In Mila Vulchanova & Tor A. Åfarli (eds.) 2005. Grammar and Beyond— Essays in honour of Lars Hellan. Oslo: Novus Press. pp. 179-195

What is Right and Wrong about Little v Stephen Wechsler University of Texas

1. Introduction. The idea that something akin to causation or agency is introduced into the meaning of an utterance within the syntax, rather than coming from lexical meaning, has recurred since the early days of generative grammar (Chomsky 1970, Lakoff 1970, McCawley 1968). This proposal allows for an account of the causative alternation in (1) in which exactly the same verb close appears in both the causative and intransitive sentences.

The sentences are derived from d-structures like those in (2)a and b,

respectively. To get the intransitive variant the object NP moves into the empty subject position, while to get the causative variant the verb close amalgamates with the abstract (silent) verb CAUSE, forming a verb with the causative meaning ‘cause to close.’

1. a. The door closed slowly. b. John closed the door slowly.

2. a. [ __ [closed [the door] ] slowly ] b. [ John CAUSE [close [the door] ] slowly ]

1

Somewhat more recently, Kratzer (1996, 2002) pushed this idea one step further by proposing that all agent arguments (indeed, all external arguments) are lacking from the predicate argument structure of the verb, and are introduced instead by a semantically contentful silent element similar to CAUSE above. This includes not only verbs like close, for which the agent participant is not ontologically necessary, as in 1a: we can conceive of a door closing on its own. Kratzer extends this proposal even to the subjects of verbs like buy (to use one of Kratzer’s examples). For Kratzer the verb buy has only an argument corresponding to the theme (the doll in 3a) and a Davidsonian event variable. It lacks a ‘buyer’ argument, as shown in the lexical entry in 3b (x corresponds to the object NP and e is the event variable):

3. a. Sue bought the doll. b. buy:

λxλe [ buy(x)(e) ]

The agent thematic role is assigned instead by a silent light verb which functions as a secondary predicate. Notating the abstract light verb with lowercase v (Chomsky 1995:315), the underlying syntactic structure of Sue bought the doll is shown in (4a). The semantics of little v appears in 4b, and the resulting logical form for the sentence is given in 4c.

4. a. [ Sue [ v [ bought the doll] ] ] b. v:

λxλe [ Agent(x)(e) ]

c. λe [ buy(the-doll)(e) & Agent(Sue)(e) ]

2

Kratzer hypothesized that in composing the semantic form in (4c) from the sentence structure in (4a), the expression buy(the-doll, e) comes from the verb bought (and its argument, the NP the doll), while the expression Agent(Sue, e) is provided by v (and its argument, the NP Sue).

A compositional principle identifies the event variables

contributed by v and the verb (Event Identification, Kratzer 1996:122). This use of the abstract light verb is similar to the use of semantic primes such as CAUSE

in theories like Generative Semantics that compose word meaning in the syntax

(Lakoff 1970). One innovation introduced in this most recent revival of Generative Semantics has been the attempt to assimilate the abstract semantic primes into the system of functional heads in extended X-bar theory. Kratzer comments that her proposal that agent arguments are not selected by the verb ‘has radical consequences for the syntactic theory of argument structure.’ (p.111) The little v hypothesis has been adopted in some circles, but remains highly controversial (see Krifka 1999, Horvath and Siloni 2003, and Kiparsky 1997 for critiques). The present paper further undercuts the little v hypothesis by pointing out a weakness in a key argument often adduced in its favor. Kratzer argued that her proposal explains certain subject/object asymmetries noted by Marantz (1984). I will argue that the proposal actually fails to predict the purported asymmetry. But I will also try to assess what is right about the proposal. I conclude that there is some plausibility to the idea of introducing semantic primes such as CAUSE or AGENT in the syntax, or at least supralexically, that is, outside of the meanings of particular

3

verbs. But, I will suggest, such primes should not be understood as the source of external arguments.

2. Marantz’s subject/object asymmetry. Marantz (1984) observed that while English has many idioms and specialized meanings for verbs in which the internal argument is the fixed part of the idiom and the external argument is free, the reverse situation is considerably rarer. To put it differently, the nature of the role played by the subject argument often depends on the filler of the object position, but not vice versa. To take Kratzer’s (1996:114) examples:

5. a. kill a cockroach b. kill a conversation c. kill an evening watching TV d. kill a bottle (i.e. empty it) e. kill an audience (i.e., wow them)

On the other hand, one does not often find special meanings of a verb associated with the choice of subject, leaving the object position open (examples from Marantz 1984:26):

6. a. Harry killed NP. b. Everyone is always killing NP. c. The drunk refused to kill NP. d. Silence certainly can kill NP.

4

e. Cars kill NP.

Marantz and Kratzer conclude that ‘these facts follow if external arguments are not true arguments of their verbs.’ (Kratzer 1996:114) For example, in Kratzer’s system one could capture the variants of kill in 5a and c, in which it means, respectively, ‘cause to die’ and ‘waste a period of time’, by placing a restriction on the sole NP argument of kill:

7. If a is a time interval, then kill(a,e) = truth if e is an event of wasting a If a is animate, then kill(a,e) = truth if e is an event in which a dies …etc.

The agent argument is added via little v, as shown above. Crucially, we cannot similarly restrict the function based on the external argument, so the asymmetry follows. Furthermore, as discussed in the following section, Kratzer shows that under a traditional account in which the verb selects its external argument, the order of composition is inadequate to predict this asymmetry. It is not enough to say that the verb composes with the internal argument before the external one, since restrictions on the function can be stated with respect to either argument, irrespective of the order of composition. Nunberg et al (1994) argue that Marantz did not identify the correct generalization, which instead involves animacy rather than subjecthood or objecthood per se: animate NPs tend to occupy the open positions of idioms rather than to be incorporated into idiom chunks. They offer an explanation that follows from general properties of idiomatization. Kiparsky (1987) argues that the subject/object asymmetry is

5

only one part of a larger generalization that idiom chunks tend to correspond to constituents in lexicosemantic structure. Without taking sides on this issue, I would like to accept Marantz’s asymmetry for the sake of argument, and ask whether Kratzer’s proposal really predicts it, as claimed.

3. Does the little v hypothesis predict the asymmetry? Consider Kratzer’s (1996) attack on Bresnan 1982 and Grimshaw 1990. Bresnan and Grimshaw had suggested that Marantz’s asymmetry ‘can be equally well explained by any theory that has it that external arguments are semantically processed after internal arguments.’ (Kratzer 1996:114) Kratzer counters that

Any argument of a verb could trigger a particular interpretation of that verb. Here is a fictitious example where the highest argument does so. Suppose that the (traditional) denotation of some two-place predicate is a function f that yields the following output for individuals a in its domain:

If b is a time interval, then f(a)(b) = truth iff a exists during b. If b is a place, then f(a)(b) = truth iff a is located at b. If b is a person, then f(a)(b) = truth iff b is the legal owner of a. …etc.…

It is not true, then, that one could capture the subject/non-subject generalization simply by giving the subject a distinguished role as final argument in the semantic

6

composition of the sentence (contra Bresnan and Grimshaw).

There is no

technical obstacle to having verb meanings like the function f if external arguments are true arugments of their verbs.

If they are not, Marantz’s

generalization is expected, however. (Kratzer 1996:115-6)

Kratzer’s reasoning is sound: there is no ‘technical obstacle’ to placing semantic conditions on the final argument to be composed. But neither is there a technical obstacle to placing such conditions on the external argument under the little v hypothesis. We can recast Kratzer’s fictitious example under her own theory by substituting ‘the Agent of e’ for her ‘b’. In the following reformulation, a is the internal argument (as above) and e is the event variable:

8. If the Agent of e is a time interval, then f(a,e) = truth iff a exists during the Agent of e. If the Agent of e is a place, then f(a,e) = truth iff a is located at the Agent of e. If the Agent of e is a person, then if(a,e) = truth iff the Agent of e is the legal owner of a. …etc.…

Here the selectional restriction placed on the event argument (that its Agent have certain properties) has the effect of restricting the subject. One might counter that such restrictions are not natural.

Perhaps there is

something wrong with the particular form of the restriction.

Under Kratzer’s

7

assumptions the agent is not an argument of the verb, so our restrictions involve inherent (‘individual level’) properties (time interval, place, person) not of an argument but of a non-argument, even though they are technically stated via the event argument. Can predicates place i-level selectional restrictions on non-arguments? Apparently they can. The noun strike denotes a range of different properties depending on whether the context is a baseball game (‘pitched ball that is counted against the batter’), a game of bowling (‘the knocking down of all the pins in the first bowl of a frame’), a labor dispute (‘work cessation in support of demands upon the employer’), geology (‘direction of a horizontal line in the plane of an inclined structural feature such as a rock bed or vein’), military conflict (‘air attack upon a single group of targets’), and so on.1 But the determining context is not an argument of the noun strike, at least not under Kratzer’s assumptions about nouns. It is not necessary to show that we are forced to accept this hypothetical proposal under Kratzer’s assumptions, but only that it encounters no ‘technical obstacle’. As far as I can tell, there is no principle to rule it out. If the neodavidsonian treatment of the agent argument does not account for Marantz’s asymmetry, then what does? As noted above, there are several other proposals on the market to consider. I will leave the matter there. Next consider the vexing problem, under the little v hypothesis, of selecting the right ‘agent’ argument from the various participants in the verb-denoted event. For example, buying and selling events necessarily involve a Goal (buyer), a Source (seller),

1

These definitions are from the American Heritage Dictionary, New College Edition. 8

and a Theme (the commodity that changes hands). But the verb buy selects the Goal as its official Agent while the verb sell selects the Source as its official Agent:

9. a. Sue sold the car to John. b. John bought the car from Sue.

How do we explain this fact under the little v hypothesis? One way is to build it into the relation denoted by v:

10. If e denotes a buying event, then agent(a,e) = truth iff a is the goal of e. If e denotes a selling event, then agent(a,e) = truth iff a is the source of e. …etc.

The other, perhaps more plausible, approach is to build it into the relation denoted by the heavy verb: the buy’ relation specifies that its agent is its goal argument, while sell’ specifies that its agent is its source. If this is the strategy, then notice that any tightening of the theory to prevent selectional restrictions on the external argument (as in 8 above) must nonetheless allow buy’ and sell’ to be distinguished.

4. What is right about little v. While the proposal to introduce external agents via an abstract light verb is not the right approach in my opinion, there is some plausibility to the idea that causation or agency can be introduced supralexically, that is, in a manner that is not tied to particular

9

words. Evidence comes from three familiar observations. First, agency tends to be optional for a given verb, except where it is intrinsic to the verb meaning.2 Second, agency is often detachable from the more verb-specific roles such as Goal and Source, apparently layered over the thematic structure on an autonomous level (recall the buy/sell example 10 above). And third, agency is typically associated with the subject position. A modified version of the little v hypothesis should explain these observations. Let us return to the traditional view that the external argument, like the internal ones, is an argument of the main verb. Without spelling out the details, I will assume a lexical decomposition account along the lines of Dowty 1979, Rappaport-Hovav and Levin 1998, Jackendoff 1990, Bierwisch and Schreuder 1992, or Wünderlich 1997. Let us further assume that the arguments are ordered, perhaps as an automatic consquence of their depth of embedding within the decomposition structure, along the lines of Wünderlich 1997. In a model of this kind the causative alternation (example 1 above) can be seen as a consequence of the optionality of the causative portion of the lexical representation of some verbs (Rappaport-Hovav and Levin 1998, inter alia). It has long been observed that agency (in these cases, volitional agency) can optionally float from the external argument to the internal one, when a transitive is decausativized (Jackendoff 1990:128, inter alia). In Jackendoff’s example 11c Bill can be a pure theme, exerting no force to cause the rolling, or he can be interpreted as intentionally rolling himself. Similarly, in 12c the addition of the purpose clause tends to force the intentional reading.

2

I realize this is somewhat circular. Perhaps a better statement is just that agency is often optional. See below. 10

11. a. JohnAGENT rolled the ball down the hill. b. The ball rolled down the hill. c. Bill(AGENT) rolled down the hill.

12. a. MaryAGENT dissolved the sugar / the committee. b. The sugar dissolved. c. The committeeAGENT dissolved (in order to avoid further bloodshed).

An agentive or causer interpretation seems almost always possible for the intransitive theme, modulo pragmatics:

13. a. JohnAGENT is cooling off with an iced mojito. b. WeAGENT closed/opened early (to do the inventory). c. BillAGENT tried to sink to the bottom of the pool (by exhaling).

What this suggests is that the internal / external causation distinction (Smith 1970, McKoon and MacFarland 2000) is not a property of particular lexical items. Even the alternating verbs, which are ‘external causation’ change of state verbs in the transitive variant, allow ‘internal causation’ interpretations in the intransitive variant. More generally, agency can be optionally attributed to most subject arguments, even those of non-alternating transitives like enter and surround:

11

14. a. Smoke entered the room through the ventilation duct. b. The thiefAGENT entered the room through the ventilation duct.

15. a. Grass surrounded the tombstone. b. The historiansAGENT surrounded the tombstone.

For example, 15a is a stative depiction of the location of the grass relative to the tombstone. Call the relation denoted by surround in 15a surroundSTATIVE. This sentence is ‘causativized’ in 15b, yielding a sentence meaning roughly ‘The historians cause themselves to surroundSTATIVE the tombstone.’ Note that no argument NP is added to the clause as a result of this causativization. Non-alternating intransitives like arrive, fall, disappear, etc. also allow optional agency under the right circumstances: e.g. John always disappears when work is mentioned. A key development in the theory of verb meaning has been the move from traditional thematic role theories, in which each argument is classified into a thematic role type such as Agent, Theme, Goal, Instrument, Source, etc., to lexical decomposition theories in which a single argument can play several roles at once. This insight is made explicit in tier-based theories such as Grimshaw 1990 and Jackendoff 1990; but in a sense it is inherent in any lexical decomposition strategy where a variable can appear more than once.3 Let us tentatively posit a general (optional) rule to attribute agency to any subject argument, modulo pragmatics. This rule would be responsible for agency in a great many

3

See also Dowty 1989 for discussion. 12

cases, although with many verbs agency is inherent in the lexical meaning anyway. The question arises now as to whether the rule operates at the level of syntax, hence marking the subject; or rather at the level of argument structure, marking the thematically highest argument (which is destined to be the subject of an active sentence or by-phrase of a passive). The two alternatives are as follows:

16. Subject Rule: Optionally interpret the subject as AGENTIVE. 17. Highest Argument Rule:

Optionally interpret the highest argument in the

argument structure as AGENTIVE.

Evidence for the Highest Argument Rule comes from passives. Agency seems to stick with the highest argument, even when it is expressed in the by-phrase:

18. a. The tombstone was being surrounded by the historiansAGENT b. The tombstone was surrounded by grass.

Can our Agency-adding rule apply to the subject of the passive? Most speakers find it difficult:4

19. a. The demonstratorsi were arrested (on purpose) (?*PROi to protest the war). b. The demonstratorsi were arrested (PROARRESTER to silence themi).

4

I assume that get assigns the agent role in get-passives like The demonstratorsi got arrested on purpose (PROi to protest the war). 13

However, in some cases it seems possible:

20. a. The doctorAGENT vaccinated Mary (to protect her/*herself against rabies). b. Mary(AGENT) was vaccinated (to protect herself/her against rabies).

Perhaps both rules are present in the grammar. I will not try to settle the issue here, except to say that the Subject Rule alone is clearly inadequate, since agency on the byphrase exactly parallels that of the subjects of the corresponding active sentences.5 We have seen a difference between the behavior of anticausatives (x sank) and passives (x was sunk): for anticausatives, volition can stick with the subject, moving to the so-called theme of the intransitive; while for passives volition tends to stick with the passivized argument, adhering to the by-phrase in just the same range of cases as with subject of the corresponding active. This follows from the Highest Argument Rule, assuming that the suppressed or demoted argument of the passive remains in the highest position, while the anti-causative has only one role in its argument structure. (This is supported by the familiar constrasts with respect to instruments, purpose clauses, and so on.)6

5

Further evidence for the Highest Role Rule comes from syntactically ergative constructions in languages such as Balinese. See Wechsler and Arka 1998. 6 Korean has two passive-like constructions, the so-called hi-passive and ci-passive. Both employ verb morphology, both demote the subject argument, and both license an optional by-phrase. Interestingly, they differ in this respect: volitionality moves to the by-phrase for the ci-passive, while it goes to the new subject of the hi-passive. That is, with respect to volition, the ci-passive acts like English passive while the hi-passive acts like the anticausative. (Hong 1991: 212ff) 14

We showed above that agency can be added to a verb without increasing its valence. Can our agency-adding rule (say, the Highest Argument Rule) be enlisted to also account for the causative alternation by adding the agent argument? This idea is tempting since it might seem to get us this alternation for free. But actually it would not be free. We would need to restrict such a rule, since many verbs refuse to causativize (*The magician disappeared the rabbits; *John arrived the students; *Mary bloomed the flowers; etc.). While the Highest Argument Rule is very general, transitivization is not. In the next section we present a somewhat more precise account of agency.

5. Relative flexibility of agent role. The account sketched above does not address the problem of flexibility: what exactly counts as an agent? Agent roles are flexible to a degree that varies from verb to verb, and from language to language. In a discussion of this issue, Kiparsky (1997:4767) asks what it takes to build a house:

‘Doing all the work by yourself? Active

participation in the construction? Interested supervision? Paying for it?’ He notes that this flexibility is ‘a language-specific fact of usage, not true of the Finnish synonym, for example.’ (p. 467) Some verbs are quite rigid. English verbs denoting actions that involve the human body tend to specify their agents narrowly. For example, many speakers would accept 21a as true even if Bush himself did not open his mouth but rather let his press secretary do the talking. But 21b requires that the words came from Bush’s mouth:

15

21. a. Bush announced a new program to put the elderly back to work. b. Bush spoke about a new program to put the elderly back to work.

The verb announce is neutral with respect to the means of announcement, while speak is specific to human speech. To take another example, compare 22a and 22b as possible reports of a dance performance.

In 22a Robert had some sort of agentive role, either the role of a

choreographer, a dancer, an accompanist, or perhaps even a critic. But in 22b the role narrows to the dancer:

22. a. Robert did a dance in Berlin last week. b. Robert danced in Berlin last week.

Again, the verb dance involves actions of the body, and the role is narrowly circumscribed. Finally, let us consider buy and sell in more detail.

Sentences containing

predicate VPs of the form buy x and sell x seem to pick out the same set of events respectively, namely commercial transactions where money or some other currency is exchanged for x. Every event of buying x is an event of selling x, and vice versa. Or is it? Parsons (1990:84) argues that mutually entailing sentence pairs like 9a and 9b above, repeated here, describe different events:

16

23. a. Sue sold the car to John. b. John bought the car from Sue.

He notes, for example, that it could be true that Sue quietly sold the car to John, while false that John quietly bought the car from Sue. Also the instruments employed in buying and selling are different, as we will see below. Let posit a core transaction event that corresponding buying and selling events such as 23 hold in common, consisting of the pure commercial transaction, the exchange of possession of the commodity and payment. Following Jackendoff (1990:191, inter alia), buy and sell differ in their respective orientations towards the transaction: for buy the agent is the Goal of the commodity while for sell the agent is the Source, and vice versa for the Payment. Here are first approximations of HPSG lexical signs for buy and sell (cp. Jackendoff 1990:191 examples 20a,b):

24. Lexical entry for buy: (First Version)

" % $ ' NPg $SPR ' $COMPS NPt ,(PP[ from]s ),(PP[ for] p ) ' $ ' "INDEX e %' $ $ ' $SEM #RESTR {agent(g,e),transact(s,g,t, p,e)}&'& # 25. Lexical entry for sell: (First Version)

!

" % $ ' NPs $SPR ' $COMPS NPt ,(PP[to]g ),(PP[ for] p ) ' $ ' "INDEX e %' $ $ ' $SEM #RESTR {agent(s,e),transact(s,g,t, p,e)}&'& # 17

!

where:

26. For individuals s (‘Source’), g (‘Goal’), x, and t (‘Theme’), sum of currency p (‘Payment’) and event e, •

transact(s,g,t,p,e) is true iff: before e occurs, s owns t, and g owns p; after e occurs, g owns t, and s owns p; and e occurs by mutual consent of s and g.

•

agent(x,e) is true iff: x brings about e.

Note that the main difference between the lexical entries for buy and sell is the variable occurring as first argument of the agent relation, which in both cases is coindexed with the SUBJect NP: for buy it is the Goal (g) while for sell it is the Source (s). However, this description so far is a simplification. We have only considered cases where the subject of buy x is the new owner of x, and the subject of sell x the previous owner of x. But these verbs are more flexible than that. Suppose Brian, the secretary of the linguistics department, buys new filing cabinets for the department. Then we can truthfully say that Brian bought the filing cabinets, even if he never owned them or possessed them in any way. He is a purchasing agent acting on behalf of the new owner. Similarly, a salesperson does not normally own the items that he or she sells. Interestingly, the prepositions to and for in construction with these verbs have a range of interpretations similar to the corresponding subjects (putting aside inanimate instruments for now; see below). Thus, 27 has the four interpretations gotten by crossing the two options shown in (i) with the two shown in (ii). Likewise for 28.

18

27. John bought a car from Sue for $500. i. John is new owner / John is purchasing agent acting on behalf of new owner ii. Sue is previous owner / Sue is sales agent acting on behalf of previous owner

28. Sue sold a car to John for $500. i. John is new owner / John is purchasing agent acting on behalf of new owner ii. Sue is previous owner / Sue is sales agent acting on behalf of previous owner

What other participants can appear in subject position? Instruments can, for both verbs:

29. a. Five hundred dollars bought this car. b. Five dollars buys a lot of bagels. c. John’s hard-earned money bought him a vacation in New Zealand.

30. a. Clever advertising sold this car. b. Sex is what sells cars to college students. c. Hard work sells cars.

Not surprisingly, the instrument must be associated with the appropriate participant in the transaction, namely the Source for sell and Goal for buy:

19

31. a. (*)Five hundred dollars sold this car. (OK if money was spent by the seller, e.g. for ads) b. (*)Clever advertising bought this car. (OK if advertising was used by the buyer as payment)

The subject of sell can also express features of the commodity as in 32a or even the commodity itself as in 32b:

32. a. The whitewall tires are what sold the car. b. The car sells for fifteen thousand dollars.

Again, substituting buy for sell does not work:

33. a. *The whitewall tires are what bought the car. b. *The car buys for fifteen thousand dollars. (cp. John bought the car for fifteen thousand dollars.)

Summarizing our findings: The role played by the subject of buy must be either the Goal, or a deputy or instrument of the Goal. The role played by the subject of sell must be either the Source; a deputy or instrument of the Source; or the Theme or some property thereof. There is an intuitively obvious explanation for the breakdown between possible subjects of buy versus sell. Subjects of buy are all participants under the control of the

20

Goal, while subjects of sell are affiliated with the Source. We can think of these two groups as teams responsible for bringing the two sides into the commercial transaction:

34. Buyer’s team: Goal (new owner) Buyer (Goal or deputy thereof) InstrumentB (currency exchanged for Theme)

35. Seller’s team: Source (previous owner) Salesperson (Source or deputy thereof) InstrumentS (advertising, etc.) Theme (commodity or properties thereof)

The range of roles that can be played by the subject of buy is extended from the Goal to any member of the Goal’s entourage, so to speak; mutatis mutandis for sell and Source. My placement of the Theme on the Seller’s team is debatable; this may not be the right way to look at the middle construction (32b). But this view does, at the very least, explain the fact that sell but not buy allows the middle, as noted by Smith (1970:104), but heretofore unexplained, as far as I know. To account for these facts, let us formalize the respective teams as events that cause the transaction event.

21

36. Lexical entry for buy: (Final Version)

" % $ ' NPx $SPR ' $COMPS NPt ,(PP[ from]s ),(PP[ for] p ) ' $ ' "INDEX e %' $ $ ' $SEM #RESTR {cause(e',e),control(g, x,e'),transact(s,g,t, p,e)}&'& # 37. Lexical entry for sell: (Final Version)

!

" % $ ' NPx $SPR ' $COMPS NPt ,(PP[to]g ),(PP[ for] p ) ' $ ' "INDEX e %' $ $ ' $SEM #RESTR {cause(e',e),control(s, x,e'),transact(s,g,t, p,e)}&'& #

! Where:

38. For individuals s, g, x, y and t, sum of currency p and events e and e’, •

control(y,x,e’) is true iff: y controls x in bringing about e’

•

cause(e’,e) is true iff: e’ causes e

Instruments and deputies (indicated by the variable x) are under the control of the Goal (for buy) or Source (for sell), so they can appear as subjects of their respective verbs. To account for the simple cases where the subject is the Goal or Source herself, acting on her own behalf, we can assume a principle to the effect that the control relation is necessarily reflexive on the first two arguments:

22

39. Principle of Self-Control An individual who controls anyone, (also) controls himself/herself/itself. ∀x∀e [[∃y.control(x,y,e)] → control(x,x,e)]

This account predicts that the argument of transact who is enlisted by the verb to be the controller of the causing event— the Goal or Source, depending on the verb— determines how far back one may go in the causal chain in picking a subject. This follows since the subject argument must be under the control of that Goal or Source. To understand this claim, recall by way of contrast the discussion of build a house above: as we go back up the causal chain that led to the construction of the house, there seems to be no definite limit on who can claim credit for the event. In contrast to buying a house, which results in someone owning the house, building a house has no determinate effect on any individual except the house. Hence the semantics of the verb build are approximately the following (note the neodavisonian treatment of the agent):

40. build: λyλxλe∃e’[ cause(e’,e) ∧ agent(x,e’) ∧ build(y,e) ]

(If we want more uniform representations across different verbs, perhaps we can think of agent(x,e’) as a shorthand for control(x,x,e’).) This places no restriction on how the subject argument causes the house to be built. In contrast, our representation for buy predicts that the subject is under the control (perhaps self-control) of the Goal argument. This predicts that merely acting as a cause of buying is insufficient to qualify an individual as subject of buy: the subject NP must specifically denote a deputy acting on

23

behalf of the Goal. For example, suppose that Mary orders her secretary John to buy a new tie for himself (with his own money), and he does so. Then we can truthfully say 41a but not 41b:

41. a. Mary caused a new tie to be bought. b. Mary bought a new tie.

Exactly the same applies to sell, as the reader can verify by substituting sell for buy in the scenario and the example sentences. Verbs like dance and speak (21-22) also have this property, as predicted, since nothing can causally mediate between the dancer and the dance.7 There seems to be no generalized notion of Causer or Agent that will correctly pick out the subject roles of different verbs, without explicit reference to verb-specific semantic roles. The challenge for the little v hypothesis is to account for this fact. What is right about the little v hypothesis and its predecessors is the idea of a generalized grammatical source for agency or causation, apart from the inherent lexical meanings of particular verbs. The passive facts show that the level at which this optional agency applies is argument structure, not phrase structure. What is wrong is tying this to valence increasing or the selection or assignment of thematic roles.8

7

However, this entire discussion applies to English, but not necessarily other languages. Recall Kiparsky’s observation about English versus Finnish ‘build.’ 8 Pylkkänen (1999) put forth a similar idea within the ‘little v’ framework, proposing a system with three functional heads, CAUSE (for causation but not valence), APPL (applicative, for adding the causee), and the external argument marking θEXT (cp. little v). She hypothesizes that languages vary in how these features are ‘bundled’, claiming for 24

References Bierwisch, Manfred and Robert Schreuder (1992). ‘From concepts to lexical items.’ Cognition 42. 23-60. Bresnan, Joan (1982). ‘The passive in lexical theory.’ In Joan Bresnan (ed.) The Mental Representation of Grammatical Relations. Cambridge, MA: MIT Press. Chomsky, Noam (1970). ‘Remarks on Nominalization.’ In Roderick A. Jacobs and Peter S. Rosenbaum (eds.) Readings in English Transformational Grammar. Waltham, Massachusetts: Ginn and Co. 184-221. Chomsky, Noam (1995). ‘Categories and transformations.’ In Noam Chomsky, The Minimalist Program, Cambridge, Massachusetts: MIT Press. 219-394. Dowty, David (1979). Word Meaning and Montague Grammar. Dordrecht: Reidel. Dowty, David (1989). ‘On the semantic content of the notion “thematic role”.’ In Gennaro Chierchia, Barbara Partee and Ray Turner (eds.) Properties, Types, and Meaning— Volume II: Semantic Issues. Dordrecht: Reidel. 69-129. Grimshaw, Jane (1990). Argument Structure. Cambridge, Massachusetts: MIT Press. Hong, Ki-Sun (1991). Argument Selection and Case Marking in Korean. Doctoral dissertation, Stanford Linguistics Department.

example that causation is divorced from the external argument in some languages, but not in English. But as we saw above, even in English adding causation does not necessarily add an argument. That is, I believe that Pylkkänen had the right idea but did not go far enough. 25

Horvath, Julia and Tal Siloni (2003). ‘Against the Little-v Hypothesis.’ In Yehuda N. Falk (ed.) Proceedings of the Israeli Association for Theoretical Linguistics 19, Ben Gurion University of the Negev, Israel. Jackendoff, Ray. 1990. Semantic Structures. MIT Press, Cambridge, MA. Kiparsky, Paul (1987) Morphology and grammatical relations. Manuscript, Stanford. Kiparsky, Paul (1997) ‘Remarks on denominal verbs.’ In Alex Alsina, Joan Bresnan, and Peter Sells (eds.) Complex Predicates. Stanford: CSLI Publications. 473-499. Kratzer, Angelika (1996). ‘Severing the external argument from its verb.’ In J. Rooryck and L. Zaring (eds.) Phrase Structure and the Lexicon. Dordrecht: Kluwer Academic Publishers. 109-137. Kratzer, Angelika (2002). The Event Argument and the Semantics of Verbs. Draft, http://semanticsarchive.net/ Krifka, Manfred (1999). ‘Manner in dative alternation.’ In Sonya Bird, Andrew Carnie, Jason D. Haugen, and Peter Norquest (eds.) Proceedings of the West Coast Conference on Formal Linguistics 18, Somerville, Massachusetts: Cascadilla Press. Lakoff, George (1970). Irregularity in Syntax. New York: Holt, Rinehart, and Winston. (Indiana University doctoral dissertation, 1965). Marantz, Alec (1984). On the Nature of Grammatical Relations. Cambridge MA: MIT Press. McCawley, James (1968). ‘Lexical insertion in a transformational grammar without deep structure’ Proceedings of the Chicago Linguistic Society 4, 71-80.

26

McKoon, Gail and Talke MacFarland (2000). Externally and internally caused change of state verbs. Language 76.4, 833-858. Nunberg, Geoffrey, Ivan A. Sag, and Thomas Wasow (1994). ‘Idioms.’ Language 70.3, 491-538. Parsons, Terence (1990). Events in the Semantics of English: A Study in Subatomic Semantics. Cambridge MA: MIT Press. Pylkkänen, Liina (1999). ‘Causation and external arguments.’ Papers from the UPenn/MIT Roundtable on the Lexicon, MITWPiL 35, 161-183. Rappaport-Hovav, Malka and Beth Levin (1998). ‘Building verb meanings.’ In M. Butt and W. Geuder (eds.) The Projection of Arguments: Lexical and Compositional Factors. Stanford: CSLI Publications. 97-134. Searle, John R. 1984. Intentionality––An Essay in the Philosophy of Mind. Cambridge University Press, Cambridge. Smith, Carlota (1970). ‘Jespersen’s ‘Move and Change’ class and causative verbs in English.’ In M. Ali Jazayery, Edgar Polomé, and Werner Winter (eds.) Lingusitic and Literary Studies in Honor of Archibald A. Hill, Vol. II: Descriptive Linguistics. The Hague: Mouton Publishers. 101-110. Wechsler, Stephen and I Wayan Arka (1998). ‘Syntactic Ergativity in Balinese: an Argument Structure Based Theory.’ Natural Language and Linguistic Theory 16:387-441. Wunderlich, Dieter (1997). ‘Cause and the structure of verbs’ Linguistic Inquiry 28.1. 2768.

27