On-line papers

“The (dis)organization of the grammar: 25 years”, to appear in Linguistics and Philosophy, 25.6, 2002. (Special 25th Anniversary issue of L&P)

This paper has two goals: (a) to show explicaitly why the hypothesis of direct compositionality represents a simplification over competing theories of the organization of the grammar, and (b) to argue that its abandonment in much current literature was premature. There is, for example, a certain amount of literature arguing against what I call “strong direct compositionality” - this is the hypothesis of direct compositionality combined with the claim that the syntax makes use only of concatenation operations (i.e., operations equivalent to context-free phrase structure rules). One might question whether these arguments actually go through - but the point here is that even if they did, they certainly would not necessitate the abandonment of direct compositionality in general nor would they lead to the conclusion that there must be some kind of representational level like Logical Form. There are other versions of the hypothesis of direct compositionality (as, for example, in the work of Montague) which are still much simpler than the currently fashionable view that the syntax first “builds” representations which are then “sent” to the semantics. Thus the general program direct compositionality has, in much work, been rejected for no empirical or theoretical reasons.

[a link to the Kluwer on-line version will be added]

back to top

“Direct Compositionality and Variable-Free Semantics: The Case of Binding into Heads”, in B. Jackson (ed.), Proceedings of the 12th Conference on Semantics and Linguistic Theory, Ithaca, NY: Cornell University CLC Publications, 2002.

Conventional wisdom has it that a bound pronoun must be c-commanded by a “binder”. When this requirement is not visibly met on the surface, conventional wisdom posits a level of representation - a “reconstruction level” - at which certain “moved” material and/or relative clause heads are in the position of a gap. This allows the c-command restriction to be met at the relevant level - and so “binding” takes place at that level. A typical case involves the apparent binding of the bold-faced pronoun within the complex head in (1):

(1) a. The relative of his that no man would forget to invite to his wedding is his mother.

b. The (only) woman that he loves that every man invites to his wedding is his mother.

This paper examines this construction in some detail, and argues first that a reconstruction solution is actually of no help: there is no simple way to get the correct semantics by invoking a reconstruction level. In a variable-free semantics, on the other hand “binding” is not a relationship between two NPs/DPs/variables/traces in a c-command relationship - with the consequence that there is no particular reason to think that, e.g., no man in (1a) must c-command his at some level. Under the variable-free program we can assign the right meaning to these cases by means of a very simple and natural type-shift rule, which shifts a 2-place relation into a set of functions.

back to top

“Binding without pronouns (and pronouns without binding), to appear in R. Oehrle and G-J. Kruiff (eds.), Binding and Resource Sensitivity, by Kluwer Academic Press.

Within the general program of Variable-Free semantics, there are a few different proposals as to how to handle binding. One - proposed in, among others, Szabolcsi (1988) and in recent work within the Type Logical Tradition - builds the binding effect into the meaning of the pronoun. Another approach - which I have explored in much of my work - uses a type-shift rule operating on the meaning of the verb. At first glance, the “binding as pronoun meaning” view is attractive - for any theory needs to assign some meaning to the pronoun and so this removes the need for an additional type-shift rule. Nonetheless, a closer examination of the realm of “binding effects” indicates that this view cannot be correct - there are many cases where we find binding effects but where no pronoun is present (examples: functional questions; the analysis of

Deletion). And of course there are pronouns which are not bound (free pronouns), a fact which significantly complicates the “binding-as-pronoun-meaning” view. The larger conclusion here is that the domain of binding provides evidence that natural language grammars do make use of a small set of type-shifting operations, but I argue that the particular operations proposed here are extremely simple and natural.

back to top

“Paychecks and Stress”, in Proceedings of the Tenth Conference on Semantics and Linguistic Theory. Ithaca, New York: Cornell University, CLC Publications, 2000.

This paper is concerned with the interaction of pronouns and contrastive stress. In earlier work, Sauerland (1999) argued that the fact that bound pronouns can sometimes be contrastively stressed provides evidence that variable names are crucial (his idea: stress is possible if the pronouns correspond to different variable names). This would thus present a challenge to a variable-free semantics - which claims that variable names (and indices) play no role in grammar.

Hence this paper argues that the full set of facts surrounding contrastively stressed pronouns actually can’t be accounted for by invoking variable names. There are cases where pronouns should be able to correspond to different variables, but where stress is nonetheless impossible. The key generalization seems instead to have to do with the fact that the domain of quantification of their “binders” contrast - and this in turn can be accounted for in a variable-free semantics by allowing pronouns to be the identity function over different domains.

This approach has striking payoffs with respect to the interaction of contrastive stress with paycheck pronouns. (A paycheck pronoun is a pronoun which exhibits sloppy identity, as in “The woman who put her paycheck in the bank was wiser than the one that put it in the Brown University Employees Credit Union”. Note that it here is not an ordinary bound pronoun, nor is it an ordinary free pronoun (it doesn’t pick up some contextually salient individual.) In the “standard” (variable-ful) view, paycheck pronouns have either complex meanings and/or complex representations at “LF”. However, when only part of that meaning/representation is in contrast, the paycheck pronoun can’t be stressed:

(1) Every THIRD grade boy LOVES HIS teacher, but every FOURTH grade boy HATES HIS teacher.

(2) Every THIRD grade boy LOVES HIS teacher, but every FOURTH grade boy HATES her/*HER.

Actually, the situation is even more complex - the paper shows that sometimes stress is allowed if only part is in contrast. The paper shows that the relevant facts fall out under the domain restriction view combined with the variable-free account of paycheck pronouns.

back to top

“Paycheck Pronouns, Bach-Peters Sentences, and Variable-Free Semantics", in Natural Language Semantic 8.2, 2000. pp. 77-155.

This paper focusses on a related group of phenomena: the interaction of paycheck pronouns with Weak Crossover effects and i-within-i effects. These interactions were first studied in Jacobson (1977) as they show up in Bach-Peters sentences. In the earlier work, I took these facts to provide evidence for the view that paycheck pronouns have a complex representation at LF - but here I show that actually all of the observations in this earlier work are compatible with the hypothesis of direct compositionality. The key tool is the adoption of a variable-free semantics where constraints on binding are constraints on the semantic combinatory operations and not constraints on representations.

In addition to the general consequences that this has for the organization of the grammar, there are two other main results. First, new arguments for a variable-free semantics is given in that under this view, the paycheck reading of a pronoun comes “for free”. (Note that in other theories it is a complete accident that a pronoun has both its ordinary meaning and its paycheck meaning - here this is non-accidental, for the paycheck meaning follows from mechanisms motivated independently in the variable-free program). Second, I reiterate one of the central points in Jacobson (1977) - which is that the first pronoun in a Bach-Peters sentence is indeed a paycheck pronoun, and hence nothing special needs to be said about these sentences nor do we need any new machinery in order to account for their existence.

(click on Volume 8, 2000. Issue 2)

back to top

"Extraction out of tough", in Snippets 1, 2000.

A brief note questioning the conventional wisdom that tough constructions are not extraction islands. They are if the tough gap is in the VP immediately embedded under the tough adjective - but if the gap is sunk further down then we find what looks like run-of-the-mill island effects.

back to top

"Towards a Variable-Free Semantics", Linguistics and Philosophy 22, 1999. pp. 117-184.

This paper argues for a semantics making no use of variables or assignment functions. The paper contains both a technical and a more empirically oriented section: the first works out the details of a variable-free semantics and the second shows that this simplifies the analysis of a variety of constructions.

As to the technical details, the meaning of a pronoun is taken to be the identity function over individuals, and two very simple type-shift rules are proposed in order to effect “binding” and to allow an expression containing a pronoun to enter into the semantic composition. I show that these rules allow for cases with any number of pronouns, any number of binders, and binding in any order. In other words, there is no problem in handling the full range of cases which are normally handled by variable names and/or indices.

I argue that this view has several theoretical and empirical advantages over the “standard” view. First, it simplifies the semantic machinery, dispensing with assignment functions which have no motivation as meaningful objects. More importantly, a variety of empirical phenomena whose existence is surprising under standard views become completely expected here. This includes functional questions, certain kinds of unexpected (or “sloppy”) inferences, and cases of across-the-board binding. Under the standard account, all three require extra apparatus - here these come “for free” from the very mechanisms used for binding effects in general. Most importantly, I argue that the variable-free view allows for strict, direct compositionality - where each local expression is assigned a meaning as it is “built” in the syntax.

click on Volume 22 (1999) Issue 2

back to top

“Antecedent Contained Deletion and Pied-Piping: Evidence for a Variable-Free Semantics”, in proceedings of the Eighth Conference on Semantics and Linguistic Theory, Ithaca, NY: Cornell University CLC Publications, 1998.

This paper is concerned, first, with providing a direct compositional account of Pied-Piping: this turns out to be straightforward and require no extra mechanisms under the variable-free approach. I then examine some interactions of Antecedent Contained Deletion (ACD) and Pied-Piping: these center on the kind of asymmetry found in (1):

(1) a. John voted for every candidate the father of whom Bill had.

b. ??Every candidate the father of whom Bill had, John voted for.

c. *Every candidate the father of whom Bill had voted for, John did.

(The awkwardness of (b) can be attributed to the fact that an ellipsis site generally doesn’t like to precede its antecedent; the surprise here is that this is much better than (c) which is absolutely impossible.)

I show that these facts all fall out assuming: (a) the analysis of ellipsis in Rooth, 1994 according to which ellipsis is subject to a “focus” condition as well as an identity condition; (b) the analysis of ACD as transitive verb (phrase) ellipsis which is proposed in Jacobson, 1992a, 1992b; and (c) the variable-free analysis of Pied-Piping. What is most interesting is that the grammaticality of cases like (1a) is quite problematic under the standard view of ACD and of Pied-Piping - there simply is no way to construct a representation and/or meaning for the elided VP which is identical to that of the matrix VP. But under the view that ACD is actually an ellipsis of a 2-place relation (i.e., a transitive verb phrase - rather than ellipsis of a VP type meaning - the analysis of (1) is unproblematic.

back to top

“The Locality of Interpretation: The Case of Binding and Coordination” in Proceedings of the 6th Converence on Semantics and Linguistic Theory. Distributed by Cornell University Working Papers in Linguistics, 1996.

There is a close connection between variable-free semantics and the broader hypothesis of direct compositionality, and this paper elucidates how and why these are connected. The connection is exemplified by an examination of cases in which there is “across-the-board binding” - this can happen in, for example, a Right Node Raising construction, as in:

(1) Every man loves but no man wants to marry his mother.

I show that this phenomena comes for free under the Dowty (1987)/Steedman (1989) approach to coordination and Right Node Raising (which is a direct compositional/Categorial Grammar account) combined with variable-free semantics. No abstract levels nor indexing conventions are needed to account for this. Moreover, there are some interesting constraints on where this kind of across-the-board binding is possible (if there is binding out of one conjunct there must be binding out of both) which also follows immediately from the account proposed here.

back to top

“Binding Connectivity in Copular Sentences”, in Proceedings of the 4th Conference on Semantics and Linguistic Theory. Distributed by Cornell Working Papers in Linguistics, 1994.

Higgins (1974) first noticed a number of connectivity effects in specificational copular sentences, and a common strategy to account for these is to posit a level of representation at which the post-copular constituent is surrounded by material from the pre-copular constituent. A typical example involves the apparent binding of the bold-faced pronoun in (1) (this example due originally to Geach, 1974):

(1) The (only) woman that every man loves is his mother.

This paper argues that the apparent binding here does not necessitate positing any silent material or abstract levels. It follows immediately under a direct compositional approach from a straightforward generalization of the anlaysis of functional questions given in Groenendijk and Stokhof (1983) and Engdahl (1986). The pre-copular constituent denotes a functional NP, and the post-copular constituent denotes a function of type <e,e>. Strikingly, just about everything needed for this result comes for free in a variable-free semantics from the general mechanisms used for binding. (For example, it follows immediately that his mother denotes a function of type <e,e> - and the functional reading of the NP is automatic except for the need to adopt one additional type-shift rule which shifts woman into a set of functions whose range is women.

Moreover, the paper also comments on the apparent connectivity found with reflexive pronouns - both reflexives which are sunk within VPs as in (2a) and bare reflexives as in (2b):

(2) a. What John did was wash himself.

b. The only person John likes is himself.

(2a) is straightforward under proposals for reflexives such as that in Bach and Partee (1981) or Szabolcsi (1989); (2b) is unproblematic given observations developed in Pollard and Sag (1991). None of these constructions therefore require abstract levels or silent elements.

back to top

“i-within-i effects in a Variable-Free Semantics and a Categorial Syntax”, in P. Dekker et al. (eds.), Proceedings of the 9th Amsterdam Colloquium. Amsterdam: ILLC, University of Amsterdam 1994. pp. 349-368.

“I-within-i” effects are exemplifed by impossible NPs like *the wifei of heri childhood sweetheart. Most past accounts of this simply involve some kind of stipulation - such as a constraint on co-indexation in a given configuration (cf., Chomsky 1981). However, this effect follows naturally given the account binding making use of a type-shift rule (as in, e.g., Jacobson, 1992) combined with a Categorial Grammar syntax and combined with one other independently motivated assumption: common nouns do not have a syntactic subject slot, even though they are arguably of type <e,t>. (For example, even in in small clause environments they do not occur with subjects.) If correct, this provides evidence both for the type-shift account of binding and for a theory in which the semantic shift is closely matched with a corresponding syntactic operation, as would be expected in Categorial grammar.

back to top

"Where (if anywhere) is transderivationality located?", in P. Culicover and L. McNally (eds.), The Limits of Syntax. Academic Press (Syntax and Semantics Series), 1997, pp. 303-336.

Within the tradition of Generative Semantics in the 60’s and early 70’s - and now again within the “Minimalist Program”, it is often proposed that the grammar has access to a set of competing derivations and that - from these - it picks the “simplest”. Call this economy-driven transderivationality. But this view is problematic: assuming that a set of competing derivations is computed it would be completely accidental for the “simplest” one (the one with fewest steps) to have any privileged status.

This paper pursues this in some detail, and tries to argue for two main points. First, simplicity-based effects have to be relegated to processing and/or production principles - not to grammatical principles. But this too must be approached with caution - to say that the processor “computes’ or “picks” the simplest/shortest derivation would also be to attribute mystical powers to it unless one thinks of this in terms of a race model (compute all in parallel, stop when a sensible meaning is reached). Second, I examine one well-known case of economy-driven transderivationality which does not in fact lend itself to a processing explanation - the account of the Sag (1976)/Hirschbuhler (1979) scope and ellipsis interactions explored in Fox (1995). I argue that the transderivational analysis makes several incorrect predictions and does not handle the full relevant range of facts, and suggest that the actual effect derives from some subtle interactions of the fine-grained semantics with discourse considerations.

back to top

On-line handouts

"On Apparent Exceptions to Weak Crossover in a Variable-Free Semantics", Conference on (Non-Lexical) Semantics, University de Paris 7 (Jussieu), Paris, June 1996.

Lasnik and Stowell (1991) note some surprising exceptions to Weak Crossover which show up in both the tough construction and in parasitic gaps. Under the standard account, the surprise here is that a pronoun and a gap need to correspond to the same variable name - but in general if the pronoun precedes the gap this is impossible. And yet we find cases of this type in the tough construction and in parasitic gaps.

This paper argues that these exceptions are unsurprising under the variable-free account of binding and Weak Crossover. There is no use of variable names - “binding” is the result of a type-shift rule which, in effect, semantically “merges” argument positions, and Weak Crossover effects are a consequence of how this rule is formulated. (Thus Weak Crossover is a constraint on the semantic combinatorics.) In the cases which look like exceptions to Weak Crossover, it turns out that the gap and the pronoun are not directly merged Put very informally, they are “co-bound” by another argument slot. In both the tough construction and in the parasitic gaps case, there is a higher verb which can undergo the relevant binding type-shift rule twice; when this happens the pronoun and the gap end up being “merged”. These apparent exceptions are thus just what we would expect to find under this view of binding.

back to top

“Binding, Extraction, and Coordinate Structure Constraint Effects”, Symposium on Coordination, European Summer School in Language, Logic, and Information, Prague, August, 1996.

This paper is concerned with accounting for the following group of facts surrounding extraction, binding, and coordinate structures: (a) across the board binding is possible out of two conjuncts, but (b) not out of just one; (c) binding into just one conjunct is possible, (d) binding of one conjunct is possible; (e) leftward extraction out of just one conjunct is sometimes possible (a fact noted in, e.g., Goldsmith, 1985, Lakoff, 1986 and many others); (f) but rightward extraction out of just one conjunct is in general impossible - yet it is possible in sentences like those of Lakoff (1986) just in case the rightmost conjunct contains a gap. These facts are accounted for by adopting a variable-free semantics, assuming that and is Curry’ed and listed in the lexicon (see also Munn, 1993), adopting a type-shift rule to account for cases of iterated conjunction (with no visible and or or) and adopting the view of conjunction put forth in various literature in Categorial Grammar (cf., Oehrle, 1990).

back to top

“Pied Piping Semantics in Relative Clauses”, Workshop on the Syntax and Semantics of Relative Clauses, Tel Aviv University, June, 2001.

This talk defends the program of direct compositionality against some obvious challenges drawn from the domain of Pied-Piping. These include Pied-Piping semantics in gneral, and the interaction of direct compositionality with ACD; the interaction of Pied-Piping with Weak Crossover effects, and the interaction of Pied-Piping with “Principle C” effects which have been discussed in a lot of the literature on reconstruction. While there remain open questions and challenges for direct compositionality, I try to provide a preliminary sketch of some of these domains and some reasons to believe that non-direct compositional solutions will not turn out to be correct. The paper also explores the consequences of this domain for the semantics of relative pronouns and - in turn - for the “radical lexicalist” hypothesis which tries to put as much burden as possible on lexical meaning (the paper argues against this). Finally I consider a puzzle surrounding the lack of de dicto readings of Pied-Piped material in certain relative clauses.

back to top

“Direct Compositionality (and Variable-Free Semantics): The Case of Antecedent Contained Deletion."

Antecedent Contained Deletion has often been taken as one of the strongest pieces of evidence for a level of Logical Form (see, e.g., Sag 1976): the received wisdom is that there is no way to construct an antecedent for the “missing” VP ubnless one pulls an object Np out of the representation and leaves a variable in its place. But this is based on the assumption that the missing material does indeed have a VP type meaning. In fact, though, one can instead take the missing material to be a 2-place relation (a transitive verb phrase type meaning). Given the Categorial Grammar account of relative clause composition proposed in, e.g., Steedman (1987) this is exactly what should in fact be supplied. Under this view, the antecedent containment paradox disappears, and one can easily provide a direct compositional analysis of run-of-the-mill ACD cases.

But there is one case discussed originally in Bouton (1976) which appears to remain problematic for the transitive verb phrase ellipsis view:

(1) John voted for every woman who thought he would [vote for her].

What appears to be missing here is the VP-meaning vote for her. And if that is correct, then once again we need LF - for we need to “raise” the object NP of the matrix sentence in order to get a variable in object position. I show that this reasoning is also mistaken - it crucially assumes that pronouns corresond to variables. Under the variable-free view the meaning of a VP like vote for her is in fact just a 2-place relation - and so the transitive verb phrase ellipsis analysis can be maintained without difficulty. (Additional cases of ACD are also discussed.) The result is that ACD provides no evidence for LF.

back to top