The Meaning of ECM Keir Moulton TOM - 6 March 2010

1 The syntactic construction of meaning A trusim: word meaning and syntactic structure interact, via the compositional semantics. (1)

Double Object Alternations: double object implies possession Green (1974) a. Send a letter to Canada vs. Send Canada a letter. b. I threw the ball to first base vs. I threw first base a ball.

Verb meaning determines, in part, the syntactic expression of arguments (see Levin (1993) and many others). Or does the syntactic context shape word meaning? Verbs are ”squishy” and can be ”pushed” into the double object frame and so acquire that construction’s distinctive meaning. (2)

Lexical Squishiness a. Sweep John a ball ≈ cause John to have the ball by sweeping b. Nonce Nancy a noodle ≈ cause Nancy to have the noodle by noncing

Modern Decompoisiton: Factor out what looks like part of the word meaning into meaningful functional structure. 1

• argument structure/theta roles (Hale and Keyser (1993), Kratzer (1996), Pylkk¨anen (2002), Beck and Johnson (2004)) • lexical aspect (Ramchand (2008), Travis (2009)) • modifiers (Morzycki (2005b)) • flavors of modality (Hacquard (2006)) • decompose everything! Borer (2005). 1.1 ECM: A Curious Case of Constructed Meaning Borkin (1984) uncovered some subtle effects that ECM has on the meaning of the embedding verb. (3)

The doctor has told Sam that his cat has an illness a. . . . but Sam won’t believe that she is sick. b. # . . . but Sam won’t believe her to be sick. (adapted from Borkin 1984:83)

Borkin (1984:83): ”[ECM] describe[s] a self-initiated, original perception or opinion rather than the recognition of the truth of a proposition formed by someone else.” 1.2 Outline • probe Borkin’s intuitions first with less subtle cases, where ECM changes meaning dramatically (and these often involve ”fuzzy” data). • a unifying characteristic of ECM is that it forces verbs to report beliefs • Proposal: decompose! a functional head in ECM constructions contributes the belief component • we will further probe the intuition that ECM must report beliefs that are based on inferences; speculation that this is the attitude verb analog of the evidential requirements of epistemic modals (Karttunen (1972), von Fintel and Gillies (2010)) Not discussed today: the syntax-semantics of ECM - how exactly these particular structures correspond to these particular meanings (ask me about it). 2

2 When ECM means believe There are a number of verbs that do not seem to inherently ascribe belief but do so when they select ECM complements.

2.1 Consider Consider+DP: the object position is extensional (i.e. there is no implication of belief). Substitution of extensional equivalents will always guarantee truth. (4)

a. John considered Mary’s new car. (. . . He thought it was cheap) b. Mary’s new car was expensive. c. ∴ John considered Mary’s new expensive car.

ECM-taking consider, however, reports a belief.1 (7)

a. John considered the car to be to be expensive . . .#but he didn’t think it was expensive. b. John considered Mary’s expensive car, . . .ok but he didn’t think it was expensive.

2.2 Hear Hear with a finite clause doesn’t imply the hearer believes the content of the complement clause (8a). But with ECM it does (8b). (8)

Hear

1 Consider

can, somewhat marginally for some speakers (see Postal (1974), take CP complements:

(5)

Particularly when one considers that some images were crossed out. (www.lotsofessays.com/essay search/considers deep.htm)

(6)

...when one considers the fact that some images were crossed out.

These report much what DP-taking consider does: thinking about some fact.

3

a. Martha heard that Fred is mocking her, but she doesn’t believe that he is. b. Martha heard Fred to be mocking her, #but she doesn’t believe that he is. Instead, hear+ECM reports a belief about something heard: (9)

Fred was imitating Martha’s Boston accent. a. Martha heard Fred to be mocking her, but I thought he was being quite sweet about her dialect. b. # Martha heard that Fred was mocking her, but I thought he was being quite sweet about her dialect.

The belief component is unique to ECM. Moreover, hearsay is difficult to report with hear+ECM, but this is just what hear+finite clauses report. (10) [Discussing a performance after the fact] I know you thought she played well, but . . . a. I heard her to be out of tune. b. I heard that she was out of tune. (11) [Discussing an on-going concurrent performance we can’t hear] How are things going on in the other room at Mary’s recital? a. I hear that she is playing out of tune. b. # I hear her to be playing out of tune. In fact, in (11) it feels like you just actually need to hear Mary playing to use ECM!

Further evidence that ECM can’t report ”hearsay”: it doesn’t tolerate a source argument like hearsay hear: (12) A source Argument for hear

4

a. b.

It was from the newspaper that I heard that she was playing out of tune. ? It was from the newspaper that I heard her to be playing out of tune.

A generalization? ECM turns a non-belief verb into a belief verb

3 The fuzzy data: pushing verbs into ECM The semantic effect of ECM helps us understand some of the more ”fuzzy” data. Some verbs are weird with (in situ)2 ECM. (13)

a. *? The doctors said John to be sick. b. *? The doctors wagered John to be sick.

Intuitions about ECM are unstable. Some quotes: ”there is an unfortunate degree of fuzziness in people’s judgments concerning ECM” (Pesetsky 1991) ”we do not find Postal’s facts [citing Postal’s work on ECM] . . . to be at all clear; that is we accept ECM with almost all the cited verbs.” (Kayne (1984)p121,ft 15) There are attested examples of say and wager with ECM. The ones that strike me as especially good are those with a modal, such as would, in the matrix clause. 2 However, ECM is possible as long as the ECM’d item is extracted (e.g. by passive, as in John was said/wagered to be sick) Postal (1974). These are known, since Pesetsky (1991) as Wager-class verbs. What makes these verbs good with ECM (see below in (17) is not needed for the extraction cases. That is, John was said by his doctors to be sick doesn’t imply belief on the part of the doctors.

5

(14) Attested Examples 3 a. I have seen about 10 movies from Colin and I would say him to be a height of between 5ft9.75” . . . and 5ft10” . . . b. After seeing a lot of his vids I would say him to be a lil over 9NBP by a little over 6. c. Some would say her to be more than handsome, if not pretty. d. I’d wager him to be on the wrong side of fifty and youre young enough to be his daughter. When say and wager take finite clauses, they do not necessarily imply belief (15). These verbs can be used, however, to report beliefs (16).4 (15)

a. If I were you, I’d (just) say that he’s about 5 feet . . . even though we know he’s not. b. John wagered that Sheila would win . . . even though he thought she wouldn’t.

(16) How tall do you think John is? a. I’d say that he is about 5 foot. . . . # but I know he’s not. b. I’d wager that he is about 5 foot. . . . # but I know he’s not. But with ECM, these verbs must ascribe belief. (17)

a. I’d say him to be about 5 foot. . . . # but I know he’s not. b. I’d wager him to be about 5 foot. . . . # but I know he’s not.

Once again . . . ECM turns a non-belief verb into a belief verb 3 Found on the web at: http://www.celebheights.com/s/Colin-Farrell-96.html; http://www.mattersofsize.com/forum/showthread.php?t=11872; http://forums.worldofwarcraft.com/thread.html?topicId=14990770355sid 4 Modal would helps (although they are not necessary) to bring out this use of say, wager.

6

4 Decomposition: the belief component is in the complement 4.1 De Re Belief and ECM Taking hear+ECM as a representative case, what ECM reports is a de re belief about what was heard. (18) Rodney heard Fred to be mocking his accent. ≈ Rodney believes of what he heard that is was Fred mocking his accent De Re belief involves, following Kaplan (1968) and Lewis (1979), 3 arguments: a subject (x), a res argument (y) and a property argument a subject (P). (19) De Re Belief: quick version J Believe(P)(y)(x)(w) K = 1 iff x believes P de re of y in w. To say what ”believes P de re of y” really means, we need the more sufficient formulation in (20). It involves a suitable ”Acquaintance Relation” that holds between the attitude holder and the res (Lewis (1979)). (20) J Believe(P)(y)(x)(w) K = 1 iff a. ∃R such that R(y)(x)(w) b. R is a suitable Acquaintance Relation c. ∀ hx’,w’i(hx’,w’i ∈ Dox(x)(w) → P(ιz.R(x’)(z)(w’))(w’))

7

4.2 Proposal Hear does not carry any modal quantification. The belief component is in a separate head we’ll call B, which is just like De Re belief in (19)/(20). Its ”meaning” is defined semi-informally below. (21) J B K = λP.λy.λx.λw. x believes P de re of y in w The embedding verb hear describes a hearing relation, whatever that might be (in most cases, it looks just like it describes an auditory perception, but see Appendix A). (22) J hear K = λy.λx.λw. hear(y)(x)(w) Combine hear and B to form the ”compound verb” hear-believe. Hear is a literally de re believe plus hear, where the object of hear is the res. (23) Rodney heard Fred to be mocking his accent. ≈ Rodney hears something s and believes de re of that something that is is Fred mocking his accent

4.3 What is the res? We don’t want the ECM subject to be the res argument, because the ECM’d DP can take scope below the belief operator, something res arguments by definition do not do: (24) Rodney hears a violin to be playing . . .but it was really a cello.

8

This does not have to mean that Rodney heard of an actual violin that it was playing. Instead, Rodney hears some eventuality and believes of it that it is an eventuality of a violin playing. Proposal: The res is an eventuality (state or event proper). That means the property argument is a property of eventualities.5 (25) J B K = λPev,st .λsev .λxe .λws . x believes P de re of s in w 4.4 Some Compositional Details B, after its property complement is saturated, combines with hear via whole-sale predicate modification: both are of type hevhehs,tiii. VP: λs.λx.λw. hear(s)(x)(w) & B’(Q)(s)(x)(w) (by PM)

(26)

V: λs.λx.λw. hear(s)(x)(w)

BP: λs.λx.λw: B’(Q)(s)(x)(w) B: λP.λs.λx.λw: B’(P)(s)(x)(w)

Q (a property)

With some judicious ∃-closure of the res argument (s above), we can add the subject:

VP: λw.∃s [hear(s)(subj)(w) & B’(Q)(s)(subj)(w)]

subj

λx.λw. ∃s [ hear(s)(x)(w) & B’(Q)(s)(x)(w)]

∃ VP: λs.λx.λw. hear(s)(x)(w) & B’(Q)(s)(x)(w)]

Computing the pieces (see Appendix B for the actual compositions of the meanings; here is the ”quick version” for de re belief): 5s

is variable over eventualities and ev is their type

9

(27) Rodney heard Fred to be mocking him = ∃s [hear(s)(Rodney)(wo ) & Rodney believes in wo de re of s that s is an eventuality of Fred-mocking-him. ≈ Rodney hears s and believes s is Fred mocking him. Summary • The belief component is in the complement • The matrix verb provides a res argument • The structure itself (via B) provides the interpretation a` la the Constructed Meaning Approach outlined above. • the syntax of B must condition ECM (ask me about this if you’re interested in the syntax)

4.5 Say and Wager verbs These verbs introduce res arguments, and so are suitable things for B to combine with (and this ensures that they are belief-ascribing): (28) Rodney would say of John that he is about 5 ft tall. (29) I would wager of/about Mary that she is no older than fifty. How much decomposition? I leave aside whether the verb roots say and wager have any modal properties themselves as well as B. It is consistent with this account that they do. Not any verb, however, that describes just any relation will work with ECM. They need to be verbs that can generally describe belief states. See Moulton (2009) for how to do this by making reference to states with propositional content. This will allow us to make B accept verbs like say and hear and believe but not verbs like sit.

10

5 ECM beliefs involve inference ECM has the effect of ”creating” a belief verb from other classes of verbs. One class of predicates, verbs that are factive with finite complements, are non-factive belief ascriptions with ECM. (Kiparsky and Kiparsky (1970))6 (32) See a. Martha saw that John was driving too fast, #but he wasn’t. b. Martha saw Fred to be driving too fast, but he wasn’t. (33) Remember a. Martha remembered that John was bald, #but he wasn’t. b. Martha remembered John to be bald, but he wasn’t. (after Kiparksy and Kiparsky 1970) (34) Understand a. Martha understood that Fred was mocking her, #but he wasn’t. b. Martha understood Fred to be mocking her, but he wasn’t. (35) Know a. She never knew that Fred was a liar. #So he must not have been. b. She never knew Fred to be a liar. So he must not have been. 6 Note that there are sometimes ways of making know not factive. I find a certain pitch accent on it helpful in canceling the factive ”presupposition.” The ”canceling” or ”defeasibility” of factive presuppositions is an old topic. Chierchia and McConnell-Ginet (2000) cite some examples: the factive presupposition doesn’t project out of the usual holes (negation, conditionals) in the following cases:

(30)

I don’t know that Kim is a spy.

(31)

If I discover that Bill is a spy, there will be trouble.

Some take factive presuppositions to be basically pragmatic (conversationally triggered rather than conventionally triggered).It doesn’t look like ECM know gets around the factivity of its complement this way (but this needs more work to be sure). For instance, I don’t need to invoke the same intonation or context for ECM.

11

5.1 ECM belief must be belief with inference Not only do these verbs become mere belief ascriptions (rather than knowledge ascriptions) with ECM, the nature of the belief is interestingly different from garden-variety belief. (36) We are explaining that our grandmother kept much hidden about the grandfather we never knew. After her death, we uncover why: he was a liar and a cheat. a. Because my grandmother never told us much about our long dead grandfather, we never knew that he was a liar and a cheat. b. # Because my grandmother never told us much about our long dead grandfather, we never knew him to have been a liar and a cheat. One feels that the ECM requires old Gramps to have been alive at some point during our lifetime for us to ”know” him this way.7 Grandma can say: (37) I knew him to be a liar and a cheat. And someone who met Grandpa can deny this with ECM: (38) I never knew him to be a liar and a cheat. What we cannot do, however, is ascribe a belief using know-ECM if that belief is merely arrived at by believing something second-hand. (39) Horace was demoted and he reacted badly to it. We are trying to keep this a secret from the boss, Rita. Unfortunately, Rita was told by another employee that Horace was upset. Someone asks me: What things does the boss know about Horace’s reaction to his being demoted? 7 Tense is a confounding factor here. Let me un-confound things in two ways. First, we can put the event/state time of the infinitive ”before” the ”knowing” time with perfect have. We still get oddness in the above scenario: #we never knew him to have been a liar. The examples in (39) decound things by making the finite complement simultaneous.

12

a. b.

Rita knows that he’s upset about it. # She knows him to be upset about it.

ECM comes out as plainly ungrammatical when embedded in a sentence that makes it clear that the knowledge was imparted second-hand. (40) What did you tell Rita in the meeting? a. I let her know that Horace was upset. b. # I let her know Horace to be upset. A speculation ECM reports a belief based on inference from direct evidence

5.2 Back to the Borkin cases The Borkin example cited in (3) tells a similar story. A variation on Borkin’s examples: (41) A woman walks into the emergency room. She complains to the triage nurses of a pain in her stomach, and says to them ”I am sick.” She fills out forms, waits, and eventually gets very impatient because no one is seeing her. Patient: Why isn’t a doctor seeing me? Do you think I am lying that I am sick? Nurse: #No, we believe you to be sick and we wish we could do something. We just don’t have a lot of staff. Im sorry. Versus a finite clause (42) Nurse: No, we believe that you are sick . . . The nurse does not believe that the patient is sick because the nurse inferred anything. Rather, she merely accepted what the patient said. With some inference or reasoning involved in arriving at the belief, ECM is ok: 13

(43) A woman comes into the emergency room with a sprained ankle. The doctors and nurses treat the ankle, but they find something startling about her blood tests, which they ran because of routine procedure. They leave her waiting while they run more tests. Woman: Why are you keeping me here? My ankle is fine now. Nurse: I am sorry. We believe you to be quite sick. Inference can be based on either direct evidence or indirect evidence: (44) The Poor but Clever RCMP The FBI has sent the Royal Canadian Mounted Police (RCMP) all their documents pertaining to a certain MacDonald, whom the RCMP are searching for. The RCMP, however, know much about MacDonald’s psychology and motives, but nothing about MacDonald where he’s escaped to. The FBI has accumulated evidence, not about MacDonalds’s location per se, about where his wife is, where he made a phone call from last month, etc. RCMP officers are able to make pretty solid conclusions based on this information. The RCMP believe MacDonald to be in Winnipeg. The RCMP’s beliefs are based here not on any ”direct” evidence they have gathered. Their ”ECM” belief follows from some inference from indirect (reported) evidence. A generalization ECM reports a belief based on inference

5.3 Epistemic Modals Epistemic modals have an evidentiality component that requires inference (Karttunen (1972), Kratzer (1991) and von Fintel and Gillies (2010). (45) Looking out the window at a rainy day #It must be raining 14

(46) People come in with wet umbrellas It must be raining von Fintel and Gillies (2010): the modal base for an epistemic modal contains, in addition to the usual ”what is known at w in context c”, a class of ”privileged” information they call kernels. The kernel K is a set of propositions that represents direct information in a context. The inference requirement: must presupposes that the kernel K does not ”diretly settle” the (non-modalized) proposition. (47) von Fintel and Gillies (2010) (Definition 5) Fix a c-relevant kernel K a. J must φ Kc,w is defined only if K does not directly settle JφKc b. J must φ K c,wT= 1 if BK ⊆ JφKc where BK is K What ”directly settled” means is detailed in their work. Importantly, ”directly settles” is distinguished from entailment. The basic idea is that in (48), the context provides a kernel that directly settles whether it is raining, which conflicts with the presupposition of the modal that it doesn’t. (48) Looking out the window at a rainy day #It must be raining The kernel appropriate to the context of (49) entails that it is raining, but no one of the propositions in the kernel ”settles” it. (49) People come in with wet umbrellas It must be raining 15

5.4 Speculations about Kernels and Attitudes Recall the Nurse examples. We suspected that what was wrong with ECM here is that the complement proposition doesn’t follow from anything the nurse believes. (50) A woman walks into the emergency room. She complains to the triage nurses of a pain in her stomach, and says to them ”I am sick.” She fills out forms, waits, and eventually gets very impatient because no one is seeing her. Patient: Why isn’t a doctor seeing me? Do you think I am lying that I am sick? Nurse: #No, we believe you to be sick and we wish we could do something. We just don’t have a lot of staff. Im sorry. What we want is to privilege the proposition ”that the patient is sick” so that ascribing belief of that proposition is odd with ECM because it directly settles things.

6 Conclusion • ECM means something: belief! • to capture this non-lexical property verbs, we decomposed and put the belief component in a separate head, B • turns out, too, that B requires not just ordinary belief, but belief that involves inferencing of the kind epistemic modals do. The speculation is that we build that requirement into B as well. • among the open questions: what is the syntax of B that triggers ECM?

16

7 Appendix A: hear-ECM with hearsay There appear to be examples which might be non-belief hear, i.e. hearsay hear. (51)

a. ”Indeed,” replied Elinor, ”I believe that you will find him, on farther acquaintance, all that you have heard him to be . . . Jane Austin, Sense and Sensibility b. His charitable actions are noteworthy, and as a person I have heard him to be a 100% good guy. posted by ”meathead320” on http://forums.steroid.com/archive/index.php/t-112054.html

To the extent that these report hearsay, rather than belief of the sort in (9), our prediction is that they require not just hearing one piece of reported information. Rather the proposition must follow from several pieces of reported information (thanks to Jon Nissenbaum for suggesting this as a possiblity). The presence of perfect have in (51a) suggests that there is more than one report that the conclusion is based on.

8 Appendix B: Compositional details hear-ECM (52) Rodney heard Fred to be mocking him = λw.∃s ∃R [hear(s)(Rodney)(w) & R(s)(Rodney)(w) & R(Acquaintance) & ∀ hx’,w’i(hx’,w’i ∈ Dox(Rodney)(w) → [Fred-mocking-him](ιs’.R(s’)(x’)(w’))(w’)] ≈ There is some eventuality s and some Acquaintance Relation R s.t. Rodney hears s and believes de re of s that it is an eventuality of Fred mocking him. a. J [VP3 ] K = λw.∃s∃R[hear (s)(Rodney)(w)&R(s)(Rodney)(w) & R(Acquaintance) & ∀ hx’,w’i(hx’,w’i ∈ Dox(Rodney)(w) → [Fred-mocking-him](ιs’.R(s’)(x’)(w’))(w’)) b. J [VP2 ] K = λx.λw. ∃s ∃R [hear(s)(x)(w) & R(s)(x)(w) & R(Acquaintance) & ∀ hx’,w’i(hx’,w’i ∈ Dox(x)(w) → [Fred-mocking-him](ιs’.R(s’)(x’)(w’))(w’))

17

c. J [VP1 ] K = λs.λx.λw. ∃R [hear(s)(x)(w) & R(s)(x)(w) & R(Acquaintance) & ∀ hx’,w’i(hx’,w’i ∈ Dox(x)(w) → [Fred-mocking-him](ιs’.R(s’)(x’)(w’))(w’)) d. J [BP] K = λs.λx.λw. ∃R [R(s)(x)(w) & R(Acquaintance) & ∀ hx’,w’i(hx’,w’i ∈ Dox(x)(w) → [Fred-mocking-him](ιs’.R(s’)(x’)(w’))(w’)) e. J B K = λP.λs.λx.λw. ∃R [R(s)(x)(w) & R(Acquaintance) & ∀ hx’,w’i(hx’,w’i ∈ Dox(x)(w) → P(ιs’.R(x’)(s’)(w’))(w’)) f. J [TP] K = λs.λw. Fred-mocking-him(s)(w) *

References Beck, Sigrid, and Kyle Johnson. 2004. Double objects again. Linguistic Inquiry 35:97–123. Borer, Hagit. 2005. In name only. structuring sense, volume 1. Oxford University Press. Borkin, Ann. 1984. Problems in form and function. Norwood, NJ: Ablex. Chierchia, Gennaro, and Sally McConnell-Ginet. 2000. Meaning and grammar: An introduction to semantics, 2nd edition. Cambridge: MIT Press. von Fintel, Kai, and Anthony Gillies. 2010. Must...stay...strong. Natural Language Semantics . Green, Georgia M. 1974. Semantics and syntactic regularity. Bloomington, Indiana: Indiana University Press. Hacquard, Valentine. 2006. Aspects of modality. Doctoral Dissertation, Massachusetts Institute of Technology. Hale, Kenneth, and Samuel Jay Keyser. 1993. On argument structure and the lexical expression of syntactic relations. In The view from building 20, ed. Kenneth Hale and Samuel Jay Keyser, 53–110. Cambridge, Massachusetts: MIT Press. Kaplan, David. 1968. Quantifying in. Synthese XIX:178–214. Karttunen, Lauri. 1972. Possible and must. In Syntax and semantics I, ed. J. Kimball, volume I, 1–20. Academic Press. Kayne, Richard S. 1984. Connectedness and binary branching. Dordrecht, Holland: Foris Publications. Kiparsky, Paul, and Carol Kiparsky. 1970. Fact. In Progress in linguistics, ed. M. Bierwisch and K.E. Heidolph, 143–73. The Hague: Mouton. 18

Kratzer, Angelika. 1991. Modality. In Semantics: An international handbook of contemporary research, ed. Arnim von Stechow and Dieter Wunderlich, 639–650. de Gruyter. Kratzer, Angelika. 1996. Severing the external argument from its verb. In Phrase structure and the lexicon, ed. Johan Rooryck and Laurie Zaring, 109–137. Dordrecht, The Netherlands: Kluwer Academic Publishers. Levin, Beth. 1993. English verb classes and alternations. Chicago: University of Chicago Press. Lewis, David. 1979. Attitudes de dicto and de se. Philosophical Review 88:513–543. Morzycki, Marcin. 2005b. Mediated modification: Functional structure and the interpretation of modifier position. Doctoral Dissertation, Graduate Linguistics Students Association, University of Massachusetts at Amherst. Moulton, Keir. 2009. Natural selection and the syntax of clausal complementation. Doctoral Dissertation, Doctoral Dissertation, University of Massachusetts, Amherst. Pesetsky, David. 1991. Zero syntax volume ii: An essay on infinitives. MIT. Postal, Paul. 1974. On raising. Cambridge, Massachusetts: MIT Press. Pylkk¨anen, Liina. 2002. Introducing arguments. Doctoral Dissertation, Massachusetts Institute of Technology. Ramchand, Gillian Catriona. 2008. Verb meaning and the lexicon. Cambridge Studies in Linguistics. Cambridge, United Kingdom: Cambridge University Press. Travis, Lisa. 2009. Inner aspect. Kluwer.

19