The phoneme as a viable phonological unit for capturing relevant surface contrasts has been rejected within generative phonology. However, certain diachronic and synchronic effects can only be explained in terms of the phoneme. The historical phenomena of denasalization in French, depalatalization in Rumanian, delabialization in Romance, palatalization and labialization in Nupe, and palatalization in Japanese are due to surface contrast rather than to statable morphophonemic or phonetic processes. From these observations I conclude that the phoneme must be recognized as a phonological entity. Synchronically, a phonemic representation is not to be discovered by applying a set of procedures to a phonetic representation, nor does it exist as an autonomous level within a generative phonology; rather, it is to be characterized as a representation of relevant surface contrasts which is deducible from the function of the rules within a generative phonology. Thus the phoneme does in fact have a place in generative phonology without in any way changing the theoretical basis.

A line must be drawn between meanings which are the reference of a word and other meanings which are inferred from its use in particular contexts, even though the inferential meanings may in time become referential. The verb remind is referentially akin to make think, only inferentially akin to resemble. Shared restrictions can be used to prove either case, but the proofs for inference (between remind and resemble) are weightier.

Certain linguistic items, capable of contrast in some environments, have the property of being apparently free variants in others. This can be accounted for, if in a given item, one or more semantic features can be regarded as ‘latent’, i.e. susceptible of being activated in some contexts and suppressed in others. The technique of ‘forced-choice’ selection testing was used to investigate the phenomenon, and the hypothesis was found to be valid in respect of (a) the two preterit forms of wet, (b) the infinitive and present participle in environments where both can occur, and (c) pairs of quasi-synonyms.

This paper investigates several Avestan words alleged to continue the reflex of Indo-European *ə as i (ī) between consonants in non-initial syllables. A critical analysis of the forms shows that such an interpretation is unjustified.

Rules of stress like those described by Chomsky & Halle 1968 are examined for their relevance in determining metricality (using Halle & Keyser's [1966] stress maximum concept) over the period of modern English. The Nuclear Stress Rule frequently locates stress maxima in unmetrical positions, and the Compound Rule occasionally does. A late, optional rule is proposed—the Stress Exchange Rule—which reverses unmetrical alignments resulting from earlier rules. An Alternating Stress Rule is also proposed, to account for non-lexical stress. Thus stress maxima are created in different ways. The possible rule combinations (lexical, phrasal; obligatory, optional) discriminate degrees of metricality, and provide stylistic parameters.

Pre-16th century Portuguese had a slight raising of post-stress /e o/ to [ẹ ọ] (but not to [i u]) in pre-pause position, and a full raising of /e/ to [i] under nasalization in initial position. The first of these developments is based on a decrease in average energy of articulation after the final stress of a breath group, while the second depends on several layers of confusion of prefixes. Later changes found in several geographically separate dialects are explicated in terms of the pre-16th century situation and various aspects of the process of grammar construction, including several formal mechanisms of generalization. For the back vowels, Continental Portuguese and Brazilian share one innovation, but differ in that the former shows a generalization to mirror-image form. Ceylon Portuguese goes one step further by generalizing this raising to front vowels as well; but in the first two dialects, raising of the front series is a result of the pre-16th century confusion of prefixes, generalized in an interesting way.

An alternation of shwa with zero which is characteristic of a wide variety of morphemes is shown to support a generative phonological treatment of Hindi-Urdu in which aspirated consonants are analysed as units, while nasalized vowels are derived from underlying sequences of vowel plus nasal consonant. Some observations concerning the distribution of nasal segments at the level of systematic phonetics provide the basis for rules by which the five nasal consonants [m n ṇ ɲ ŋ] can be derived from underlying representations in which only two nasal segments are distinguished.

This paper uses data from eight Polynesian languages to sketch the typological characteristics of this group of related languages by means of statistical methods. In addition to specific problems of Polynesian phonotactics, two points of theoretical interest are discussed. First, an attempt is made to distinguish universal phonotactic features of morph structure from specifically Polynesian ones; and second, similarity sequences checked for statistical significance are suggested as a device leading from the set of similarity coefficients to a reliable typological classification.