Browse Books

Reviewer: Gerald David Chandler

One of the most frequently used avenues for experimenting with natural language understanding is the creation of front-ends to databases systems. This book gives a description of one such effort, the QPROC system, which was the result of the authors PhD research. In QPROC, a response to a natural language question is formulated in two steps: First the natural language is translated to Wallaces “D&Q” (Description and Qualifier) formalism and, second, the D&Q representation is translated to a relational database language that resembles SQL. The system, written entirely in PROLOG, is composed of four modules: (1) DCG, an extension of PROLOGs Definite Clause Grammar; (2) SYNTAX, a parser which dynamically accesses the data dictionary; (3) DIC, a core dictionary in the form of PROLOG facts; and (4) EX, a program that executes the formal queries, expressed in D&Qs. This is a small book (only 170 pp.). One-third (Chapters 1, 2, and 7 (the last)) surveys other natural language work and sets the context for the current work. Just under a third (Chapter 3 and 4 and Appendices 2 and 3, which make up 50 pages) introduces and gives PROLOG specifications for D&Qs and their translation into ICLs Queymaster, a relative of SQL. Chapters 5 and 6 (39 pp.) are primarily concerned with the difficulties in parsing of natural language. The major interest of the book comes from the detailed presentation of the D&Q formalism, which is introduced by Wallace as being preferable to a standard parse tree representation of sentences. Effectively, in D&Q formalism a Description represents a verb and a Qualifier represents a noun, adjective, or modifying phrase. It is an interesting approach to the problem of specifying formal determiners (“the”, “a”, “which”), quantifier hierarchy (e.g., the range of “and”), and reference evaluation (pronoun references). Detailed PROLOG programs for translating D&Qs are given. The specific details of how QPROC parses natural language are omitted. It is unfortunate that at least some programming details, comparable to those given for translating D&Qs, were not given. However, Chapter 5, titled Semantics, does cover most of the pitfalls that must be avoided in parsing. An ironic illustration of some of the problems of interpreting natural languages is the phrase, “. . .ICL, Europes leading computer manufacturer,” which appears on the cover of the book. Does “leading” here mean the largest selling, the most technologically advanced, or the closest to going broke Wallace gives a better, classic example: “I saw a man on a hill with a telescope.” In this case the meaning of every word is known, and there is no ambiguity; still there are three possible interpretations—“I”, the man, or the hill could have the telescope. I found the organization and style of writing of the book rather displeasing. I cant find any justification for the relegation of so much material to the appendices; it should be integrated with the short chapters to which it belongs. The book is written in what I consider an outline style—far too many details and transitions are omitted for it to be easy reading or good pedagogy. For example, on p. 71 we read (there is no ellipsis here) 3. CONVERTING D&Q TO SIMPLFIED RDBMS LIST COMMANDS 3.1 In this section we will use three relations for the example query It is a long time before the reader understands how the first sentence relates to the title. Only on p. 74 do we learn anything about the title by the appearance of the sentence, “This formula is a simplification of ICLs RDBMS List Command. ” I find it hard to formulate a recommendation for the potential purchaserreader. On the one hand, the recommended price of $29.95 is high for a slim, 170-page book; on the other, it is lower than the standard charges for the reproduction of a thesis, and this was originally a thesis. I think that it is a book that should be available in a library; but that the non-specialist should seek an introduction to the subject with some other book, perhaps Lis book [1], which, although it has its flaws, is much more readable.