Phonology is not psychological and speech processing is not linguistic ∗

Experimental evidence about human speech processing and linguistic memory shows that words are not spelled from letter-like units, whether thought of as phones or phonemes. Linguists, like others with a Western education and a lifetime of literacy, identify speech quite automatically with a sequence of letter-sized units. Consonants and vowels seem like directly observable units in speech. However, the evidence is clear that we actually use high-dimensional spectro-temporal (i.e., auditory) patterns to support speech production, perception and memory in real time. Thus abstract phonology (with its phonemes, syllable types, distinctive features, etc.) needs to be re-conceived as a social institution – a system of patterns that evolves over historical time in some community playing almost no role in practical speech processing. There are really two sciences of language: First there is real-time Speech-Language Processing, the rich code used to store and manipulate information about linguistic patterns. Second is a proposed new Phonology that characterizes patterns in the corpus of a linguistic community (but makes no claims about psychological processing). This distinction cuts orthogonally across Chomsky's Competence vs. Performance since each has some properties of competence and some properties of performance. Experimental results force us to differentiate the language as a social institution (thus not necessarily`present' in each speaker's brain) from the actual psychological processing as we talk to each other. Chomsky thought speakers must``know'' their language, but they do not becausè`the language'' is an abstract, speaker-independent system that exists only at the level of the community. Speakers exhibit or perform a language, but do not need tò`know'' it.