Gorrell presents a psycholinguistic model of human parsing in which specific properties of syntactic knowledge play a crucial role in determining ambiguity-resolution behavior. The proposal builds on work in determinis­tic parsing, relating it to grammar-based ac­counts of human syntactic preferences and garden-path behavior. Other grammar-based approaches have attempted to derive per­formance data from the content of particu­lar syntactic constraints and how well they are satisfied on-line. Gorrell instead proposes that the significant factor for human process­ing is the distinction between different types of structural information that are brought to bear in syntactic analysis. The classification of information types determines a natural modularization of the human parser, which is argued to account for the behavioral prop­erties observed. The approach has potential interest for computational linguists who are interested in the grammatical underpinnings of natural language understanding systems, and in the problem of deriving the best parse for an ambiguous utterance. In particular, issues of information encapsulation and its consequences for incremental interpretation are of great importance for cognitive mod­els of language as well as practical NLU sys­tems. Unfortunately, computational linguists are likely to be less than satisfied by the lack of precision in Gorrell's specification of his model, and by the absence of a firm compu­tational foundation for his claims.

The focus of Gorrell's proposal is the di­vision of structural information into primary and secondary relations. The primary rela­tions are the primitives that are necessary to a specification of the parse-tree structure (i.e., dominance and precedence). The sec­ondary relations are grammatical concepts that are defined in terms of the resulting tree (e.g., c-command and government). The distinction is reflected in the processing of the two types of information: Primary re­lations are asserted by a structure-building module, and secondary relations are evalu­ated within the resulting tree by a separate structure interpreter. Given the distinction between building a structure and verifying relations within that structure, Gorrell argues for a new kind of determinism in parsing, called structural determinism, which applies only to the structure-building module. The constraints on the operation of the structure builder underlie Gorrell's account of human disambiguation.

The book is thorough in its review of Government-Binding theory (the grammat­ical framework assumed by Gorrell) and of GB-based models of human parsing. Other aspects of the book are less complete, leav­ing a number of questions unanswered. For example, given that assertions of primitive relations cannot be retracted in normal pars­ing, the mechanism for determining which assertions to make is of critical importance. Gorrell proposes that the structure-building module is driven to "[attempt] incrementally to satisfy principles of grammar." As stated, this entails that the structure builder needs to judge the satisfiability of syntactic con­straints in order to make assertions about the parse-tree structure. However, evaluat­ing the structural conditions under which those constraints are satisfied is the purview of the structural interpreter, for which it in turns needs the assertions about the basic tree structure. The strict modularization of the proposed model is the foundation of Gor­rell's proposal, but it is not clear how the division can be maintained, given that the operation of each module depends on infor­mation provided by the other.

The book focuses on a topic in psycholin-guistics that should be of great importance in building parsers that can mimic human be­havior and respond to human expectations. However, the relevance of the particular pro­posal for computational linguists is less es­tablished, given the loose development of the specific components of the model. Computa­tional linguists who model human sentence-processing behavior will find an interesting discussion of some familiar issues in the field (such as form- versus content-based prefer­ence strategies, and modularity). The impact of the central proposal awaits a more thor­ough and precise development of the de­tails of the model.—Suzanne Stevenson, Rut­gers University

Jennifer Cole, Georgia M. Green, and Jerry L. Morgan (editors)

(University of Illinois at Urbana-Champaign) "The papers in this volume were presented at a workshop held in June 1991 at the Univer­sity of Illinois at Urbana-Champaign, in cele­bration of the twenty-fifth anniversary of the Department of Linguistics. The broad theme of the workshop was Computational Linguis­tics and the Foundations of Linguistic Theory, and its purpose was to provide a forum in which computational linguists, syntacticians, mor-phologists, and phonologists could come to­gether and discuss important issues of com­mon interest.

The areas of linguistic research represented in this volume are diverse, yet they are united under the notion that the princi­ples of grammatical well-formedness and the principles of language processing are interdependent____ The fact that many of the papers included here span several ar­eas within linguistic theory underscores the importance of communication between re­searchers in diverse areas."—From the preface

The contents of the volume are as follows:

"Formal devices for linguistic generaliza­tions: West Germanic word order in LFG" by Annie Zaenen and Ronald M. Kaplan

"Stratified feature structures for multistratal relational analyses" by David E. Johnson and Lawrence S. Moss

"Feature-based grammars as constraint grammars" by Alan M. Frisch

"A quarter century of computation with transformational grammar" by Robert C. Berwick and Sandiway Fong

"Chunks and dependencies: Bringing pro­cessing evidence to bear on syntax" by Steven Abney

"Some open problems in head-driven gen­eration" by Dale Gerdemann and Erhard Hinrichs

"Construction of LR parsing tables for gram­mars using feature-based syntactic cate­gories" by Tsuneko Nakazawa

"Phonology and computational linguistics— a personal overview" by John Coleman

"Eliminating cyclicity as a source of complex­ity in phonology" by Jennifer S. Cole

"Pitch accent prediction from text analysis" by Julia Hirschberg and Richard Sproat

Mary Dalrymple, Ronald M. Kaplan, John T. Maxwell III, and Annie Zaenen (editors)

(Xerox Palo Alto Research Center and Stan­ford University) "We have pulled together into a single vol­ume a set of papers by ourselves and our colleagues that address some of the devel­opments [in Lexical-Functional Grammar] of the past years. This book outlines work in formal issues in LFG theory in the twelve-year period from 1982 to 1994. We have in­cluded papers on a range of topics that have been central in LFG research during this pe­riod. In particular, we have tried to include those papers which have been most influen­tial, as well as those that have previously been most inaccessible."—From the introduc­tion