Naturalness in Generative Phonology

Abstract

We raised the problem of the abstractness of underlying representations in the interlude and suggested there that, in some cases (e.g. the postulated high back, unrounded vowels of Hungarian), we might question the supposed psychological reality of those representations. The objection is to underlying representations which appear not to be directly induced by the sounds to which the speaker is exposed. Let us examine that line of argument. One very clear expression of this objection to abstract underlying representations is given by Hooper (1976). We have argued all along that the combined rules and representations we have formulated are psychologically real, and that they allow us to characterise the knowledge which enables speakers to decode a multiplicity of different speech sounds. Hooper claims that standard generative phonology cannot sustain these claims to psychological reality for analyses involving absolute neutralisation (AN) analyses and extrinsic ordering. She accordingly imposes two general conditions on phonological analyses which, she argues, will help to guarantee their psychological plausibility. Those conditions may be stated as follows:

The True Generalisation Condition (TGC) No phonological generalisation is a true one unless it is true at the level of surface phonetic representation.