Ling 522 Dowty: weighted argument structure linking cont Feb. 13, 2002

pages 566-end

13. The question Dowty proposes to address: Argument selection, elsewhere called the "mapping problem"…"what principles languages use to determine, for each argument of an n-place relation that is denoted by a predicate, which argument (intuitively speaking) can be expressed by which grammatical relation."

That is, what semantic properties tell you what's going to be the SUBJECT, what the OBJECT, etc.

14. Some assertions about How to Proceed:

(a)We should only consider "event-dependent" notions when constructing theories of thematic roles (and, even abstracting away from this specific proposal, any theories of linking). Especially, perspective-dependent considerations such as "Figure" and "Ground" should be eliminated from our box of basics.

(b) We should consider those notions that are demonstrably relevant to argument selection (somewhere in some language).

(c) One such notion is "Incremental Theme", where an argument measures-out the event through a series of homomorphic part-whole relationships.

15. An argument for (14a):

(i) In an adequate linguistic description, grater relative degrees of connectedness to previous discourse, givenness, etc., must be explicitly specified as a semantic correlate of grammatical subject denotations (see examples in 16 below)

(ii) All putative instances of perspective-dependent thematic roles and other 'perspective-indicating' lexical entailments of words can be shown to be instances of (i) when properly analysed.

(iii) Therefore, perspective-dependent thematic roles are superfluous and unnecessary, and all roles are event-dependent in meaning.

16. Consider, e.g., a proposal where a verb assigns a "Figure" and a "Ground' theta role to its external and internal argument, respectively. The facts in a-d, below work right.

What happenend to the tree?

a. ??The truck hit the tree.

b. The truck hit it.

c. The tree was hit by the truck (perhaps ? due to lack of anaphor).

d. It was hit by the truck

What happened to the truck?

e. The truck hit the tree. (perhaps ? due to lack of anaphor).

f. It hit the tree.

g. ??The tree was hit by the truck.

h. ??The tree was hit by it.

If the "lexical subject" of "hit" were bearing some sort of "Figure" role here, then (h) ought to be natural. The fact that it isn't demonstrates that it's the notion of syntactic subject that is linked to the perspective-implicating role "Figure" (cf. the naturalness of 16d). Similarly for If I was Bill, I'd take the job vs. #If Bill was me, I'd take the job.

17. Incremental Themes: 14(b)

Dowty (and most people who work on telicity) wants to accord special status to a particular kind of argument that has not been accorded theta-role status up til now, which he calls "Incremental Theme".

Recall that themes are generally objects which undergo changes of state or location. Incremental themes are cases where the degree of state-change that the object has undergone corresponds with the degree of doneness of the event — Incremental Themes "measure-out" a telic event, in Tenny's terms, or allow an "event-object homomorphism" in Dowty's (Krifka's) terms. Consider:

a. John ate the apple.

b. John mowed the lawn. ß Incremental Themes because you can tell

how far done the event of mowing the

lawn is by looking at the lawn, ditto

for apple

c. John raised the thermostat ß theme but not Incremental Theme

because you can't tell how far done

the event of raising the thermostat is

just by looking at the thermostat --

in principle, it's potentially always

further raisable and the event's

over when John decides it is.

Tenny calls this "measuring-out" because the objects are playing the same role in these sentences as true measure phrases are in sentences like John ran a mile. (Although note: not in stative measure predicates like weigh 3 pounds!)

18. With incremental themes, there's a homomorphism between the degree of affectedness of the argument and the doneness of the event. Subparts of the argument correspond to subparts of the event:

e_{eating the apple} = e_i + e_j +…. e_n each of which are examples of e_{eating a part of the apple}

x_apple = x_i + x_j +… x_n each of which are examples of x_{a part of the apple}

(of course you get Zeno-like subdivisions, but what's crucial is that you get them for both domains, the event domain and the object domain).

19. There's a few trickinesses with incremental themes. The first is very relevant to (indeed, homomorphic with!) the questions that Massimo will raise later:

a) The problem with plurals. Consider:

i) John visited Atlanta.

ii) John visited 25 cities.

iii) Three hundred people visited Atlanta.

iv) Three hundred people visited 25 cities

Note that while i) does not involve an Incremental Theme, in (ii) the degree of doneness of the event is measured-out by the # of cities John's visited, in (iii) by the # of people who've visited Atlanta (on the distributed plural interpretation, not the collective, of course) and in (iv) by both.

Dowty says: "Apparently a quantified NP argument along with almost any distributive telic verb (and some collective telics) can be understood homomorphically, because this combination generates reference to a set of individual events, one for each entity referred to by the quantifier. It is the 'meta-event' combining all these individual events that has subparts corresponding to the individual entities picked out by the quantifier NP [i.e. the cities, people, or whatever] "

b) The problem with PATHS: while by definition events with real incremental themes are telic, not all telics have obvious incremental themes. Consider:

i) Mary walked to New York in a week

ii) Bill pushed the cart to New York in a week

Here, it's Mary's or the cart's progress along the path whose endpoint is across the desert or to New York that truly measures-out the event; if anything is an incremental Theme, the path is. Sometimes the path is explicitly realized, cf.

iii) Mary crossed the desert in a week

iv) Mary hiked the Appalachian trail in 6 months

but more often it's not. Nonetheless, there is an appropriate homomorphism between path sub-parts and event sub-parts that meets the criteria for an Incremental Theme.

c) The question about subparts of motion-verb subjects:

i) John entered the icy water in 4 minutes ß measured-out by

John's body

ii) The iceberg took several minutes to pierce the ship's hull

to this depth.

iii) The winning turtle crossed the finish line in 42 seconds.

iv) The crowd exited the auditorium (in 21 minutes)

Here, you get a kind of distributed reading over the singular, where the homomorphism is defined by subparts of the singular subject (in i-iii; iv just shows the plural distribution effect again, only with a semantic plural, not a syntactic plural). These are also Incremental Themes.

20. fn. 15: "Some will suggest that the subjects of the unaccusatives are derived by Unaccusative Advancement froom underlying direct objects, hence that at that level they conform to the claim that all Incremental Themes are direct objects. This is less plausible for "John entered the water gradually", which has a visible, independent direct object. Even here, of course, one can imagine a suggestion that the water originates as an underlying oblique and is advanced to direct object after John is advanced from direct object to subject. At that point, of course, one would have a right to ask whether the invariant association of Incremental Theme with syntactic direct object still had any empirical content or had been elevated from empirical hypothesis to methodological assumption, i.e. that one was in actuality prepared to postulate any syntactic abstractness necessary to maintain a uniform semantic association with a certain syntactic position…."

21. The Actual Proposal: Entailments relevant to mapping of arguments

(a) PROTO-AGENTS

i. volitional involvement in the event or state

ii. sentience and/or perception

iii. causing an event or change of state in another participant

iv. movement relative to the position of another participant

(v. exists independently of the event named by the verb).

(b) PROTO-PATIENTS

i. undergoes change of state

ii. incremental theme

iii. causally affected by another participant

iv. stationary relative to movement of another participant

(v. does not exist independently of the event, or not at all).

22. Dowty maintains that you can identify each of these separately, i.e. that there are verbs whose arguments exhibit just one of each of the above properties.

ia John is ignoring Mary / John is being polite to Bill

What he did was not eat for two days. /John fasted for 2 days.

iia John knows/believes.is disappointed at the statement

John sees/fears Mary

iiia His loneliness causes his unhappiness

Unemployment causes delinquency

iva The rolling tumbleweed passed the rock

The bullet overtook the arrow

He accidentally fell

va John needs a new car

ib John made a mistake / erased the error

John moved the rock

iib John crossed the driveway

John filled the glass with water

iiib Smoking causes cancer

ivb The bullet entered the target / overtook the arrow

vb John built a house / erased the error

This situation constitutes a major dilemma for us

John needs a car/seeks a unicorn/lacks money

23. I have problems with a number of these. Consider first (ai):

(ai) counterexamples: The dike retains the Atlantic No volition, no movement

John suppressed his burp Volition, no movement -- but

surely not no action!

John pushed ineffectually at the stalled car same as above

Surely refraining from eating or from saying impolite things is resistance against a pressure -- a metaphorical pressure, but Dowty isn't against appealing to metaphor in cases like John gave Bill a pep talk or a cold. Doesn't much matter in this case, but I just want to point out that isolating volition from agentivity is not easily done. (Actually I think it can be done but just that Dowty hasn't done it -- see my papers on have on my website if interested)

(aii and a,b iv): (When Uhura powered the transporter) Scotty gradually appeared

The image gradually appeared as the photo developed.

Again, here's something that doesn't seem to work quite right, depending on whether or not you think "appearing" is motion relative to other participants (such as?). But basically in these sentences the only criterion that seems relevant is Incremental Theme, which these subjects obviously are. Basically, I think that there's way too many contingencies about real-world situations for these kinds of entailments to be universally useful. For more discussion, see the Problems section below. For now, let's move on to seeing the system in action.

24. (a) ARGUMENT SELECTION PRINCIPLE: In predicates with grammatical subject and object, the argument for which the predicate entails the greatest number of Proto-Agent properties will be lexicalized as the subject of the predicate; the argument having the greatest number of Proto-Patient entailments will be lexicalized as the direct object.

(b) COROLLARY 1: If two arguments of a relation have (approximately [!!! -ed]) equal numbers of entailed PROTO-AGENT and PROTO-PATIENT properties, then either or both may be lexicalized as the subject (and similarly for objects). [this will allow for alternations]

(c) COROLLARY 2: With a three-place predicate, the nonsubject having the greater number of entailed Proto-Patient properties will be lexicalized as the direct object and the nonsubject argument having fewer entailed Proto-Patient properties will be lexicalized as an oblique or prepositional object (and if two nonsubject arguments have approximately equal numbers of entailed P-Patient properties, either or both may be lexicalized as direct object). [again the second clause is what's going to allow alternations]

(d) NONDISCRETENESS: Proto-roles, obviously, do not classify arguments exhaustively (some arguments have neither role) or uniquely (some may share the same role) or discretely (some arguments could qualify partially or equally for both proto-roles).

25. This is based on the notion of semantic entailment: "A set of entailments of a group of predicates with respect to one of the arguments of each". This allows us to group verbs into classes, where, say, class 1 involves entailment of property (10ai and 10aii) for its subjects, but none of the others (10aiii-iv).

26. Examples of how the system works to predict alternations (NOTE: Dowty is assuming a monostratal syntax; for him, all argument-structure alternations must be controlled from the lexicon, by lexical rules. This is of particular interest in his discussion of the Unaccusative Hypothesis, since he can't appeal to movement as an explanation of the properties of unaccusative verbs):

(a) buy/sell: both merchant and customer must be volitional in an event of this type; they aren't distinguished by any relevant argument selectional properties, above. Hence there are two different lexicalizations of this event, one with one argument as subject, one with the other.

(b) Psych predicates: like/please, fear/frighten, suppose/seems to, regard/strike as…: The "stimulus" argument and the "experiencer" argument each have some Agenty properties (causation in the first instance, sentience in the second). As they're tied, again, alternate lexicalizations of the events are possible.

(a) "partially symmetric interactive predicates"

i. The drunk embraced the lamppost

ii. #The lamppost and the drunk embraced.

iii. The truck collided with the lamppost

iv. The truck and the lamppost collided (both moving)

(b) spray/load alternations

i. Mary loaded hay onto the truck

ii. Mary loaded the truck with hay

Either "hay" or "truck" can be incremental themes, and hence are tied as Proto-Patients.

(c) hit/whack alternations, vs. no such break alternations:

i. John hit the fence with the stick

ii. John hit the stick against the fence.

iii. John broke the fence with the stick.

iv. *John broke the stick against the fence (on a reading equal to iii.)

In hit verbs, both Instrument and Patient are tied for Proto-Patient possibilities (see problem (f) below, though), while with break verbs, the Patient undergoes a change of state, forcing it to map to direct object in preference to the Instrument.

Note: he argues that a superficially similar seeming class of non-alternators (smash, wallop, club, flog… ) that require the Patient-Object frame and ones that require the Instrument-Object frame (dash, throw, lob, loft, bounce…) each entail a change-of-state in the Patient or Instrument, respectively, unlike the hit class above, and hence don't alternate (like the break classs).

27. Problems

(a) chase/flees

(b) What's the difference between "alternative lexicalizations" and "different meanings for the same verb that are recorded as independent items in the lexicon, perhaps connected by lexical rules"? I.e., why aren't buy and sell the same verb, like spray in each of its frames?

(c) Object-experiencer verbs (w/ "stimulus" subject) are all eventive:

i. What happened to Mary was that the birthday party surprised her.

ii. *What happened to Mary was that she liked the birthday party.

These are perhaps not a problem, given that the Experiencer argument in (i) but not (ii) is undergoing a change of state. But see (d) below:

(d) "Nonstandard lexicalizations": receive, inherit, come into, undergo, sustain, suffer from, submit to, succumb to, and tolerate, get-passives. "sentience might in some cases be a sufficient entailment to license an argument's lexicalization as subject, no matter how many P-patient entailments it has" or "historical semantic drift can result in a predicate that violates selection principles." The former might take care of the caused-motion cases mentioned below, but it wreaks havoc on the eventive psych-verb generalization

(e) In the spray/load verbs, while either argument can be the Incremental Theme and undergo a change of state, only one argument moves: the other is always "stationary relative to movement of another participant", a Proto-Patient property. Hence, "the truck" ought to be the preferred object. Here we see why he hedges ("approximately equal") above! In fact, cross-linguistically, as we saw last time, it's the "ground-object" frame that is rarer, when Dowty predicts the opposite. We can see a reflex of this property in (i) and (ii) below:

i. Paint sprayed the wall.

ii. *The wall sprayed with paint.

Why is it relevant for subjecthood/objecthood here, but not for direct vs oblique below?

(Also, his answer to the "why don't fill/cover verbs alternate" is pretty much unsatisfactory: "One might speculate that the existence of the morphologically related adjective full and noun cover, both entailing complete occupancy/coverage of the space in question, help maintain the restriction of these verb meanings to a locative Incremental Theme, and have prevented the child's temporary innovation from surviving into the adult language, over the many centuries these forms have existed in English".) He doesn't even discuss the nonalternating figure-object verbs: pour, etc.

(f) Same problem as for spray/load above: in any "hitting" or "breaking" scenario, the Instrument always moves, and the Patient is always "stationary relative to movement of another participant"; hence, if all the criteria are weighted equally, the Patient should always be the object, in the hit cases as well as the break cases.

An observation: when the Instrument is a body part, in the Instrument-Object frame, one gets an "accidental" implication, although not in the Patient-Object frame:

i. John hit his hand against the fence. (seems like an accident)

ii. John hit the fence with his hand.

iii. John hit his head on/against the post. (accident)

iv. John hit the post with his head (not so much an accident?)

(g) Dowty gives i. below as an example of a movement-related PROTO-AGENT property, hence explaining the association of the moving thing with subject position:

i. The bullet overtook the arrow

ii. Water filled the boat.

Presumably he would wish to ascribe the subjecthood of the moving thing in (iii) and (iv) to this constraint as well:

iii. The ball rolled down the hill.

iv. The remote-control car raced around the track

However, compare (v) and (vi)

v. John rolled the ball down the hill.

vi. Sue raced the remote-control car around the track.

Here, causation is the only Proto-Agent property of the subject (not sentience: compare the wind rolled the tumbleweed down the hill), while the object has another Proto-Agent property. Why aren't they tied? (Perhaps in these cases the subject moves relative to the position of another participant? What then about (vi)?)

(h) Unaccusativity as a semantic, rather than syntactic, phenomenon. (Syntactic correlates of unaccusativity, including ne-cliticization, auxiliary selection, lack of himself+resultative construction, etc. etc., do not reflect a derivational fact, but are correlates of the semantic cluster of unnacusative properties mapping to the syntax (special linking principles for unaccusatives).