Extended Finite State Models of Language (review)

Interest in finite-state analysis of language is still increasing despite Chomsky’s famous proof (in Syntactic structures, The Hague: Mouton 1957) that finite-state processes can’t do the job. The reason is that practical natural language processing systems need to be efficient more than they need to capture linguistic generalizations. Besides, any phrase-structure or transformational grammar can be transformed mechanically into a finite-state process with limits on depth of embedding.

The nineteen papers in this volume comprise the proceedings of aworkshop held at the European Conference on Artificial Intelligence (ECAI), Budapest, 1996. The postal and e-mail addresses of all 33 contributors are included; 11 have corporate rather than academic affiliations. (And given the nature of the industry, it’s a sure bet that several of them have moved!)

On p. 1 the editor urges us to make use of the enclosed CD ROM, which contains still more papers as well as software (ready to experiment with, not necessarily ready for nonprogrammers to run), and also to read Volume 2 of Natural language engineering, which contains different versions of many of the same papers.

The papers average just thirteen pages each, and many would benefit from fuller exposition. As is often the case with conference proceedings, the whole thing goes by rather fast. Many of the papers are reports of implementations rather than substantial scientific advances, and in some cases one gets the impression that authors in industry are not at liberty to describe their accomplishments as fully as they would like.