Vujović, Ramscar & Wonnacott (pre-print). Language learning as uncertainty reduction: The role of prediction error in linguistic generalization and item learning
Discriminative theories frame language learning as a process whereby prediction error reduces uncertainty about the meaning of an utterance. Previous work proposed that learning a suffixing language promotes prediction error, and thus generalization, to a greater extent than prefixing, which in turn tended to promote item-learning. We explored this in two large-scale web-based artificial language learning experiments with adult learners (total N = 434). Although we found no overall benefit of suffixing on generalization, participants in the prefix condition were more affected by feature frequency, and were more likely to incorrectly overgeneralize a high frequency, but non-discriminating (in terms of affix use) feature than those in the suffix condition. We demonstrate computationally that this behaviour is in-line with the predictions of a naïve discriminative learning model which treats affixes as cues to nouns under prefixing, and nouns as cues to affixes under suffixing. For item learning, we did not see the predicted benefit of prefixing, although there was overall better item-learning of low type-frequency items, which we discuss in terms of differences in the entropy of individual items. The results demonstrate the crucial role of prediction error in linguistic generalization, and have implications for how generalization interacts with item-learning.