Chapter 5 VERB PHRASE ELLIPSIS

1.

Introduction

The kind of meaning multiplication that LLC is designed to model is not restricted to pronominal anaphora. Another pervasive instance of anaphora is ellipsis. To be precise, meaning multiplication is characteristic for the subclass of elliptical construction that Hankamer and Sag, 1976 subsume under “surface anaphora”. These are anaphoric constructions that require a linguistically realized antecedent. The following contrast (from Hankamer and Sag, 1976) may serve to illustrate this point: (1) a. [Hankamer attempts to stuff a 9-inch ball through a 6-inch hoop] Sag: #It’s not clear that you’ll be able to. b. [Same context] Sag: It’s not clear that you’ll be able to do it.

Even though there is a salient and pragmatically plausible interpretation available in both anaphoric constructions above, bare to cannot be interpreted without an overt antecedent while do it can. The former is an instance of surface anaphora and the latter one of deep anaphora. Surface anaphora thus requires the semantic re-use of linguistic resources. Surface ellipsis typically involves two phrases (usually clauses) that exhibit a certain kind of parallelism. Crucially, in one of the two clauses some syntactic material is missing, and this missing material is identified with the parallel material from the other phrase. The incomplete phrase is called the ellipsis site or the target and the other one the source.

183

184

ANAPHORA AND TYPE LOGICAL GRAMMAR

Ellipsis can be classified according to the syntactic category of the missing material, the remaining material in the target clause, and according to the structural relation that holds between source and target. Some well-studied examples of ellipsis are (the list is not supposed to be exhaustive in any way):

Right node raising. Source clause and target clause are conjoined, the target precedes the source, and the remaining material is on the left periphery of the target.
(2) a. John likes and Bill detests corduroy. b. Every man loves but no man wants to marry his mother.

Gapping. Source clause and target clause are conjoined, the source precedes the target, the missing material consists of the verb (both auxiliaries and the main verb in case of periphrastic forms), possibly together with a verb-adjacent object in double object constructions.
(3) a. b. c. d. e. f. John met Mary and Bill Sue. John has invited Mary and Bill Sue. John gave a flower to Sue and Bill a CD to Anna. John gave a flower to Sue and Bill to Anna. John gave Sue a flower and Bill Anna a CD. He gave Sue a flower and she a CD.

Stripping. The remnant in the target clause is an argument of the verb; no special constraints on the structural relation between source clause and target clause.
(4) a. Bill opened a bottle of wine, and Harry too. b. Bill opened a bottle of wine. Harry too.

VP ellipsis. No special constraint on the syntactic relation between source clause and target clause (coordination or subordination within one sentence, or different sentences). The missing material is an infinite VP.
(5) a. b. c. d. e. f. John John John John John John left, and Bill did too. left, but Bill didn’t. left, and Bill wants to, too. left before Bill did. is tall, and Bill is too. is tall. Bill is too.

Verb Phrase Ellipsis

185

Antecedent contained deletion. The target clause is a relative clause that modifies the object of the main clause, the missing material is an infinite transitive VP.1
(6) a. John read every book that Bill did. b. John showed Bill every place that Harry already had.

Sluicing. No special requirement on the relation between source clause and target clause, target is a constituent question, everything except the wh -phrase is missing.
(7) a. John read a book, but I don’t know which one. b. They wanted to hire somebody who speaks a Balkan language, but I don’t know which one. (from Merchant, 1999) c. They wanted to hire somebody who speaks a Balkan language. Which one?

As said above, this list is by no means intended to be exhaustive. These kinds of ellipsis can be grouped into three super-categories. Most of the kinds of ellipsis shown above are triggered by the presence of some lexical item (like a coordination particle in coordination ellipsis, or a wh -word in sluicing). This does not hold for stripping. Nor does stripping seem to be conditioned by a particular type of grammatical environment. Given this, this kind of ellipsis does not seem to lend itself easily to a compositional grammatical analysis.2 Among the remaining cases, we may distinguish between bounded and unbounded kinds of anaphora. All instances of coordination ellipsis, i.e., right node raising, gapping etc. are confined to coordinate structure. Therefore it seems plausible to locate the source of the meaning multiplication that comes with anaphora in the lexical entry of the coordination particle. As discussed in Chapter 1, there is a straightforward Categorial treatment of all cases where the remnant in the target clause is a continuous substring of the reconstructed clause, and there are proposals
1 Most work on antecedent contained deletion subsumes it under VP ellipsis, but under a purely descriptive perspective, what is missing is a transitive VP. See Jacobson, 1992a for arguments that this seemingly naive view is in fact the correct one. 2 A possible strategy for an analysis within the present framework could run as follows: We assume a lexical rule like x : A/B ⇒ x : A|B

which would transform a lifted NP like Bill with category s/(np\s) and meaning λP.P bill’ into a sentence that needs a VP antecedent, i.e., something of category s|(np\s) that accesses a property P from the context and assumes the meaning P bill’. Whether such an analysis is viable of course depends on whether appropriate restrictions on such a lexical rule can be formulated. For the time being I have to leave this issue open.

P : (np\s)|(np\s) . Therefore the natural candidate as lexical entry for did (and other auxiliaries) is (9) did – λP. but it requires a VP as an antecedent. In the present chapter. This simplification being made. and an analysis that locates the trigger for ellipsis in the relative pronoun seems conceivable (even though to my knowledge it hasn’t been tried yet). The theory comes in two variants. (8) VPE: The Basic Idea John walked. an analysis which locates the job of meaning multiplication in the lexicon would be as complex as the corresponding theories of pronominal anaphora. I take it that VPE is triggered by the auxiliary. wh -words for sluicing). Therefore I will propose a somewhat more complex theory variant in the sections 5 and 6 which extends to cases that are problematic for a Sag style account—admittedly at the price of overgeneration in certain respects. and so does the one to be developed here. The remaining cases. I am just concerned with the syntax and semantics of VPE here.186 ANAPHORA AND TYPE LOGICAL GRAMMAR to extend this kind of treatment to gappping as well (see for instance Steedman. I will develop an analysis of VPE using LLC. Let us start the discussion with a simple instance of VPE like Both did and too are prima facie candidates for the lexical trigger of ellipsis. An analysis of sluicing with the apparatus of LLC requires certain assumptions about the semantics of indefinites that are to be introduced in the following chapter. while there are evidently pragmatic adequacy conditions that constrain ellipsis further. Therefore. It is well-known from the literature that Sag’s theory undergenerates in certain respects. The meaning of the target VP is identical to that of the source VP. This makes these kinds of ellipsis candidates for a modeling in terms of |E . Since an auxiliary (or—in the case of infinite target clauses—to ) is obligatory for VPE while the presence of too isn’t (as can be seen from this sentence). However. Antecedent conained deletion is likewise clause bounded. in the next section I will introduce a fairly simple version that is similar in spirit to the theory of Sag. 1976 (who however uses a transformational syntax). and they are in principle unbounded. 1990 and Morrill and Solias. 1993). and Bill did too. the auxiliary did occupies the position of a VP. 2. including the semantic impact of tense and aspect. verb phrase ellipsis (VPE henceforth) and sluicing are both triggered by certain lexical items (auxiliaries or the infinitive marker to for VPE. I will ignore all issues pertaining to verbal inflection here.

only Modus Ponens is involved. auxiliaries are treated as “pro-verbs” here. it will be coreferential with the elided pronoun in the target. Besides.. i lex John john’ np lex walked [walk’]i np\s lex walk’john’ λq. During the subsequent discussion. The source VP. has the appropriate category to antecede did and thus enables the application of |E . 1993) from deletion theories (like for instance Fiengo and May. we will in fact encounter examples where reconstruction leads to wrong predictions. except that they occupy a VP position rather than an NP position. i. Such a coreferential reading is also possible if the pronoun in the source is bound to the local subject. In the latter case. . Derivation of (8) 3. (I ignore the contribution of too as inessential for the semantic composition. The derivation of a construction like (8) is entirely analogous to pronominal anaphora resolution.e.Verb Phrase Ellipsis 187 In words.) did Bill and lex \E λpq.P (np\s)|(np\s) walk’ np\s walk’bill’ s \E /E \E lex |E.1. and they require a VP rather than an NP as antecedent. If the pronoun in the source is free.q ∧ walk’bill’ s\s walk’john’ ∧ walk’bill’ s Figure 5. 1994). it is also possible that the elided pronoun refers to the subject of the target clause. walks.1.q ∧ p (s\s)/s bill’ np λP. As a consequence. They are analyzed completely in parallel to pronouns. Their denotation is the identity function over VP denotations. ellipsis resolution leads to a three-way ambiguity. over properties. This approach thus does not assume any internal syntactic structure of the elliptical VP. The derivation is given in Figure 5. Interaction with Pronominal Anaphora If the elided VP contains a pronoun that can be bound by the local subject. This distinguishes this approach (and all other proverb approaches like for instance Hardt. I do not expect that there is always a non-elliptical counterpart to a VPE construction with exactly the same meaning.

This three-way ambiguity arises naturally from the possible interactions of |E for the VP anaphor with the interpretation of the pronoun. and Billj did revise hisk paper. Finally. and Billj did revise hisi paper. the anaphora slot corresponding to the pronoun may be eliminated via |E . Here we have two options. Johni revised hisk paper. This leads to the derivation in Figure 5. Johni revised hisi paper.P (np\s)|(np\s) rv’(ppr’x) np\s rv’(ppr’x)b’ s \E /E lex |E. John. which in turn serves as antecedent for ellipsis resolution. and Bill did (too). I treat his paper as a lexical unit with category np|np which denotes the Skolem function paper’ which maps individuals to their papers. This rule involves the temporary replacement of a premise of type np|np by a hypothesis of type np.2. Derivation of (10b) Alternatively.rv’(ppr’x)j’ ∧ rv’(ppr’x)b’ s|np Figure 5. as antecedent for his. Johni revised hisi paper. his paper revised John j’ np rv’ (np\s)/np [rv’(ppr’x)]i np\s rv’(ppr’x)j’ s paper’ np|np ppr’x np \E lex k /E and λpq. and Billj did revise hisj paper. John revised his paper.q ∧ p (s\s)/s lex Bill b’ np lex did λP. we may fill the subject slot of the verb in the source clause with a hypothetical subject which serves as antecedent for his. Its anaphora slot is inherited by the sentence as a whole via application of |I . They are indicated in (10) (where italicized material is to be understood as being elided). To keep things simple. After . As a first option. c. We may choose the subject of the source clause.188 ANAPHORA AND TYPE LOGICAL GRAMMAR The former two readings are called strict and the latter one sloppy in the literature. The derivation is given in Figure 5. corresponding to the strict reading (10c).3 on the next page. k rv’(ppr’x)j’ ∧ rv’(ppr’x)b’ s λx. b.q ∧ rv’(ppr’x)b’ s\s |I. (10) a. d. This hypothesis participates in the construction of the source VP. i \E lex lex λq. we may leave the pronoun unbound.2.

4. In other words.rv’(ppr’x)x]j np\s rv’(ppr’j’)j’ s \I \E ppr’ np|np ppr’x np \E and λpq.q ∧ p (s\s)/s lex Bill b’ np lex did λP. it does not due to the intervening |E step. While this sequence of a \E followed by \I seems to lead to an η -normalization configuration.4. i /E and λpq. Derivation of (10c) pronoun resolution.rv’(ppr’x)x np\s rv’(ppr’b’)b’ s \E /E lex |E.q ∧ p (s\s)/s lex lex lex |E. This leads to the sloppy reading (10d). This gives the type np\s for the source VP. i /E Bill b’ np lex did λP. j \E 1 [x]i np John j’ np lex λq. the pronoun is not bound to some overt antecedent but to the subject slot of the verb.q ∧ rv’(ppr’j’)b’ s\s rv’(ppr’j’)j’ ∧ rv’(ppr’j’)b’ s Figure 5.q ∧ rv’(ppr’b’)b’ s\s rv’(ppr’j’)j’ ∧ rv’(ppr’b’)b’ s Figure 5. In the present system. j \E lex lex [rv’(ppr’j’)]j np\s rv’(ppr’j’)j’ s λq. this hypothesis gets discharged via \I . which is thus a suitable antecedent for ellipsis resolution.P (np\s)|(np\s) λx. The derivation is given in Figure 5.P (np\s)|(np\s) rv’(ppr’j’) np\s rv’(ppr’j’)b’ s \E /E 189 lex |E. In this derivation.Verb Phrase Ellipsis his paper revised John [j’]i np rv’ (np\s)/np ppr’ np|np ppr’j’ np \E lex |E. the ambiguity between a strict and a sloppy reading for a pronoun under ellipsis is not a property . the source VP receives the interpretation to revise one’s paper. his paper revised rv’ (np\s)/np rv’(ppr’x) np\s rv’(ppr’x)x s [λx. Derivation of (10d) Cascaded ellipsis.3.

c. Here. To this end. Gawron and Peters. (11a) is four-way ambiguous. and Bill revised Bill’s paper before the teacher revised Bill’s paper. pronoun resolution is combined with hypothetical reasoning for the subject of revised as in the example derivation . (11) a. So the pronoun his participates in two ellipses.190 ANAPHORA AND TYPE LOGICAL GRAMMAR of the pronoun per se but arises from the interaction between anaphora resolution and other operations of semantic composition. the inner ellipsis receives a strict interpretation and the outer ellipsis a sloppy one. John revised John’s paper before the teacher revised John’s paper. It is thus possible that the same pronoun receives a strict and a sloppy construal simultaneously if it is part of two ellipses. and Bill revised John’s paper before the teacher revised John’s paper.. the meaning to revise John’s paper. Using this VP as antecedent for the first did and then assembling the source VP of the outer VP leads to the meaning before’(rv’(ppr’j’)teacher’)(rv’(ppr’j’)) for the matrix VP revised his paper before the teacher did.e. b. The source VP of the outer ellipsis contains an embedded clause with an elided VP (which refers back to the matrix VP of the source clause). and Bill didi . In addition to the reading in which his remains free—which I omit from the following discussion—both ellipses may receive a strict construal (reading (11b)) or they may both receive a sloppy construal (paraphrased in (11c)). Using this VP as antecedent for the second did leads to reading (11b). the inner VP is assembled and assigned the Curry-Howard term rv’(ppr’j’). The sentence (11a) involves cascaded ellipsis. In the indicated ellipsis pattern. too. To obtain the strict-strict reading. The derivation of the first two readings in LLC are straightforward. The interesting reading is given in (11d). d. and Bill revised Bill’s paper before the teacher revised the teacher’s paper. After this. John revised John’s paper before the teacher revised John’s paper. |E is applied to the pronoun with the np John as antecedent. 1990 discuss an example with this property. The strict-sloppy reading is obtained if the inner ellipsis is given a sloppy construal. John [[revised his paper]j before the teacher didj ]i . i. John revised John’s paper before the teacher revised the teacher’s paper.

only the three readings in (11) are derivable in LLC. Using this construal to resolve the first occurrence of did and then assembling the matrix VP leads to the interpretation before’(rv’(ppr’teacher’)teacher’)(λx.4 on page 189). amounts to the derivability of the following sequent in LLC: LLC x : np. John revised John’s paper before the teacher revised the teacher’s paper. bef : (np\s)\(np\s)/s. bef : (np\s)\(np\s)/s.Verb Phrase Ellipsis 191 of a sloppy construal above (Figure 5. and Bill revised John’s paper before the teacher revised the teacher’s paper. r : (np\s)/np. To see how this reading is derived. For the time being. Nevertheless the majority opinion tends towards the assessment that (12a) is possible while (12b) is not. Native speaker judgments are notoriously shaky in examples as complex as this.bef (d(r(px))t)(r(px))x : np\s In linguistic terms. which are paraphrased below: (12) a. . t : np. b. t : np. d : (np\s)|(np\s) ⇒ λx. there is the critical reading in which the outer ellipsis receives a sloppy construal and the inner ellipsis a strict one. Dalrymple et al.rv’(ppr’x)x) for revised his paper before the teacher did.This leads to the meaning λx.rv’(ppr’x)x for revised his paper. p : np|np. this means that the VP revised his paper before the teacher did can have the interpretation to revise one’s paper before the teacher revised one’s paper. p : np|np. Using this VP meaning as antecedent for the second ellipsis gives us reading (11c). John revised his paper before the teacher did. we get the mixed reading (11d). 1991 consider two more readings for (11). there seems no obvious remedy for this undergeneration.. With the given type assignment. and Bill revised Bill’s paper before the teacher revised John’s paper. d : (np\s)|(np\s) ⇒ bef (d(r(px))t)(r(px))x : s Applying \I to this sequent yields the equally derivable sequent LLC r : (np\s)/np. Finally. John revised John’s paper before the teacher revised the teacher’s paper. observe that the strict reading of the first conjunct. If this VP meaning is used as antecedent for the resolution of the outer ellipsis.

to give the ellipsis a sloppy construal and simultaneously use her mother as antecedent for she. [np]i \E \I. According to the definition of proof trees. (both examples from Shieber et al. It seems impossible to interpret (13a) as (13b). 1990) b.192 ANAPHORA AND TYPE LOGICAL GRAMMAR Let us next turn our attention to a puzzle that was brought up by Gawron and Peters. Therefore her mother cannot enter an anaphoric relationship with an anaphor outside the scope of this application of \I such as she in the derivation above.5 1 revised [heri mother]k ’s paper np\s s M.5. Under the type assignment np|np for she. Madeline revised [her mother]i ’s paper before shei did. Their example is (14) a. 1996. 1996 point out that this kind of reading does in fact exist if the example is changed in such a way that the reading in question becomes pragmatically plausible. 1990 and received further attention by Shieber et al. np lex [np\s]j s Figure 5.. 1996) .e. Maryi heard about the layoffs from [heri manager]j shortly after hej did.. this reading is in fact excluded in our theory.. Thus. (Gawron and Peters. \I operates on a proof tree with a single conclusion. A derivation of this reading would schematically look like the one in Figure 5. However. Ronniei criticized [hisi predecessor]j ’s policy just as hej did when hej assumed office. 1 np\s \E before shek didj (np\s)\np\s \E (Illicit) Derivation of (13b) This is not a licit proof tree because her mother is anaphorically related to the np-hypothesis with the label “1”—otherwise we do not get a sloppy reading—and thus the np-node resulting after applying |E to her mother and the hypothetical np dominate the note \I where the hypothesis 1 is discharged. b.. i. this reading cannot be derived with the given type assignments. *Madeline revised Madeline’s mother’s paper before Madeline’s mother revised Madeline’s grandmother’s paper. Shieber et al. Consider the example (13a). (13) a.

This is not entirely unsatisfactory because the same seems to hold for paycheck readings in general. 1998). 1997. and Bill said Bill talked to Bill’s mother. John said John talked to John’s mother. they all share the intuition that we are dealing with a minimality effect here. Here. c. Fox. 1994. and Bill said John talked to Bill’s mother. John said John talked to John’s mother. we have to resort to the assumption that pronouns are lexically ambiguous between an individual reading and a paycheck reading. Sem. b. *John said John talked to John’s mother. This means its type is (np|np)|(np|np). Fiengo and May.g. and Bill did. One of these readings is impossible though. The idea runs roughly as follows: The source clause is spuriously ambiguous because the second pronoun may refer . Kehler. The pronouns may be both strict (paraphrased in (15b)) or both sloppy (cf. Williams. An instance in given in (15). too. e. and if they are both coreferent with the source subject in the source VP. and using Ronnie as second antecedent results in the reading in question. 1973 and which is sometimes called the “many pronouns puzzle” in the literature. So there are four logically possible readings. a Skolem function and an individual (i. where the first pronoun is strict and the second one sloppy (as in (15e)) is excluded though. Despite differences pertaining to framework and implementation. The literature contains quite a few proposals to cope with this fact (see e. and it requires two antecedents for resolution. and Bill said John talked to John’s mother.Verb Phrase Ellipsis 193 In (14a). John said John talked to John’s mother. 1994. Furthermore. d. 1993. John said he talked to his mother. It becomes derivable in our system if we assume a paycheck reading for the pronominal subject of the target clause. In this reading. and Bill said Bill talked to John’s mother. The fourth reading.e. the reading in which Ronnie (Reagan) criticized the policy of Carter just as Carter criticized Ford’s policy is possible. I close this section with a problem that was originally brought up in Dahl. (15c)).. it is possible that the first pronoun is sloppy and the second one strict (reading (15d)). (15) a. his predecessor in (14a) can perfectly well antecede he. we expect a strict/sloppy ambiguity for both in the ellipsis. To account for the contradictory empirical evidence in these constructions. but that the latter one is strongly dispreferred and only pops up if it is pragmatically enforced. the source VP contains two pronouns. an np containing an as yet unresolved pronoun and an np).

The following example (due to Hardt. 1993:119) It is questionable. the anaphoric link from his to he blocks the anaphoric link from his to John. According to the mentioned theories. readily admits the reading corresponding to (15e). and Bill also said he talked to his mother. but BILL does too. If his is linked to he in the source and he receives a strict construal. which is structurally parallel to (15). and it gains further support from the fact that it can be overruled by an appropriate contextual setting. The sentence (17) has the same range of readings as (15). even though this reading is grammatically permitted.194 ANAPHORA AND TYPE LOGICAL GRAMMAR back either to the first pronoun or to the matrix subject. While both pronouns can in principle be either bound or coreferential. So it seems that the interpretation λx. however. (16) (John is suspected of murdering Bill’s mother. the following sentence is only three-way ambiguous: (18) Only John said he talked to his mother. 1993). The intuition that we are dealing with a blocking effect here is certainly appealing. the sentence has no reading where John is the only person x such that x said John talked to x’s mother. since source antecedent and target antecedent have the same index.. A similar effect can be observed under de-accenting and in connection with focus. (17) John said he talked to his mother. Bill has claimed that John was visiting Bill’s mother on the night in question. But John has presented as his alibi that he was home with his own mother that night. sloppy readings arise because under ellipsis resolution.] So where WAS John last night?) John says he was at his mother’s house. and this makes it a dispreferred antecedent for ellipsis resolution. Since the second pronoun is closer to the first pronoun than to the matrix subject John. Likewise. whether it is really the ellipsis resolution module wherein this blocking effect is rooted. even though no ellipsis is involved. the referential index of a pronoun may either be maintained or replaced by the index of a parallel element of the source antecedent.. his cannot be sloppy.x said John talked to x’s mother is a highly marked interpretation for the source VP said he talked to his mother in the context of (15). . in reference to the case against John: [. The district attorney says. Here the italicized material is meant to be pronounced with flat intonation. (from Hardt.

In its first reading.before’(meet’xbill’)(meet’x)john’) s John lex Figure 5. the quantifier hypothesis may participate in the composition of a VP which serves as antecedent for ellipsis resolution. i every’(λx. An example of this pattern is (18). The proof tree is given in Figure 5. one in the source VP and one in the target VP. 1976.6. As we saw above. c. Here. and this VP is used as antecedent for ellipsis resolution. s) i x np /E before Bill didj before’(meet’xbill’) (np\s)\np\s \E john’ before’(meet’xbill’)(meet’x) np np\s \E before’(meet’xbill’)(meet’x)john’ s qE. and this hypothesis can be used as an antecedent of anaphora resolution. John met everybodyi before Bill met himi . and in this section I will briefly discuss some key examples and show how LLC copes with them. and the second step of qE is delayed until after ellipsis resolution. The second reading is paraphrased in (19c). John met everybody before Bill met everybody.) (19) a.6. Derivation of reading (19c) of (19a) . s. This results in a reading where the quantifier binds two variables. This results in a reading that is synonymous with (19b).Verb Phrase Ellipsis 195 4. A similar pattern arises in connection with VPE as well. This pattern results in bound readings for pronouns. quantifier scoping involves the temporary introduction of an np-hypothesis. everybody met lex meet’ (np\s)/np meet’x [np\s]j lex every’ q (np. the object quantifier everybody takes scope over the source VP. Interaction of VPE and Quantification The interaction of VPE with quantification has received much attention in the literature on ellipsis. John met everybody before Bill did. Here. b. (An analogous example was first discussed in Sag. the quantifier everybody is replaced by a hypothesis in the process of scoping. The sentence (19a) is ambiguous.

including extraction from non-peripheral positions. It is even possible to construct examples that do not have a non-elliptical counterpart at all. goes beyond the scope of this book. the man who Mary met before Sue did b. the man who Mary met before Sue met him A wh -operator like who above triggers the introduction of a hypothesis.196 ANAPHORA AND TYPE LOGICAL GRAMMAR Here the hypothesis x : np that is introduced by the quantifier is part of the source VP that serves as antecedent for ellipsis resolution. Therefore the corresponding qE -step must be dominated both by the hypothesis and the ellipsis site. and this hypothesis can participate in the composition of the source VP of VPE. As mentioned before. including both the source and the target VP. This hypothesis must be discharged by qE . Measure phrases are an example. a detailed discussion of wh -constructions in the context of TLG. 3 The fact that (20b) itself is of questionable grammaticality is inessential for my point here. just like a quantifier. (21) How many miles are you prepared to walk if the people want you to? This provides evidence for an interpretative theory of VPE which does not assume that ellipsis arises from the phonological deletion of syntactically present material in the ellipsis site. This happens if a wh -operator binds a gap of a category that has no proform. the counterpart of the elided VP in the closest non-elliptical paraphrase is not identical with the source VP. Nevertheless.) Note that here and in the previous example. (20) a.3 for an example like (20a). and this rule is only applicable if it operates on a proof tree with a single conclusion. The final discharging of the hypothesis leads to a reading such as that paraphrased in (20b) for an example like (20a). In linguistic terms. this means that such a reading is only possible if the quantifier takes scope over the whole construction. we need limited access to the structural rule of Permutation to carry out such a derivation. where the wh -operator binds two variables. . it should be mentioned in passing that a pattern comparable to the one discussed above arises in connection with wh -movement as well. (Since in the example above this hypothesis would occur after met and thus in a non-peripheral position. The example in (20) illustrates this.

The interesting case is the reading (22d) where the pronoun his is bound by the quantifier but nevertheless strict. there is an interpretation where the pronoun is bound but nevertheless strict (d). It demonstrates that the ambiguity of bound versus coreferential interpretation of pronouns on the one hand and the strict/sloppy ambiguity on the other hand are independent phenomena: (22) a. In the present system. even though the pronoun is unambiguous. x))) . x) ∧ recommend’(m.ϕ(x)). b. Every studenti revised hisi paper before the teacherj revised hisi paper. The qE -scheme basically says: Whenever an a : np occurs in the context of an s with meaning ϕ(a). ellipsis resolution and quantifier scope may interact and thus constrain each other. Every studenti revised hisi paper before the teacherj revised hisj paper. all three readings fall out immediately. the sloppy reading in (22c) is comparable to the sloppy reading of (10). The existence of this reading follows from the intuitive idea behind Moortgat’s treatment of quantification here. (23) a.7 on the next page. Sentence (22a) has three readings (paraphrased in (22b-d)). Every studenti revised hisj paper before the teacherk revised hisj paper. the derivation is analogous to the one given in Figure 5. d. Using the strict reading of (10) as input to this operation yields the critical reading (22d). (∃x(book’(x) ∧ hate’(a. If the pronoun is free. The full derivation is given in Figure 5.Verb Phrase Ellipsis 197 Let us next consider a construction that is related to the example of cascaded ellipsis discussed in the previous section. x))) c. 1990 as well. you can replace the np by a quantifier with meaning Q and obtain an s with meaning Q(λx. The following sentence comes from Gawron and Peters. pronoun resolution. c. x) ∧ recommend’(a. Likewise.2 on page 188. 1990) illustrates this. Every student revised his paper before the teacher did. combined with quantifier scoping. b. x) ∧ recommend’(m. The following example (from Gawron and Peters. (∃x(book’(x) ∧ hate’(a. x))) < (∃x(book’(x) ∧ hate’(m. role-linking as in (c). Alice recommended a book that she hated before Mary did. x))) < (∃x(book’(x) ∧ hate’(a. Gawron and Peters therefore assume a three-way ambiguity of pronoun uses—referential as in (b). Next to the unproblematic cases where the pronoun is either free and strict (b) or bound and sloppy (c). x) ∧ recommend’(a. and co-parametric as in (d). Last but not least.

i np np\s lex lex \E rv’ ppr’x before’ rv’(ppr’x)b’ everybody ( np \ s ) /np np (( np \ s ) \ np \ s ) /s s lex /E /E ∀ [rv’(ppr’x)]j before’rv’(ppr’x)b’ np\s q (np.) The formula in (23d) represents the reading where the pronoun is strict and the quantifier takes wide scope. It is discharged after quantifier scoping. while the sentence is in fact only three-way ambiguous.P his paper Bill ( np \ s ) | ( np \ s ) lex lex |E. Its derivation (sketched in Figure 5. a parallel quantifier appears in the interpretation of the target clause. The sloppy reading with narrow scope of the indefinite is paraphrased in (23b). x) < recommend’(m.7. The current example furthermore involves pronoun resolution . the subject slot of recommended is filled by a hypothetical subject which serves as antecedent for the pronoun. Here. (The introduction of a hypothetical subject is nevertheless necessary to give the quantifier VP scope. s) 1 (np\s)\np\s \E [x]i before’(rv’(ppr’x)b’)(rv’(ppr’x)) np np\s \E before’(rv’(ppr’x)b’)(rv’(ppr’x))x s qE. the object quantifier is scoped over the s that is composed with this hypothetical subject. Furthermore.6 on page 195. In both cases. j ppr’ b’ rv’(ppr’x) revised bef ore np|np |E. a quantifier in the object position of the source VP takes wide scope over the whole construction and thus binds a variable both in the source clause and in the target clause. the only difference is that in the latter case. ∃x(book’(x) ∧ hate’(a. So in effect the quantifier takes scope only over the source VP.198 ANAPHORA AND TYPE LOGICAL GRAMMAR did lex λP. 1 ∀(λx. x))) The sentence (23a) contains two sources of ambiguity.9 on the facing page) is structurally analogous to the derivation in Figure 5. and after ellipsis resolution. The pronoun she may receive a strict or a sloppy construal with respect to the VP ellipsis. x) ∧ (recommend’(a. So we expect four logical interpretational possibilities. Derivation of (22d) d.before’(rv’(ppr’x)b’)(rv’(ppr’x))x) s Figure 5. the overt NP Alice serves as antecedent for the pronoun. The derivation of the narrow scope/strict reading (23c) is completely analogous. and the indefinite NP a book that she hated may have narrow scope or wide scope with respect to before. s.

It is given in Figure 5.8. this rule takes two independent proof trees as inputs: one which has q (np. s.10 on the following page. a lex q (np. Derivation of (23d) The interesting point about this example is that there is no sloppy reading where the quantifier has wide scope. s) j before Mary didk (np\s)\np\s \E \E book that shei hated n /E Figure 5.and the quantifier is scoped at the end of the derivation. and one which has the corresponding hypothetical np as premise. However. resulting in a sloppy reading. there is an anaphoric link (indicated by the index i) connecting a . s. s) as conclusion. the hypothetical subject of the source VP is used as antecedent for she. it seems possible to design a proof tree which represents this reading. Derivation of (23b/c) (with the overt subject Alice as antecedent) as part of the composition of the quantifier. According to the qE -scheme. s)/n recommended 1 (np\s)/np s s Alice [np]l lex lex np\s qE. At a first glance. there is no derivation that would result in this proof tree. s)/n recommended (np\s)/np Alice lex [np]i s qE. j s lex np [np\s]k /E np\s q (np. s. Here. In the derivation in Figure 5.10 on the next page. s.Verb Phrase Ellipsis a q (np. j \I. This leads to wide scope of the quantifier. 1 np\s s \E \E lex book that shei/l hated n j /E 199 q (np. s) np /E [np]i before Mary didk (np\s)\np\s \E [np\s]k Figure 5.9. s.

2 the source clause is a main clause. I limited my attention to cases of VPE where 1 the subjects of source VP and target VP are either proper nouns or simple quantifiers. the term corresponding to the relative clause. Let us take stock. s)/n recommended 1 (np\s)/np s lex np\s [np\s]k \I. j \E \E lex book that shei hated n j /E q (np. thus LLC does not admit it. This point is possibly further clarified by the fact that the Curry-Howard label of the conclusion of the illicit proof tree would be ∃z. In the preceding two sections. contains an occurrence of the variable x that is unbound.10. together with the type assignment (np\s)|(np\s) . they are essential to establish the equivalence between the proof tree format on the one hand and the two sequent formats on the other hand.recommend’zx)alice’ Here. Illicit derivation for (23) discharged assumption of the second input with a node in the first input. s) np /E [np]i Alice np before Mary didk (np\s)\np\s \E lex Figure 5. s. s. and 3 the target clause is either also a main clause or else directly subordinated to the source clause. i np\s s s qE. This is illicit since the two inputs to qE must be independent from each other. there is no sequent derivation corresponding to the wide scope sloppy reading. Within this fragment.book’z ∧ hate’zx ∧ before’(recommend’z mary’) (λx. In other words. on the contrary. the logic LLC. Licit derivations in LLC (or any other substructural logic) never produce Curry-Howard terms containing free variables that do not correspond to some premise. hate’zx. It should be stressed again that the constraints on proof tree formation are not ad hoc rules.200 ANAPHORA AND TYPE LOGICAL GRAMMAR a q (np.

1976. and I will demonstrate that an identity-of-meaning approach can be maintained if we admit a limited polymorphism in the lexicon. where VPE is basically seen as involving identical syntactic structure which is not pronounced in the elliptical part. too. in a manner akin to the standard Categorial treatment of coordination. there is one American and one Canadian flag per window. This would render the example analogous to . the present theory massively undergenerates when we look at cases that lie outside the fragment defined above. (24) A Canadian flag was hanging in front of each window. This revision is conservative though: As long as we restrict attention to the fragment that was covered up to now. In the preferred reading. These issues will be discussed in the remainder of this chapter. Hirschb¨ uhler considered the option that this reading arises because the object each window scopes over the whole construction. the revised theory makes exactly the same predictions as the original one. the system handles all key constructions from the VPE literature in an largely empirically adequate way. the subject can take wide scope in both conjuncts.Verb Phrase Ellipsis 201 for the auxiliary in the target clause. VPE and Polymorphism The approach to VP ellipsis presented in the last section belongs to the family of “identity-of-property” theories for VPE. This idea is in sharp contrast with theories like that proposed by Fiengo and May. Paired with the TLG analyses of pronominal anaphora and quantifier scope that were introduced in the preceding chapters. Following basically Sag. and an American one was. I will discuss several problems for an identity-of-property approach that have been discussed in the literature. leads to a theory of VPE with considerable empirical coverage. these theories assume that the source VP and the elliptical VP express the same property at some level of derivation or representation. including the conjunction. 5.1 The Hirschb¨ uhler Problem Hirschb¨ uhler. and I will propose a revision of the lexical type assignment for the auxiliary. and the latter problem is arguably independent from VPE. 1994. The only mispredictions occur in connection with Gawron and Peter’s (1990) example of cascaded ellipsis (11)—where we predict three readings while there are probably four—as well as in connection with the many pronoun puzzle. While this is a fairly satisfactory result. In the sequel. 5. 1982 notes that in the following example.

We observe a similar reading in (25). a phrase like was hanging in front of many windows may receive different meanings with different types. The former meaning assignment leads to a reading where the subject has wide scope in both conjuncts.(many’ windows’(λy. so the Kempson/Cormack style treatment is easy to incor- .Ryx)). Flexible meaning assignment is an essential aspect of any Categorial Grammar.T (λx.many’ windows’(λy. It maintains the basic intuition that it is the meaning that is shared between source and target in VPE construction rather than syntactic structure. such a solution would fail. the only meaning of type e. where R stands for the meaning of hanging in front of. The Hirschb¨ uhler problem effectively falsifies this group of ellipsis theories. the VP in question is ambiguous between the lifted properties λT. The preferred reading here is the one where the object takes scope over the subject in both conjuncts. 1983. However. none of them was really successful. The key example for this more flexible treatment is Kempson and Cormack. To see why. The syntactic antecedent in the last example is was hanging in front of many windows. while the latter one gives the critical Hirschb¨ uhler reading. (25) A Canadian flag was hanging in front of many windows. as Hirschb¨ uhler points out. This VP is entirely unambiguous. To be somewhat more specific. Even though several attempts have been undertaken to treat this kind of example within an identity-of-property approach. Under a flexible approach to meaning assignment. but the conjunction still takes scope over both objects. one has to consider what potential antecedent properties the source clause supplies here. something which has been called “identity-of-meaning” theories. It does not falsify a somewhat more general setup.202 ANAPHORA AND TYPE LOGICAL GRAMMAR (19).Ryx))) and λT.(many’ windows’(λy. A VP containing a quantified object will be ambiguous in this type. too. but it possibly gives up the assumption that this has to be a property.T (λx. They claim that the piece of meaning that is shared between source VP and ellipsis site is not a property of individuals but a property of quantifiers. and an American one was.was hanging in front of’xy )) Combining this meaning with either the source subject or the target subject yields inevitably the subject wide scope reading. t that can be derived from it is the one where the object scopes over the VP: λx. which in turn leads to the Hirschb¨ uhler ambiguity in ellipsis. Identity-of-property approaches to VPE are unable to derive this reading.

This is due to the fact that argument lowering is a theorem of L (and thus of LLC . even though Categorial meaning assignment is flexible. s) every patient’ np x /E 2 s/(np\s) T 1 /E s T visit’x qE.Verb Phrase Ellipsis 203 porate into the present theory of ellipsis resolution.. however.11. visited every patient – object wide scope It should be noted that due to the built-in flexibility of Type Logical Grammar. every patient visited (np\s)/np visit’ lex np\s visit’x q (np. s.e. the category-to-type correspondence between syntax and semantics is strict.T visit’x) \I. To start with. a functor that consumes a subject quantifier to its left to yield a clause. and a nurse did too. In the present setup.every patient’(λx. (27) A doctor visited every patient. A reading where the subject takes wide scope in the source clause and narrow scope in the target clause is excluded. this approach overgenerates. but only if it occurs both in the source clause and the target clause. The obvious candidate is (s/(np\s))\s. paired with the appropriate syntactic type. 1 (s/(np\s))\s λT.x : ((s/(np\s))\s)|((s/(np\s))\s) The derivation of the lifted source VP of the Hirschb¨ uhler reading of (27) is given in Figure 5.11. The Hirschb¨ uhler examples admit scope inversion. i. 2 s every patient’(λx. So assigning the string was hanging in front of many windows a meaning of a higher type implies assignment of a more complex syntactic category. I assign it the identity function over properties of quantifiers.T visit’x) Figure 5. So the only adjustment that is necessary to adopt Kempson and Cormack’s analysis is a modification of the lexical assignment for the auxiliary in VPE construction: instead of the identity function over properties. such crossed readings are derivable. So the modified lexical entry is (26) did/was – λx.

(28) a. This will result in a reading where the object has wide scope in the source but narrow scope in the target clause. If Harry was having trouble in school.. 1989. quite independently of ellipsis. John’s coach thinks he has a chance. . and Bill’s coach does too. (after Wescoat. Some evidence for such an architecture will be collected in the next subsection. (after Reinhart. I would help him. while pragmatic constraints that are basically independent of the ellipsis module are responsible for the fine tuning. (after Hardt. and the one who arrested Bill did too.2 Non-subject Sloppy Readings Even more problematic for an identity-of-property approach are cases where the antecedent for a sloppy pronoun is not the subject of the source VP. The policeman who arrested John failed to read him his rights. 1991) NPs embedded in a topicalized constituent. One might wonder though whether the kind of parallelism effects that we observe here should really be treated as a property of ellipsis resolution as such. People from LA adore it and people from NY do too. 1992) b. (Rooth. 1983) c. I wouldn’t.x(λz. there is a parallelism constraint in coordinate construction anyway. This would lead to a kind of hybrid theory of VPE.204 ANAPHORA AND TYPE LOGICAL GRAMMAR x : (s/(np\s))\s ⇒ λy. while pragmatics filters them out. 1993) NPs from superordinated clauses. 5. combine it with the source subject directly. So it is possible that grammar in fact admits crossed readings. where the interface between syntax and semantics is fairly liberal as far as admissible readings are concerned. Arguably. but let the copy undergo argument lowering before we combine it with the target subject. cited by Dalrymple et al.zy ) : np\s Now suppose we assign the source VP the object wide scope reading as in the sample derivation in Figure 5. (29) If Bill was having trouble in school. Possible antecedents can be NPs embedded in the subject.11 on the previous page.

and their antecedents by underlining. What the examples above do show is that whatever governs the distribution of non-subject sloppy readings. (32) First John told Mary that I was bad-mouthing her.Verb Phrase Ellipsis 205 (30) I didn’t know that Bill was a bigamist. On the other hand. If John was having trouble in school. these data are not overly problematic for an identity-of-meaning program if pronouns are analyzed in a variable free way. Let us take the intuition “the elliptical VP has to find an antecedent in the pre-VP material of both clauses” seriously. and today Bill’s boss did. which may be enriched by some notion of “implicational bridging” (Rooth. as the following example from Fiengo and May. 1994). The first descriptive hypothesis about sloppy readings that comes to mind in view of these data is that the two antecedents of a sloppy pronoun must occupy structurally parallel positions in the source clause and the target clause (this is for instance assumed in Fiengo and May. I would help him. however. To put this idea slightly differently. Yesterday the guy John works for told him to shape up. 1994) The sloppy pronouns are marked by italic font. as the example (30) demonstrates. c. This approach. (The source clause and the target clause need not be the local clauses. Mary just said he’s married to her. 1994 demonstrates. 1992 ((a) and (b)) and Hardt. 1993 (c): (31) a. 1992) to cover cases like (31a). b. (from Fiengo and May. I do not have a novel account of the structural / semantic / pragmatic relation that has to hold between source and target in VPE to offer here. So apparently a notion of semantic rather than structural parallelism is called for. what is shared between source clause and target clause in a VPE construction is the meaning of a VP that may contain a series of pronouns which are bound inside the source clause and in the target clause respectively. this is shown to be too rigid by Rooth. The only (trivial) grammatical constraint seems to be that the sloppy pronoun has to find an antecedent in the pre-VP material of both clauses. First John told Mary that I was bad-mouthing her. turns out to be too narrow too. I doubt if I would. and Sally did.) . too. Even though an identity-of-property approach to VPE is incapable of covering any non-subject sloppy reading. However. it is certainly not determined by grammar in the narrow sense. and then Sue heard that I was. and then Sue behaved as though I would. if Bill was having trouble.

This piece of meaning serves as antecedent for ellipsis resolution. except for the fact that the lifted VP is not lexically founded but retrieved from the source clause via |E . Parallelism Versus Source Ambiguity Speaking in somewhat less technical terms. paired with the meaning λT. the lexical entry for the auxiliary given in (34)—paired with the general approach to anaphora presented in this book—leads to a constraint on sloppy readings of VPE: a given sloppy pronoun has to find its antecedents in the pre-VP material of the source clause and the target clause (or clauses in the case of multiple ellipsis) respectively. the category of a VP containing one pronoun is (np\s)|np.T help’. In an intermediate step of the derivation. it is not very surprising that most examples that are discussed in the literature can be derived in such a system. (34) λx. the VP in question does not contain unresolved pronouns any longer. Let us abbreviate this category with vp1 . and Bill’s father does too.12 on the facing page.206 ANAPHORA AND TYPE LOGICAL GRAMMAR Let us restrict the discussion to cases with one pronoun for the moment. To enforce binding of the pronoun within a superordinate clause. Accordingly. The general type scheme for lifted VPs is then (s/vpn )\s for arbitrary natural numbers n. Since this is a very mild constraint indeed. . The derivation of the target clause runs completely in parallel. 6. This can be generalized to an arbitrary number of pronouns in a simple way: Let us say that vp0 = np\s and vpn+1 = vpn |np. Basically. A notoriously difficult one is due to Dahl. The derivation of the source clause is given in Figure 5. namely the identity function over all instances of lifted VPs. this type has to be lifted to (33) (s/vp1 )\s Note that after lifting. 1973.x : ((s/vpn )\s)|((s/vpn )\s) Note that the proposal made in the last subsection is just a special case of this where n = 0. To see how this proposal works. the string helps him is assigned the lifted VP category (s/vp1 )\s. consider a simple example like (35) John’s father helps him. So the meaning of the target clause winds up being (help’ b’(of’ b’ father’))—Bill’s father helps Bill. I assume a polymorphic lexical entry for the auxiliary.

Source clause of John’s father helps him. We may analyze both ellipses as taking the first clause as antecedent and receiving a sloppy construal.Rj’(of’ j’ father’) s/vp1 T /E /I. let us briefly return to the general issue: Does VPE involve identity of meaning? I have argued above that such a theory has to be paired with some theory of parallelism to cope with the problem of overgeneration. Given this. Under an identity-of-meaning theory. it is not entirely unconstrained. 1976 style theory. it predicts a fundamental asymmetry between VPE in coordination and subordination. even though his wife does. Bill fails to realize that Bill is a fool. The critical reading is the one where John realizes that John is a fool.12. 1 s (T help’) (s/vp1 )\s λT. . In subordinative constructions. 1991... 2 \E s (help’ j’(of’ j’ father’)) Figure 5. and Shieber et al.Verb Phrase Ellipsis John [np]i j’ lex s np\np/n of’ lex \E f ather n father’ lex /E 1 |E \E 2 207 np/n (of’ j’) [vp1 ]i R vp (Rj’) helps him vp1 help’ np (of’ j’ father’) s (Rj’(of’ j’ father’)) s/vp1 λR. Another analysis is possible though. while the third clause is anaphoric to the second clause and strict. The second ellipsis is extremely sloppy because it takes the possessor of the subject as antecedent of the sloppy pronoun rather than the subject itself. it is as restrictive as the traditional Sag. it might be suggested that we totally trivialize the operation of ellipsis resolution (“fill in whatever gives you a sentence”) and locate all interesting generalizations inside the parallelism module. 1992.T help’ \I. In particular. but Bill does not. So apparently the second clause takes the first clause as antecedent and receives a sloppy reading. Rooth. To place this aspect in the right perspective. but Bill’s wife realizes that Bill is a fool. Liberal though the present theory may be. This idea has been pursued by many authors. this configuration should be impossible. most prominently by Dalrymple et al. 1996. and Bill’s father does too (36) John realizes that he is a fool.

that de-accenting need not be licensed by overt material.208 ANAPHORA AND TYPE LOGICAL GRAMMAR As far as the syntax-semantics interface goes. in the following exchange. which is interpreted as the function λg. This variable is instantiated by means of a pragmatic resolution process that takes parallelism constraints into account. the free variable approach corresponds to the variant of the present LLC-based approach where VPE slots always remain free. it is infelicitous to de-accent the VP likes Sarah in the second sentence: (37) a. however. bound readings of anaphors are preferred over those where the interpretation of the anaphor is supplied by the context. The main effect of admitting binding of VPE slots is to induce a preference ordering over possible resolutions. For instance. 1991 and Shieber et al. say. B: No. A: John likes Mary. 1996 can be directly extrapolated to the analysis of de-accenting. Gardent illustrates this point with Lakoff’s (1971) example: (39) First John called Mary a republican. b. (example from Gardent. While constraints on resolution that are predicted by the LLCanalysis of VPE are relevant to the syntax-semantics interface. and this is the identity function over properties. the parallelism constraints operate on the discourse level. 2000. and then SHE insulted HIM. . This distinction is worked out clearly for instance in Gardent. Inferred propositions can serve as licensors as well. PETER does. Suppose we would try to incorporate such an approach into the overall Categorial machinery. So the VP anaphor does would be translated as a variable P . 2000) Likewise. Arguably. the meaning of a VP anaphor like does in a VPE construction is simply a free variable over properties in such an approach. Recall that in a system using free variables.. A: John likes Mary. B: No. it is pointed out that the constraints on de-accenting are of a similar nature as the constraints on ellipsis resolution. a resolution of does in (38b) as likes Sarah is totally infelicitous. (38) a. There. PETER [likes Sarah]. So in a sense. The closest counterpart of a free variable is a function that consumes only those components of the assignment function g that are relevant for the evaluation of P . It is well-known. the formal machinery to state the parallelism constraints on VPE resolution that is proposed in Dalrymple et al.. the meaning of a free variable is a function from assignment functions to values. As Gardent shows.g (P ). b.

the mentioned theories require only that replacing Bill by John in the target clause leads to the same meaning as the source clause. An identity-of-meaning approach has to assume that the source VP is ambiguous between to revise John’s paper and to revise one’s own paper. it can be argued that the assumption of a spurious ambiguity is unavoidable. The former proposition is not explicitly expressed but contextually inferred from the first conjunct. it has to be shown that there are structural constraints on VPE resolution beyond parallelism. The fact that the comparative clause faster than Bill did cannot be attached to the matrix S node can be seen from the fact that it is impossible to give the comparative operator scope over the matrix subject. independently of the semantic derivation of the source clause. To see why. but it leads to different truth conditions for the target clause in VPE. This suggests that parallelism is a constraint on interpretation that operates on the discourse level. So a parallelism based theory does without the assumption of spurious ambiguity. and Bill did too. consider the following example. using traditional category labels. (42) Betsy collected more rose hips than all the boy scouts. It may constrain the resolution of VPE as a side effect. Consider a comparative construction like (41) John revised his paper faster than Bill did. Reconsider a simple strict/sloppy ambiguity like (40) John revised his paper. So an adequate account of VPE has to be hybrid between syntax/semantics and pragmatics to some degree.Verb Phrase Ellipsis 209 Here the parallelism that licenses the de-accenting of insulted holds between the propositions John insulted Mary and Mary insulted John. A purely parallelism based approach can do without this kind of spurious ambiguity. is given in Figure 5. this ambiguity is spurious. both the strict and the sloppy readings fulfill this requirement. Clearly. To motivate a structural account of VPE like the present one. Alongside the fact that these theories are unified—only the parallelism constraint matters—this is another strong argument in their favor.13 on the next page. Outside ellipsis construction. However. The syntactic structure of this sentence. This sentence is ambiguous between a collective reading of all the boy scouts (where Betsy outperformed the joint efforts of all the boy scouts) . Informally put. as soon as we turn attention to subordination constructions.

Given this.13. Here the quantifier all the girl scouts can only have a distributive interpretation. Syntactically speaking. Second. Compare this to (43) All the girl scouts collected more rose hips than Tom. This implies that the matrix subject in a comparative construction must take wide scope with respect to the comparative operator. The target VP is ambiguous between a strict and a sloppy reading. since the target clause is included in the source clause. it is impossible to establish parallelism between source and target clause in (41). the subjects are excluded from parallelism. So all the boy scouts may take wide scope or narrow scope with respect to the comparative operator. But this means that the meaning of source VP and target VP must be identical. but the space of possible interpretations there is neither totally free nor totally . First. thus there must be a spurious ambiguity in the source VP. not between clauses. So if parallelism plays a role here. That much spurious ambiguity is inevitable. Two conclusions have to be drawn from this. but must be analyzed as a VP adjunct. it can only be a parallelism between VPs.210 ANAPHORA AND TYPE LOGICAL GRAMMAR S NP John VP revised his paper VP AdvP faster than Bill S VP did Figure 5. the interface between syntax and semantics has to supply the option of a pronoun being bound “sloppily” to the subject argument place of a superordinate verb before the overt subject is supplied. and a distributive reading (where she was just better than every individual boy). since the parallelism constraint in whatever form is unable to say anything about constructions like (41). this means that a comparative clause like collected more rose hips than Tom cannot be construed as being attached to the matrix S node.

Verb Phrase Ellipsis 211 restricted. As for the embedded clause. the subject Bill has to combine with the “copy” of the lifted VP to form a sentence.Ryx) ⇒ s Rb’b’ So we correctly predict there to be a sloppy reading in subordinating VPE constructions. .T (λxλy. But after that. Binding the pronoun to John right away leads to the unproblematic strict reading.Ryx) ⇒ np\s λx.T R λT. I assume the lexical assignment (44) faster’ : ((np\s)\(np\s))/s for faster than.Rxx So for the matrix. both solutions for the lifted type lead to the same result: np b’ (s/vp1 )\s λT. the unresolved VP has to be lifted to the ellipsis type (s/vp1 )\s to supply an appropriate antecedent for the target clause. We are interested in the reading where Bill’s lawyer defended Bill. it has to be lowered to the ordinary VP type np\s to serve as argument of the operator faster than.T R λT. Now let us see what the present theory has to say about the sloppy reading of examples like (41). it can only be bound to the subject of the matrix verb. There are two derivations for the first part. we derive the expected reading where John revises his own paper. Here again. Matters become more interesting if we combine the kind of non-subject sloppy scenario discussed in connection with (35) with subordination. we need a non-trivial theory of VP ellipsis beyond parallelism. What I said about the two goal types of the source VP above applies here as well. First.T (λxλy. The construction requires that we derive successively two goal types for the source VP revised his paper while leaving the pronoun unresolved. It goes without saying that this sentence has a strict reading where John was defended both by his own and by Bill’s lawyer. Reproducing the phrase structure given above in Categorial terms. If the pronoun him is not bound before ellipsis resolution (which leads to a strict reading). But we also correctly predict a sloppy reading. (45) John’s lawyer defended him better than Bill’s lawyer did. but they lead to the same result in the second part: vp1 ⇒ R (s/vp1 )\s λT. An analysis of VPE as hybrid in nature appears to be inevitable.

VPE in subordinated clauses follows the predictions of the original. To cope with these phenomena. Despite its simplicity. the resulting theory does a good job if we limit our attention to cases where either the target VP is subordinated or both source clause and target clause are main clauses and the subjects are simple. There is no way to derive a genuine nonsubject sloppy reading. while VPE in coordinated constructions is subject to few structural constraints. For the latter kind of cases. our final theory overgenerates considerably and has to resort to the assumption that ellipsis resolution is constrained by pragmatic factors. And this reading in fact does not exist. . Our main aim was to demonstrate that the Categorial logic LLC is a suitable base for a linguistically informed theory of VPE. Let us summarize the findings of this chapter. as an item with the type vp|vp that denotes the identity function. treating the stranded auxiliary in the target clause of VPE constructions as a pro-VP. In other words. Here the predictions derived with the “lifted” entry for the auxiliary are not different from the much simpler theory of Section 2.212 ANAPHORA AND TYPE LOGICAL GRAMMAR This leads to the subject-sloppy reading which is excluded here because it violates Binding Principle B.e. simple theory. pronoun resolution and quantifier scope was shown to be largely empirically adequate with respect to this fragment.q The account of the interaction between the three factors of VPE resolution. i. I first considered a very simple implementation. I modified the lexical entry of the auxiliary to a polymorphic proform over lifted VPs.. The resulting theory correctly predicts a strong asymmetry between coordination and subordination. This theory proved to be too restrictive though for cases involving inverted scope or constructions in which the antecedents of a sloppy pronoun are not the subjects of the source clause and the target clause respectively. we correctly predict sloppy readings in subordination constructions to be strictly limited to subjects.

Sign up to vote on this title
UsefulNot useful