Analogical modeling of language By Royal Skousen (review)

420 LANGUAGE, VOLUME 67, NUMBER 2 (1991) effect as a level of semantic representation) by analogy to collocational patterns in the corpus. Detection of analogy thus serves to disambiguate the original expression. S's system works by brute force: it directly calculates the relative overlap of collocational sets. For example, instead of storing hyponymy relations directly, S extracts them from collocational patterns. As the object of mendi 'buy', for example, the noun ligno 'wood' overlaps most collocationally with materialo 'material'; this can be calculated when necessary to determine the most appropriate translation. Similarly , if both source and target language collocations are taken into account, the best translation of a phrase can be selected by analogy to model translations in the corpus. The book is divided into two parts. Part 1 describes a working prototype with a knowledge bank (S's term) limited to immediate contextual information. The results are suggestive: while the system failed to achieve accuracy levels anywhere near that of a human translator, its collocational calculations of semantic relations were very close to the semantic judgments of human consultants. Part 2 describes the author's plan for a full-scale implementation, in which the knowledge bank would contain a large corpus of linguistically analyzed texts matched point by point to Esperanto translations. The approach embodied in this book has theoretical significance beyond the field of machine translation. In effect, it contains linguistic representations but no linguistic rules. Rules (in this case, rules of disambiguation and translation ) emerge from the interaction of the knowledge bank with the system of analogical reasoning. [Paul D. Deane, University ofCentral Florida.) Individual differences in second-language learning. By Peter Skehan. London: Edward Arnold, 1989. Pp. viii, 168. Paper $17.95. The publication of Skehan's book on individual differences (ID) in second-language learning is a timely publication and a welcome addition to the field. As S points out early in the text, the study of ID is well established in other fields of study, such as psychology, but is still relatively new in the field of second-language learning . It is not that any one of these areas is new to second-language learning, but that the organized study of many of the variables of ID is not as strong within our profession as perhaps it is in others. S begins by reviewing the theoretical foundations for ID. He establishes this foundation by reviewing models of second-language acquisition : Krashen's monitor model, the good language learner model of Naiman et al., John B. Carroll's interactional model of school learning, and S's own disjunctive model. S's survey provides a good foundation for discussing the concepts of ID, because no one model seems to deal adequately with the issues of individual learner differences. Ch. 2. 'Methodological considerations in ID research', provides a good statistical reference for teachers and scholars involved in ID research ; here S covers principles of questionnaire design and scale construction, bivariate relationships, and multivariate statistics. In Chs. 3-7 S reviews the literature on ID research in second language learning, covering the topics of language-learning aptitude, motivation, language -learning strategies, cognitive and affective influences on language learning, and interactions among these variables. These same topics are currently the center of research by others within the profession. S offers a synthesis and critique of the major studies within each of these areas. Perhaps a disappointing aspect of the book is the lack of attention given to the research technique of using think-aloud protocols to identify ID. This particular technique, which has been widely used outside the fields of linguistics and applied linguistics, is not without criticism, but it is one of the most valuable ways of assessing ID. In his final chapter, 'Conclusions and implications ', S indicates directions for future ID research . He suggests that the use of better research designs, replication of research, more longitudinal studies, and greater familiarity with and use of statistics will result in more significant research. He thus provides a research agenda for the coming years in the area of individual learner differences. [Neil J. Anderson. Ohio University.) Analogical modeling of language. By Royal Skousen. Dordrecht: Kluwer , 1989. Pp. xii, 212...