The question examined here is how a linguistic change takes place in the behavior of the individual speaker. Examples are chosen primarily from lexical change, though phonological and semantic changes are also mentioned. The opinion is expressed that every speaker has a (usually unconscious) norm against which variants operate, change taking place in two steps: (1) when one variant is promoted to tolerated deviation, (2) when a tolerated deviation replaces the earlier norm (called here KYRIOLEXIA in the case of lexical change). The topic of gradual change is also touched upon.

Labov 1981 argues that, in principle, sound changes may be implemented in two ways: through lexical diffusion (when the change is phonetically discrete), or by gradual sound change. It is contended here that although lexical diffusion is an important process, one must also account for discrete sound changes without such diffusion. The mode of progress of such changes is shown to involve an interplay between the production and the perception of the language user. Further, it is demonstrated that gradual sound change can be studied in perception, and that the relation between production and perception in gradual change is in some ways similar to that in discrete change. In both cases, the change in production norm is crucial.

The Saipan dialect of Chamorro displays certain transderivational relationships that hold optionally between complex words and the non-complex forms from which they are derived. These relationships are examined, and it is argued that phonological theory must either allow optionally-cyclic rule application, or else recognize a class of transderivational phenomena for which the cycle cannot account.

Valid tests for metrical theories can be constructed by removing the effect of ‘poetic’ syntactic transformations and observing the metrical consequences: e.g. Thy edge should blunter be is converted to Thy edge should be blunter (a metrical violation). Applying this approach to the over 500 inversions in Hamlet and the Sonnets provides considerable support for the metrical theories developed by Halle–Keyser and Kiparsky, with K faring slightly better. However, inversions affecting words like divine and enough suggest that minor revisions may be needed in K's rules. As each such revision becomes necessary, the likelihood increases that the set of permissible iambic pentameter lines is ‘fuzzy’ rather than well-defined.

The reduced forms of the auxiliaries is and has are excluded from many more constructions than have previously been noted. This article extends the range of ungrammatical reductions for which any theory of Auxiliary Reduction must account, and argues that many of them are already predicted under a liberal reading of earlier constraints on the context following the AUX. However, it is shown that the preceding context must also have a particular form. Auxiliary Reduction is a cliticization, and imposes a categorial restriction on the host which precedes the clitic: it must be an NP.

Sentence stress is not always a sufficient condition for interpretation as focus. An insightful analysis of the appropriate generalizations can be accommodated under a ‘modular’ approach to grammatical theory. Certain observations concerning the stress properties of WH questions are shown to be consistent with the assumptions of trace theory as developed in, e.g., Chomsky & Lasnik 1977, where the relationship between focus and stress is mediated by S-structure. The notion of focus has no consistent pragmatic characterization; it is, rather, a grammatical notion. The interpretation of this grammatical notion in particular discourse contexts is provided by rules of Discourse Grammar using the predicate ‘c-construable’, which is here defined.