Published online by Cambridge University Press: 22 April 2026
1 The monograph reviewed here is a version of some parts abstracted from a much larger work now in preparation, together with summaries of other materials previously published and unpublished. The reviewer was privileged to read a first version of the larger work— The logical structure of linguistic theory, pp. 752 + li (Cambridge, Mass., 1956)—and now finds it difficult to refrain from referring to topics and results which appear there but not in the book under review. This discussion may therefore serve in part as a preview of the author's detailed statement of his theory of language.
The reader will profit also from three other publications of Chomsky's: Systems of syntactic analysis, Journal of symbolic logic 18.242-56 (1953); Semantic considerations in grammar, Report of the sixth annual Round Table Meeting on Linguistics and Language Teaching (= Monograph series on languages and linguistics, No. 8) 141-50 (1955); Three models for the description of language, IRE transactions on information theory, Vol. IT-2, No. 3 (1956). See also his doctoral dissertation, Transformational analysis (MS; University of Pennsylvania, 1955).
The reviewer is a member of the Research Laboratory of Electronics, Massachusetts Institute of Technology, where his work is supported in part by the U. S. Army (Signal Corps), the U. S. Air Force (Office of Scientific Research, Air Research and Development Command), and the U. S. Navy (Office of Naval Research), and in part by the National Science Foundation,
2 It is interesting to note that Hayward Keniston, who describes himself as a humanist, criticizes linguistics as not speculative enough and too bound to a description of physical entities; see Monograph series on languages and linguistics 7.146-52 (1954). While criticisms of this sort on the part of humanists may often be interpreted to mean something like ‘Linguistics is too scientific, not vague enough’, it would be far healthier for us to take it quite literally to mean, ‘Linguistics is not scientific enough, too atomistic and taxonomic‘
3 Pursuit of this goal is seen in perhaps its best and most resolute form in the works of Zellig S. Harris: Discontinuous morphemes, Lg. 21.121-7 (1945); Morpheme to utterance, Lg. 22.161-83 (1946); Methods in structural linguistics (Chicago, 1951); Discourse analysis, Lg. 28.1-30 (1952); Discourse analysis: A sample text, Lg. 28.474-94 (1952); Distributional structure, Word 10.146-62 (1954); Phoneme to morpheme, Lg. 31.190-222 (1955); Co-occurrence and transformation in linguistic structure, in the same number of Language as this review. Following Harris, Rulon Wells attempted to provide a firm basis for discovery of grammatical structure: Immediate constituents, Lg. 23.81-117 (1947); but he denied there (93) that it would be an entirely mechanical procedure.
4 Wells, Immediate constituiente §46.
5 Charles F. Hockett, in a review of Shannon and Weaver's The mathematical theory of communication, Lg. 29.69-93, esp. 87-8 (1953), proposed that morphemic segmentation might be mechanized by a statistical technique. Similar suggestions are attempted by Harris, Phoneme to morpheme, and by Seymour B. Chatman, Immediate constituents and expansion analysis, Word 11.377-85 (1955).
6 Cf. Carl G. Hempel, Fundamentals of concept formation 36-7 (International encyclopedia of unified science 2.7; Chicago, 1952).
7 See below for a case in which superfluous rules can be eliminated by rejecting this principle of bi-uniqueness. Cf. Bernard Bloch, Phonemic overlapping, Amer, speech 16.278-84 (1941); Hockett, review of Martinet's Phonology as functional phonetics, Lg. 27.340 (1951). Chomsky, Halle, and Lukoff have chosen to reject the principle of bi-uniqueness entirely in their article On accent and juncture in English, For Roman Jakobson 65-80 (The Hague, 1956).
8 At the same time the whole concept of linguistic level can be reformulated much more rigorously.
9 Cf. William G. Moulton, Juncture in modern standard German, Lg. 23.225 fn. 14 (1947) ; H. A. Gleason Jr., An introduction to descriptive linguistics 66, 175 (New York, 1955); On quite different grounds, grammatical criteria for phonemic analysis have been advocated by Kenneth L. Pike, Grammatical prerequisites to phonemic analysis, Word 3.155 (1947), More on grammatical prerequisites, Word 8.106-21 (1952). In a review of Harris's Methods, Lg. 28.507 fn. 8, Murray Fowler erroneously construes Harris's references to the use of meaning criteria in that work as the major cause of his failure to provide a discovery procedure (though Harris's methods in fact make no use of semantic criteria), and uses this charge of circularity as his strongest criticism of distributional analysis. Cf. also Eugene A. Nida, The analysis of grammatical constituents, Lg. 24.173 (1948); Hockett, review of Martinet 341.
10 Cf. David L. Olmsted, review of Psycholinguistics, Lg. 31.50 (1955).
11 See Hockett, A manual of phonology 3-14 (Baltimore, 1955). Cf. also Claude E. Shannon and Warren Weaver, The mathematical theory of communication (Urbana, Ill., 1949); Hockett, review of Shannon and Weaver 86-7; Benoit Mandelbrot, Structure formelle des textes et communication, Word 10.1-27 (1954); id., Simple games of strategy occurring in communication through natural languages, Trans. IRE, PGIT-3. 124-37 (1954).
12 Shannon and Weaver 13-5; George A. Miller, Language and communication 83-6 (New York, 1951); Kellogg Wilson, Psycholinguistics: A survey of theory and research problems 46 (Baltimore, 1954).
13 Three models (cf. fn. 1).
14 E.g. S. C. Kleene, Introduction to metamathematics, esp. Chap. 13 (Princeton, 1952).
15 Ibid.; see also J. G. Kemeny, Man viewed as a machine, Scientific American, April 1955, 58-67, for a particularly graphic description of the Turing machine. Even if such a machine were powerful enough to reproduce all of human behavior, as Kemeny suggests, it does not seem to this reviewer any more fruitful as an approach to an understanding of this behavior than computer theory is revealing in the explication of grammatical facts.
16 The independence of phrase-structure from forms, order, and construction has also been pointed out by Hockett, Two models of grammatical description, Word 10.218-20 (1954). Wells also notes the independence of sequence from ‘construction’ (= phrase-structure) in IC analysis.
17 Cf. Floyd G. Lounsbury, A semantic analysis of the Pawnee kinship usage, Lg. 32.159-62 (1956); Sol Saporta, Morph, morpheme, archimorpheme, Word 12.9-14 (1956).
18 The inadequacy of IC analysis of ambiguous sentences is also recognized by Hockett, Two models 218.
19 But it is also interesting to note that, in a sense, transformational analysis is essentially a formalization of a long-accepted, traditional approach to grammatical relations. To cite only a single example of classical grammatical thought which is basically a kind of transformation theory: ‘It is different when we come to such a combination as an early riser, which it is quite impossible to turn into a riser who is early. Here the adjunct is a shifted subjunct of the verb contained in the substantive riser: he rises (vb) early (adv) = he is an early (adj) riser (sb)‘—Otto Jespersen, A modern English grammar 2.283 §12.12. Here Jespersen correctly perceives that the difference in the way that we understand, on the one hand, he is an early riser, and, on the other, he is an early bird can be accounted for by regarding the former as a transform of he rises early, the latter as a transform of the pair he is an X and early bird (which in turn is a transform of the bird is early).
20 Harris, Lg. 28.18-25; Co-occurrence and transformation.
21 For example, suppose that the representation of a sentence, NP + VP (i.e. noun-phrase plus verb -phrase), is to be developed next, and is rewritten directly as Dogs + can + swim, then there is no specification of the phrase-structure in the result, since there is no way to tell which of the developed elements originated as NP and which as VP. But if the development proceeds in two steps—S = NP + VP, NP = dogs, VP = can + swim, expanding each phrase separately—the phrase-structure is preserved at each step: S = NP + VP = dogs + VP = dogs + can + swim, and we can say that dogs is a NP and can + swim is a VP.
22 Personal communication. Halle will adopt this view in his forthcoming work The sounds of Russian.
23 Gleason, Introduction 92-5.
24 For similar discussions, see Yehoshua Bar-Hillel, Logical syntax and semantics, Lg. 30.230-7 (1954), and an answer by Chomsky, Logical syntax and semantics: Their linguistic relevance, Lg. 31.36-45 (1955).
25 Meaning in three linguistic theories, read before the Linguistic Forum, University of Michigan, in 1953, and again before the Washington Linguistic Club in 1954.
26 Note that some expressions do not seem to have any meaning, in its ordinary sense, e.g. to in he likes to sing. Other expressions have no reference, e.g. the average American and mermaid, unless we are willing to people the universe with countless imaginary objects. Cf. also C. E. Bazell, The choice of criteria in structural linguistics, Word 10.132 (1954); Miller, Language and communication 160.
27 See Nida, A system for the description of semantic elements, Word 7.1-14 (1951); id., The identification of morphemes, Lg. 24.430 (1948) ; Norman A. McQuown, review of Harris's Methods, Lg. 28.501 (1952) ; and possibly Harris, Methods 7 fn. 4, 365 fn. 6, and Distributional structure 155-8. But per contra cf. Bar-Hillel's article 233 and Chomsky's reply 44, as well as Miller, Language and communication 112.
28 It is interesting to note, however, that no author has yet taken such a distributional criterion really seriously and examined all possible frames for substitutability, whether to establish synonymy or grammatical equivalence, i.e. form-class membership. What is really used is substitution in ‘diagnostic’ frames; and these are chosen quite arbitrarily— which is sufficient to show the futility of a discovery procedure for grammatical analysis by substitution techniques.
29 Cf. Gleason, Introduction 182.
30 Bloch has attempted to eliminate the need for the informant's response entirely by basing an apparently mechanical analysis scheme upon distributional criteria ; see A set of postulates for phonemic analysis, Lg. 24.3-46 (1948); Studies in colloquial Japanese IV. Phonemics, Lg. 26.89-90 (1950). That his postulate system does not provide a true discovery procedure can be seen at least from its use of nonmechanizable notions such as ‘general definition’; see Contrast, Lg. 29.59-60 (1953). Lounsbury, while defending complete independence of semantics and linguistics, still assumes that informant response is a semantic criterion; see A semantic analysis of the Pawnee kinship usage, Lg. 32.190-1 (1956).
See also Einar Haugen, Directions in modern linguistics, Lg. 27.219 (1951); Eli Fischer-Jørgensen, The phonetic basis for identification of phonemic elements, JASA 24.611, 615 (1952); Lounsbury, Oneida verb morphology 16 fn. 8 (New Haven, 1953); Paul L. Garvin, review of Jakobson, Fant, and Halle's Preliminaries to speech analysis, Lg. 29.476 (1953). Harris avoids this confusion in phonemic analysis by advocating the use of pair-tests (Methods 31-2), and alludes to a distinction between informant response and semantic criteria in the phrase ‘meaning-like distinction between utterances which are not repetitions of each other’ (7 fn. 4); see also 29 fn. 1, but 173, 363 (appendix to §12.41), and 365 fn. 6. The confusion is also avoided by Halle, The strategy of phonemics, Word 10.200 (1954), and by Hockett, Manual 144-5.
31 Semantic considerations (cf. fn. 1).
32 The starred examples are taken from Eli Fischer-jørgensen's discussion of this question in her article The commutation test and its application to phonemic analysis, For Roman Jakobson 140-51 (The Hague, 1956).
33 For a clear statement of the pair-test technique in phonemic analysis, see Harris, Methods 31-2; id., Distributional structure, Word 10.158-9 (1954); Halle, Strategy, Word 10.200 (1954); Chomsky, Semantic considerations.
34 It hardly seems necessary to document the use of a semantic criterion in the identification of morphemes, since this principle is almost universal in contemporary linguistic works. See for instance Bloch, English verb inflection, Lg. 23.399-418 (1947); Hockett, Problems of morphemic analysis, Lg. 23.341 (1947); Moulton, Juncture, Lg. 23.218 (1947); Nida, System, Word 7.2 (1951); Lounsbury, Oneida verb morphology 11; C. F. Voegelin, Distinctive features and meaning equivalence, Lg. 24.133 (1948); John Lotz, Speech and language, JASA 22.713-4 (1950); Bar-Hillel, Logical syntax, Lg. 30.230 (1954); Gleason, Introduction 54-5, 77, 79, 109.
35 See for instance Nida, System 6, Identification 431-2.
36 See Nida, System 9; Ward H. Goodenough, Componential analysis and the study of meaning, Lg. 32.207-8 (1956); but Harris, Distributional structure 152.
37 Nida, Identification 437 fn. 40; also Harris, Distributional structure 157.
38 See Bloch, English verb inflection 399-400 §1.2; Lounsbury, Oneida verb morphology 18, Pawnee kinship usage 189; Martin Joos, Description of language design, JASA 22.708 (1950) ; Gleason, Introduction 55.
39 Especially Henry Lee Smith Jr. and George L. Trager. See for example Smith, review of Carroll's The study of language, Lg. 31.61 (1955); Linguistic science and the teaching of English 11-12 (Cambridge, Mass., 1956).
40 It is precisely this Sprachgefühl, this intuitive notion about linguistic structure, which, together with the sentences of a language, forms the empirical basis of grammatical analysis ; and it is precisely the purpose of linguistic science to render explicit and rigorous whatever is vague about these intuitive feelings. This is apparently what Hockett means in his discussion of ‘empathy’ in phonemic analysis, Manual 146-7.
41 See for instance Joos, review of Locke and Booth's Machine translation of languages, Lg. 32.294 (1956); but per contra cf. Wilson, Psycholinguistics 46.
42 There have been many popular treatments of information theory since Shannon's original papers in the Bell Journal. Both Shannon and Weaver point out the necessity of distinguishing information from meaning; see The mathematical theory of communication 3 and 99. See also E. Colin Cherry, A history of the theory of information, Proc. Inst. Electr. Engrs. 98.383 (1951) ; Wilson, Psycholinguistics 46; Hockett, review of Shannon and Weaver, Lg. 29.89-90 (1953); Miller, Language and communication 41.
43 A good start has already been made in this direction in Chomsky's The logical structure of linguistic theory, Chap. 8 (cf. fn. 1).
44 This is not the same question as that of a discovery procedure, which concerns the automatic generation of grammars. This is the weaker notion of automatic generation of the constituent structure of a given sentence, assuming the grammar.
45 It already seems quite likely that no finite automaton, such as an electronic computer, will be adequate to generate the nested and recursive properties of natural language.
46 Hempel 23-4, 32-9 (cf. fn. 6 above).