The Neogrammarian Principle, that sound change must be phonetically conditioned and exceptionless, has developed into a well-accepted heuristic that is the foundation for the comparative method. However, during the 20th century, researchers discovered a number of changes in sounds that had developed perfectly regular correspondences, but without the required phonetic conditioning environment, apparent sound-changes-in-progress that were not entirely regular, and confirmed that speech sounds are highly variable from utterance to utterance, and from talker to talker. A terminological dispute erupted over what counted as sound change, and was settled by the introduction of new terminology, such as the initiation or actuation and spread or diffusion of sound change. Because of the usefulness of the Neogrammarian principle to historical and comparative linguists, researchers who examine other aspects of sound change have largely avoided direct confrontation, allowing the gradual narrowing of the meaning of the term sound change to refer to something that is very small in scale in order to maintain its phonetic conditioning and exceptionlessness, even though the phenomena to which it first referred were very broad and sweeping changes.
While there is debate about just what may be considered sound change, the larger goals of all linguists studying sound change is to understand how sound change works, to isolate the causes and mechanisms of change, to better understand language, and hopefully one day be able to make predictions about sound change. In this dissertation, I outline a general model of phonetically gradual sound change, using experimental evidence to support its premises. This general model holds that while the basis for sound change is phonetic variation, the actuation period of sound change begins when the phonetic variation is associated with some other co-occurring factor, whether phonetic, phonological, physical, social, grammatical, syntactic, etc., or is generalized as being a characteristic of the sound category itself. In this way, all sound change is cognitive; although, the associations may range from below the level of consciousness (as in phonetic and some phonological associations) to conscious and purposeful adoption (as in socio-indexical markers), with an associated difference in the amount and type of cognitive effort required to adopt or abstain from a sound change.
Perceptual learning (e.g., Norris, McQueen and Cutler 2003) is a mechanism in which the perception of a sound category shifts to include a previously ambiguous range of exemplars. Shadowing techniques (e.g., Goldinger 1998) can induce a change in the production of speech sounds, as talkers adapt their speech to become more similar to the speech of another talker. I use perceptual learning and shadowing to recreate sound change in the laboratory, in order to test the hypothesis that non-phonetic factors may play a key role in shaping sound change, including phonetically-conditioned sound change, and that regularity, as we have discovered with so many other linguistic concepts, is not truly a dichotomous characteristic, but rather a continuous one, relying on more than just phonetic universals.
Using this laboratory sound change paradigm, phonetically-conditioned sound change was replicated, with differences in perception and production based on, among other things, word familiarity and participant gender. These changes in perception were extended to novel talkers, words, and even phonologically different environments, showing that the paradigm was effective, but also suggesting that phonetic conditioning environments may not be as specific as we had thought. Part of the series of experiments attempted to condition a sound change with non-phonetic environments, which may be associated with differences in sound much in the way that phonetic environments are, with the result that talker and participant gender interact in much more complicated ways than simple association. I uncover evidence that the early stage of phonetically-conditioned sound change is not exceptionless, and is influenced by non-articulatory factors. “Universal” cognitive processes underlying perception are still bound by talker- and listener-specific factors (word familiarity and frequency are slightly different for each person, access to different speech partners, one’s own speech production feedback loop, age, sex, linguistic attitudes, etc.), which are active from the beginning of a sound change, and may influence even the most pared down phonetically-conditioned sound change.