Constraint-Based Syntax and Semantics

Natural Sciences and Engineering Research Council of Canada
Discovery Grant, 2009-2014
Grant #371969

Principal Investigator: Ash Asudeh, Carleton University

Computational natural language processing has typically been caught between two complementary forces: 1. the desire for broad coverage, which is useful for natural language technologies like internet search, and 2. the desire for deep understanding, which is useful for natural language technologies like text summarization, translation and dialogue systems. Automated natural language understanding ultimately relies on the development of computational semantics. The work supported by this grant focuses on computational semantics for Lexical-Functional Grammar (LFG), a constraint-based theory of natural language syntax. Constraint-based syntactic theories have formed the heart of the deep grammar engineering approach for nearly thirty years.

The LFG grammatical architecture provides a number of unique opportunities for pursuing both theoretical and computational issues about natural language syntax and semantics and the relationship between them. First, the architecture supports hybrid approaches that have the potential to combine advantages of two distinct traditions in how semantic representations are built. Second, LFG's syntactic representations use a rich vocabulary that encodes syntactic cues that are directly relevant to certain important semantic-pragmatic categories, such as presupposition and implicature. The computational treatment of these sorts of pragmatic effects is of growing importance, since they are pervasive in natural language and any system that aims for deep textual understanding must be able to handle them. Third, the long-standing Parallel Grammar and newer Parallel Semantics projects not only provide a broader natural community for the research, they also bring to the fore questions of cross-linguistic semantics, which relate this work to machine translation.