Chapter 3 Sluicing. 3.1 Introduction to wh-fragments. Chapter 3 Sluicing in An Automodular View of Ellipsis

Transcription

1 1 Chapter 3 Sluicing 3.1 Introduction to wh-fragments (1a, b) below are examples of sluicing, which was first discussed in Ross (1969). In these examples, a wh-phrase (XP[WH[Q]]) is interpreted as a full wh-question (CP[WH[Q]]). See AMEG 6.1 for the details on wh-questions. We will henceforth call sluiced wh-phrases such as those in (1a, b) wh-fragments in view of striking similarities between wh-phrases in sluicing and fragments such as (1c, d). Fragments will be discussed in Chapter 5. (1) a. John is drinking something, but I don t know what. (wh-fragment) b. John is drinking something. What, I wonder. (wh-fragment) c. John is drinking something. Beer, I guess. (fragment) d. A: What is John drinking? B: Beer, I guess. (answer fragment) e. Mary said John was drinking something, but I don t remember what. (wh-fragment) In (1a), the wh-fragment what is interpreted in two ways, as either (but I don t know) what he is drinking (what we call direct interpretation) or (but I don t know) what it is, where it refers to something in the first conjunct (what we call short-circuited interpretation). This is exactly the same as how the fragment Beer in (1c, d) is interpreted: either as He is drinking beer (direct interpretation) or It is beer (short-circuited interpretation). Direct interpretation of a wh-fragment is obtained from the preceding utterance, which we will call the antecedent of the wh-fragment, by replacing the relevant phrase in it by the wh-fragment in question. For example, in (1a), the direct interpretation is obtained by replacing something in the preceding utterance by what. On the other hand, short-circuited interpretation of a wh-fragment is to interpret it as if it is in a wh-question consisting of wh-phrase + pronoun (it/they) + be, in which the pronoun refers to the most salient NP (topic) in the prior context. In (1e), the wh-fragment can be interpreted in three ways: (but I don t remember) what Mary said John was drinking (direct interpretation), (but I don t remember) what it was (short-circuited interpretation), and (but I don t remember) what he was drinking (localized interpretation). The third way of wh-fragment interpretation, what we call localized interpretation, is to interpret a wh-fragment relative to the local domain of its antecedent (e.g., within the innermost clause of the antecedent or within its own conjunct ignoring other conjuncts). Note that Mary said John was drinking something entails that I (= speaker) remember John was drinking something if the speaker believes what Mary said about John. The localized interpretation obtains based on this entailment. It is not true that the three interpretations are always available to every instance of wh-fragment. In (2a), the direct interpretation is impossible while in (2b, c), it is semantically odd. (2) a. Bill wondered how many papers Sandy had read, but he didn t care which ones. (Chung et al. 1995:257) direct interpretation: *(but he didn t care which ones) [he wondered (how many papers) she had read] localized interpretation: (but he didn t care which ones) [she had read] short-circuited interpretation: (but he didn t care which ones) [they were]

2 2 (they refers to a certain number of papers Sandy had read in the presupposition of how many papers Sandy had read) b. Joan said she talked to the students. Fred couldn t figure out which ones. (Chung et al. 1995:266) direct interpretation: #(Fred couldn t figure out which ones) [she said she talked to] localized interpretation: (Fred couldn t figure out which ones) [she talked to] short-circuited interpretation: (Fred couldn t figure out which ones) [they were] c. He announced he would marry the woman he loved most. None of his relatives could figure out who. (ibid.) direct interpretation: #(None of his relatives could figure out who) [he announced he would marry] localized interpretation: (None of his relatives could figure out who) [he would marry] short-circuited interpretation: (None of his relatives could figure out who) [she was] The examples in (3) are extreme cases in which all the three interpretations are not available. Chung et al. (1995:253) observed that a universally quantified NP (e.g., everybody in (3a) and each of the performers in (3b)) cannot serve as part of the antecedent of wh-fragment. (3) Chung et al. 1995:253 (30) a. *She said she had spoken to everybody, but he wasn t sure who. direct interpretation: #(but he wasn t sure who) [she said she had spoken to] localized interpretation: #(but he wasn t sure who) [she had spoken to] short-circuited interpretation: *(but he wasn t sure who) [he was] (he is an unbound variable, not c-commanded by the quantified NP) b. *Each of the performers came in. We were sitting so far back that we couldn t see who. direct interpretation: #(we couldn t see who) [came in] short-circuited interpretation: *(we couldn t see who) [he was] (he is an unbound variable, not c-commanded by the QNP) In the direct and localized interpretations of (3a, b), who presupposes the existentially quantified propositions (she said) she had spoken to somebody in (3a) and some performers came in in (3b). These existential presuppositions conflict with the universally quantified NP in each antecedent. Interpreting a wh-fragment is very similar to interpreting fragments, which we will discuss in Chapter 5. To interpret an answer fragment such as (1dB), we need to find a propositional function P(x) such that the answer fragment a is interpreted as P(a). The propositional function P(x) is provided by the immediately preceding wh-question. In the same way, in order to interpret a wh-fragment, we need to find a propositional function P(x) with which the wh-fragment is interpreted as [ QPx [ Q WH] x = ] [ PROP P(x)] (in F/A terms). For example, the wh-fragment in (1a, b) is interpreted as For which thing x, [ P(x) John is drinking x] (direct interpretation) or For which thing x, [ P(x) it is x] (short-circuited interpretation), and the propositional function of the former (direct)

3 3 interpretation is provided by the prior utterance John is drinking something. Because of the similarities between (answer) fragments and wh-fragments, we would like to explore the explanation of sluicing by treating wh-fragments essentially as the same as fragments and relying as much as possible on what is needed to interpret fragments. In addition to interrogative wh-fragments, exclamative wh-fragments are also observed. They are interpreted with the understood null subject/topic and be (i.e., short-circuited interpretation). (4) Huddleston and Pullum 2002:921 a. What nonsense! b. What a strange thing for him to say! c. How fantastic! d. How incredibly unlikely! 3.2 The syntax of wh-fragments The internal syntax of wh-fragment is the same as that of an interrogative wh-phrase XP[WH[Q]] (AMEG (3) and 6.1.7). As for its external syntax, Ross (1969), Merchant (2001:40ff.), and Culicover and Jackendoff (2005: ) showed that a wh-fragment behaves syntactically as a wh-interrogative clause (CP[WH[Q]]). The first piece of evidence is about subject-verb agreement (1a). Although the subject wh-phrase is in the plural, the agreeing verb is in the singular, just like a subject wh-clause. The second piece of evidence is about the possibility of extraposition of a wh-fragment (1b). The wh-phrase is extraposed, again just like a subject wh-clause. (1) Culicover and Jackendoff 2005: 268 (52) We were supposed to do some problems for tomorrow, a. but [which problems] {isn t *aren t} clear. (agreement) cf. but [which problems we were supposed to do for tomorrow] {isn t *aren t} clear. (direct interpretation) cf. but [which problems they were] {isn t *aren t} clear (short-circuited interpretation) b. but it isn t clear [which problems]. (extraposition) cf. but it isn t clear [which problems we were supposed to do for tomorrow] (direct interpretation) cf. but it isn t clear [which problems they were] (short-circuited interpretation) The third piece of evidence is about its distribution with respect to a verb-particle construction. In (2a) below, the wh-phrase what, which is apparently an NP, cannot occur between the verb and the particle, as is expected of NPs. (2) a. He was doing something illegal, but I never {found out [what] *found [what] out}. (Culicover and Jackendoff 2005: 269 (54a)) cf. I never {found out [what he was doing] *found [what he was doing] out} (direct interpretation) cf. I never {found out [what it was] *found [what it was] out} (short-circuited interpretation) b. He was doing something illegal, and I discovered {yesterday immediately} what. cf. He was doing something illegal, and I discovered yesterday {*the thing *that *it}.

4 4 The fourth piece of evidence is based on the distribution of NPs: that in English, a light NP object must immediately follow a transitive verb, and that an adverb cannot intervene between a transitive verb and its light NP object, as in I discovered {the thing yesterday *yesterday the thing}. In (2b), although the light NP object cannot occur after yesterday, the interrogative what can, just like a wh-clause. This shows again that the wh-fragment what functions not as an NP but as a clause. The fifth way to show that a wh-fragment is a CP[WH[Q]] is to use a verb that can take a CP[WH[Q]] complement but cannot take an NP object. The best example of such a verb is wonder. (3) He was doing something illegal, but I wonder [what]. cf. but I wonder [{what he was doing what it was}] The sixth way is coordination. A wh-fragment can be coordinated with a full wh-question. (4) a. There are guidelines in the company dress code for [ CP[WH[Q]] when women can use perfume] and [ CP[WH[Q]] how much]. b. Languages differ widely on [ and [ CP[WH[Q]] CP[WH[Q]] what sort of material can appear in the COMP field] under what circumstances]. (Merchant 2001:61) c. Mary invited someone, but I don t know [ CP[WH[Q]] who] or [ CP[WH[Q]] when she invited them]. The seventh piece of evidence is case-marking (Ross 1969, Merchant 2001:42-43). In (5), the German verb schmeicheln ( flatter ) takes a dative NP as its complement and wissen ( know ) either an accusative NP or a clause. (5) Er will jemandem schmeicheln, aber sie wissen nicht, {wem *wen}. In (5), the dative wem is grammatical but the accusative wen is not, which shows that the wh-fragment is interpreted as the dative object of schmeicheln as in Sie wissen nicht, wem er schmeicheln will. Because of these pieces of evidence, we conclude that a wh-fragment (syntactically XP[WH[Q]]) is dominated by a CP[WH[Q]] node in syntax. 3.3 Interpreting wh-fragments The modular representations of a wh-fragment are given in (1) below. As was shown in the previous section, the wh-fragment XP[WH[Q]] is dominated by a CP[WH[Q]] in syntax. It corresponds to a QP x headed by the quantifier [ Q WH] in F/A (AMEG (1)) and a certain ROLE in RS if the wh-fragment is interpreted as an argument. Note that the speech act (SA) superstructure with the illocutionary force (IF) #inquire# (cf. AMEG (1), 7.1 (6b)) is only present when the wh-fragment is not embedded.

5 5 (1) modular representations for sluicing (wh-phrase may or may not be embedded) syntax F/A RS CP[WH[Q]] PROP SA XP[WH[Q]] (S?) QP x (PROP?) IF [AG, SO] [PT, GO] TH inquire SP AD (EV?) ROLE In the syntax of (1), the interrogative clause (CP[WH[Q]]) only consists of a single daughter, a wh-fragment (XP[WH[Q]]). Therefore, there is no possibility for a cliticized element or complementizer agreement to appear on a wh-fragment (Merchant 2001:66-69). This explains Merchant s Sluicing-COMP generalization (2001:62 (71)), which says that in sluicing, no non-operator may appear in COMP. In order to properly interpret a wh-fragment, we need to find a propositional function with which the wh-fragment is interpreted, namely, in (1) above, the S that goes with the wh-fragment, the PROP that goes with the QP x, and the EV in which the wh-fragment carries a ROLE. We claim that this is achieved by the extragrammatical process of inference, in the same way as the interpretation of VP ellipsis (2.2) and fragments (5.2 (4)), that is, the same three steps of extragrammatical inference that is involved in interpreting VP ellipsis and fragments are also involved in interpreting wh-fragments. Extragrammatical inference treats incoming information successively and accumulatively. Note that step (iii) is pure grammatical computation. (2) extragrammatical inference for fragment interpretation (i) Find from linguistic or nonlinguistic context the information that can serve as a potential antecedent. (ii) Infer the contextually appropriate S based on the wh-phrase and the potential antecedent found in step (i). (iii) Based on the syntactic structures of the inferred S obtained in step (ii) and hence the CP[WH[Q]], compute the corresponding structures in the other modules F/A and RS. In (2), inferring the contextually appropriate S in syntax that goes with the wh-fragment in question (step ii) must precede the computation of the corresponding F/A and RS (step iii). For example, to properly interpret C s wh-fragment in (3) below, B s inferred S Jane pulled his/someone s leg too must be determined first. Without this inferred S, the idiom interpretation of whose cannot be recovered. (3) (cf. 5.4 (1)) A: Mary pulled someone s leg. B: Yeah. Jane did too. (ambiguous) Yeah. Jane pulled his leg too. or Yeah. Jane pulled someone s leg too.

6 C: Tell me whose? (ambiguous) Tell me whose leg you SG/PL pulled Wh-fragments and inference Just as is the case with fragments in general, it is true that the propositional function that goes with a wh-fragment is not always available from the immediately preceding utterance, but rather it must be inferred from pieces of information spread across the preceding discourse. (1) is such an example (cf. 5.2 (1)). (1) The participants A, B, and C share the knowledge that John used to have a drinking problem. A: Have you heard about a friend of yours? B: (About) which friend of mine? Which friend of mine are you asking about? A: (About) John. I m asking you about John. He has started again. He has started drinking too much again B: Which alcohol this time? Which alcohol has he stared drinking too much of this time? Whisky? Has he started drinking too much of whisky this time? A: No, beer. No, he has started drinking too much of beer this time B: When? When did he start drinking too much of beer? A: Two weeks ago. He started drinking too much of beer two weeks ago. C: {I didn t know that That s true That s false}. C s that refers to the message John started drinking too much of beer two weeks ago. Note that the verb drink in the message does not appear in the discourse and only exists in the shared knowledge that John used to drink too much. In each utterance in (1), its intended message is shown next to the symbol. (2) is another example, in which the intended interpretation of Guess who at the end is Guess who has not both read the textbook and finished the assignment. (The other possibility is that Guess who is interpreted as Guess who he is, in which he refers to someone who hasn t (short-circuited interpretation).) The inferences taking place as dialogue (2) progresses are shown next to the symbol. (2) A: Have you read the textbook for today? B: Yes, I have. [VP ellipsis] I have read the textbook for today. And I have finished the assignment too. How about you? Have you read the textbook and finished the assignment? C: Of course, I have. [VP ellipsis] Of course, I have read the textbook and finished the assignment. I m sure everyone has. [VP ellipsis] I m sure everyone has read the textbook and finished the assignment. D: That s not true.

7 7 I know someone who hasn t. [VP ellipsis] I know someone who hasn t both read the textbook and finished the assignment. Guess who. [wh-fragment] Guess who hasn t both read the textbook and finished the assignment. (3) is another example in which the interpretation of a wh-fragment cannot be recovered from the immediately preceding utterance. (3) A: I m sure John isn t seeing any girl. B: But I hear Tom is. [VP ellipsis] A: Yeah. B: But do you know where? [wh-fragment] A: In his apartment, I guess. [answer fragment] C: Tell me which girl. [wh-fragment] The intended interpretation of (3C) is Tell me which girl Tom is seeing in his apartment. Note that which girl in C asks about a girl in the antecedent of the VP ellipsis in B s Tom is, namely, Tom is seeing a girl (cf. missing antecedent phenomenon (2.5)). (4) is an example similar to (3). (4) A: John isn t seeing any girl but Tom is. [VP ellipsis] B: Do you know how often? [wh-fragment] A: I suppose twice a day. [answer fragment] C: Do you know where? [wh-fragment] The intended interpretation of C is Do you know where Tom is seeing {the girl her *a girl} twice a day? In (5) below, just like 5.2 (3), real world knowledge is needed to infer what the bartender meant, which is Which beer would you like this time? or something like it. The discourse-initial, non-embedded wh-fragment in (5) receives its interpretation directly from nonlinguistic context, just like the examples in (6). (5) [A regular customer enjoys various kinds of beer at the bar every night. He holds out his empty glass to the bartender without saying anything. The bartender says to the customer:] Which one? (6) Chung et al. 1995:264 (cited from Ginzburg 1992) a. [said by a taxi driver] Where to, lady? b. [said by a distraught homeowner staring at the ashes of his house] Why? In examples (5) and (6), there is no prior utterance in the discourse that provides the information needed to infer an

8 8 intended message. The relevant information comes only from nonlinguistic context. Therefore, this shows again that we need to assume the extragrammatical process of inference, which is at work in constructing an inferred message. What inference needs to do is to arrive at a propositional function that goes with the wh-fragment, namely, to arrive at the S?, the PROP? and the EV? in 3.3 (1) from the relevant linguistic and nonlinguistic discourse context. 3.5 Need for short-circuited interpretation Short-circuited interpretation of wh-fragments was defined and illustrated in 3.1. In this section, we will present various pieces of evidence that we need to posit this type of interpretation. First, there are cases of fragments whose interpretation requires short-circuited interpretation. For example, in (1a, b), short-circuited interpretation of fragments is required. (For more discussion on short-circuited interpretation, see ) (1) a. A: What did John buy? B: A book? direct interpretation: #Did John buy a book? short-circuited interpretation: Was it a book? (it refers to something in the presupposition of the wh-question Pat bought something ) b. A: John bought something. B: A book? direct interpretation: #Did John buy a book? short-circuited interpretation: Was it a book? (it refers to something) c. A: John bought a book written by a famous scientist. B: By whom? direct interpretation: *By whom did John buy a book written? localized interpretation: By whom was it written? (it refers to a book written by a famous scientist) In (1a, b), B s fragment must be interpreted as short-circuited interpretation. The direct interpretation in (1a, b) is pragmatically odd. Incidentally, (1c) shows the need to posit localized interpretation of wh-fragments. Here, the direct interpretation is impossible. Here is a German example. (2) receptionist: Ach, Moment. Sie sind Herr Aoki? Hier wurde etwas für Sie abgegeben. Aoki: Für mich? [fragment] Von wem? [wh-fragment] receptionist: Von Frau Professor Schütz. [fragment] Eine Geige, glaube ich. [fragment] (from Ota, Tatsuya and Marco Raindl Das Geheimnis der Geige. NHK. p.154)

9 9 The first fragment Für mich? and the following wh-fragment Von wem? are interpreted not as Hier wurde etwas für mich abgegeben? or Von wem wurde etwas für mich hier abgegeben? (direct interpretation) but simply as Ist das für mich? and Von wem ist das? (short-circuited interpretation). The last two fragments in (2) Von Frau Professor Schütz and Eine Geige, glaube ich are interpreted not as Hier wurde für Sie eine Geige von Frau Professor Schütz abgegeben or Hier wurde für Sie eine Geige abgegeben, glaube ich (direct interpretation) but simply as Das ist von Frau Professor Schütz and Das ist eine Geige, glaube ich (short-circuited interpretation). These fragments are of the most frequent type, in which a fragment carries new information (focus) about an understood topic that is phonologically null. This is short-circuited interpretation. In (2), the understood topic is etwas ( something that was handed in for Aoki ). Second, Chung et al. (1995:273) pointed out that some PPs that do not allow preposition stranding in wh-question (e.g., (3a, b)) allow it in sluicing as in (3c, d). This was first observed by Rosen (1976). (3) Chung et al. 1995:273 a. *What circumstances will we use force under? cf. Under what circumstances will we use force? b. *What sense is this theory right in? cf. In what sense is this theory right? c. We are willing to use force under certain circumstances, but we will not say in advance which ones. short-circuited interpretation: (but we will not say in advance which ones) [they are] (they refers to certain circumstances) d. This theory is surely right in some sense; it s just not clear which exactly. short-circuited interpretation: (it s just not clear exactly which) [it is] (it refers to some sense) The direct interpretation of (3c, d) would have involved preposition stranding. The short-circuited interpretation is available which does not depend on preposition stranding. Third, in wh-question such as (4a), the variable of the Wh-QP who, whose F/A is [ QPx [ Q WH] [ PROP x = PERSON]] (AMEG (2)), ranges over its own domain, namely, [ PROP x = PERSON]. (4) inheritance of content (Chung et al. 1995: ) a. Joan said she talked to some students, but I don t know who she talked to. b. Joan said she talked to some students, but I don t know who. localized interpretation: (but I don t know who) [she talked to] short-circuited interpretation: (but I don t know who) [they were] (they refers to some students) c. A: Joan talked to some students. B: Who? short-circuited interpretation: (Who) [were they]? (they refers to some students)

10 10 Chung et al. (1995: ) observed that in wh-fragments such as (4b), the variable ranges over (i.e., the answer space of the variable is) the intersection of its own domain and the domain defined by the expression in the antecedent that corresponds to the wh-fragment (their inner antecedent (p.241)), some students in (4b). They said (p.260) that restrictions on the range of the variable are determined jointly by the content of the inner antecedent (some students in (4b)) and that of the wh-fragment (who in (4b)). This is due to the fact that in the case of wh-fragments such as (4b), as opposed to full wh-questions, the speaker s intention in uttering a wh-fragment is to ask about the inner antecedent, as in (4c). Therefore, the answer space for the wh-fragment in question is jointly determined by both the wh-fragment and the inner antecedent. This inheritance of content is captured in our approach by the short-circuited interpretation. For example, in (5a, b), the answer space for the wh-fragment is the set of male guests (at the party). (5) a. A: Mary talked to some men at the party. B: Which guests? short-circuited interpretation: Which guests were they? (they refers to some men in A) b. I know Mary talked to some men at the party, but I don t know which guests. c. I know Mary talked to some men at the party, but I don t know which guests she talked to. Chung et al. (1995:261) observed that this inheritance of content is particular to sluicing, and full wh-questions do not show similar effects. In (5c), the wh-phrase which guests is not necessarily intended to identify the preceding NP some men. On the other hand, this intention is clear with the wh-fragment in (5a, b). Fourth, we need short-circuited interpretation when the adjective in question is only used attributively such as drunken. (6) I saw a drunken man, but I don t remember how {*drunken drunk}. short-circuited interpretation: (but I don t remember how {*drunken drunk}) [he was] (he refers to a drunken man in the antecedent) Fifth, short-circuited interpretation is required when the meaning of the adjective in question is different in attributive and predicative uses. (The symbol # indicates semantic/pragmatic anomaly.) (7) a. *I saw a hard worker, but I don t remember how hard. short-circuited interpretation: #(but I don t remember how hard) [he was] (he refers to a hard worker in the antecedent) b. I saw a hard worker, but I don t remember how hard a worker. direct interpretation: (but I don t remember how hard a worker) [I saw] short-circuited interpretation: (but I don t remember how hard a worker) [he was]

11 11 (8) a. *I met a big fan of Andy Griffith, but I don t know how big. short-circuited interpretation: #(but I don t know how big) [he was] (he refers to a big fan of Andy Griffith in the antecedent) b. I met a big fan of Andy Griffith, but I don t know how big a fan. direct interpretation: (but I don t know how big a fan) [I met] short-circuited interpretation: (but I don t know how big a fan) [she was] To account of the unacceptability of (7a) and (8a), short-circuited interpretation is needed. Sixth, when a wh-fragment takes its own complement, we need short-circuited interpretation. (9) a.?i saw many proud parents in the graduation ceremony, and I know how proud of their children. short-circuited interpretation: (and I know how proud of their children) [they were] b.?i saw a proud winner in the ceremony, and I know how proud of himself. short-circuited interpretation: (and I know how proud of himself) [he was] Seventh, when the adjective in question modifies a plural count noun or a mass noun. (10) a. She writes thorough reports, and wait till you see how thorough! (Merchant 2001:167) short-circuited interpretation: (and wait till you see how thorough) [they are]! (they refers to thorough reports in the antecedent) cf. *How thorough reports does she write? (Merchant 2001:165) b. He bought expensive jewelry, but he wouldn t say how expensive. (Merchant 2001:167) short-circuited interpretation: (but he wouldn t say how expensive) [it was] (it refers to expensive jewelry in the antecedent) cf. *How expensive jewelry did he buy? (Merchant 2001:165) We conclude that because of these pieces of evidence, short-circuited interpretation is independently motivated. 3.6 Ambiguity in wh-fragment interpretation Just like fragments in general (5.4 (9)), wh-fragments can be ambiguous, when there are multiple potential antecedents available in the prior discourse. (1) a. Mary said she has five cats, but I don t know how many dogs. (ambiguous) direct interpretation: but I don t know how many dogs Mary said she has (step i) antecedent = Mary said she has five cats (step ii) inferred S = (but I don t know how many dogs) [Mary said she has] localized interpretation: but I don t know how many dogs she has (step i) antecedent = she has five cats

12 12 (step ii) inferred S = (but I don t know how many dogs) [she has] b. The newspaper has reported that they are about to appoint someone, but I don t remember who. (Chung et al. 1995:256) direct interpretation: (but I don t remember who) [the newspaper has reported that they are about to appoint] localized interpretation: (but I don t remember who) [they are about to appoint] short-circuited interpretation: (but I don t remember who) [he is] Note that localized interpretation such as the one in (1a) is obtained based on an entailment from the antecedent. For example, the first part of (1a) Mary said she has five cats entails that I know she has five cats, if the speaker believes what Mary said. The localized interpretation but I don t know how many dogs she has obtains based on this entailment. This entailment is real, which is shown by the fact that Mary said she has five cats, but I don t know how many cats she has is felt to be contradictory. Here is another example of ambiguous wh-fragment. (2) A: Mary said she is learning a new foreign language. (cf. 5.4 (9)) B: Which language? direct interpretation: (Which language) [did she say she is learning]? localized interpretation: (Which language) [is she learning]? In the direct interpretation of (2B), B asks A what Mary said about her new foreign language. On the other hand, in the localized interpretation, B asks A what A knows about Mary s new foreign language. Note that A s Mary said she is learning a new foreign language entails A s knowledge that I know she is learning a new foreign language. In (3), the first conjunct is ambiguous between the scope relations of MOST > A and A > MOST. (3) Chung et al. 1995:256 a. Most columnists claim that a senior White House official has been briefing them, and the newspaper today reveals which one. direct interpretation: (and the newspaper today reveals which one) [most columnists claim has been briefing them] localized interpretation: (and the newspaper today reveals which one) [has been briefing them] short-circuited interpretation: (and the newspaper today reveals which one) [he is] (he refers to a senior White House official based on its wide scope) b. Most columnists claim that a senior White House official has been briefing them, but none of them will reveal which one. (under the scope relation MOST > A) direct interpretation: (but none of them will reveal which one) [they claim has been briefing them]

13 13 localized interpretation: (but none of them will reveal which one) [has been briefing them] (= Chung et al. 1995:257 (44)) short-circuited interpretation: *(but none of them will reveal which one) [he is] The most likely interpretation of (3a) is the one in which the first and second conjuncts are interpreted as A > MOST and WHICH > MOST, respectively, namely, there is a certain senior White House official who most columnists claim has been briefing them, and the newspaper reveals which one it is that most columnists claim has been briefing them. This interpretation is compatible with any of the direct, localized, and short-circuited interpretations. On the other hand, in (3b), the first conjunct is interpreted as either A > MOST or MOST > A. When it is interpreted as A > MOST, the interpretation of the second conjunct is compatible of any of the direct, localized, and shot-circuited interpretations, just like (3a). However, when it is interpreted as MOST > A, namely, different unidentified White House officials are claimed to have been briefing different columnists, the interpretation of the second conjunct is compatible only with the direct and localized interpretations. The short-circuited interpretation is not possible, because he is nothing but an unbound variable of the narrower scope a senior White House official. 3.7 Multiple wh-fragments Although English is not a multiple wh-fronting language (AMEG (1)), it allows multiple wh-fragments when the antecedent of the multiple wh-fragments has an appropriate pair-list reading (Merchant 2001:112). Just as multiple wh-questions (AMEG (3), (4)), the first wh-fragment has scope over the second wh-fragment and serves as the generator of pair-list reading. As shown in (1e), multiple fragments are also possible. (1) a. Everyone brought something (different) to the potluck, but I couldn t tell you {(?)who what *what who}. (Merchant 2001:112, 113) direct interpretation: (but I couldn t tell you) who brought what to the potluck b. Everybody said he d bring something different to the potluck. But I can t remember who what. (Merchant 2001:113) direct interpretation: (But I can t remember) who said he d bring what to the potluck localized interpretation: (But I can t remember) who would bring what to the potluck c. In French, we have noticed that some intransitive V permit Extraposition of Indefinite, while others permit Impersonal Passive. Which which? (Merchant 2001:112, attributed to Paul Postal) direct interpretation: Which (intransitive verbs) permit which? d. Someone said that once, but I don t remember who, what, when, or where. (from the web) e. A: Everyone brought their favorite food to the potluck, didn t they? B: Yes. John sandwiches, Tom sushi, and Mary an apple pie. We simply assume the syntactic structure for these multiple wh-fragments in which a single CP[WH[Q]] node directly dominates multiple XP[WH[Q]] nodes and the scope relations are indicated by left-to-right order.

14 Aggressively non-d-linked the hell An aggressively non-d-linked wh-phrase (e.g., who the hell) generally cannot occur in sluicing (Merchant 2001:121), which is shown in (1a, b, c).. (1) a. I wish I knew who (*the hell)! (Merchant 2001:122) b. A: Someone bought a car yesterday. B: Who (*the hell)? cf. Who the hell bought it? c. Mary bought something yesterday, but I don t know what (*the hell). direct interpretation: (but I don t know what) [she bought] short-circuited interpretation: (but I don t know what) [it was] (it refers to something) There seem to be two reasons why the aggressively non-d linked wh-phrases who/what the hell are unacceptable in (1). First, the interpretation of wh-fragment is entirely dependent on the antecedent, and in this respect, wh-fragments are D-linked and the aggressively non-d-linked the hell is not appropriate to them. This is similar to the fact that the hell is not appropriate to a wh-phrase with D-linked which. (2) (Pesetsky 1987, cited in Lasnik and Saito 1992:173) a. [ NP[WH[Q]] What the hell book] did you read that in? b. *[ NP[WH[Q]] Which the hell book] did you read that in? Secondly, an aggressively non-d-linked wh-phrase must c-command a gap (XP[G]) (i.e., its extraction site). This explains why an in-situ wh-phrase cannot occur with the aggressively non-d-linked the hell (AMEG (4)). (3) a. Who read what? b. *Who read what the hell? (Lasnik and Saito 1992:173) c. [ NP[WH[Q]] Who the hell] [ S[PAST] NP[G] read what]? However, Merchant (2001) observed that the hell is acceptable if it is adjoined to a wh-phrase that undergoes sluiced wh-phrase inversion with preposition swiping. (4) a. He was talking, but God knows who (the hell) to. (Merchant 2001:65 note 14) b. A: I m getting married. B: Who the hell to? A: Who do you think? Rita. (from the web) c. I m battling that on top of sleep deprivation. I m almost on my knees now. What is this? Am I meant to surrender? If so, who the hell to? And what then? (from the web) d. CP[WH[Q]

15 15 NP[WH[Q]] PP[to] NP[WH[Q]] NP? P[to] NP[G] who the hell to We assume the syntactic structure (4d), in which the wh-phrase who the hell c-commands the corresponding NP gap. 3.9 Access to lexical entries The extragrammatical process of inference involves accessing the lexical entries of the words uttered and at the same time accessing the lexical entries of the closely related words. The following example, just like the similar example of fragment in 5.3 (1), leads us to assume that when the lexical entry for intransitive verb drink is accessed, the lexical entry for transitive drink is also activated and becomes available for the construction of the inferred message but I don t know what Harriet is drinking again. (1) Culicover and Jackendoff 2005: 266 (48b) Harriet is drinking again, but I don t know what. [sprouting sluicing] Here is another example. When the lexical entry for intransitive flirt in (2A) is accessed (syntax: V in [ ]), the related entry with PP[with] complement is also activated (syntax: V in [, PP[with]]). Therefore, in (2B), the presence of the preposition with is obligatory (cf. 5.3 (4a)) and no other preposition is possible. (2) Culicover and Jackendoff 2005: 271 (58b) A: Harriet s been flirting again. B: Who *(with)? Note that if B had simply said Who? without with, it would have been interpreted as Who is Harriet?, which is not B s intention. However, there seems to be a restriction between different lexical entries of the same verb on the way which lexical entry can activate which one. As shown in (3a, b), matching sluicing (i.e., the type of sluicing in which the wh-fragment can be matched with an overt phrase in the antecedent (Culicover and Jackendoff 2005:257)) presents no problem, because there is an antecedent in the preceding context about which the wh-phrase asks and also because the same lexical entry of the same verb ([V, NP, NP] in (3a) and [V, NP, PP[to]] in (3b)) is used both in the clause that contains the antecedent of the wh-fragment and in the inferred message of the wh-fragment. In contrast, sprouting sluicing (i.e., the type of sluicing in which the wh-fragment has no matching constituent in the antecedent but functions as a suppressed/unassociable role (i.e., an implicit argument) or an adjunct in the antecedent (Culicover and Jackendoff 2005:257)) is restricted, as shown in (3c, d, e).

16 16 (3) Chung at al. 1995:248, Merchant 2001:33, and Culicover and Jackendoff 2005: 258, note 17 a. They served someone fish, but I don t know who. [matching] direct interpretation:??(but I don t know who) [they served fish] short-circuited interpretation: (but I don t know who) [he was] b. They served something to the students, but I don t know what. [matching] c. She served the soup, but I don t know to whom. [sprouting] d. She served the students, but I don t know what. [sprouting] e. *She served the soup, but I don t know who. [sprouting] short-circuited interpretation: *(but I don t know who) [he was] Note in passing that the restriction on the extraction of the first object in a double-object construction (e.g.,??who did she serve the meal? (Merchant 2001:34 note 16)) is absent in sluicing as in (3a). This is because the interpretation of the wh-fragment in (3a) is obtained not from its direct interpretation but from its short-circuited interpretation, which does not involve the wh-movement of the first object of the ditransitive verb. This is supported by the fact that the short-circuited interpretation of wh-fragment in (3e) is impossible. (4) four partial lexical entries for serve a. syntax: V in [, NP] RS: TYPE serve in [ [AG, SO; SERVER] [PT, GO; DINER] <[TH; FOOD]>] b. syntax: V in [, NP, NP] RS: TYPE serve in [ [AG, SO; SERVER] [PT, GO; DINER] [TH; FOOD]] c. syntax: V in [, NP] RS: TYPE serve in [ [AG, SO; SEVER] [PT, TH; FOOD] <[GO; DINER]>] d. syntax: V in [, NP, PP[to]] RS: TYPE serve in [ [AG, SO; SERVER] [PT, TH; FOOD] [GO; DINER]] When an addressee hears the first conjunct of (3c), she accesses the lexical entry (4c), which in turn activates (4d), which is used to infer the intended message of the second conjunct ( but I don t know to whom she served the soup ). Lexical entry (4d) is a natural extension of (4c) in the sense that the suppressed/unassociable role in (4c), namely, <[GO; DINER]>, becomes overt/associable in (4d) and the outrank relations of the roles ([AG, SO] > [PT, TH] > GO) are kept intact between the two lexical entries (cf. Chung et al. 1995:248, 262). The situation about (3d) is quite similar. Again, lexical entry (4b), which is used for the second conjunct of (3d), is a natural extension of (4a), which is used for the first conjunct of (3d), in exactly the same sense. Generally, when a verb has two lexical entries L 1 and L 2, we define the binary relation, L 2 is a natural extension of L 1, iff (i) the RS of L 1 and that of L 2 have the same set of roles (and, therefore, the same outrank relations), and (ii) the lowest role in outranking is suppressed/unassociable in the RS of L 1 but overt/associable in the RS of L 2. For example, if a verb has two lexical entries L 1 and L 2 with three roles R 1, R 2, and R 3, and the

17 17 outranking relations R 1 > R 2 > R 3 between them and their RSs are: the RS of L1 [ EV TYPE R 1 R 2 <R 3 > ] and the RS of L2 [ EV TYPE R 1 R 2 R 3 ], then the lexical entry L2 is a natural extension of the lexical entry of L 1. Note that of the four lexical entries of the verb serve in (4), (4b) and (4d) are natural extensions of (4a) and (4c), respectively. If a verb has two lexical entries L1 and L 2, and L 2 is a natural extension of L 1, access to L 1 can activate L 2. For example, access to (4a) and (4c) can activate (4b) and (4d), respectively. The intended message of but I don t know who in (3e) is??but I don t know who she served the soup, which needs lexical entry (4b). However, in the first conjunct of (3e), lexical entry (4c) is accessed, but access to (4c) in the first conjunct of (3e) cannot activate (4b) for the second conjunct, because (4b) is not a natural extension of (4c). This explains the ungrammaticality of (3e). The verb read offers another example of natural extension of lexical entry and how it is activated. This verb has at least two lexical entries (5a, b) and (5b) is a natural extension of (5a). Therefore, access to (5a) in the first conjunct of (6a) can activate (5b) for the second conjunct, which makes sprouting sluicing possible in (6a). On the other hand, sprouting sluicing is impossible with the verb bathe, which is shown in (6b). (5) a. syntax: V in [ ] RS: TYPE read in [ [AG; PERSON] <[PT, TH; INFORMATION]>] b. syntax: V in [, NP] RS: TYPE read in [ [AG; PERSON] [PT, TH; INFORMATION]] (6) a. She was reading, but I couldn t make out what. [sprouting] (Chung et al. 1995:249) b. *She was bathing, but I couldn t make out who. [sprouting] (ibid.) b. *They were marching, but I couldn t make out who. c. Someone was bathing, but I couldn t make out who. [matching] c. Some prisoners were marching, but I couldn t make out who. d. She was bathing someone, but I couldn t make out who. [matching] d. The guard was marching some prisoners, but I couldn t make out who. Although bathe has both intransitive and transitive usage, the transitive bathe (8b) is not a natural extension of the intransitive bathe (8a). The former is a causativization of the latter, just like the intransitive and transitive march, as in The prisoners marched and The guard marched the prisoners. Compare bathe with wash, whose intransitive usage is a (semantic) reflexive use of the corresponding transitive verb (7). Also, John bathed Mary entails Mary bathed but John washed Mary does not entail Mary washed. (7) a. John bathed. / John bathed him. /??John bathed himself. b. John washed. / John washed him. / John washed himself.

18 18 In (6b), the intended message of the second conjunct is I couldn t make out who she was bathing with the transitive bathe. When an addressee hears the first conjunct of (6b), she accesses the lexical entry of intransitive bathe (8a), which does not activate the transitive bath (8b), because the latter is not a natural extension of the former. Note that there is no unassociable role (i.e., implicit argument) in (8a) that becomes associable in the corresponding transitive (8b) and that successful sprouting depends on. (8) partial lexical entries for bathe a. intransitive bathe syntax: V in [ ] RS: [ TYPE bathe ] in [ [AG, TH]] b. transitive bathe syntax: V in [, NP ] RS: [ TYPE cause 2 -bathe ] in [ VA [ TYPE cause 2 ] AG [ PT [ SITU [ TYPE bathe ] [AG, TH]]]] cf. cause is a two-place predicate that takes an agent (as causer) and a caused 2 situation and denotes a volitional action (VA). See AMEG 4.7 (3). The same restriction on activation of lexical entries (i.e., that access to a lexical entry can only activate the lexical entry of the same verb that is a natural extensions of the former) applies to fragments. This shows a strong similarity between sluicing interpretation and fragment interpretation. (9) restriction on lexical entry activation in fragments a. A: They served John fish. B: No, Mary. [matching; cf. (3a)] b. A: They served fish to the students. B: No, chicken. [matching; cf. (3b)] c. A: She served the soup. B: Yes, to John. [sprouting; cf. (3c)] d. A: She served the students. B: Yes, fish. [sprouting; cf. (3d)] e. A: She served the soup. B: *Yes, John. [sprouting; cf. (3e)] f. A: She was bathing. B: *Yes, her baby. [sprouting; cf. (6b)] Chung et al. (1995: ) observed that implicit arguments can serve as the antecedent for a wh-fragment.

19 19 According to Fillmore (1986), some verbs such as eat and read allow novel implicit arguments whereas others such as apply and contribute allow familiar implicit arguments. In (10a), the implicit argument (PP[for]) of apply is familiar to the speaker, and, therefore, his asking about it by what conflicts with this presupposition. On the other hand, in (10b), the implicit argument of eat is novel to the speaker, and, therefore, he can ask about it by what. (10) a. *They applied yesterday. I wonder what for. (Chung et al. 1995:267) a. *He has already contributed $100. I wonder to what organization. (ibid.) b. They were eating. I wonder what. (ibid.) c.?they claimed to us that they had applied, but they refused to say for which jobs. (Chung et al. 1995:268) d.?he revealed that he had already contributed $100, bu t he would not reveal to what organization. (ibid.) In (10c, d), the verbs with a familiar implicit argument is embedded in a context that neutralizes the familiar/novel distinction of presupposition, the sentences become more acceptable Island-insensitivity Chung et al (1995:273), Merchant (2001:87), and Culicover and Jackendoff (2005:258, note 17 and ) showed that wh-phrases in sluicing are not sensitive to strong islands by giving examples that would violate island constraints if a purely syntactic account such as wh-movement followed by deletion were resorted to. Ross already observed the weakening of island violation effects in sluicing in his seminal paper (1969: (71)-(73)). (See 5.5 (1)-(3) and 5.8 (7), where fragments will be shown to be island-insensitive.) These examples show that purely syntactic accounts do not fare well in accounting for sluicing. (1) a. Bob found a plumber who fixed the sink, but I m not sure with what. [CNPC] (Culicover and Jackendoff 2005:258) direct interpretation: *(but I m not sure with what) [Bob found a plumber who fixed the sink] localized interpretation: (but I m not sure with what) [he fixed it] b. That Tony is eating right now is conceivable, but I m having a hard time imagining what. [Sentential subject] (Culicover and Jackendoff 2005:258) direct interpretation: *(but I m having a hard time imagining what) [that Tony is eating right now is conceivable] localized interpretation: (but I m having a hard time imagining what) [he is eating right now] c. A: Harriet drinks scotch that comes from a very special part of Scotland. B: Where? [CNPC] (Culicover and Jackendoff 2005:267) direct interpretation: *(Where) [does Harriet drink scotch that comes from]? localized interpretation: (Where) [does it come from]? d. A: John met a guy who speaks a very unusual language.

20 20 B: Which language? [CNPC] (Culicover and Jackendoff 2005:267) direct interpretation: *(Which language) did John meet a guy who speaks? localized interpretation: (Which language) does he speak? Although the direct interpretation of each of the wh-fragments in (1) is not available due to an island violation, each localized interpretation is available, which hides the island violations of wh-fragments. Chung et al. (1995: ) claimed that sluicing involving sprouting are subject to the standard array of island effects, and gave the following examples. (2) Chung et al. (1995:279) (102) a. Wh-island *Sandy was trying to work out which students would speak, but she refused to say {who to to who(m)}. b. Wh-island *Agnes wondered how John could eat, but it s not clear what. c. Subject Condition *That Tom will win is likely, but it s not clear which race. d. CNPC *Bob found a plumber to fix the sink, but it s not clear what with. e. CNPC *Tony sent Mo a picture that he painted, but it s not clear with what. (2d, e) are in sharp contrast with (1a), in which there is no CNPC effect, although it involves sprouting sluicing. Also, (2c) is in sharp contrast with (1b), in which there is no Subject Condition effect. We tentatively conclude, following Culicover and Jackendoff (2005), that wh-fragment is not sensitive to islands, regardless of whether it involves matching sluicing or sprouting sluicing. Although wh-fragment is island-insensitive, VP ellipsis is island-sensitive (Chung et al. 1995:275, Merchant 2001:4-5, 114, Merchant 2004:705). (3a) is an example of CNPC violation. (3b) is a VP ellipsis example of CNPC violation. (3c) is a wh-fragment example of CNPC violation. (3) a. *They want to hire someone who speaks a Balkan language, but I don t remember which they want to hire someone who speaks. (Merchant 2001:5) b. *They want to hire someone who speaks a Balkan language, but I don t remember which they do. (Merchant 2001:5) c. They want to hire someone who speaks a Balkan language, but I don t remember which. (Merchant 2001:4) In our AMG approach to unbounded dependency, island facts are captured by the dominance path conditions

21 21 (AMEG (1)), which formulate licit wh-movement paths by means of dominance relations, in which the top of the dominance path is CP[WH[Q]] (i.e., an interrogative wh-cp) and the bottom of the path is XP[G] (i.e., a gap with the syntactic category XP). The simplified version of dominance path conditions are repeated here as (3). As for the full version of the dominance path conditions, see AMEG (1). (4) the dominance path conditions on XP[WH[Q]] (simplified version) a. non-subject gap: CP[WH[Q]] {S, VP, CP[that]}* VP {NP, N, PP}* XP[G] b. embedded subject gap: CP[WH[Q]] {S, VP, CP[that]}* VP S NP[G] c. matrix subject gap: CP[WH[Q]] S[FIN] NP[G] In (4a, b), {S, VP, CP[that]}* is to be interpreted as the set of dominance paths consisting of any number (including zero) of S, VP, CP[that] in any order. (The correct order of S, VP, CP[that] in each instance of wh-movement is determined separately by the relevant PS rules and lexical entries.) For example, the dominance path condition (4b) says that a licit wh-movement dominance path must be a member of the set of all the dominance paths that start with the top node CP[WH[Q]] and end with the string VP S NP[G] and in between there can be any number (including zero) of S, VP, CP[that] in any order. The dominance path conditions (3a, b) account for many of the well-known islands. For example, (5a) is an example in which the Complex NP Constraint (CNPC) (the relative clause subcase) is violated. The CNPC says in derivational terms that derivational steps in which material is extracted from the S of a complex NP are excluded (McCawley 1998:523). (5) a. Complex NP Constraint *What topics does Mary like to read books that are about? b. CP[WH[Q]] S[INV] S[BSE] VP[BSE] VP[to] VP[BSE] NP N CP[that] S[FIN] VP[FIN] PP[about] NP[G] The dominance path involved in (5a) is given in (5b). Every time wh-movement moves a wh-phrase out of a complex NP, we have as part of its dominance path the sequence NP N CP S. For instance, if a wh-phrase was extracted out of a relative clause (e.g., the man who [ S XP[G] ]), its dominance path would contain the sequence NP N CP[WH[R]] S, because the constituent structure of the relevant part is [ NP DET [ N N [ CP[WH[R]] NP[WH[R]] [ S XP[G] ]]]]. However, (3a, b) do not allow any dominance path containing the sequence NP N CP S. A wh-fragment, which is syntactically [ CP[WH[Q]] XP[WH[Q]]] (3.1 (7)), does not trigger the dominance path conditions, because the dominance path conditions are encoded in the lexical entry of moved wh-phrase whose external syntax is [ CP[WH[Q]] XP[WH[Q]], S] (AMEG (1)), which is not satisfied by a wh-fragment. On the