Executable Attribute Grammars for Modular and Efficient Natural Language Processing.

摘要

Language-processors that are constructed using top-down recursive-descent with backtracking parsing are highly modular, and are easy to implement and maintain. However, a widely-held inaccurate view is that top-down processors are inherently exponential for ambiguous grammars and cannot accommodate left-recursive syntax rules. It has been known that exponential time and space complexities can be avoided by memoization and compact graph-structured representation, and that left- recursive productions can be accommodated through a variety of techniques. However, until now, memoization, compact representation, and techniques for handling left-recursion have either been presented independently, or else attempts at their integration have compromised modularity and correctness of the resulting parses.;Specifying syntax and semantics to describe formal languages using denotational notation of attribute grammars (AGs) has been widely practiced. However, very little work has shown the usefulness of declarative AGs for constructing computational models of natural language. Previous top-down approaches fall short in accommodating ambiguous and general CFGs with arbitrary semantics in one pass as executable specifications. Existing approaches lack in providing a declarative syntax-semantics interface that can take full advantages of dependencies between attributes of syntactic constituents to model linguistically-motivated cases.;This thesis solves these shortcomings by proposing a new modular top-down syntactic and semantic analysis system, which is efficient and accommodates all forms of CFGs. Moreover, this system provides notation to declaratively specify semantics by establishing arbitrary dependencies between attributes of syntactic categories to perform linguistically-motivated tasks such as: building directly-executable natural-language query processors, computing meanings of sentences using compositional semantics, performing contextual disambiguation tasks, modelling restrictive classes of languages etc.