Systemic Functional Grammar in Natural Language Generation: Linguistic Description and Computational Representation

Teich's book will be of interest to three groups of readers. First, specialists in Systemic Functional Grammar (SFG) will find a detailed description of the use of SFG theory for a specific application, a discussion of some fundamental problems in the theory that are revealed by the application, and proposals for a modification of SFG theory to handle these problems. Second, those working in natural language generation (NLG) will find a detailed discussion of why SFG is effective for NLG, an interesting comparison of the SFG-based Penman system with three other generation systems, and a description of a fully implemented SFG-based surface realization system for German. The third group is less likely to read the book, but I recommend it to specialists in Head-driven Phrase Structure Grammar (HPSG). After identifying some fundamental problems in SFG theory, the author explains that these problems have been fully solved in HPSG. However, rather than abandoning SFG and adopting HPSG, the author prefers to try to integrate the HPSG-style solutions into the existing SFG theory. This conviction that the wider SFG framework has major advantages, so significant that they outweigh the identified problems, should be respected as deeply challenging for HPSG theory. The basic approach to NLG, using SFG and the Penman system, has already been described in an earlier book (Matthiessen and Bateman 1991) from the same series, with examples from English and Japanese. Teich's distinctive contributions are the development of a large SFG-based grammar for German using the KOMET-Penman system, and a detailed analysis of the problems encountered due to the absence of any notion of head-daughter dependency within SFG theory. Comparing SFG and HPSG, Teich proposes a modification of SFG theory to allow the inclusion of dependency relations. Chapter 1, "Introduction", presents the motivation for the book. The terminology used within SFG theory sometimes appears to be unnecessarily daunting. For example, we learn on p. 3 that the background is "linguistic theory as metasemiosis." Chapter 2, "Theory and linguistic representation: Systemic functional linguistics," is intended as a short introduction to SFG theory. This is very well written for SFG specialists, but it covers an enormous amount of ground at great speed (even summarizing the theoretical differences between Hudson, Huddleston, Henrici, Halliday, Fawcett, and Berry) and is unsuitable as a first introduction to SFG. Instead, I recommend Berry (1975 and 1977). Although this old textbook was never intended to be used for any computational implementation, it gives a very clear account of how SFG actually works, and is especially good on realization rules.