The authors summarize their study of semantic-associations networks as these relate to machine translation and to artificial intelligence. They review their published efforts (beginning in 1969), as well as research by their students at the University of Kansas, to adapt Roget’s International Thesaurus to machine-readable form. Most of the summary deals with the pragmatics of prefixation and efforts to develop algorithms that produce appropriate associations among thesaural root groups.

The authors review [1] and [2]. Examination of 56 randomly selected root groups produced pairs with “helpful” and “harmful” information. Unfortunately, the summary does not explain why only certain prefixes are included, nor does it articulate the notion of thesaural levels, other than by physical proximity. It should be noted that on p. 75, column 1, lines 1–9, a segment of text is apparently missing. Next, a study by Harris [3] is glossed over, with reformatting problems, especially hyphens, highlighted.

Work by Bryan [4] develops an implicit structure that characterizes thesauri as “abstract entities about which we know only what an observer with no knowledge of English would know” (p. 78). Using entry, word, and category sets, Bryan’s ten levels of logical linkage are described; their semantic equivalencies are not. However, application of a type 6-e chain from [5] suggests neighborhoods based on word pairs extrapolated from the Wechsler-Bellevue intelligence test. Yet intersections between intuitive pairs such as wagon and bicycle are by no means guaranteed, and archaic words increase the confusion. A better result is reported from Patrick [6] that focuses on type 9 and type 10 chains. In these categories, according to the authors, Patrick reports strong semantic connectivity and effective distinctions between homographs.

Although the results reported by their students offer various levels of success, the authors conclude with cautious optimism that effective machine translation remains a viable goal. I was somewhat disappointed by the fragmented nature of the present summary, particularly the terse descriptions of actual semantic tests. Perhaps a more complete record would be useful.