This article examines the role of distinctness between adjacent segments in consonant deletion. On the basis of five stop-deletion patterns, it establishes a correlation between the likelihood of cluster simplification and the level of similarity between the consonants in the cluster. This correlation is motivated on perceptual grounds, and an OT analysis of similarity avoidance is provided in which perceptual factors are integrated in the grammar through both faithfulness and markedness constraints. This perceptual approach improves in two ways on previous analyses, notably the OCP. First, it integrates similarity avoidance within a more general perception-based framework, which accounts naturally for its gradient nature. Second, it uncovers a distinction between absolute and contextual similarity avoidance between adjacent segments, depending on whether similarity avoidance is established without reference to the context in which the segments appear or relative to the quality of the perceptual cues available to the segments.

I introduce 1460 lines of Chinese regulated verse and offer an analysis of the data. I also compare Chinese with English and discuss two approaches to variability in linguistic patterns (such as regular vs. exceptional forms, or perfect verse lines vs. lines with metrical tension). I argue that, whereas word stress is often more important than phrasal stress in English, it is crucial to understand the latter in Chinese. However, stress maxima play a central role in both languages. This suggests that metre is probably less variable cross-linguistically than previously thought. Moreover, while a correlation is thought to exist between metrical tension and frequency, it is difficult to see it in the present corpus. I argue that non-phonological factors can influence frequency patterns and that the presence of variable patterns does not necessarily imply the presence of marked forms. Rather, even fully well-formed patterns may occur rarely.

In this paper I explore the theoretical significance of phonologically conditioned gaps in word formation. The data support the original approach to gaps in Optimality Theory proposed by Prince & Smolensky (1993), which crucially involves MPARSE as a ranked and violable constraint. The alternative CONTROL model proposed by Orgun & Sprouse (1999) is found to be inadequate because of lost generalisations and technical flaws. It is shown that a careful distinction between various morphophonological effects (e.g. paradigm uniformity effects, phonological repair and ‘stem selection’) is necessary to shed light on the morphology–phonology interface. The data investigated here support affix-specific constraint rankings, but argue against any stratal organisation of morphology.