This paper describes a new interpretation of Tree Adjoining Grammars (TAG) that allows the embedding of TAG in the unification framework in a manner consistent with the declarative approach taken in this framework. In the new interpretation we present in this paper, the objects manipulated by a TAG are considered to be descriptions of trees. This is in contrast to the traditional view that in a TAG the composition operations of adjoining and substitution combine trees. Borrowing ideas from Description Theory, we propose quasi-trees as a means to represent partial descriptions of trees. Using quasi-trees, we are able to justify the definition of feature structure-based Tree Adjoining Grammars (FTAG) that was first given in Vijay-Shanker (1987) and Vijay-Shanker and Joshi (1988). In the definition of the FTAG formalism given here, we argue that a grammar manipulates descriptions of trees (i.e., quasi-trees); whereas the structures derived by a grammar are trees that are obtained by taking the minimal readings of such descriptions. We then build on and refine the earlier version of FTAG, give examples that illustrate the usefulness of embedding TAG in the unification framework, and present a logical formulation (and its associated semantics) of FTA G that shows the separation between descriptions of well-formed structures and the actual structures that are derived, a theme that is central to this work. Finally, we discuss some questions that are raised by our new interpretation of the TAG formalism: questions dealing with the nature and definition of the adjoining operation (in contrast to substitution), its relation to multi-component adjoining, and the distinctions between auxiliary and initial structures.

this paper is to motivate the use of inheritance in lexical specification. To do this, we take a narrowly circumscribed phenomenon in English grammar--that of vp-complementtaking adjectives, as in hard + to deliver--and spell out the lexical specifications a thorough treatment demands. The sheer complexity of these specifications cries out for a redundancy-eliminating approach, and we propose a structured lexicon treatment. The grammatical analysis not only serves to motivate the general approach, it also illustrates several key issues in the design of structured lexicons, such as the use of default inheritance, the need for lexical rules, and the range of phenomena amenable to this sort of treatment

This paper begins with a review of Bresnan's proposals concerning inflectional systems, and evaluates the extent to which they make empiricallytestable predictions. While Bresnan's current proposal is not yet precise enough to enable one to determine just what the range of possible agreement systems is predicted to be, it does seem that several putative linguistic universals concerning agreement can be stated in her framework.

The development of large coverage, rich unification- (constraint-) based grammar resources is very time consuming, expensive and requires lots of linguistic expertise. In this paper we report initial results on a new methodology that attempts to partially automate the development of substantial parts of large coverage, rich unification-(constraint-) based grammar resources. The method is based on a treebank resource (in our case Penn-II) and an automatic f-structure annotation algorithm that annotates treebank trees with proto-fstructure information. Based on these, we present two parsing architectures: in our pipeline architecture we firstextract a PCFG from the treebank following the method of [Charniak, 1993; Charniak, 1996] , use the PCFG to parse new text, automatically annotate the resulting trees with our f-structure annotation algorithm and generate proto-f-structures. By contrast, in the integrated architecture we firstautomatically annotate the treebank trees with fstructure information and then extract an annotated PCFG (A-PCFG) from the treebank. We then use the A-PCFG to parse new text to generate proto-fstructures. Currently

This paper discusses an approach to feature logic called the structural approach. The method consists in first viewing feature structures as models (or relational structures) in the standard model theoretic sense, and then going on to consider various languages that can be interpeted on these structures and the way these languages are interrelated. It will be shown that the structural approach offers...

Montague's famous characterization of the homomorphic relation between syntax and semantics naturally gives way in computational applications to constraint-based formulations. This was originally motivated by the close harmony it provides with syntax, which is universally processed in a constraint-based fashion. Employing the same processing discipline in syntax and semantics allows that their processing (and indeed other processing) can be as tightly coupled as one wishes -- indeed, there needn't be any fundamental distinction between them at all. In this paper, we point out several advantages of the constraint-based view of semantics processing over standard views. These include (i) the opportunity to incorporate nonsyntactic constraints on semantics, such as those arising from phonology and context; (ii) the opportunity to formulate principles which generalize over syntax and semantics, such as those found in head-driven phrase structure grammar; (iii) a characterization of semantic ...

Abstract not found

This paper contrasts two different linguistic models of agreement in natural language. The first is an account of agreement commonly given in complex-feature, "unification-based" theories of Grammar (UBG), where agreeing categories impose constraints on some shared agreement value, and the construction is well-formed only if these constraints are satisfiable or mutually consistent. The second is a simple theory of agreement inspired by Lambek Categorial Grammar (LCG) presented in Bayer and Johnson (1995). In this theory agreement phenomena are modelled in terms of the requirement that arguments must be subsumed by, or logically imply, the corresponding argument specification of a predicate or functor category. We will see that this model, when embedded in the logic of LCG, makes a number of interesting and linguistically correct predictions that are not made by the UBG account. 2 Agreement in UBGs

This paper investigates whether the fundamental linguistic insights and intuitions of Lexical Functional Grammar (LFG), which is usually presented as a "constraint-based" linguistic theory, can be reformulated in a "resource sensitive" framework using a substructural modal logic. In the approach investigated here, LFG's f-descriptions are replaced with expressions from a multi-modal propositional logic (with permutation and possibly limited contraction). In effect, the feature structure "unification" basis of LFG's f-structures is replaced with a very different resource based mechanism. It turns out that some linguistic analyses that required non-monotonic devices in LFG (such as the "constraint equations" in the Andrews (1982) analysis of Icelandic) can be straightforwardly expressed in the framework presented here. Moreover, a Curry-Howard correspondence between proofs in this logic and -terms provides a semantic interpretation as a by-product of the process of showing syntactic we...

Four different kinds of grammars that can define crossing dependencies in human language are compared here: (i) context sensitive rewrite grammars with rules that depend on context; (ii) matching grammars with constraints that filter the generative structure of the language, (iii) copying grammars which can copy structures of unbounded size, and (iv) generating grammars in which crossing dependencies are generated from a finite lexical basis. Context sensitive rewrite grammars are syntactically, semantically and computationally unattractive. Generating grammars have a collection of nice properties that ensure they define only “mildly context sensitive” languages, and Joshi has proposed that human languages have those properties too. But for certain distinctive kinds of crossing dependencies in human languages, copying or matching analyses predominate. Some results relevant to the viability of mildly context sensitive analyses and some open questions are reviewed.