Evolution of a Rapidly Learned Representation for Speech

Evolution of a Rapidly Learned Representation for Speech Ramin Charles Nakisa and Kim Plunkett Department of Experimental Psychology Oxford University South Parks Road Oxford OX1 3UD, UK {ramin. nakisa,kim, plunkett }@psy. ox. ac. uk Abstract Newly born infants are able to finely discriminate almost all human speech contrasts and their phonemic category boundaries are initially identical, even for phonemes outside their target language. A connectionist model is described which ac- counts for this ability. The approach taken has been to develop a model of innately guided learning in which an artificial neu- ral network (ANN) is stored in a "genome" which encodes its architecture and learn- ing rules. The space of possible ANNs is searched with a genetic algorithm for net- works that can learn to discriminate hu- man speech sounds. These networks per- form equally well having been trained on speech spectra from any human language so far tested (English, Cantonese, Swahili, Farsi, Czech, Hindi, Hungarian, Korean, Polish, Russian, Slovak, Spanish, Ukranian and Urdu). Training the feature detec- tors requires exposure to just one minute of speech in any of these languages. Cate- gorisation of speech sounds based on the network representations showed the hall- marks of categorical perception, as found in human infants and adults. 1 Introduction Precocious abilities in newborn infants are fre- quently taken as evidence for pre-specification of the representations that support those abilities. The prespecifications of these representations is innately determined, presumably in the genome of the indi- vidual. One such ability is that of newborn infants to be universal listeners, able to discriminate speech contrasts of all languages. This is all the more re- markable since the low-pass filtered speech sounds that foetuses hear in utero vary widely between dif- ferent languages. Eimas et al. (1971) showed that 1-4 month old infants displayed categorical perception of the syl- lables/ba/and/pa/. That is to say, infants carve up the phonetic space into a set of categories with sharp boundaries. Variants of a phoneme, such as /b/, are not discriminable, even though they dif- fer acoustically by the same amount as /p/ and /b/ (although see (guhl, 1993)). More recent re- search has shown that the categories are univer- sal, so that English-learning infants can discriminate non-native contrasts in Czech (Trehub, 1973), Hindi (Werker, Gilbert, Humphrey, ~ Tees, 1981), Nth- lakampx (Werker & Tees, 1984a), Spanish (Aslin, Pisoni, Hennessy, ~ Percy, 1981).and Zulu (Best, McRoberts, ~ Sithole, 1988). This suggests that in- fants develop an initial representation of speech that is universal and largely insensitive to the particular language to which they are exposed. The ability to discriminate some non-native speech contrasts de- clines after the age of 10-12 months (Werker ~ Tees, 1984a). Such rapid learning can be defined in terms of a taxonomy developed in the field of animal be- haviour. Mayr (1974) suggested that programs of development form a continuum of flexibility in their response to environmental stimulation. He distin- guished between "open" and "closed" programs of development. "Closed" programs of development rely …

Content truncated to 500 words