Effect of inconsistent distinctiveness of artificial semantic features on retrieval speed

A purported universal of language is the generalization of distinctive features, for example, in phonology and semantics, within a language or an individual. This could be due to a tendency either to use or to ignore a feature depending on whether or not it has previously been distinctive. This mechanism is shown to operate in retrieval of well-learned artificial concept names. Several simple models of performance cannot account for this effect. The apparent inefficiency of possible mechanisms which could account for this result suggests that it may reflect a special predisposition.