Theoretical issues in natural language processing Edited by Yorick Wilks (review)

424 LANGUAGE, VOLUME 67, NUMBER 2 (1991) remarkable resemblances in clause structure among these languages. The articles on specific problems in Armenian consonantism. verbal morphology, and syntax, on Georgian ablaut, phonemic structure, case system, verb morphology, syntax, and etymology , on Ossetian case, and on Kartvelian prehistory all remain valuable contributions both in their reliable data and in their careful methodology . Vogt's memoir on Ubyx is a fascinating history of the study of this language, which is remarkable, among other things, for the largest known consonant inventory in the world. His observations on Hjelmslev's and Jakobson 's case theories are valuable commentaries that have not lost their relevance. To be sure, the articles contain dated and obsolete elements; it would be a sad comment on lack of progress in our field if they did not. But what is far more striking is the amount of insightful and useful data and analyses. This is a highly worthwhile collection of a part of Vogt's work that has been scattered and difficult to find, as the editors point out in their brief preface . It deserves to be consulted by linguists both in and outside the Caucasian and Armenian areas. [Victor A. Friedman, University of North Carolina, Chapel Hill.] Theoretical issues in natural language processing. Ed. by Yorick Wilks. Hillsdale, NJ: Erlbaum, 1989. Pp. 243. Cloth $36.00, paper $17.50. This volume contains position papers and selected discussion from the third Tinlap (Theoretical Issues in Natural Language Processing) Workshop held in January 1987 in Las Cruces. New Mexico. The contents reflect a research frontier in a state of mild disarray, or, in the words ofthe editor, ' ... an underlying worry that the field has not progressed in the linear fashion that was hoped for at the time of the first workshop in 1975' (xiii). This worry is not necessarily bad (although many of the participants at the conference seemed to think so), but it is something that emerges clearly from a reading of the conference report. A panel entitled 'Why has theoretical NLP made so little progress?' indicates the tone. With 30 position papers, it is impossible to do justice to the full volume. Instead I will give an overview and then give more detail on one of the panels that seemed to elicit the most interest and discussion—'Unification and the new grammaticism '. The book is divided into 9 chapters, each one containing from 3 to 5 position papers. Three chapters have appended discussion. The chapters and their contents are (1) 'Word and world representation' (papers by Donald E. Walker, Branimir K. Boguraev, Robert A. Amsler, Jerry R. Hobbs. and Judy Kegl); (2) 'Unification and the new grammatism' (Gerald Gazdar. Steve Pulman, and Aravind K. Joshi); (3) 'Connectionist and other parallel approaches to language processing' (David L. Waltz, Garrison W. Cottrell, Eugene Charniak, James L. McClelland, and Wendy G. Lehnert); (4) 'Discourse theory, goals and speech acts' (C. Raymond Perrault and Robert Wilensky); (5) 'Why has theoretical NLP made so little progress?' (Norman K. Sondheimer and Larry Birnbaum); (6) 'Formal versus common sense semantics' (David Israel, Yorick Wilks, and Karen Sparck Jones); (7) 'Reference: The interaction of language and the world' (Douglas E. Appelt, Deborah A. Dahl, Amichai Kronfeld, and Bradley A. Goodman); (8) 'Metaphor' (Dedre Gentner, Brian Falkenhainer, Janice Skorstad, Andrew Ortony & Lynn Fainsilber, and Edwin Plantinga); and (9) 'Natural language generation' (Aravind K. Joshi, David D. McDonald, and Douglas E. Appelt). In the chapter on unification, Gazdar's paper, 'COMIT => * PATR IF, compares Victor Yngve's 1950s-vintage COMIT with the 1980s PATR II and finds similarities of purpose (both designed for specifying NL grammars for use in parsers) and design (both allowing feature-valued categories); but Gazdar also notes a number of ways in which the PATR II formalism is significantly different (e.g.. unification is the only permitted operation on features in rules in PATR II). Pulman's paper, 'Unification and the new grammatism', in addition to providing an interesting summary of some of the uses to which the unification formalism has been put, argues that the chief role of unification with respect to grammatical formalisms will be to 'provide ... an assembly language into which different linguistic theories...