Computational approaches to morphology and syntax (review)

Computational approaches to morphology and syntax provides an overview of the fields of computational morphology and syntax (i.e. parsing), covering both classic techniques and the state of the art. After a general introduction, the book divides cleanly into two halves, ‘Computational approaches to morphology’ and ‘Computational approaches to syntax’, allowing readers to focus on one without reading the other.

Ch. 1, ‘Introduction and preliminaries’ (1–19), provides background information on core elements of computational processing: finite-state automata and transducers. It moves at a brisk pace, but offers a solid platform for the material in the rest of the book, particularly that of computational morphology. Composition is discussed more thoroughly, since this material especially will be important for relating different approaches to morphology. Of particular interest to linguists is the end of this chapter, which lays out a viewpoint that shapes the rest of the book: namely, the utility of finite-state methods for analyzing morphology and syntax, despite their proven ‘inadequacy’ in theoretical research.

Ch. 2, ‘The formal characterization of morphological operations’ (23–61), outlines the formal properties of morphological operations and demonstrates how composition can account for various phenomena. The thorough coverage of this chapter is impressive: diverse morphological issues are tied together nicely in one framework, and one can see how the methods can apply to larger-scale work in computational morphology. The chapter covers various types of syntagmatic variation, including simple concatenation, prosodically governed concatenation, subsegmental morphology, subtractive morphology, and root-and-pattern morphology. Each of these operations is treated via function composition: that is, the suffix is an operation that applies to the stem. In cases like root-and-pattern morphology, the authors provide an insightful comparison of their approach with other prominent approaches to accounting for nonconcatenative morphology, such as that of Beesley and Karttunen (2003). The rest of the chapter deals with paradigmatic variation, that is, relating complex forms to each other, and the difficult problem of reduplication, as morphological reduplication requires one to go beyond finite-state methods.

Building off of the description of treating morphological operations through composition, Ch. 3, ‘The relevance of computational issues for morphological theory’ (62–99), is the most relevant for linguists, since it delves into what computational morphology can say about morphological theory. In particular, the authors argue that long-standing debates in morphology (e.g. incremental vs. realizational approaches) make very little difference from a computational perspective. To illustrate their point about the equivalence between different approaches, they provide three grammar fragments based on Stump 2001, which show that all three approaches can be implemented with equal ease. The discussion surrounding these three fragments is enlightening, and the authors provide an appendix that contains a full description of the lextools toolkit, which would allow one to quickly begin working with morphological tools.

The advantage of the treatment of morphology in this book is further on display in Ch. 4, ‘A brief history of computational morphology’ (100–115). This chapter focuses mainly on one way of considering computational morphology. Because of this, the reader will understand the issues very well but may not gain a broader perspective. The chapter, as the authors state, is really a selective history of computational morphology that focuses on the KIMMO two-level morphological analyzer developed by Kimmo Koskenniemi (Koskenniemi 1983, Beesley & Karttunen 2003). This selection is justified, however, because two-level morphological analysis—built upon finite-state transducers—has dominated computational morphology and has been used for large-scale morphological analyzers for different languages. The authors also provide some pointers to more ad hoc approaches to morphology (e.g. the Porter stemmer). The main difference with generative phonology (at least in the early 1980s) was that KIMMO directly specifies the relation between a lexical and a surface form, instead of involving a series of ordered rewrite rules. Thus, the chapter nicely ends with a comparison of two-level morphology with an approach...