Published online by Cambridge University Press: 22 April 2026
The purpose of this paper is to present the initial results of a study of certain regularities in the structure of written English words. Our objective will be to show that it is possible to extend the phonemic concepts of consonant, vowel, and syllable to written forms and that there is considerable utility in doing so. In particular, we shall show that a major proportion of the one-syllable words, graphemically defined, have the same part-of-speech assignments when checked against standard dictionaries. In this sense we shall claim that the one-syllable words form a part-of-speech class. We further conjecture that graphemic syllabification is intimately related to part-of-speech assignment in the set of all written English words.
* This work was supported by the Lockheed Independent Research Program.
The authors are grateful to J. J. Finkelstein and G. M. Schramm for their encouragement and critical comments; A. F. Brown for providing an early edition of the Pennsylvania Word List; Mark Halpern, F. W. Householder, and R. L. Venezky for reading an early draft of this paper and for helpful suggestions; S. R. Fox, T. A. Lauer, and Mrs. Louise Watson for performing the many calculations and for assisting in supplementary work necessary to the paper; and E. G. Graziano for preparing bibliographies of work done in this area (‘Algorithmic studies of English monosyllabics for prediction of parts of speech: An annotated bibliography’, SB-62–29, Lockheed Missiles & Space Company, October 1962; ‘Further syllabic studies for algorithmic prediction of English parts of speech: An annotated bibliography’, SB-63–3, Lockheed Missiles & Space Company, March 1963).
1 Normal and reverse English word list, compiled by A. F. Brown, vols. 1–8.
2 For these purposes, we ignored the lexes for abbreviations (lb., Mr., ms, etc.); the letters themselves (b, c, d, etc.); the digraphs (ch, ph, th, etc.); and the interjection st.
3 F. W. Householder has suggested that a more sophisticated approach would have been to eliminate only forms ending in Cs, es, or ys. Such a rule would leave in the corpus the forms bus, gas, his, was, has, this, as, us, is, etc., not all of which are irregular. We prefer to leave the rule as given because it is simpler and because most of these forms are either irregular or of some syntactic interest and should be studied separately. In any case, the number of forms involved is not large enough to cause us to modify any of our subsequent conclusions.
4 Many researchers have examined the analogous phonological question. Cf. Otto Jespersen, ‘Monosyllabism in English’, Proc. British Acad. 14.341–68 (1928).
5 K. Malone, ‘The phonemic structure of the English monosyllables’, American speech 11.205–18 (1936), and ‘The phonemes of current English’, Studies for William A. Reed 133–65 (1940).
6 J. Kramsky, ‘On the quantitative phonemic analysis of English mono- and disyllables’, Philogica 8.45–59.
7 L. Bloomfield, Language (New York, 1933).
8 B. J. Wallace, A quantitative analysis of consonant clusters in present-day English (Univ. of Michigan thesis, 1950).
9 D. P. Harris, Phonemic patterning of the initial and final consonant clusters from late Old English to the present (University of Michigan thesis, 1954).
10 P. Menzerath, ‘Typology of languages’, J. Acoust. Soc. Am. 22.698–701 (1950).
11 S. Saporta, ‘Frequency of consonant clusters’, Lg. 31.25–30 (1955).
12 One could argue heuristically that in speech s appears frequently in word final position since it is a popular inflectional ending, and that, accounting for the difficulty of distinguishing word-final and word-initial positions aurally, it would be reasonable to anticipate the common occurrence of s in word-initial position. The simple graphemic distinction made in the text shows that this argument leads to the correct conclusion, although it does not certify the validity of the argument. It would be interesting to know precisely how the hearer determines word-final positions aurally.
13 The unique character of h in two-letter strings is not so obvious from the data of this study, since other letters occur in second position in considerable numbers. In a separate study, ‘The graphic structure of word breaking’, presented at the annual meeting of the Association for Machine Translation and Computational Linguistics (Denver, August 1963); abstract in Mechanical translation 7:2 (August 1963), we show that in multisyllable words where a syllabic split in the written word occurs in the sequence vowel-consonant-consonant-vowel, h plays a distinctive role. In some sequences, such as vowel-bl-vowel, the split occurs either between or before the two consonants. In others, such as vowel-nt-vowel, the split occurs between or after the two consonants. But when the second consonant is h, the ambiguity resolves into an occurrence before or after the two consonants, except in certain compounds such as pan-handle, sheeps-head, meat-head.
14 The OED notes that the OE initial hr is rarely represented in later periods by rh.
15 Jespersen, A modern English grammar (1954).
16 Both kn and gn are in the process of reduction to n in initial position; see H. Kökeritz, ‘The reduction of initial kn and gn in English’, Lg. 21.75–86 (1945).
17 Note that we refrain from basing these arguments on the total number of words with one-letter initial consonant strings versus the total for two-letter and three-letter strings. The numbers are 4118, 2682, and 212 respectively. Clearly, the constant ratio does not hold in this case, and for good reason. As we have already shown, the one-letter strings (and hence, presumably, the two-letter and three-letter strings) can be partitioned into several meaningful subgroups. The totals then reflect a complexity of behavior where the maxima are characteristic of particular subsets. A more refined analysis would require a formal breakdown of each set to obtain comparable ratios.
18 Most of the words ending in h are interjections.
19 The lack of distinction between the forms ie and ei is not unrelated to the notorious spelling difficulties associated with words involving this pair. At least in our dialects, most of the words that occur outside the arc are one-syllable words, while many of those that occur inside the arc are not. It seems reasonable to argue that in the process of learning English spelling, one comes early to the recognition that ai usually signals a single syllable but that ia does not. There are obvious counter examples involving suffixes. The speller presumably has no great difficulty learning which is which. This distinction is largely lost in the ie–ei words, and the speller must search for other clues or rules for guidance.
20 We cannot help conjecturing that a deeper study will show that the hierarchy should be extended to include phrases, clauses, sentences, and so forth.
21 The teacher's word book of 30,000 words (1944).
22 If the argument is inverted, these numerical results can be used in a rather different way. Suppose, for instance, that we assume that all of the elementary words have both noun and verb uses. Then the percentages given can be interpreted as estimates of the probability that the dictionary maker will find and list the two parts of speech as a function of the frequency of use of the word. If the probability of finding a noun use is independent (in the probabilistic sense) of the probability of finding a verb use, the percentage given in the last column should be the product of the percentages given in the first two columns. These product percentages are 76.3, 82.2, 83.7, and 90.2—figures which are well within the statistical variation expected for samples of the size used here. Such an analysis would indicate that it is either more difficult to find verb uses or that the dictionary maker is less likely to list them, or perhaps a bit of both. At least on very superficial grounds, this seems to be a reasonable conclusion.
23 It is well to recall at this point that we had to produce a rather elaborate structure in order to separate the Cle words from the elementary words. Since the phonetic distinction between these two classes is rather obvious, this graphemic complexity seems to be an argument against English spelling habits. We now see that from the part-of-speech point of view, these are, indeed, the two largest subsets of the class of noun-verbs. Thus, a choice had to be made, in spelling practice, between agreement with phonetic principles and agreement with the practice of written grammar; in this case the latter was chosen. The same phenomenon is observed in English word-breaking practice, where grammatical considerations generally outweigh phonetic considerations if the two cannot be satisfied simultaneously.
24 It would be more satisfying to identify the diminution ratio with the base of the natural logarithm system, e* 2.718 …, tying the entire problem to the questions of information theory and Poisson distribution. Certainly much can be accomplished in constructing a theory along such lines. For the distribution of topological genus in alphabets, which is closely related to the diminution ratio considered here, see the recent contribution of Osovec, ‘Some statistical regularities in the writing of letters of the alphabet’, Soviet mathematics 3.1096–7 (1962), translated from Doklady.