Book Reviews Synonymy and Semantic Classification

The book Synonymy and Semantic Classification by Karen Sparck Jones was published in 1986 as the first volume of Edinburgh Information Technology Series on computer science and artificial intelligence. The book is interesting both in its content and in some circumstances of its publication. The main part of the book comprises Sparck Jones's Ph.D. thesis, approved at the University of Cambridge as long ago as 1964. No changes in the text of the thesis have been made for this publication. All that has been added is a new chapter entitled "Twenty years later: A review", in which Sparck Jones acts as her own reviewer and considers her own work from the viewpoint of the results achieved in linguistics during the past 20 years. The book consists of seven chapters and two appendices. It reflects the complex paths along which linguists have to proceed when searching for the semantic structure of a vocabulary, for semantic primitives, and for the establishment of the parts they play in the automatic process of natural language text understanding. To realize the idea of machine translation of natural language texts, it is necessary, among other things, to have a dictionary that makes it evident what meanings each word-sign possesses, and to have a procedure for recognizing the meaning in which the word-sign is used in the context. For designing this kind of procedure, the author considers it important that a discourse is connected by one idea that recurs several times in the discourse. It should be discovered what kind of information is required to detect this idea. The author asks the question as to whether the existing thesauri can fulfill these requirements, and arrives at the conclusion that they cannot. She suggests a different approach: group all the elements of a vocabulary on the basis of the interrelations the words have with each other and define the meanings of words with the aid of these interrelations. The relations between words that a linguist stores in the dictionary are supposed to be semantic relations (i.e., relations between words), not factual relations (i.e., extra-linguistic references of words). After studying several types of relations (e.g., hyponymy, contrast, likeness, implication), the author arrives at the conclusion that synonymy as the exemplar of likeness of meaning is a semantic relation because it is dependent on the way in which the mechanism of using signs is functioning. It is therefore possible, on the basis of synonymy, to define the idea of the semantic structure of a vocabulary. The author gives analyses of several existing definitions of synonymy; however, she does not find them quite suitable for designing machine translation dictionaries and suggests a definition of her own, which is linked with the terms sentence, length of the sentence, ploy (a primitive notion, representing roughly the meaning, application, and form of a sentence), row (a set of elements that can replace one another with respect to a position n in a sentence s, without changing its ploy), word-sign, word-use, and word: "Two or more word-uses j are synonymous, and therefore form a row, if their word-signs are mutually replaceable at some position in some sentence, without any change in the ploy of the sentence" (p. 71). To each word-use it is possible to assign one and only one row consisting of all the word-signs mutually replaceable in a context, hence synonymous. Let us take, for example, word-signs A, B, C, D, E, F, and G, and let us consider word uses represented by the word-sign A. We may find, for instance, that A has three word uses characterized by three rows: