Book Reviews: Computational Linguistics and Formal Semantics

This is a collection of excellent papers from a workshop, chaired by the editors, held in Lugano at IDSIA, an institute of the Dalle Molle Foundation. I believe the workshop was held in September 1988, but the book is not explicit. The "formal semantics" (FS) of the title refers to the theories of NL meaning derived from model theory and in standard use in theoretical linguistics. The title (and the preface) suggest a thematic cohesion among papers that is actually lacking. Only some of the papers actually focus on relating FS to computational linguistics. There are honorable exceptions, however (which this review pays the most attention to), and the papers are generally of excellent quality, but on computational linguistics or formal semantics. "Unification" by Martin Kay is a lucid introduction to feature-based linguistic description, with Prolog examples included. It is additionally interesting for Kay's history of unification-based theories, which he traces not to the use of feature-based linguistic theories but rather to the difficulties of using a single grammatical specification for parsing and generation in ATNs. "Representations and Interpretations" by Jens Erik Fenstad, Tore Langholm, and Espen Vestre is best read as two papers. There is first a 40-page overview of the mathematics of feature structures (graphs), feature description languages (terms), and the attempt to find canonical data structures for these that allow the definition of efficient algorithms. This half of the paper is marred by incoherence in the only complicated example (2.48-49 on p. 52), but it is one of the best articles at an accessible level on this still very current topic. An updated version with discussion of typed theories would be welcome. The second half of Fenstad et al.'s paper is a 24-page development of situation semantics. It first introduces a proof theory for a situation-theoretic logic using a generalization of a Gentzen calculus (needed to deal with strong and weak negation simultaneously). The theorem prover is put to use in a question-answering system, but the report on the nonsemantic details of the system (morphology, parser, generator) unfortunately leaves little room for the development of a number of intriguing points on the potential and implementation of situation theory for computational semantics.