¹ Presidential address delivered at the thirty-third annual meeting of the Linguistic Society of America in New York, 29 December 1958. On zero see now W. Haas, Studies in linguistic analysis 33-53 (Oxford, 1957), with rich references to older literature from R. Jakobson, Mélanges Bally 143-52 (1939), to W. S. Allen, Indian linguistics 16.106 ff. (1955).

² Ideally, unlimited idiolect corpora. The question how to approximate, sample, or otherwise represent ‘all possible’ discourses in a language constitutes a well-known uncertainty. The restriction to idiolects is of course no guarantee of uniformity, since individuals take part in replacement processes, i.e. go through language change, within their life spans.

³ This depends of course on what manner of uniqueness is expected of our ‘code’.

⁴ More specifically, if still sketchily: if, when the comparative method is applied to two languages A and B, the common ancestor turns out to be identical with A (or to exhibit certain other special characteristics with regard to A), A is the earlier stage and B the later stage. The terms ‘earlier’ and ‘later’ accord with empirical facts; where the evidence is sufficient, A is found to precede B in time. J. Holt, Proceedings of the Seventh International Congress of Linguists 98 (London, 1956) seems to hint at a similar view.

⁵ The comparison between written messages (in archives) and spoken discourses (in language corpora) is no more than an analogy. Therefore the use of written records as a source of information on discourses does not enter into the comparison. Discourses (connected texts) are stretches of utterance within the corpus such that no element within one restricts any element within another. Discourses are the true minimal free forms. On transformations see Z. S. Harris, Lg. 33.283-340 (1957).

⁶ This concept meets some of the grave problems raised by W. S. Allen, TPS 52-108, especially 100 ff. (1953). See also J. Ellis, Lingua 7.134-74 (1958). Cf. C. F. Hockett, A course in modern linguistics 394-5 (New York, 1958), on competitive and noncompetitive innovations.

⁷ Granting that indeed they have.

⁸ This reflects an impression, probably correct, that morphemic change typically affects immediate constituents. Generally, the dividing line between environments in which a morpheme is replaced in different fashions is often not blind or random, but already recognizable as synchronically significant at the earlier stage. See §3, end.

⁹ For instance a morpheme or a phoneme. It goes without saying that all this applies equally well to simultaneous components and other extensions from segmentation. In the charts which follow, the entities of the earlier stage are listed on the horizontal axis, those of the later stage on the vertical axis. The replacement processes are represented by entries at the appropriate intersection. See SiL 6.81-94 (1948), and E. Haugen, SiL 7.63-6 (1949), where the application is to sound change.

¹⁰ That is, separate sets of alphabetic symbols are used: a to l for the earlier stage, m to z for the later stage.

¹¹ Data on earlier stages (aside from occasional explicit description) are obtained from the interpretation of written records, from reconstruction (internal or comparative), or from both. Proto-Semitic is, of course, reconstructed. The positing of [p]-like phones for *p is largely based on wide consensus among the sister languages of Arabic. The main factors in the picture are these:— (1) Existing contrasts are left intact; there is neither increase or decrease. (2) Phonetically, the replacing phone is appreciably different. (3) If distinctive features—S(toppage), F(ricativity), L(abiality)—are studied instead of their segment-size combinations, the replacement appears as follows: F > F, L > L, S > S, except that S combined with L > F. In this particular combination (which corresponds to the phoneme *p), contrast between S and F remains neutralized throughout the change—whereas t continues to contrast with θ. The one phone in question is assigned to S (× L) in Proto-Semitic, to F(× L) in Arabic, each time on the basis of ‘phonetic similarity’. In terms of Figure 7 below, we have

But since the contrast between S and F failed to extend to combination with L, there is no true merger. (4) Typologically, the entities S, F, L of the earlier stage and the identically labeled entities of the later stage may be regarded as homologous and may thus be used as a fixed framework. In this sense, the change *p > f has phonetic and typological standing.

¹² The symbol > is not ordinarily used in morphology. Read either: the (phone or) phoneme sequence of uncle replaces that of eme, eam in certain environments, e.g. in +]—[+ (between word-boundary junctures), but not in squ]—[ish; or: the morph uncle replaces the morph eme; or: the morpheme uncle replaces the morpheme eme one-to-one, so that in effect it remains the ‘same’ morpheme.

¹³ It would again be wrong to claim that the two morphs actually functioned as simple positional allomorphs in the intervening period.

¹⁴ See fn. 11.

¹⁵ See Lg. 22.140-1 (1946) = Readings in linguistics 140-1 (Washington, 1957).

¹⁶ In Figures 7, 8, 10, 11, and 12, only the environments are entered at the intersections. The parties to the replacement process are to be supplied from the axes.

¹⁷ Compare the somewhat different graphic treatment in L. Bloomfield, Language 430 (New York, 1933).

¹⁸ Leaving aside the usual allomorphic stress differences; see B. Bloch, Lg. 23.404 (1947) = Readings in linguistics 246.

¹⁹ For the present purpose it makes no difference whether this disappearance is morphological in nature—one construct replacing another—or a side product of sound change—the same phonemic sequences being affected everywhere regardless of morph boundaries.

²⁰ If indeed the stress pattern is, or may be, the same for both meanings. On shortening etc. see G. Stern, Meaning and change of meaning 237-81 (Göteborg, 1931).

²¹ As shown by Edgerton, Lg. 19.86 (1943), an older Indo-European *str with its semivowel r after two consonants should have had allophonic [] and hence could not have led to a triple cluster in most daughter languages. But this automatic distribution was already disappearing in late Indo-European, and there is special reason to believe that the s-clusters in particular no longer precluded a nonsyllabic [r] in subsequent position.

²² See fn. 11.

²³ More precisely: is equivalent, in the core corpus, to.

²⁴ Including, again, fractions of the distribution of earlier nil.

²⁵ See C. F. Hockett, Manual of phonology 43-4 (Baltimore, 1955) on macrosegments.

²⁶ See E. H. Sturtevant, The pronunciation of Greek and Latin ² 21-9 (Philadelphia, 1940). Reassignment of neutralized phones often does leave orthographic traces even though it involves no merger. In general, see H. Penzl, Studies presented to Joshua Whatmough 193-208 ('s-Gravenhage, 1957).

²⁷ See A. Martinet, Économie des changements phonétiques 77 (Berne, 1955).

²⁸ See the summary confrontation of phonological processes in Martinet, Économie 174-5.

²⁹ See Lg. 26.364 (1950) = Readings in linguistics 301-2.

³⁰ When alternating, nil becomes a true, synchronically defined zero. See fn. 1.

³¹ See the opinion quoted by F. Edgerton, Lg. 10.264 (1934).

³² These examples, to be sure, show the correspondence in question after # C, an environment which may be assumed to have had, in part, the same effect as the environment CC].

³³ Even , here and in the next few lines.

³⁴ The split of nil, like the split of t in méli(t), leads to alternations: Greek pít(nāmi) ‘I spread’: (pé)pt(atai) ‘it has been spread’.