Why are some sound patterns more prevalent than others? As common cross-language sound patterns are found to have physical phonetic origins, the rare occurrence of certain sound patterns has often been attributed to the low probability of the corresponding phonetic effects being phonologized through sound change (e.g., Ohala, 1993, Blevins, 2004). Yet, given the wealth of articulatory and acoustic evidence of systematic variation in speech, it remains a puzzle that not all variations are phonologized into sound patterns or have analogs in the diachronic domain (cf. Moreton, In press). As there would be no change at all unless there were some structural or social advantage, however small, in the new form, a theory of change and its directionality must elucidate common, perhaps universal, selective advantages that characterize the new variant (Kiparsky 1995).

In this talk, I identify four selective pressures (cognitive, structural, social, and ecological) that affect the directionality of variation and change. I argue for the centrality of speaker/listener's evaluative strategies in modulating all non-ecological pressures. Experimental results from three case studies will be presented in support of this conclusion. Experiment 1 tests Cantonese speakers' judgment of the wellformedness of syllables that are absent either systematically or accidentally in Cantonese. Speakers' evaluation of such unattested syllables is found to be gradient, contrary to the predictions of classical theory of phonotactics and grammaticality; speakers rely mainly on the lexical statistics of the language in making their goodness judgments. Experiment 2 examines the role of phonetic precursor robustness in explaining an apparent case of differential phonologization (i.e. vowel-height-to-vowel-height dependencies are more common than vowel height-consonantal voicing dependencies; Moreton In press). Production results suggest that this instance of typological asymmetry is best explained by differential phonetic precursor robustness. That learners have more difficulties with learning height-voicing dependencies than with height-height dependencies (Moreton In press) can be explained by their lack of experience in attending to phonetic dependencies between vowel height and voicing. Experiment 3 tests the role perceptual compensation plays in explaining the underphonologization of dependencies between tone height and length in comparison to the overphonologization of dependencies between contour tone and length. Experimental results show that, while English speakers perceptually compensate for the articulatory effect of pitch height on syllable duration, the effect of pitch movement on duration is left unaccommodated. Speakers' differential compensatory responses to phonetic variation might explain the ubiquity of differential phonologization. Taken together, the results of these experiments suggest that the way speakers deal with linguistic inputs is heavily affected by the evaluative mechanisms that "filter" the input. Thus in order to understand what types of variation and change are possible, it is not enough to identify the seed of variation, as it is traditionally done. Intrinsic variation in speech production and perception provides only the necessary but not the sufficient condition for change. Only through the identification of the evaluative mechanisms that deal with these new variants can the explanation of phonologization be complete.