On the Naturalness of Unnatural Rules

In the 1970’s the approach of Natural Phonology focused on the observation that phonetically motivated (“natural”) processes are common in the phonologies of the world’s languages, while phonetically unmotivated (“unnatural”) ones are uncommon (cf. Stampe 1969, Vennemann 1974, Hooper 1976, Donegan and Stampe 1979; for reactions at the time see, for example, Hellberg 1978 and Anderson 1981). Much recent work, especially in Optimality Theory (Prince and Smolensky 1993), revives this general philosophy: the phonological system incorporates phonetic constraints on outputs, ranging from perceptibility of acoustic cues in particular contexts to the general principle of least effort (Flemming 1995, Steriade 1997, Myers 1997, Boersma 1998, Kirchner 1998). I present evidence that this approach makes false predictions regarding the historical development of phonological systems (cf. Ohala 1974, 1981, 1997, Kaye 1989, Hale and Reiss 1998, in press, Hyman 1998, Kingston 1999). I argue that while phonetically driven approaches correctly identify the factors that lead to common alternations, it is problematic to incorporate these factors into the grammar itself. From the learner’s (and the grammar’s) point of view, the original cause of an alternation is irrelevant: the learner’s only goal is to reproduce the language she hears around her.