Results 11 - 12 of 12
"... Despite our strong intuitions that language is represented in memory using some kind of alphabet, phones and phonemes appear to play almost no psychological role in human speech perception, production or memory. Instead, evidence shows that people store linguistic material with a rich, detailed audi ..."
Abstract - Add to MetaCart
Despite our strong intuitions that language is represented in memory using some kind of alphabet, phones and phonemes appear to play almost no psychological role in human speech perception, production or memory. Instead, evidence shows that people store linguistic material with a rich, detailed auditory and sensory-motor code that tends, in its details, to be unique for each speaker. The obvious phonological discreteness of languages reflects conventional categories of pronunciation but not discrete symbols. In learning to read, we all master the Speech-Letter Blend, so that letters can be effortlessly interpreted as speech when reading. This Linguistic Intuitions Linguistics, for at least the last century, has almost invariably subscribed to the following assumption: Speech in any language consists of a sequence of words which are composed from a sequence of phonemic (or phonetic) symbols. This is the standard view of language as the archetypal symbol system and is shared by neighboring disciplines, such as psychology of language and language development. Of course, this assumption is shared by lay people as well since we all have strong intuitions that words and phonemes are just the right units for description of any language. This paper will argue that these intuitions primarily reflect conventions about alphabetical orthographies that have been assumed to be true of spoken language as well, but without sufficient examination (Port, 2007; 2010). In fact, many kinds of familiar data have been incompatible with this assumption for at least 60 years (Pisoni, 1997). But linguists and others have refused to consider seriously the idea that speech may demand much higher dimensionality and thus much richer memory for utterances than has been presumed by the standard view. Interestingly, engineers, despite similar intuitions, began over 40 years ago to turn toward speech recognition systems that seek models of whole words and phrases (rather than phonemes) and to specify them in terms of spectral detail rather than discrete symbols (Jelinek, 1969; Huckvale, 1999). There have been a few attempts to apply these insights to models of human speech perception, such as Klatt’s