How do we get there? Evaluating transformer neural networks as cognitive models for English past tense inflection

Anonymous

Abstract: Neural network models have achieved good performance on morphological inflection tasks, including English past tense inflection. However whether they can represent human cognitive mechanisms is still under debate. In this work, we examined transformer models with different size and distribution of training data to show that: 1) neural model's performance correlates with the adult behavior, but not children's behavior; and the model with small-size training data that matches parents' input distribution has the highest correlation; 2) neural models' errors are not human-like; however, the errors on the regulars and irregulars show a clear distinction. Therefore, we conclude that the current transformer models exhibit some resemblance of human behavior, but is insufficient as a cognitive model of learning morphological rules.

Paper Type: long