A Relational Syntax-Semantics Interface Based on Dependency Grammar

A Relational Syntax-Semantics Interface Based on Dependency Grammar Ralph Debusmann Denys Duchier∗ Alexander Koller Marco Kuhlmann Gert Smolka Stefan Thater Saarland University, Saarbrücken, Germany ∗LORIA, Nancy, France {rade|kuhlmann|smolka}@ps.uni-sb.de, duchier@loria.fr, {koller|stth}@coli.uni-sb.de Abstract We propose a syntax-semantics interface that realises the mapping between syntax and se- mantics as a relation and does not make func- tionality assumptions in either direction. This interface is stated in terms of Extensible De- pendency Grammar (XDG), a grammar formal- ism we newly specify. XDG’s constraint-based parser supports the concurrent flow of informa- tion between any two levels of linguistic rep- resentation, even when only partial analyses are available. This generalises the concept of under- specification. 1 Introduction A key assumption of traditional syntax-semantics interfaces, starting with (Montague, 1974), is that the mapping from syntax to semantics is functional, i. e. that once we know the syntactic structure of a sentence, we can deterministically compute its se- mantics. Unfortunately, this assumption is typically not justified. Ambiguities such as of quantifier scope or pronominal reference are genuine semantic am- biguities; that is, even a syntactically unambigu- ous sentence can have multiple semantic readings. Conversely, a common situation in natural language generation is that one semantic representation can be verbalised in multiple ways. This means that the relation between syntax and semantics is not func- tional at all, but rather a true m-to-n relation. There is a variety of approaches in the litera- ture on syntax-semantics interfaces for coping with this situation, but none of them is completely sat- isfactory. One way is to recast semantic ambiguity as syntactic ambiguity by compiling semantic dis- tinctions into the syntax (Montague, 1974; Steed- man, 1999; Moortgat, 2002). This restores function- ality, but comes at the price of an artificial blow- up of syntactic ambiguity. A second approach is to assume a non-deterministic mapping from syntax to semantics as in generative grammar (Chomsky, 1965), but it is not always obvious how to reverse the relation, e. g. for generation. For LFG, the oper- ation of functional uncertaintainty allows for a re- stricted form of relationality (Kaplan and Maxwell III, 1988). Finally, underspecification (Egg et al., 2001; Gupta and Lamping, 1998; Copestake et al., 2004) introduces a new level of representation, which can be computed functionally from a syntac- tic analysis and encapsulates semantic ambiguity in a way that supports the enumeration of all semantic readings by need. In this paper, we introduce a completely rela- tional syntax-semantics interface, building upon the underspecification approach. We assume a set of linguistic dimensions, such as (syntactic) immedi- ate dominance and predicate-argument structure; a grammatical analysis is a tuple with one component for each dimension, and a grammar describes a set of such tuples. While we make no a priori function- ality assumptions about the relation of the linguistic dimensions, functional mappings can be obtained as a special case. We formalise our syntax-seman- tics interface using Extensible Dependency Gram- mar (XDG), a new grammar formalism which gen- eralises earlier work on Topological Dependency Grammar (Duchier and …

Content truncated to 500 words