This article represents the Presidential Address delivered at the Golden Anniversary Meeting of the LSA in New York, 29 December 1974, and attempts to provide a personal answer to the question as to how one goes about doing linguistics. Although the author recommends certain moves—e.g., diligent search for illuminating leading ideas, attention to formal aspects of the description, care and sophistication in dealing with data—he is at pains to point out there is no hope of ever finding a procedure for discovering insightful solutions to linguistic problems. This puts the linguist into the same boat with practitioners of every other science, and suggests that progress in linguistics (as in every other science) is the result of a combination of good ideas, powerful formulations, and careful attention to facts.

It is argued here that lexical redundancy rules (and all phonotactic constraints) must apply to phonological strings that consist of complete syllables (i.e. to words rather than formatives), and further that there is no significant difference between phonetically motivated phonological rules and lexical redundancy rules. It follows that there are two possible forms of a phonological grammar: either all rules act on lexical entries as they are stored, and they are stored as whole words (of complete syllables); or else lexical items are formatives, and NO phonetically motivated rules apply until formatives are concatenated into larger strings of complete syllables. The latter position, which requires redundancy-free or archi-segmental lexical representations, is adopted, and the consequences for natural generative phonology are discussed.

Recent work in generative phonology has shown that the need to preserve underlying contrast, or to avoid surface homonymy among underlying distinct forms, may sometimes affect the application of rules. This can take the form of blocking rule application, or of ordering rules in such a way as to avoid merging of contrasts. Another possible way in which grammars might avoid the derivation of homonymous surface forms has remained unexplored, however: it is conceivable that a language might have special rules whose function is to differentiate underlyingly identical forms on the surface. It is shown here that, in the Bantu language Tunen, although many phonemically identical nouns are tonemically distinct underlyingly, a residue of pairs of nouns remains which are phonemically and tonemically identical in underlying form. These nouns do not remain homonymous in all surface environments, however, but are sometimes distinguished tonally through the operation of a special rule. Despite the surface tonal differences which sometimes appear, it is not possible to introduce underlying contrasts by taking the different tonal alloforms as the underlying representations in these cases. The Tunen situation seems in some sense very unusual, though in another sense it can be said to be expected. Certain theoretical conclusions are drawn at the end of the paper.

The way stress is patterned in English verse depends on word and phrase structure, according to strict rules which are not accounted for by either traditional or more recent metrics. This paper is a contribution to the exploration of these rules, and an attempt to develop a formal metrical theory capable of expressing them. The presentation is organized as follows. After an introductory review of some of the problems and proposed solutions in metrics, I give (§1) a sketch of the theory to be defended here. §2 is devoted to justifying this theory on the basis of a fairly close examination of Shakespeare's verse. Then the scope of the investigation is extended in several directions: to other varieties of iambic pentameter (§§3.1–2), trochaic meter (§3.3), ternary meters (§3.4), and finally (§3.5) to some similarities and differences between the metrics of English and other languages, on which the presen: approach sheds new light.

The position taken in this paper is that significant progress in the study of semantics and phonology—and probably also of syntax—cannot be made without concomitant study of derivational morphology. Examples are given to support this position, and a proposal for directed research in derivational morphology is outlined. Various types of problems—some real and some apparent—are mentioned.

This paper proposes a theory of the lexicon consistent with the Lexicalist Hypothesis of Chomsky's ‘Remarks on nominalization’ (1970). The crucial problem is to develop a notion of lexical redundancy rules which permits an adequate description of the partial relations and idiosyncrasy characteristic of the lexicon. Two lexicalist theories of redundancy rules, each equipped with an evaluation measure, are compared on the basis of their accounts of nominalizations; the superior one, the FULL-ENTRY THEORY, is then applied to a range of further well-known examples such as causative verbs, nominal compounds, and idioms.

The syntax of the relative clause in the Saraswat Brahmin dialect of Konkani, an Indo-Aryan language, has been virtually Dravidianized because of the impact of the Dravidian Kannada language, operating through bilingual speakers. Konkani has two versions of the relative clause—a native Indo-Aryan type, and a borrowed Kannada type; the latter is gaining predominance at the cost of the former, which has come under various kinds of constraints. The mechanism of this syntactic change is examined here. The peculiar features of the Konkani–Kannada bilingual situation are described, and an attempt is made to provide an explanatory account of the syntactic change in Konkani.

Evidence from a comparison of the usage of 20 morphemes in each of 43 languages suggests that when morphemes are borrowed from a socially dominant language into a pidgin, and extended in usage as in a creole, the major factor determining the direction of such extension is the linguistic background of the speakers of languages other than the dominant one. Features inherent in pidginization and creolization themselves play only a subsidiary role. The role of semantic universals appears to be even less significant. Monogenetic hypotheses of the origin of pidgins and creoles are not supported by the evidence.

Grammarians since Saussure have insisted that language symbols are arbitrary, though conventionalized, in form. Sign languages in general, however, and American Sign Language (ASL) in particular, have been noted for their pantomimic or iconic nature. This paper examines some historical processes in ASL, and shows that there is a strong tendency for signs to change in the direction of arbitrariness, rather than maintaining a level of iconicity. Changes at the formational level can be seen as contributing to language-internal consistency, at the expense of transparency.