Modelling phonological category learning

The term " phonological categories " refers to the discrete elements that make up a phonological representation, i.e. elements of its temporal organization (e.g. foot, syllable, mora, segment, or autosegment) and elements of its internal content (e.g. phonemes such as /p/ and /n/, or feature values such as [+nasal] and the high tone H). From the mere fact that the term " phonological category learning " appears in the title of this contribution, the reader can already infer that the editors of this Handbook, and/or the present author, assume that at least some of these phonological categories can be learned. This assumption is opposite to the assumption held by most generative phonologists, which is that all phonological categories are innately given to the human infant. Thus, Chomsky & Halle (1968: 4) state that " phonetic features " belong to the " substantive universals " , which are a subgroup of " linguistic universals " , which are " available to the child… as an a priori, innate endowment ". Likewise, Prince & Smolensky (1993 [2004: 2–3]) state that " Universal Grammar " , i.e. the innate language endowment, " consists largely of a set of constraints on representational well-formedness " ; in their examples, such innate constraints often refer to substantive phonological elements, which therefore have to be innate a fortiori. the generative assumption of innate categories comes with a learnability problem, namely the linking problem. In the phonology and phonetics domain this means that even if all phonological categories were innately given, the language-acquiring child would still have to connect at least some of these innate categories to auditory events available in the incoming speech data, and this is a problem because the mappings between some phonological categories and auditory events vary crosslinguistically and cannot therefore be innate. After all, the hypothesis of innate categories presupposes the universal existence of e.g. the phoneme category /u/ (or of the feature values [+back], [+high] and [+round]), but since a phonological element representable as /u/ (or as the feature bundle [+back, +high, +round]) is typically pronounced slightly differently in every language, the mapping between this phonological category and auditory events must be language-specific and cannot therefore be innately given. The reasoning in the previous paragraph may not convince many generative phonologists. After all, a generative phonologist could object that an innate category /u/ could correspond to a region of auditory events, e.g. …