locate a situation in time relative to the 'now- of the act of speaking or writing. Aspect is also a grammatical system relating to time, but here the speaker may choose how to describe the internal temporal nature of a situation. If the situation, is in the past, for example: does the speaker portray it as a closed completed event, as in 5.14 below, or as an ongoing process, perhaps unfinished, as in 5.15? 5.14 David wrote a crime novel. 5.15 David was writing a crime novel. This is a difference of aspect, usually marked, as with tense, by grammatical devices. Tense and aspect áre discussed in sections 5.2.4-5 and we discuss 120 Semantic Description the problems of comparing the aspectual systems of different languages in 5.2.6 Finally section 5,2.7 is a brief look at how these dimensions combine to allow speakers to portray different situations. 5.2.2 Verbs and situation types We saw in the last section that certain lexical categories, in particular verbs, inherently describe different situation types. Some describe states, others are dynamic and describe processes and events. In this section wire describe elements of the.meaning of verbs which correlate to differences of situation type. Stative verbs In the last section we saw examples of inherently stative verbs like be, have, know and fcw.These verbs allow the speaker to view a situation as a steady state, with no internal phases or changes. Moreover the speaker does not overtly focus on the beginning or end of the state. Even if the speaker uses a stative in the past, e.g. 5.16 Mary loved to drive sports cars. no attention is directed to the end of the state. We do not know from 5.16 if or how the state ended: whether Mary's tastes changed, or she herself is no longer around. All we are told is that the relationship described between Mary and sports cars existed for a while. We can contrast this with a sentence like 5.17 below, containing a dynamic verb like learn: 5.17 Mary learned to drive sports cars. Here the speaker is describing a process and focusing on the end point: at the beginning Mary didn't know how to drive sports cars, and at the end she has learnt. The process has a conclusion. Stative verbs display some grammatical differences from dynamic verbs. For example in English progressive forms can be used of dynamic situations like 5.18a below but not states like 5.18b: 5.18 a. I am learning Swahili. b. *I am knowing Swahili. As noted by Vlach (1981) this is because the progressive aspect, marked by -mg above, has connotations of dynamism and change "which suit an activity like learn but are incompatible with a stative verb like know. We discuss the English progressive in sections 5.2.5-6 below. Similarly it usually sounds odd to use the imperative with statives; we can compare the following: • Sentence Semantics 1: Situations 121 5.19 a. Learn Swahili! b. ?Know Swahili! Once again, we can speculate that imperatives imply action and dynamism, and are therefore incompatible with stative verbs. It may be however that the distinction between state and dynamic situations i& not always as clear-cut. Some verbs may be more strongly stative than others; remain for example, patterns like other stative verbs in not taking the progressive, as in 5.20b below, but it does allow the imperative, as in 5.20c: 5.20 a. The answer remains the same: no! b. *The answer is remaining the same: no! c. Remain at your posts! It is important too to remember that verbs may have a range of meanings, '• some of which may be more stative than others. We can contrast the stative and non-stative uses of have, for example, by looking at how they interact with the progressive:2 5.21 a. I have a car. b. *I am having a car. c. I am having second thoughts about this. 5-22 a. She has a sister in New York. b. *She is having a sister in New York c. She is having a baby. .Dynamic verbs Dynamic verbs can be classified into a number of types, based on the semantic distinctions, durative/punctual and tellc/atelic which we will discuss below. These different verb types correlate to different dynamic situation typ«. One possible distinction within dynamic situation types, for . example, is between events and processes. In events, the speaker views the situation as a whole, e.g. 5.23 The mine blew up. while in a process, we view, as it were, the internal structure of a dynamic situation, e.g. 5.24 He walked to the shop. Processes can be subdivided into several types, for example inchoatives and resultatives. Inchoatives are processes where our attention is directed to the beginning of a new state, or to a change of state, e.g. 122 Semantic Description Sentence Semantics 1: Situations 123 5.25 The ice melted. 5.26 My hair turned grey. Resultatives are processes which are viewed as having a final point of completion: our attention is directed to this end of the process, e.g.: 5.27 Ardal baked a cake. 5.28 Joan built a yacht. One difference between these types concerns interruption. If the action of melting is interrupted in 5.25 or my hair stops turning grey in 5.26, the actions of melting' and turning grey can still be true descriptions of what went on. However if Ardal in 5.27 and Joan in 5.28 are interrupted halfway, then it is no longer true to describe them as having baked a cake or built a yacht. In some sense, to use resultatives we have to describe a successful conclusion. In this section we look at two important semantic distinctions in verbs which underlie these different dynamic situation types. The first distinction is between durative and punctual: dux-alive is applied to verbs which describe a situation or process which lasts for a period of time, while punctual describes an event that seems so instantaneous that it involves virtually no time. A typical comparison would be between the punctual 5.29 and the durative 5.30: 5.29 John coughed. 5.30 John slept. What matters of course is not how much time an actual cough takes, but that the typical cough is so short that conventionally speakers dp not focus on the internal structure of the event. In Slavic linguistics the equivalent of verbs like cough are called semelfactive verbs, after the Latin word semel, 'once'. This term is adopted for general use by C. S. Smith (1991), Verkuyt (1993) and other writers. Other semelfactive verbs in English would include flask, shoot, knock, sneeze and blink. One interesting fact is that in English a clash between a semelfactive verb and a durative adverbial can trigger an iterative interpretation, i.e. where the event is assumed to be repeated for the period described, e.g. 5-31 Fred coughed all night. 5.32 The drunk knocked for ten minutes. 4 5.33 The cursor flashed until the battery ran down. .} In each of these examples the action is interpreted as being iterative: 5.31 jT is not understood to mean that Fred spent all night uttering a single drawn-out cough! The second distinction is between telle and atelic. Telle refers to those processes which are seen as having a natural completion. Compare for example: §>, 5.34 a. Harry was building a raft. b. Harry was gazing at the sea. ■'gp If we interrupt these processes at any point then we can correctly say: 5.35 Harry gazed at the sea. but we cannot necessarily say: 5.36 Harry built a raft. As we saw earlier, telic verbs are also sometimes called resultatives. Another way of looking at this distinction is to say that gaze being atelic can continue indefinitely, while build has an implied boundary when the process will be over. Alternative terms are bounded for telic and unbounded for atelic. It is important to recognize that while verbs may be inherently telic or atelic, combining them with other elements in a sentence can result in a different aspect for the whole, as below: 5.37 a. Fred was running, (atelic) b. Fred was running in the London Marathon, (telic) 5.38 a. Harry was singing songs, (atelic) b. Harry was singing a song, (telic) This telic/atelic distinction interacts with aspectual distinctions: for example a combination of either the English perfect or simple past with a telic verb will produce an implication of completion. Thus, as we have seen, both 5.39 and 5.40 entail 5.41: 5.39 Mary painted my portrait. 5.40 Mary has painted my portrait. 5.41 ■ The portrait is finished. However, the combination of a progressive aspect and a telic verb, as in 5.42 below does not produce this implication: 5.42 does not entail 5.41 above: 124 Semantic Description Sentence Semantics 1: Situations 125 5.42 Mary was painting my portrait. Comrie (1976) gives examples of derivational processes which can create telic verbs from atelic verbs, e.g. the German pairs in 5.43: 5.43 a. essen 'eat?, aufessen 'eat up' b. kämpfen 'fight', erkämpfen 'achieve by fighting* He contrasts the following sentences: 5.44 a. die Partisanen haben fur die Freiheit ihres Landes gekämpft, b. die Partisanen haben die Freiheit ihres Landes erkämpft. The partisans have fought for the freedom of their country.' (Comrie 1976: 46-7) where 5.44b implies that their fight was successful while 5.44a does not. 5.2.3 A system of situation types Speakers use their knowledge of these semantic distinctions - Stative/ dynamic, durative/punctual, telic/atelic - to draw distinctions of situation type. We have seen that some verbs, like paint, draw and buäd, are inherently telic while others like talk, sleep and walk are atelic. Similarly some verbs are inherently Stative like know, love and resemble, while others like learn, die and kill are non-stative.We have also seen from examples like 5.37 and 5.38 above that while these distinctions are principally associated with verbs, combining a verb with other elements in a sentence, like object noun phrases and adverbials, can alter the situation type depicted. The task for the semanticist is to show how the inherent semantic distinctions carried by verbs, and verb phrases, map into a system of situation types. One influential attempt to do this is Vendler (1967). Below are the four kinds of situations he identified, together with some English verbs and verb phrases exemplifying each .type (Vendler 1967: 97-121): 5.45 a. States desire, want, love, hate, know, believe b. Activities (unbounded processes) run, walk, swim, push a can, drive a car c. Accomplishments (bounded processes) run a mile, draw a circle, walk to school, paint a picture, grow up, deliver a sermon, recover from illness d. Achievements (point events) recognize, find, stop, start, reach the top, win the race, spot someone C. S. Smith (1991), building onVendler's system, adds the situation type semelfactive, distinguishing it from achievements as follows: 5.46 Semelfactives are instantaneous atelic events, e.g. [knock], [cough]. Achievements are instantaneous changes of states, with an outcome of a new state, e.g. [reach the top], [win a race]. (Smith 1991: 28) She identifies three semantic categories or features: [stative], [telic] and [duration], with roughly the characteristics we have already described, and uses these to classify five situation types, as follows (1991: 30): 5.47 Situations States Activity Accomplishment Semelfactive Achievement Static +1 Durative w w w [-]■ H Telic n.a. 'H W H W We can provide examples of each situation type, as follows: 5.48 She hated ice cream. (State) 5.49 Your cat watched those birds. (Activity) 5.50 Her boss learned Japanese. (Accomplishment) 5.51 The gate banged. (Semelfactive) 5.52 The cease-fire began at noon yesterday. (Achievement) It is important to remember that these situation types are interpretations of real situations. Some real situations may be conventionally associated with a situation type; for example it seems unlikely that the event described in 5.53 below would be viewed other than as an accomplishment: 5.53 Sean knitted this sweater. Other situations are more open, though: 5.54 and 5.55 below might be used of the same real-world situation, but give two different interpretations of it: 5.54 as an activity and 5.55 as a state: 5.54 Sean was sleeping. 5.55 Sean was asleep. 5.2.4 Tense Tense and aspect systems both allow speakers to relate situations to time, but they offer different slants on time. Tense allows a speaker to locate a 126 Semantic Description situation relative to some reference point in time, most likely the time of speaking. Sometimes in English this information is given by a temporal adverb; compare the following: 5.56 Yesterday they cut the grass. 5.57 Tomorrow they cut the grass. Here, because the shape of the verb cut does not change, the temporal information is given by the adverbs yesterday and tomorrow. Usually in English, though, tense is marked on the verb by endings and the use of special auxiliary verbs, as in the forms of speak below: 5.58 She spoke to me. 5.59 She will speak to me. 5.60 She is speaking to me. Tense is said to be a deictic system, since the reference point for the system is usually the act of speaking. As we shall see in chapter 7, deictic systems are the ways in which a speaker relates references to space and time to the 'here and now' of the utterance. Most grammatical tense systems allow the speaker to describe situations as prior to, concurrent with or following the act of speaking. So in English we have the three tenses: past, future and present as in 5.58-60 above. These are basic tenses and we could use a diagram like figure 5.1 to represent them, metaphorically representing time as a line moving left to right, and using the clock symbol for the time of the act of speaking. More complicated time references are possible. For example the speaker can locate an event in the past or future and use that event as the reference point for its own past, present and future. To do this in English complex tenses are used. If a speaker in 1945 said, for example: 5.61 By 1939 my father had seen several arrests. Figure 5.1 Simple tenses past present -©— future act of speaking will iee Sentence Semantics 1: Situations 127 Figure 5.2 Complex past tense past ^ Secondary past !__t_ present future past event act of speaking had seen the verb had seen is one of these complex tenses, called the past perfect or pluperfect The year 1939 is in the past of the utterance of course, but the .speaker has made it the anchoring point for its own past. The father's acts of seeing are marked as being in this secondary past, as well as in the past relative to the act of speaking. Again we could represent this in a simple diagram as in figure 5.2. Complex future tenses like will have seen allow a similar creation of a past-of-a-future-event, as in an utterance now of 5.62: 5 62 By 2050 we will have experienced at least two major earthquakes. Here of course the earthquakes are portrayed as in the past relative to 2050, but in the future relative to the act of speaking. Since tense is a deictic system it may vary from language to language. Some languages, like the Bantu language Chibemba (Sharman 1956, Giv6n 1972) have more complicated systems of divisions than English: 5.63 Chibemba past tense system (Givon 1972) a. Remote past: Ba-ali-bomb-ele "They worked (before yesterday)' b. Removed past: Ba-alti-bomba 'They worked (yesterday)' c. Near past: Ba-ad-bomba They worked (earlier today)'- d. Immediate past: Ba-a-bomba 'They worked (in the past few hours)' 5.64 Chibemba future tense system a. Immediate future: Ba-dM-bomba "They'll work (in the next few hours)' b. Near future: Ba-Jei-bomba 'They'll work (later today)' c. Removed future: Ba-ka-bomba "They'll work (tomorrow)' 128 Semantic Description Sentence Semantics 1: Situations 129 d. Remote future: Ba-ká-bomba "They'll work (after tomorrow)* Here we see four degrees of remoteness from the act of speaking (Givón 2001): a few hours from now; within today; within the day adjacent to today; and beyond the day adjacent to today. Each of these projects backwards into the past and forwards into the future. Since this system includes not only intervals relative to the act of speaking but an implied measurement of the intervals) it is termed a metrical tense system by Chung and Timberlake (1985: 207). An influential system of representing the deictic nature of tense is Reichenbach's (1947) reference point theory of tense which, as shown in (5.65)j identifies three reference points in time: 5.65 Reichenbach's (1947: 290) tense reference points: S = the speech point, the time of utterance; R = the reference point, the viewpoint or psychological vantage point adopted by the speaker; E = event point, the described action's location in time. Tenses are then defined by three ordering relations between these points: at the same time (=); before (x < y); and after (x < y). Crucial.to the identification of tense are the relations (1) between reference time and speech time, and (2) between event and reference time. We can show this with the examples in (5.66—8): 5.66 5.67 5.68 'I saw Helen' (R = Eclause in sentences like 5.135-6, the condition, and the other clause, the consequent. This view of conditionals as part of the modal system neatly explains why we also find modal verbs used in consequent clauses, like would in 5.135-6 above, or should in the condition clauses below: 5.137 If you should go to Paris, stay near the river. 5.138 Should you meet Christy, there's something 1 would like you to ask him. This approach to modality is also supported by the existence of languages which have verb forms which regularly distinguish between events in the real world and events in future or imaginary worlds. This two-term modal distinction is often called a realis/irrealis modality (i.e. a reality/unreality distinction): for example, Palmer (1986: 47) describes a distinction between realis and irrealis moods in the Australian language Ngiyambaa: 5.139 a. yururj-gu rjidja-r,a.7 rain-ERG rain-PRES. 'It is raining.' (realis) b. yururj-gu rjidja-l-aga. rain-BRG rain-CM-lRREALIS 'It might/will rain.' (irrealis) In this section we have looked briefly at the semantic system of modality; in the next we look at how modality distinctions are encoded in the gram-p> mar, in particular, at mood. 5.3.2 Mood Thus far we have seen modality distinctions in English being marked by various means including adverbs and modal verbs. When such distinctions 9k 142 Semantic Description are marked by verb endings which form distinct conjugations, there is a grammatical tradition of calling these moods. Thus the distinction in the Ngiyambaa verb in 5.139 would be described as a distinction between a realis mood and an irrealis mood. In the verbal inflection of the Cushitic language Somali we find'in addition to the basic indicative mood in 5.140 a conditional mood, as in 5.141, and a potential mood as in 5.142: 5.140 Wuu sameeyey. he make.PAST 'He made it.' 5.141 Wuu sameyn lahaa. he make.mFiNmvE have 'He would make it, he would have made it.' 5.142 Show sameyee. possibly tnake.PorENTTAL 'Maybe he'll make it, it's possible he will make it.' The indicative in 5.140, which is a realis form, and the potential in 5.142 are marked by specific verb endings, while the conditional in 5.141 uses an the infinitive with an auxiliary verb 'have', rather like English.8 A more familiar example of mood is the subjunctive mood found in many European languages. The label subjunctive is applied somewhat differently in different languages, but we can identify two- opposite poles of use, with an area of misting and overlap between them. One pole is the grammatical one of syntactic subordination, i.e. subjunctive verb forms show that a verb is in a subordinate clause. The other pole is semantic, where the subjunctive marks language-specific types of irrealis mood, and is thus used for wishes, beliefs, exhortations, commands etc. At die syntactic pole, we can cite the example of Somali again where subordinated clause verbs are always differentiated from their main clause equivalents by a combination of tone and endings; compare 5.143 and 5.144 below: 5.143 Lacagta way lacag-ta waa-ay keetiaysaa . money-the CLASS-she bringjPROGRESsrvE 'She is bringing the money.' 5.144 inay lacagta keenayso in-ay lacag-ta keenayso that-she money-the bring, subjuncitve 'that she is bringing the money* In 5.143 the classifier waa identifies a main clause, while in 5.144 the complementizer m 'that' identifies a subordinate clause. As is clear, the main Sentence Semantics 1: Situations 143 clause and subordinate clause forms of the verb keen 'bring' have different ||. itonal shapes and a different endings.10 If such subordinate verb forms are termed 'subjunctive', then this use of |}v the term does not seem to have anything to do with the semantic system of ' :': modality. However in classical Greek and in Latin, the subjunctive describes ."•a verbal form that occurs in both main and subordinate clauses, though .with somewhat different applications in each. Palmer (1986: 39-43), citing :R. T. Lakoff (1968), gives six meanings of the subjunctive in Latin main : clauses: imperative, optative (for wishes), jussive, concessive, potential and .deliberative. Each of these can be identified with descriptions of unreal situations, and thus be examples of our semantic pole of unreality. They contrast \ with the indicative mood used for descriptions of factual, or real, situations. • ■, In-between positions are very common, especially in modern European languages. In many languages, the subjunctive is most commonly found in subordinate clauses, but often with some special meaning: oftea following verbs of wishing and preference, as in the Spanish example 5.145 below (Butt and Benjamin 1994: 246) and the French 5.146; for the future in Spanish 5.147 (Butt and Benjamin 1994: 241); or indirect speech as in German 5.148 (Hammer 1991: 310): 5.145 Quiero que estudies mas. want.iNDic.PRES.lsg that study.suBjUN.PRBS.2sg more 'I want you to study more.' 5.146 II vaut mieux qu'elle le sache. it worth better that+she it knDW.sUBjUM.PRBS.3sg 'It's better that she know it.' 5.147 Iremos alU cuandohaga buen tiempo . go.iNDic.FUT.lp there when have.suBjUN.PRBS.3sg good weather 'We'll go there when the weather's good.' 5.148 Sie sagte sie schreibe den Brief, she said she write.suBjuN.iMPBRF.3sg the letter 'She said she was writing the letter.' While there seems to be some shared element of modality in these uses, i.e. of non-factuality,11 the range of use of subjunctives is usually both complex and language specific. Often the choice between indicative and subjunctive moods allows speakers to make subtle semantic distinctions, as for example between the different degrees of possibility marked by the French indicative and subjunctive in 5.149 and 5.150 below (Judge and Healey 1985: 141): 5.149 Je pense qu'il viendra. I think.iNDlc.PKES that-he come.rNDlc.FUT 'I think that he'll come.' 144 Semantic Description Sentence Semantics 1: Situations 145 5.150 Je doute qu'il vienne. I doubt.iNDic.PRES that-he come.suBjUN.PRES 1 doubt that he'll come.' Before we close this section on mood, we should point out that there is another quite distinct use of the term in semantics. This applies to changes in verbal morphology associated with the different social functions or speech acts that a speaker may intend. For example a speaker may intend a sentence as a statementj a question, a command or a wish. Depending on the language, these different functions may be marked by different word orders or special intonation tunes. Some languages mark this information by particular verb forms: for example, some languages have special optative verb conjugations to express wishes like the English phrases 'may he get well', 'I hope he gets well', 'if only he would get well', etc. See for example the Nahuatl sentence (Bybee 1985: 171): 5.151 ma choca. 'If only he would weep.' Such special speech act verbal forms are often called moods: the example above would therefore be in the optative mood, and in some languages this would contrast with an imperative mood (for commands), an interrogative mood (for questions) or a declarative mood (for statements). We will discuss this grammaticalization of speech functions in chapter 8 on speech acts. See Foley and Van Valin (1984) for discussion of the relationship between this use of mood and the epistemic and deontic modality we have been concerned with here. 5.3.3 Evidentiality Under epistemic modality we looked at ways in which a speaker can mark different attitudes towards the factuality of a proposition. There is a further semantic category evidentiality which allows a speaker to communicate her attitude to the source of her information. This is possible in English of course by the use of a separate clause or by parenthetic adverbials. Compare the bare assertion in 5.152 with the various evidentially qualified versions in 5.153a-g: 5.152 She was rich. 5.153 a. I saw that she was rich. b. I read that she was rich. c. She was rich, so they say. d. I'm told she was rich. e. Apparently she was rich. f. She was rich, it seems. g. Allegedly, she was rich. TĚ,*, These qualifications allow the speaker to say whether the statement relies on personal first-hand knowledge, or was acquired from another source; and if the latter, perhaps to say something of the source. Some languages routinely mark such information grammatically, by special particles or specific verb forms, so that in these languages evidentiality is coded in the morphology. A collection of descriptions of such languages is Chafe and Nichols (1986), which contains articles both on the North and South American languages where such systems were first described and also on evidential systems in European and Asian languages. Aikhenvald (2004) provides a comparative overview of such evidential systems. We can také as an example Tariana, an Arawak language spoken in northern Amazonia, whose verbal morphology distinguishes several different sources for information (Aikhenvald 2004: 2-3): 5.154 a. Juse irida di-manika-ka José football 2sgnf-play-EEC.P.vis 'José has played football (we saw it)' b. Juse irida di-manika-m nhfca José football 2sgnf-play-KEC.P.NONVis 'José has played football (we heard it)' c. Juse irida di-manika-nihka José football 2sgnf-play-KECP.lNFR 'José has played football (we infer it from visual evidence)' d. Juse irida di-manika-sika José football 2sgnf-play-REC.p.ASSUM 'José has played football (we assume this on the basis of what we already know)' e. Juse irida di-manika-pidaka José football 2sgnf-play-KEC.P.KBP 'José has played football (we were told)' We follow Aikhenvald in marking the evidential morphemes in bold, giving us the five-fold evidential distinction between these reports of a recent past event. In a the speaker has seen the event; in b the speaker heard the noise of the football game; in c the report is an inference from visual evidence12; in d the assumption is based on previous knowledge about Jose's habits; and finally in e, the speaker has learned the information from someone else. What emerge from these studies of evidential systems are differences among languages in whether the evidential markers are obligatory in ordinary speech or an optional resource for speakers. Hardman, for example, reports that among the Jaqi languages of Peru, Bolivia and Chile the identification of what she calls 'data source' (i.e. the use of evidentials) is a central part of knowing how to communicate (1986: 114): 5.155 Accuracy on the part of the speaker is a crucial element in the public reputation of individuals; misuse of data-source is somehow somewhat less than human, or is insulting to the listener. 146 Semantic Description Sentence Semantics 1: Situations 147 Speakers of Jaqi languages, which include Jaqaru, Aymara and Kakwi, have obligatorily to signal whether the source of information for their statements is personal experience, or knowledge gained from other individuals by language, or comes from the remote past where no witnesses are available, i.e. from myths, history and religion. In other languages the use of evidentials is more voluntary, providing a speaker with creative resources to structure a point of view in a discourse, or perhaps to argue more convincingly. See Chafe (1986) for a description of evidentials in English. ' 5.4 Summary In this chapter we looked at aspects of sentence meaning which allow the speaker to classify situations. The category of situation type, for example, incorporating semantic distinctions like static/dynamic, durative/punc-tual and teUc/atelic, allows a basic classification of situations into states, activities, accomplishments, etc. The categories of tense and aspect interact with situation type to allow a speaker to relate a situation to time in two ways: to locate it relative to the act of speaking, and to portray its internal temporal shape. We saw something of how these choices are reflected in grammar. We also saw that the distinctions available to speakers may be very subtle and language specific. We also looked at the semantic categories of modality and evidentiality, which allow the speaker to assume various attitudes towards a proposition. Epistemic modality reflects various judgements of factuality and deontic modality communicates judgements of moral and legal obligation. Both can be seen as implying a comparison between the real world and hypothetical versions of it. Evldentiality is a term for the ways in which a speaker qualifies a statement by referring to the source of the information. We saw that in some languages this information is grammaticalized and therefore obligatory, implying that in these communities, calculation of evidence is assumed of speakers by their hearers. We look at the role of similar hearer assumptions, e.g. that the speaker is estimating and updating her audience's state of knowledge, in chapter 7. FURTHER READING Comrie's Aspect (1976) and Tense (1985) are concise monographs, using examples from a range of languages. C. S. Smith (1991) discusses universals of situation type and aspect and gives brief descriptions of the aspectual systems of English, French, Russian, Mandarin Chinese and Navajo. Palmer (1986) and Bybee and Fleischman (1995) contain discussions of modality systems in various languages. Bybee, Perkins and Pagliuca (1994) contains a large cross-linguistic survey of tense, aspect and modality. The marking of these semantic categories on the English verb can be seen |j in Leech (1971) and the comprehensive reference grammar Quirk et al. (1985). | Aikhenvald (2004) provides a survey of evidential systems in a wide range of languages. > ft -Jilt,*. ? A fl* ''^-W* FdX u deikt-i fff imi huActnHhi. plus*1* llH1W l *" „ if 150 Semantic Description iiiiPiili i h d t U 1/ Wmms msm. 11 i i g ■> (hin*, tb läeftlí 1 f UjoUk n oi ď t-pc o tr i til ke K ng h. i fl T tbr r ti fr pv ŕ v i U iwilr^i fl n,i di oliÉgediy j * NOTES 1 Transcription as in the original, where tone is marked as follows: a (macron) = high level tone, 6 = rising; 6 = fall-rise, 6 = falling. 2 See Dowry (1979) for a discussion of stativiry and English verbs, especially verbs like sit and stand, which act like statives in many ways but allow progressive forms. 3 See also Ogihara (1989). 4 Note that our translations here are meant to be suggestive: in fact, as my colleague Sarah Smyth has pointed out to me, the contrast between the English past progressive and past simple doesn't exactly capture the Russian distinction between imperfective and perfective. Thus 5.97 can also mean He read a letter or He has read a letter. The perfective form in 5.98 is more likely to mean He read a letter (and then threw it away) for perfective verbs in Russian suggest continuation of narrative. Sentence Semantics 1: Situations 151 10 12 The French imparfait does not of course correspond to the Russian imperfective: for example, the French perfective Tuasvuce film? would be translated into Russian as an imperfective 7y videl ewt fil'm?. We discuss this notion of possible worlds in chapter 10. In this transcription cm = 'conjugation marker", erg = ergative case. We have glossed shorn in 5.141 as 'possibly* but in fact it is a sentence type indicator, or classifier, which can only be used with verbs in the potential mood. See Saeed (1993) for more details, and chapter 8, section 8.5, where we discuss these classifiers in Somali and their status as sentence type markers. The tone markings used here are á = high tone, and a (i.e. unmarked) =• low tone. They are only marked on the first vowel of long vowels, e.g. ée. Note that such subordinate clause verbs are finite, showing inflectional marking of person, tense and aspect. Another way of viewing what these uses of the subjunctive have in common comes from the modality of speech acts, to be discussed in chapter 8. This to recognize a common element of non-assertion in these clauses. Aikhenvald gives a possible licensing context as follows: 'If one see that the football is not in its usual place in the house, and José and his football boots are gone, with crowds af people coming back from the football ground, this is enough for us to infer that José is playing football' (2004: 2). (0 v. 0) HH Q. (0 0 Sentence Semantics 2: Participants 6.1 Introduction: Classifying Participants In the last chapter we looked at aspects of sentence level semantics: how speakers may choose to characterize situations and express various degrees of commitment to the portrayal. Another set of semantic choices which face a speaker seeking to describe a situation concerns how to portray the roles of any entities involved. Take for example 6.1 below: 6.1 Gina raised the car with a jack. This sentence identifies three entities, Gina, the car and a jack, related by the action described by the verb raise. The sentence portrays these entities in specific roles: Gina is the entity responsible for initiating and carrying out the action, the car is acted upon and has its position changed by the action, and the jack is the means by which Gina is able to cause the action. Such roles have a number of labels in semantics, including participant roles (Allan 1986), deep semantic cases (Fillmore 1968), semantic roles (Ghron 1990), thematic relations (Gruber 1976Jackendoff 1972) and thematic roles (Dowty 1986, 1989, 1991, Jackendoff 1990). Given its wide usage in recent work, we will use the last term here: thematic roles. im :;f| Sentence Semantics 2: Participants 153 In this chapter we examine this notion of thematic roles. We begin by sketching the basic picture of these roles that seems to be assumed by much of the syntax and semantics literature. Thus in sections 6.2-6.4 we outline the main contenders for individual types of roles, look at the relationship between thematic roles and grammatical relations, and discuss the idea that verbs must have their thematic role requirements listed in the lexicon. In the second part of the chapter we look more critically at the idea of thematic roles: first in section 6.5 we review criticisms that have been levelled at the notion. Then in 6.6 we review the job these roles do in linguistic description. In the third part of the chapter, section 6.7, we investigate voice systems and see how they allow speakers some flexibility in the relationship between thematic roles and grammatical structure: we focus on passive voice and middle voice. In the final part of the chapter we turn our attention to semantic classification systems that are based on the inherent features of nominals rather than their roles within a predication. In section 6.8.1 we discuss classifiers and in 6.8.2 noun classes. W 6.2 Thematic Roles Each of the writers mentioned above, and others, for example Andrews (1985) and Radford (1988), have proposed lists of thematic roles. From this extensive literature we can extract a list of thematic roles like the following (where the relevant role-bearing nominal is in bold): agent: the initiator of some action, capable of acting with volition, e.g. 6.2 David cooked the rashers. 6.3 The fox jumped out of the ditch. patient: the entity undergoing the effect of some action, often undergoing some change in state, e.g. 6.4 Enda cut back these bushes. 6.5 The sun melted the ice. theme: the entity which is moved by an action, or whose location is described, e.g. 6.6 Roberto passed the ball wide. 6.7 The book is in the library. 154 Semantic Description bxpekiencbr: the entity which is aware of the action or state described by the predicate but which is not in control of the action or state, e.g. 6.8 Kevin felt ill. 6.9 Mary saw the smoke. 6.10 Lorcan heard the door shut. beneficiary: the entity for whose benefit the action was performed, e.g. 6.11 Robert filled in the form for his grandmother. 6.12 They baked me a cake. instrument: the means by which an action is performed or something comes about, e.g. 6.13 She cleaned the wound with an antiseptic wipe. 6.14 They signed the treaty with the same pen. location: the place in which something is situated or takes place, e.g. 6.15 The monster was hiding under the bed. 6.16 The band played in a marquee. goal: the entity towards which something moves, either literally as in 6.17 or metaphorically as in 6.18: 6.17 Sheila handed her licence to the policeman. 6.18 Pat told the joke to his friends. source: the entity from which something moves, either literally as in 6.19 or metaphorically as in 6.20: 6.19 The plane came back from Kinshasa. 6.20 We got the idea from a French magazine. stimulus: the entity causing an effect (usually psychological) in the experiences, e.g. 6.21 John didn't like the cool breeze. 6.22 The noise frightened the passengers. Thus to return to our first example, repeated below: Sentence Semantics 2: Participants 6.23 Gina raised the car with a jack. 155 we can describe the thematic roles by calling Gina the agent of the action, the car the theme, and the jack the instrument. There is some variation in the use of these terms: for example Radford (1988) treats patient and theme as different names for the same role. Here we adopt the distinction that patient is reserved for entities acted upon and changed by the verb's action while theme describes an entity moved in literal or figurative space by the action of the verb, but constitutionally unchanged. Thus the noun phrase the rock would be a patient in 6.24 below but a theme in 6.25 6.24 Fred shattered the rock. 6.25 Fred threw the rock. A number of tests for identifying thematic roles have been suggested. Jackendoff (1972) for example provides a test for agent: whether the phrases like deliberately, on purpose, in order to, etc. can be added to the sentence. This reflects the fact that an agent characteristically displays animacy and volition. The contrast between 6.26 and 6.27 below identifies John as an agent in 6.25 but not 6.27: 6.26 John took the book from Bill in order to read it. 6.27 ?John received the book from Bill in order to read it. Some writers (e.g. Foley and Van Valin 1984, Jackendoff 1990) have suggested that agent is a particular type of a more general thematic role actor, where actor 'expresses the participant which performs, effects, instigates, or controls the situation denoted by the predicate' (Foley and Van Valin 1984: 29). So every agent is an actor, but not the other way round: in 6.28 below the car is an actor but not agent since it presumably is neither in possession of a wish to kill nor animate: 6.28 The car ran over the hedgehog. Other simple tests suggested by Jackendoff (1990) include predicting that for an actor (X) it will make sense to ask 6.29 below, and for a patient (Y) that it will be able to occur in the frames in 6.30: 6.29 What did X do? 6.30 a. What happened to Y was... b. What X did to Y was... 156 Semantic Description So for example 6.31 below the tests would give 6.32-3, identifying Robert as the actor and the golf dub as patient: 6.31 Robert snapped the golf club in half. 6.32 What Robert did was to snap the golf club in half. 6.33 a. What happened to the golf club was that Robert snapped it in half. b. What Robert did to the golf club was snap it in half. . Some writers have suggested other thematic roles in addition to those we have discussed. For example a role of force is sometimes used instead of instrument for an inanimate entity which causes something, e.g. 6.34 a. The wind flattened the crops. b. The sea wall was weakened by the waves. A role of recipient is sometimes identified, e.g. by Andrews (1985), as a type of goal involved in actions describing changes of possession, e.g. 6.35 a. He sold me this wreck. b. He'left his fortune to the church. While these roles, actor, agent, patient, experiencer, theme, instrument etc. may seem intuitively clear, in practice it is sometimes difficult to know which role to assign to a particular noun phrase. For example, in a sentence like 6.36 below to the lightkouse is clearly a goal, and in 6.37 him is a beneficiary, but in 6.38 below is Margarita the GOAl/feeciPiENT, or the beneficiary, or both? 6.36 Fergus carried the bag to the lighthouse. 6.37 Sylvie bought him a sports car. 6.38 Margarita received a gift of flowers. Examples like these raise the difficult question of whether a single entity can fulfil two or more thematic roles at the same time; for example in 6.39 below, are we to say that Mr Wheeler is both agent and theme? 6.39 Mr Wheeler jumped off the cliff. These issues are still under investigation in various theoretical approaches. A central claim of Chomsky's Principles and Parameters theory, for example, is the Theta-Criterion, which states that there must be a one-to-one Sentence Semantics 2: Participants 157 correspondence between noun phrases and thematic roles (see Chomsky 1988, Haegeman 1994). Jackendoff (1972), on the other hand, suggested that one entity might fulfil more than one role. In Jackendoff (1990) the idea that one nominal might fulfil more than one role is elaborated into a theory of tiers of thematic roles: a thematic tier, which describes spatial relations, and an action tier which describes actor-patient type relations. His examples include the following (1990: 126-7): 6.40 a. Sue hit Fred. Theme Goal (thematic tier) Actor Patient (action tier) b. Pete threw the ball. Source - Theme (thematic tier) Actor Patient (action tier) c. Bill entered the room. Theme Goal (thematic tier) Actor (action tier) d. Bill received a letter. Goal Theme (thematic tier) (action tier) Thus Fred in 6.40a is simultaneously the goal and the patient of the action. The gaps in a tier reflect instances where the nominal has only one thematic role: thus the room in 6.40c has no role in the action tier. Presumably these tiers would divide thematic roles into two types, perhaps as follows: 6.41 a. Action tier roles: b. Thematic tier roles: actor, agent, experiencer, pattent, beneficiary, instrument, theme, goal, source, location. To these dimensions of action and space, Jackendoff also proposes a dimension of time, which we will not investigate here. The basic insight is clear: the roles that speakers assign to entities may be more complicated than a single thematic role label. For a detailed discussion of this proposal, see Jackendoff (1990: 125-51). Having identified these thematic roles, the next question we might ask is: how are such roles identified in the grammar? For our English examples above, the answer is by a combination of syntactic structure and the choice of verb. There are typical matchings between participant roles and grammatical relations. As in our original example 6.22, the subject of the sentence often corresponds to the agent, the direct object to the theme, while the instrument often occurs as a prepositional phrase. Though this is the typical case, it is not necessarily so: for example it is possible to omit the agent from the sentence and as a result have the instrument occupy subject position, e.g.: I 158 Semantic Description 6.42 The jack raised the car. We can see the effect of the choice of verb if we try to describe this same situation without either the agent or the instrument. We cannot simply allow the theme to occupy subject position as in 6.43; we have to change the verb as in 6.44: 6.43 *The car raised. 6.44 The car rose. This is because the verb raise requires an actor. The verb rise however describes a change of state without any slot for an actor so that while 6.44 above is fine, 6.45 and 6.46 below are not possible: 6.45 *Gina rose the car. 6.46 *The jack rose the car. What this simple example shows is that a speaker's choice of participant roles has two aspects: the choice of a verb with its particular requirements for thematic roles, and within the limits set by this, the choice of grammatical relations for the roles. We look at these choices in the rest of this chapter, beginning with the relationship between thematic roles and grammatical relations: first we describe how various thematic roles may occupy subject position, then we look briefly at the selection of thematic roles as part of a verb's lexical semantics. Later we discuss the role of voice in allowing speakers to alter prototypical marchings between thematic roles and grammatical relations. 6.3 Grammatical Relations and Thematic Roles We have seen that while in English there is a tendency for subjects to be agents, direct objects to be patients and themes, and instruments to occur as prepositional phrases, this need not always be the case. There are two basic situations where this is not the case: the first is where roles are simply omitted, and the grammatical relations shift to react to this, as we will discuss in this section; and the second is where the speaker chooses to alter the usual matching between roles and grammatical relations, a choice often marked by an accompanying change of verbal voice. We deal with voice later on in section 6.7. We can begin with a simple example of thematic role omission in 6.47-9 below: Sentence Semantics 2: Participants 159 6.47 Ursula broke the ice with a pickaxe. 6.48 The pickaxe broke the ice. 6.49 The ice broke. This is similar to our example 6.23 earlier: in 6.47 Ursula is the agent and subject, the ice is patient and direct object, and the pickaxe, the instrument, is in a prepositional phrase. In 6.48 the agent is omitted and now the instrument is subject; and finally in 6.49 with no agent or instrument expressed, the patient becomes subject. The verb break, unlike raise earlier, allows all three thematic roles to occupy subject position. Several writers have suggested that this process of different roles occupying the subject position is a hierarchical process, not only in English but across many languages. The observation is that when speakers are constructing a sentence, they tend to place an agent into subject position, the next preference being for a recipient or benefacttve, then thbme/pattent, then other roles. From our English examples, it seems that instrument is then preferred to location. This is sometimes described as an implicationat hierarchy. There are various versions of such a hierarchy proposed in the literature, e.g. in Fillmore (1968) and Givon (1984b), but we can construct a simple example of a universal subject hierarchy like 6.50 below: 6.50 agent > RECIPrENT/benefacttve > THEME/pATTENT > instrument > location This diagram can be read in two equivalent ways: one is that the leftmost elements are the preferred, most basic and expected subjects, while moving rightward along the string gives us less expected subjects. A second way to read this diagram is as a kind of rule of expectation, going from right to left: if a language allows the location role to be subject, we expect that it will allow all the rest. If, however, it allows the role instrument to be subject, we expect that it allows those roles to the left, but we don't know if it allows the location role as subject. The idea is that languages can differ in what roles they allow to occur as subject but they will obey this sequence of preference, without any gaps. So, for example, we should not find a language that allows agent and instrument to be subject but not themb^atient. It is a little difficult to think of English examples with location as subject, unless we include sentences like 6.51a-b below: 6.51 a. This cottage sleeps five adults, b. The table seats eight.1 but the other positions on the hierarchy occur regularly, as we can see from the following examples: 160 Semantic Description Sentence Semantics 2: Participants 161 6.52 agent subjects: The thief stole the wallet. Fred jumped out of the plane. 6.53 experiencer subjects: I forgot the address. Your cat is hungry. 6.54 recipient subjects: She received a demand for unpaid tax. The building suffered a direct hit. 6.55 patient subjects: The bowl cracked. Una died. 6.56 theme subjects: Joan fell off the yacht. The arrow flew through the air. 6.57 instrument subjects: The key opened the lock. The scalpel made a very clean cut. See Comrie (1981) and Croft (1990) for discussion of this and other im-plicational hierarchies. 6.4 Verbs and Thematic Role Grids As we saw earlier with the verbs raise, rise and drive, verbs have particular requirements for their thematic roles. Since this is part of a speaker's semantic knowledge about a verb, we might expect it to be part of the lexical information stored for verbs. Thus we need to know not only how many arguments a verb requires (i.e. whether it is intransitive, transitive, etc.) but also what thematic roles its arguments may hold. In the generative grammar literature, this listing of thematic roles is often called a thematic role grid, or theta-grid for short.2 A simple example might be: 6.58 put,V: This entry tells us that put is a three-argument, or ditransitive, verb and spells out the thematic roles the three arguments may carry. Here we show Williams's (1981) suggestion of underlining the agent role to reflect the fact that it is this role that typically occurs as the subject of the verb (or 'external argument' in Williams's terminology). Clearly this is just the start of the job that a grammatical description must do of mapping between thematic roles and grammatical categories and structures. Our thematic grid for put in 6.58 predicts that this verb, when saturated with the correct arguments, might form a sentence like 6.59: 6.59 JohnMEK1 put the boolc^ on the shelf^^.3 Of course, not all nominate in a sentence are arguments of a verb and thus specified in verbal theta-grids in the lexicon. We will make the assumption that one can employ grammatical tests to identify arguments: for example, to distinguish between the role of argument played by the prepositional phrase in the bathroom in 6.60 below andits status as a non-argument in 6.61: 6.60 [s Roland [vp put [jqp the book] in the bathroom]]] 6.61 [s Roland [w read [OTthe book]] tpin the bathroom]} The square brackets in 6.60-61 reflect the fact that while m the bathroom is an argument of the verb put, explaining why it cannot be omitted: 6.62 *Roland put the book. it is not an argument of the verb read, for example, which can form a sentence without it: 6.63 Roland read the book. In grammatical terms, while in the bathroom is an argument in 6.60, it is an adjunct in 6.61. As well as not being required by the verb, adjuncts are seen as less structurally attached to the verb, explaining why 6.64 below is a much more unusual word order than 6.65, and usually requires a marked intonation pattern: 6.64 In the bathroom Roland put a book. 6.65 In the bathroom Roland read a book. See Radford (1988) and Haegeman (1994) for discussion of the grammatical status of arguments and adjuncts. We will assume that all verbs may co-occur with adjuncts (usually adverbials of time, place, manner, etc.) and that requirements need only be listed in the lexicon for arguments. Another way of making this distinction is to distinguish between participant roles and non-participant roles. The former correspond to our 162 Semantic Description Sentence Semantics 2: Participants 163 arguments: they are needed by the predication, in the sense we have been discussing; the latter are optional adjuncts which give extra information about the context, typically information about the time, location, purpose or result of the event. Of course only participant roles will be relevant to verbal thematic grids, and our discussion in this chapter focuses on these participant roles. listing thematic grids soon reveals that verbs form classes which share the same grids. For example English has a class of transfer, or giving, verbs which in one subclass includes the verbs give, lend, supply, pay, donate, contribute. These verbs encode a view of the transfer from the perspective of the agent. They have the thematic grid in 6.66; .6.67 is an example: 6.66 V: 6.67 Barbara^ loaned the moneyTOto MichaelaE.'t Another subclass of these transfer verbs encodes the transfer from the perspective of the recipient. These verbs include receive, accept, borrow, buy, purchase, rent, hire. Their thematic grid is in 6.68, with an example in 6.69, paralleling 6.67 above: 6.68 V: recipient. theme, source> 6.69 Michael,,, borrowed the moneyra from Barbara*,. Thematic grids such as these are put to use in the literature for a variety of descriptive jobs. We can look at some of these in section 6,6, when we ask more generally: what purpose do thematic roles serve in linguistic analysis? First though we discuss some of the problems associated with the simple picture of thematic roles we have outlined so far. 6.5 Problems with Thematic Roles In our introductory discussion, we mentioned that the lists of roles given in the literature have varied from author to author. Authors disagree about what if any distinctions are to be made between patient and theme, for example, or between agent and related roles like actor, expertencer, etc. We can see these debates as reflections of two general problems with thematic roles (usually abbreviated to 'theta-roles', sometimes also called 9-roles). The first problem is really about delimiting particular roles. The extreme case would be to identify individual thematic roles for each verb: thus we would say that a verb like beat gives us two theta-roles, a beater-role and a beaten-role. This would of course reduce the utility of the notion: if we lose the more general role types like agent, patient etc., then we .538 ■m m cannot make the general statements about the relations between semantic roles and grammatical relations discussed earlier, nor put theta-roles to any of the uses we describe in the next section. But if we are to classify individual theta-roles roles like beater and beaten into theta-role types like agent and patient, we will have to find some way of accommodating variation within the role type. Let us take the example of patient in a typical grid: 6.70 V: \ \ > Patient > i > lExperiencer J (.Goal J As before, the candidates move from left to right in decreasing strength of linkage to the subject position. In this version, though, the roles themselves are not primitives but convenient labels for clusterings of the Proto-role entailments. So far we have been talking about theta-roles as explanatory devices in accounting for linkage between semantic and syntactic argument structure. A second justification for using thematic roles is to help characterize semantic verbal classes. For example we can identify in English two classes of psychological verbs both of which take two arguments (i.e. are transitive), one of which is an experiences and the other a stimulus.10 The classes differ however in their linking between these roles and subject and object position. The first class has the theta-grid in 6.84a below, and can be exemplified by the verbs in 6.84b, while the second class has the theta-grid in 6.85a and includes verbs like those in 6.85b: 6.84 Psychological verbs type 1 a. V: b. admire, enjoy, fear, like, love, relish, savour 6.85 Psychological verbs type 2 a. b. amuse, entertain, frighten, interest, please, surprise, thrill11 Thus we say Claude liked the result but The result pleased Claude. Such classifications of verbs can help predict the grammatical processes individual verbs will undergo. Thus, though the motivation for grammatical rules is often multifactorial, theta-role grids have been used to describe argument changing processes like passive, as we shall see shortly, or argument structure alternations like those in 6.86-7 below, where in each case the example sentences are in a, the link between theta-grids and syntactic arguments is given in b, and some example verbs in c: 6.86 a. He banged the broom-handle on the ceiling. He banged the ceiling with the broom-handle. She tapped the can against the window. • She tapped the window with the can. 168 Semantic Description b. V: NP NP PP c. bang, bash, beat, hit, knock, pound, tap, tap, whack" 6.87 a. The whole community will benefit from the peace process. The peace process will benefit the whole community. b. V: NP PP V: NP NP c. benefit, profit14 These alternations are just two of a large range identified for English in Levin (1993). The conditional factors, for such alternations are often a mix of semantic information, such as the verb's meaning and its theta-grid (as shown above), and its syntactic environment. We can look at one further type of justification for thematic roles which comes from another area of grammar: the claim that in some languages they play a role in the morphology of verbal agreement. Mithun (1991: 514) gives examples of the pronominal verbal prefixes in Lakhota (Siouan; USA, Canada). In the transitive verbs in 6.88a below we see a prefix tea which marks an agent argument and in 6.88b a prefix ma, which marks a patient: 6.88 a. awa?u waktekte b. ama?u maktekte 'I brought it.' Til kill him.' 'He brought me.' 'He'll kill me.' We can see that these prefixes do not mark subject or object agreement because a subject, for example, can take either prefix depending on whether it is an agent (as in 6.89a below) or.patient (as in 6.89b) (Mithun 1991: 514): 6.89 a. agent subjects wapsica 'I jumped' wahi *I came' b. patient subjects mak^e Tm sick' maxwa Tm sleepy' In other words, what would be a subject pronoun in English corresponds to either an agent or patient pronoun affix in Lakhota. Thus Lakhota morphological marking is sensitive to theta-roles rather than grammatical relations. Mithun gives similar examples from Guarani.(Tupij Paraguay, Sentence Semantics 2: Participants 169 Bolivia), and the Pomoan languages of California. The implication for our discussion is clear: if we need theta-roles to explain morphological patterns, this is strong evidence that they are significant semantic categories. We have seen then in this section a number of different motivations for identifying thematic roles: to explain linking rules in verbal argument structure, to reflect semantic classes of verbs, to predict a verb's participation in argument structure alternations, and finally to describe morphological rules adequately. For marry linguists this utility motivates their continuing use, despite the definitional problems discussed in the last section. In the next section we look at the category of voice, which, as we shall see, adds new dimensions to the relationship between theta-roles and grammatical relations. 6.7 Voice 6.7.1 Passive voice The grammatical category of voice affords speakers some flexibility in viewing thematic roles. Many languages allow an opposition between active voice and passive voice. We can compare for example the English sentences in 6.90 below: 6.90 a. Billy groomed the horses. b. The horses were groomed by Billy. In the active sentence 6.90a Bitty, the agent, is subject and the horses, the patient, is object. The passive version 6.90b, however, has the patient as subject and the agent occurring in a prepositional phrase, the structure often associated with instrument, as we saw in the last section. This is a typical active-passive voice alternation: the passive sentence has a verb in a different form - the past participle with the auxiliary verb be - and it allows the speaker a different perspective on the situation described. This passive sentence (6.90b) allows the speaker to describe the situation from the point of view of the patient rather than that of the agent. In some cases indeed passive constructions are used to obscure the identity of an agent, as in 6.91 below: 6.91 The horses were groomed. Here the agent is so far backgrounded that it becomes merely an implied participant. Many writers describe this foregrounding of the patient and backgrounding of the agent in terms of promoting the patient and demoting the agent (for example Giv6n 1990) or as reflecting the speaker's greater empathy with the patient rather than the agent (Kuno 1987). There are other lexical and syntactic strategies which alter perspective in this way. For 170 Semantic Description Sentence Semantics 2: Participants 171 example in 6.92 below the alternation relies in part on the lexical relation between in from of and behind, while in 6.93 it is accomplished by the syntactic patterns known as pseudo-cleft in a and cleft in b: 6.92 6.93 a. The house stood in front of the cliff. b. The cliff stood behind the house. a. What Joan bought was a Ferrari. b. It was Joan who bought the Ferrari. In 6.93 above the same situation is described but.in a the speaker is interested in Joan's purchase, while in b she is interested in the Ferrari's purchaser-This kind of choice of perspective presumably depends on a speaker's judgements of conversational salience. We can use the terms figure and ground15 to describe this kind of linguistic perspective: if we call the situation described a scene, then the entity that the speaker chooses to foreground is the figure, and the background is the ground. So in 6.92a above the house is the figure and tke cliff the ground, and vice versa in 6.92b. Passive constructions allow the foregrounding of roles other than patient In 6.94-6 we see English examples of theme, percept, and recipient roles occurring as the subject of passives: 6.94 This money was donated to the school, (theme) 6.95 The UFO was seen by just two people, (percept) 6.96 He was given a camera by his grandmother, (recipient) The qualifications for foregrounding in a passive in English are complex: partly grammatical, partly semantic and partly due to the flow of discourse and the speaker's choice of viewpoint. The importance of grammatical information can be shown by observing that each of the roles occurring as passive subjects in 6.94-6 above occut in object position in a corresponding active sentence: 6.97 Someone donated this money to the school. 6.98 Just two people saw the UFO. 6.99 His grandmother gave him a camera. The typical pattern is that a nominal occupying object position is fronted to subject in passives. When a theta-role normally occurs as a prepositional phrase in an active sentence, this is less likely to be foregrounded in a passive. Neither moving the full prepositional phrase nor extracting just the nominal seems to work, as shown below: 6.100 a. This house stood on the corner, (location) b. *On the corner was stood by this house. c. ?The corner was stood on by this house. 6.101 a. John built a garage for her. (beneficiary) b. *For her was built a garage by John. c. ?She was built a garage by John. 6.102 a. He opened the door with this key. (instrument) b. *With this key was opened the door by him. c. *This key was opened the door with. Some apparent exceptions to this rule are possible however, e.g. 6.103 a. Three monaTchs lived in this house, (location) b. This house was lived in by three monarchs.1' To farther underline this grammatical aspect of passives, i.e. that it is the object position that is relevant to passivization, we can look at a class of English verbs called the spray/load verbs. These verbs allow the speaker to select either their theme role (as in 6.104a and 6.105a) below, or the goal (as in 6.104b and 6.105b), to be the verb's direct object and thus be the focus of the effect of the action: 6.104 a. He sprayed paint on the car. b. He sprayed the car with paint. 6.105 a. He loaded hay on to the tractor, b. He loaded the tractor with hay. We can easily show that whichever argument occupies object position can be passivized while the argument in the prepositional phrase cannot: corresponding to 6.104 above we find the patterns: 6.106 a. Paint was sprayed on the car. b. *The car was sprayed paint on. c. The- car was sprayed with paint. d. *Paint was sprayed the car with. See Rappaport and Levin (1985,1988), Jeffries and Willis (1984) and Levin (1993) for further discussion of these spray/load verbs.'7 The discourse factors affecting passives have been described in a number of frameworks: for example, as mentioned above, Kuno (1987: 209-16) employs the notion of speaker empathy. He gives an example of a person relating a story about their friend Mary and her experiences at a party. In the narrative the speaker's empathy is with Mary and thus events are viewed 172 Semantic Description from her perspective.This explains why a passive is fine in 6.107b below but not in 6.108b (treating these as two independentreports of events): 6.107 Mary had quite an experience at the party she went to last night. a. An eight-foot-tall rowdy harassed her. b. She was harassed by an eight-foot-tall rowdy. 6.108 Mary had quite an experience at the party she went to last night. a. She slapped an eight-foot-tall rowdy in the face. b. *An eight-foot-tall rowdy was slapped in the face by her. The passive construction works in 6.107b because the fronted nominal refers to the entity the speaker empathizes with, but not in 6.108b where the other participant is fronted. Passive constructions have received a great deal of attention in the linguistics literature. This is not surprising: even from our brief discussion, we can see that while the general effect of passive is to allow a shift in linkage between theta-roles and grammatical relations, the process is subject to a complex of grammatical and discourse factors. It is this interdependence of different levels of analysis that makes passives an interesting arena for theoretical debate. 6.7.2 Comparing passive constructions across languages While many languages have passive-type constructions, the comparison of passives across languages reveals that there is considerable variation around the pattern of the English passive outlined in the last section, i.e. where the agent is demoted from subject position, a non-agent role is promoted to subject, and the verb shows a distinct form which agrees with the promoted subject: the total package being what we have called passive voice. Often languages have more than one passive construction: in English for'example, it is possible to distinguish between 4e-passives and get-passives, as in 6.109 (R. Lakoff 1971, Givon and Yang 1994): 6.109 a. Mary was shot on purpose, b. Mary got shot on purpose. As noted by Lakoff these sentences differ in the amount of control over the event associated with Mary.'8 Other languages have a special type of passive, often called the impersonal passive, which does not allow the agent to be mentioned in the sentence. In Irish, for example, we can distinguish between one type of passive associated with verbal noun constructions as shown in the active/ passive pair in 6.110 below, and another, the impersonal passive, with verbs, as is shown in 6.111 (Noonan 1994: 282-6): I -M Sentence Semantics 2: Participants 173 6.110 a. Bhi si ag bualadh Sheain. was she at hit-NOMiN John-GEN 'She was hitting John.' b. Bhi Sean a bhualadh aici. was John to+his hit-NOMiN at-her 'John was being hit by her.' 6.111 a. Thug siad Siobhan abhaile inniu. brought they Joan home today 'They brought Joan home today.' b. Tugadh Siobhan abhaile inniu. brought-iMPERS Joan home today. 'Joan was brought home today.' This impersonal passive in 6.111 does not straightforwardly correspond to the translation given: i.e. to an English passive where no agent is expressed. In 6.111b we can see how both in Irish and in the English translation the passive verb form is differentiated from the active, and how in both the agent is often omitted. However the Irish passive in 6.1 lib differs from its English translation because the themb, Siobhan, remains in its original position as an object while in the English passive Joan becomes subject. In other words, the patient is not promoted to subject in the Irish impersonal passive in 6.11 lb, but the agent is omitted. See Noonan (1994) for discussion. This example from Irish is of a transitive impersonal passive. In many languages the term impersonal passive is used to describe passives of intransitive verbs: Kirsner (1976: 387) gives the following pair of examples from Dutch: 6.112 a. De jongens fluhen. the boys whistle. 'The boys whistle/are whistling.' b. Er wordt door de jongens gefloten. there becomes by the boys whistling "By the boys (there) is whistling.' In 6.112b the agent is backgrounded, but there is no other argument to be foregrounded and subject position is taken by the word er 'there', which does not refer directly to any entity and which has no theta-role. It is also possible to delete the agent altogether in this passive, giving: 6.113 Er wordt gefloten. there becomes whistling "There is whistling/People whistle/Someone whistles.' Similar impersonal passives have been reported for other languages, including German, Welsh and Latin; see Perlmutter (1978) and Perlmutter and Postal (1984) for discussion. 174 Semantic Description Sentence Semantics 2: Participants 175 These impersonal passives imply that in comparing languages we need to separate out the two functions of the passive: firstly, the demotion of agents, and secondly, the promotion of non-agents. Thus an English passive like Spike was arrested by the police combines both functions: the agent argument is demoted to a prepositional phrase, and the patient is promoted to subject. We can see the related sentence Spike was arrested as a special case of this, where demotion reaches its extreme in the suppression of the agent. In the Dutch impersonal passives in 6.112b on the other hand we see a passive strategy which just embodies the first function: demotion of agent, with no concomitant promotion function. Since this example has an intransitive verb, the further step of suppressing the agent leaves a sentence with no theta-role bearing nominal as in 6.113. The third characteristic of English passives described in the last section was a special verb form and associated verbal agreement with the promoted subject. This too is subject to cross-linguistic variation. Passive verbs are often semantically distinguished from their active counterparts, for example by being more stative, though this is not always so, and they may show agreement with the promoted non-agent nominal (as in English), or.the demoted agent, or neither, since agreement inflections may be neutralized; see Givon (1990: 563-644) for discussion of variations along this parameter as well as along the parameters of agent demotion and non-agent promotion. One conclusion from comparing passives across languages seems to be that the phenomenon is typically a cluster of functions: in each case following the genera] pattern of allowing the speaker planning her discourse some variation in the linkage between thematic and grammatical roles, but with considerable variation in the associated semantic and grammatical elements of the cluster. In most active-passive systems the active form is usually grammatically simpler and we may ask why this should be so. It has been argued that we as humans naturally view situations from the point of view of any human beings involved, and if there are none, of other living creatures. This pref-' erence, sometimes called an animacy hierarchy (see for example Dixon 1979, Hopper and Thompson 1980), is coded into the lexical semantics of a language so that a verb like drive, for example, in 6.114 sets up a thematic role frame which requires an agent as the subject: 6.114 Ann drove the truck across the field. and since agency, as we have seen, requires wilful action, agents are typically people, or higher animals. It is difficult to think of a verb which describes the action in 6.114 from the point of view of the truck. We might say: 6.115 The truck carried Ann across the field. but this sentence has a different meaning: we have'not specified that Ann was driving. So it seems that the meaning of the verb drive is set up to prioritize the role of any human or volitional agent. Passive voice allows the speaker to get around this in-built bias, so that to switch the viewpoint from Ann to the truck, or to the field, she can use passive constructions as in 6.116-17: 6.116 The truck was driven across the field by Ann. 6.117 The field was driven across by a truck (*by Ann). We can see that in 6.117 there is no longer a slot for the agent, Ann. So passive constructions do allow a change of perspective but the conventional bias towards animate subjects means that the active drive is grammatically simpler than the passive was driven, 6.7.3 Middle voice While very many languages display this active/passive voice contrast, some languages have a three-way distinction between active, passive and middle voice. As we might expect, the use of middle voice varies from language to language but a central feature is that middle forms emphasize that the subject of the verb is affected by the action described by the verb. This affect-edness, as it is often termed (e.g. Klaiman 1991), can be of several types, and we can select four typical uses as examples: neuters, bodily activity and emotions, reflexives, and autobenefactives. Though we will use examples from several languages, to keep the discussion brief we will concentrate on two unrelated languages, well separated in space and time: classical Greek and the modern Gushitic language Somali.1' In both these languages middle voice is marked by verbal inflection. Neuter intransitives This type of middle is where the subject undergoes a non-volitional process or change of state. The external cause is not represented but can often be shown in a related active form, as shown in 6.118 below, an example from Sanskrit (Klaiman 1991: 93): 6.118 a. So namati dandam. he-nom bends-3sg active stick-Ace 'He bends the stick.' b. Namate dandah. bends-3sg middle stick-nom 'The stick bends.' Middle voice verb forms of this neuter type, where the subject undergoes a process over which it has no control, occur in classical Greek, as shown in 6.119 (Bakker 1994: 30) and Somali,20 as in 6.120: 176 6.119 6.120 phú-e-sthai tréph-e-sthaí sép-e-sthai ték-e-sthai rhěgnu-sthat kab-o qub-o dhim-o haf-o garaads-o Semantic Description 'grow' 'grow up' 'rot' 'melt1 'break' 'recover, set (of a bone)' 'fall (of leaves and fruit)' 'die' 'drown' 'reach maturity' Sentence Semantics 2: Participants 177 Bodily activity and emotion In some languages the verb occurs in a middle voice when die activity involves the body or emotions of the subject. These would seem to be clear cases of affectedness since the subject is so overtly involved. Examples of such middle voice verbs are in 6.121-2: 6.121 Classical Greek (Bakker 1994) klin-e-sthai 'lean' hed-e-sthai 'rejoice' 6.122 Somali (Saeed 1999) * fadhiis-o 'sit down' baroor-o 'mourn, wail' Reflexives In some languages the middle is used where the subject's action affects the subject himself, or a possession or body part of the subject. To take another example from classical Greek (Barber 1975: 18-19): 6.123 Lou-omai. wash lsg MIDDLE 'I wash myself.' This use means that in many languages verbs of grooming occur in the middle voice, with no need for a reflexive pronoun as object; see 6.124 for some further examples from Somali, and examples from other languages in 6.125 from Kemmer (1994: 195): 6.124 feer-o 'comb one's hair' maydh-o 'wash oneself, bathe' labbis-o 'dress up, put on one's best clothes' I m pi. 6.125 Latin Quechua Turkish Hungarian orno-r arma-ku-y giy-in mosa-kod- 'adorn oneself* 'bathe' 'dress' Vash oneself Autobenefactives This type of middle is used to signify that the action of the subject is done for his or her own benefit; Once again this use occurred in classical Greek as in 6.126 (Barber 1975: 18), and is a regular process is Somali, as 6.127 shows (Saeed 1993: 58): 6.126 a. hair-o moiran. take-lsg-AcnvE share 'I take a share.' b. hari-oumai moiran. take-lsg-MiDDLB share *I take a share for myself." 6.127 Active verbs: wad 'to drive' beer 'to cultivate' qaad 'to take' sid 'to carry' Middle verbs: wad-o 'to drive for oneself beer-o 'to cultivate for oneself qaad-o 'to take for oneself sid-o 'to carry for oneself In the examples so far, middle voice has been marked by verbal inflection. In some languages a pronoun marks middle forms, often the same form as a reflexive pronoun, e.g. German sick, French se, Spanish se, or a closely related form', e.g. Russian reflexive sebja, middle -sja, Dutch reflexive ziehzelf, middle -zetf (Kemmer 1994). In such languages the overlap between middle voice and reflexivity, seen in examples 6.121-7 above, becomes overt. In French and Spanish for example, we might identify our first three types of middle: 6.128 French middle reflexives a. neuter: b. bodily activity: emotion: c. reflexive: s'ecrouler s'evanouir s'asseoir se plaindre 8'habiller se peigner 'collapse' 'vanish' 'sit down' 'complain' 'dress oneself 'comb one's hair' 6.129 Spanish middle reflexives a. neuter: helarse recuperarse b. bodily activity: tirarse emotion: enamorarse (de) 'freeze (intr.)' 'get well' 'jump' 'fall in love (with)' 178 c.' reflexive: Semantic Description afeitarse quitarse 'shave' 'take off (clothes)' However, even in languages where the middle and reflexives are marked by the same pronoun, there are usually clear cases where the meaning distinguishes between true reflexives and the middle, e.g. in German (Kemmer 1994: 188): 6.130 Er sieht sich Er fürchtet sich *He sees himself' *He is afraid' (Reflexive) (Middle - emotion) In English there is no inflectional or pronominal marker of the middle: the distinction is only shown by alternations between transitive active verbs and intransitive middle verbs, where the agent is omitted, e.g. 6.131 a. They open the gates very smoothly, b. The gates open very smoothly. (Active) ■ (Middle - neuter) These intransitive middles in English are often used to describe the success of a non-agent in some activity, e.g. 6.132 a. These clothes wash well. b. This model sells very quickly. V c. These saws don't cut very efficiently. See Dixon (1991: 322-35) for more examples of this type of construction in English. Because of the similar suppression of the agent in this type of middle and in the passive, some writers use the term medio-passive to cover both.- 6.8 Classifiers and Noun Classes So far in this chapter we have been exploring the ways that participants may be assigned semantic roles relative to the action or situation described by the verb. In this section we look at semantic characterizations that are based on inherent properties of the entities referred to by noun phrases. Many languages have overt systems for marking how referents fit into a semantic classification system. We divide our brief discussion of these into first, classifiers, and then, noun classes. 6.8.1 Classifiers Noun classifiers are morphemes or lexical words that code characteristics of the referent of the noun, allowing the speaker to classify the referent Sentence Semantics 2: Participants 179 according to a system of semantic/conceptual categories. They may show up grammatically in different guises. Some, termed noun classifiers, occur with nouns. Dixon (1977) describes the noun classifiers of the Australian language Yidni as a closed set of around 20 members, which he divides into two general types, each containing several subtypes. The first type, inherent nature classifiers, includes as subtypes of classifiers: human; animals; vegetation; natural objects (like the classifier vialba 'stone'); and artefacts (like the classifier baji 'canoe'). The second type, functional classifiers, divides entities into: meat food; non-meat food; drinkable things; movable; habitable; and 'purposeful noise'. Dixon (1982) reports that two classifiers can be used with the same nominal as long as they come from the two different general types, for example (where cx = classifier): 6.133 bulumba walba malan cl:habitable cl:stone flat.rock 'a flat rock for camping' (Dixon 1982: 200) In many languages classifiers occur in specific grammatical constructions or locations, for example numeral classifiers, which occur when the entity is being counted, and possessive classifiers, which occur in constructions describing possession. Numeral classifiers occur in Japanese as in shown in example 6.134 below: 6.134 Classifiers in a Japanese shopping list (cited in Aikhenvald 2000: 2) Shopping list Numeral Classifier Meaning of classifier nam (eggplant) nana (7) -ho CL: small, equidtmentional kyuuri (cucumber) hachi (8) -hon Cl: elongated hamu (ham) juu (10) -mai Cl: sheetlike As we can see, these classifiers relate to a classification based on shape. Possessive (or genitive) classifiers may characterize the possessed item, as in the Fijian example in 6.135 below; or classify the type of possession relation involved, as in 6.136 from Hawaiian: 6.135 Fijian possessive classifiers (lichtenberk 1983: 157-8) a. na me-qu yaqona art ct:drtnkable-my kava 'my kava (which I intend to drink)' b. ha no-qu yaqona art cl-.gbneral-my kava 'my kava (that I grew, or that I will sell)' 180 Semantic Description 6.136 Hawaiian possessive classifiers (Iichtenberk 1983: 163) a. fc-o-'u inoa ART-CL-my name 'my name (that represents me)' b. k-a-'u inoa art-cl-my name 'my name (that I bestow on someone)' A further type is verbal classifiers, where the classifier occurs as a morpheme attached to the verb and serves to classify (intransitive) subjects or objects: see for example: 6.137 Dogrib (Athapaskan) (cited in Allen 2001: 309) a. let'e niyeh-tsi bread I.pick.up-PERF.CUFLAT.PLEXIBLE.ENTITY ' 'I pick up a slice of bread' b. let'e niyeh-?a bread I.pick.up-PERF.CL:ROUND.ENTlTY 'I pick up a loaf of bread' Wherever they are marked grammatically classifiers tend to exploit a fixed set of semantic distinctions. Though there is large variation, it is possible to identify some prototypical distinctions, as Allan (2001) does below: 6.138 Prototypical classifier categories (Allan 2001: 307) a. Material make-up: e.g. human (-like), animate, female, tree (-like) b. Function: e.g. piercing, cutting, or writing instruments; for eating, drinking c. Shape: e.g. long (saliently one-dimensional), fiat, round -.v d. Consistency: e.g. rigid, flexible, mass e. Size: including diminutives and augmentatives f. Location: inherently locative entities such as towns g. Arrangement, e.g. a row of, a coil of, a heap of h. Quanta: e.g. head of cattle, pack of cigarettes 6.8.2 Noun classes Noun classes are agreement-based noun systems that seem, at least historically, to be based on semantic classifications somewhat similar to those we have seen for classifiers. One famous example occurs in the Bantu languages of Africa, where nouns belong to a pattern of classes, related variously in the modern languages to an ancestral system that is characterized by Sentence Semantics 2: Participants 181 Aikhenvald (2000) as follows, (where class pairs 1/2 etc. are singular and plural): 6.139 Noun classes in Proto-Bantu (Aikhenvald 2000: 282) Class Semantics 1/2 Humans, a few other animates 3/4 Plants, plant parts, foods, non-paired body parts, miscellaneous 5/6 Fruits, paired body parts, miscellaneous inanimates 7/8 Miscellaneous inanimates 9/10 Animals 11/10 Long objects, abstract entities 6 Small objects, birds 14 Masses 15 Infinitives The key feature of noun class systems is that other elements in the sentence agree with the noun in terms of its class. See for example (6.140) below from the modern Bantu language Swahili: 6.140 Swahiliclass 8 (Allan 2001: 310): Vi-su vidogo viwili hi-vi amba-vy-o nili-vi-nunua rri ta'-knife w-smali w-two this-tii which-oi l.s-ui-buy be vi-kali sana . ra-sharp very 'These two small knives which I bought are very sharp' Here the noun class prefix, marked in bold, is copied as an agreement feature by other elements in the noun phrase headed by visu 'knife' and in the sentence in which the noun phrase is subject. In the modern Bantu languages the assignment of nouns to classes is not always as semantically transparent as the classes in 6.139 suggest. Often the classes are much more heterogeneous and membership may -be more conventionalized. Gender systems, familiar from Indo-European languages, in which nouns are assigned to two or thee classes - male, female and perhaps neuter - are a type of noun class system. Indeed Corbett (1991) extends the term gender to cover all norm class systems. As may be the case with more complex noun class systems, gender in languages like German or Hindi is a grammatical distinction only loosely connected to biological sex. Humans and animals may be typically (though not exclusively) assigned to genders on the basis of biological sex, but other nouns are assigned by a mixture of criteria, some of which have no semantic basis, for example phonological shape. Noun class systems may be differentiated from classifiers by a number of features, some of which are summarized by Dixon (1986) as follows: 182 Semantic Description 6.141 Differences between noun classes and classifiers (Dixon (1986) Noun classes Size Small finite set Realization Closed grammatical system Scope Marking is never entirely within the noun word Classifiers Large number Free forms Never any reference outside the noun phrase However the large degree of variation within both types of system means that any simple characterization is only suggestive of typical cases. Sentence Semantics 2: Participants 183 '!• of passive and middle voice. Dixon (1991) discusses the ways in which the grammar i. of English verbs reflects semantic distinctions, and includes sections on thematic =. roles, and voice. Levin and Rappaport Hovav (2005) provide farther discussion of ■' the problems with thematic roles identified in this chapter. Givon (1994) is a collec-'.' tion of studies on argument structure changing processes, including passive. Keenan :' (1985) reviews passive constructions in a range of languages, while Klaiman (1991) : does a similar job for middle voice. Wilkins (1988), Grimshaw (1990) and Williams :: (1994) shed light on the interaction of thematic roles and grammatical processes. : These works are quite technical, however, and require some background in syntactic ':■ theory. Aikhenvald (2000) provides a comprehensive cross-linguistic overview of X classifier systems; and Corbet: (1991) discusses noun class systems. 6.9 Summary In this chapter our main focus has been on the ways in which a speaker may portray the roles of participants in a situation. We outlined a classification of such semantic roles, termed thematic roles or theta-roles, including agent, patient, thbme, etc. and described the relationship between these roles and grammatical relations like subject and object. It has been claimed that as part of its inherent lexical specification a verb requires its arguments to be in specific thematic roles, and that this can be reflected by formulating thematic role grids, or theta-grids. We discussed the difficulties there are in fixing tight definitions for individual thematic roles, and presented one approach, from Dowty (1991), which seeks to provide a solution in terms of frizzy categories. This difficulty with precision notwithstanding, it seems that the notion of thematic roles has proved a useful descriptive tool in a number of areas of the semantics-grammar interface. The grammatical category of voice allows speakers different strategies for relating thematic roles and grammatical relations. We concentrated on relations with subject position, in particular the way in which passive voice ' allows the foregrounding of non-AGENT roles to subject and the backgrounding of AGENT roles away from subject. We also looked at middle voice, which reflects the affectedness of. the subject in the action of the verb: thus offering a different view of the relationship between subject and verb from the active voice. Finally we looked at classifiers and noun classes: systems where nouns identifying entities are classified by inherent semantic features, though membership of the relevant classes may only be partially semantically determined. FURTHER READING An important study of thematic roles is Dowry's (1991) article. Palmer (1994) is a survey of thematic roles, the different ways they are grammaticalized and the role rXFRGIsf^S*. 0 K Ob flic pn iS'ii the mf im 1 iLlinitio» in s tion o tr\ to "fin,» X* Ingle'ffcetnttii. iok. libel i > «.fi h 1 ihi ij i i s "»i MP * "• bdd rt tl Kllivtiqr entvnc s »lUlwfl do\e.tO fhe pirtv ~ « * P* HL tfltti.a'the fly vitb i *»ifw«Spapt,r jiTht b1boon«was i k^f? on thPtoofof my ctü. ^ „» e- Cafhpr>tll4sW the ßun*ifjr ^ * ^ '1 GvorgtS^phe thod«ftin\an«p'' *1 PNT » -v i. 'tHIUL * S f e smrtn * **** « 3 1 ri,i(,li pfthc-theti-rple bd<$ cwstruc 1h jE-nfeli 1 nr nc^. n le t in irgunefli hnrim, Ü at rolv'ctl9n> HMBMMHBBbMBB oonUf 186 11 id 1 £ m. "m I J'l > * l>tt V* 1 (-H.mil, 1 m n tr 1 n ih rid ] 1 jfel II IT V HUtJTl.fi il ( ttj II II i NOTES 1 One might also think of examples like: In the village stands a pump. But here the subject still seems to be a pump rather than in the village, as can be shown by the pattern of agreement in: In the village stand several pumps. But see Levin and Rappaport Hovav (1995: 261-4) for arguments, couched in the theory of Lexical-Funcrional Grammar (e.g. Bresnan 1994), that the preverbal PF is, at some level of analysis, a subject. 2 See the introductory discussion of theta-grids in Haegeman (1994: 33-73). 3 Hereafter we will use just the two first letters of a thematic role with this subscript notation, e.g. Joan,,, for Joan,^ 4 In JackendofFs (1990) two-tier representation described earlier, these 'transfer* verbs woutd have a more complicated thematic grid: we could, for example, assign both agent and source roles to Barbara in 6.67.' 5 Note that in this view, theta-roles convey a speaker's classifications of things in the world: in other words, the roIeB are bome by real-world entities rather than grammatical elements like NPs. See for example the following example and comment from Laduslaw and Dowry (1988: 63): 1 a. b. Sentence Semantics 2: Participants Fido chased Felix. Felix was chased by Fido. 187 ... The only sense in which it is reasonable to think of the subject NP of (la) as the Agent is the sense in which it is shorthand for saying that the object (in the world) referred to by the subject is the Agent in the action described by the sentence. What makes Fido an agent in the event described by (la) and (lb) is information about Fido and his role in the event, not about the grammatical category or function of anything in the sentence. For a related idea, see Foley and van Valin's (1984) theory of macro-roles, where all thematic roles fall into two mam categories: actor and undergoer. This term arises from Dowry's (1991) examination of different types of what he calls themb roles, some of which would be patient roles in our classification. He proposes a class of incremental themes for the thbme/patient roles of achievement and accomplishment verbs, e.g. mow the lamu eat an egg* build a house, demolish a building. The observation is that the action (for example, the mowing action) and the state of the associated theme/patibnt (e.g. the lawn) are in a proportional relationship: some mowing cuts some of the grass, more mowing, more of the grass, etc. until completing the action cuts all of the grass. Dowty extends this idea of incremental themes to other types of role, e.g. ««6« from England to France, where the path is incrementally affected, and memorize a teem, where there is a similar incremental relationship between the action and a representation of the theme entity. See Dowty (1991) for further details. In our discussion we focus on languages like English which have the grammatical relations, subject and object. We therefore leave aside the different pattern of mapping between theta-roles and grammatical relation shown by ergative languages. Briefly, in a typical ergative system one grammatical relation, called absolutive, is used for die single argument of an intransitive verb, whatever its theta-role (and in this resembles English subject), but is also used in ditransitive verbs for the patient argument (and here resembles English object). A second grammatical relation, called ergative, is used for the agbnt/experiencer in ditransitive verbs (as is English subject). There is therefore no correspondence between the absolutive/ergauve distinction and the subject/object distinction. They represent two different strategies for mapping between theta-roles and grammatical relations. See the following simple example of an ergative system fromTongan (Austronesian: Tonga), given by Anderson (1976): a. na'e lea 'a etalavou. past speak abs young.man The young man spoke.' b. na'e alu 'a Tevita ki Fisi. past go abs David to Fiji 'David went to Fiji.' c. na'e tamate't 'a Kolaiate 'e Tevita. . past kill abs Goliath erg David •David killed Goliath;' d. na'e ma'u 'e siale 'a e me'a'ofa. past receive erg Charlie abs dbf gift 'Charlie received the gift.' , 188 Semantic Description 10 11 12 13 14 15 16 17 Note that in these Tongan sentences the verb comes first in the sentence, and me case-marking particles (in bold) precede their nominals. Sentences a and b have intransitive verbs and the verb's only argument is in the absolutive case. Sentences c and d have transitive verbs. Here the agent in c and the recipient in d are in the ergative case. The patient in c and the theme in d are in the absolutive case. The reader may compare this with the mapping for subject-object languages like English. Ergative languages are found all over the world and include Basque in southern Europe, the Australian language Dyfrbal,Tongan from the Pacific, and the Inuit languages of Canada, Greenland, etc See Dixon' (1979) for discussion and Croft (1990) and Palmer (1994) for cross-linguistic overviews. Note that Dowry's hierarchy here has instrument and patient in reverse order to our earlier hierarchy. We won't try to arbitrate between these claims here: compare the discussion in Dowry (1991) and Croft (1990). These are labels commonly used in the literature for the thematic roles associated with these verbs. We leave aside discussion of how these roles would correlate with the Agent-properties and Patient-properties in a Dowty-style approach. See Grimshaw (1990) and Levin (1993) for discussion of these classes of psychological verbs. Here we follow Jackendoff (1990) in allowing one argument to have two theta-roles, as described earlier. See Dowry (1991: 594-5), Levin (1993: 67-8). See Levin (1993: 83). This is similar to the use of'figure' and 'ground' in the analysis of motion verbs byTalmy (1975), and others, as discussed in chapter 9. There Hat figure is the entity in motion and the background is called the ground. But only under some special conditions, which have been much debated in the literature. Levin and Rappaport Hovav (1995: 143-4), for example, discuss examples of this type like This platform has been stood on by an ex-president under the label prepositional passives. They provide a restriction on the construction in English that mixes grammatical and semantic factors: that it is only possible with unergatlve verbs which take an animate subject. Unergative is a term introduced by Perlmutter (i978) for intransitive verbs like sit and stand whose single argument is an agent and whose grammatical behaviour contrasts ' with unaccusative verbs which are intransitive verbs like grow or drawn and whose single argument is essentially a patient. Dixon (1991: 298-321) on the other hand proposes syntactic restrictions, which include the absence of a direct object in the active sentence, and a lack of an alternative active construction in which the passivized NP could occur as direct object. For an in-depth study of these prepositional passive constructions see Couper-Kuhlen (1979). Other English verbs allow alternations into object position, e.g.: He wrapped cling-film around the food. He wrapped the food in cling-ftlm. 2 a. David gave the keys to Helen, b. David gave Helen the keys. She bought some flowers for her husband. She bought her husband some flowers. Sentence Semantics 2: Participants 189 18 Alternations like 2 and 3 are often called Dative Shift. Giv6n (1984a) describes these, and similar alternations in other languages, as promotion to object, a process paralleling passive. By comparison with passive, though, the process is more restricted to particular verbs and is less likely to be marked on the verb by a distinct inflection of voice. Though this is less true of pairs like: Mary was killed. Mary got killed. See Giv6n and Yang (1994) for a discussion of the English ger-passive; and Werner and Labov (1983) for a sociolinguisric approach. 19 For a survey of the meanings of middle voice in Somali, see Saeed (1995). 20 Note that not alt neuter middles in Somali have an active form: the verbs jaboi qubo, hafo do, but garaadso does not, and the middle verb dhimo 'to die' has as its active equivalent a different lexical verb dil 'to kill'. It seems that all languages which have a middle voice have some verbs that are inherently middle and have no morphologically related active forms. See Klaiman (1991) for discussion.