Ad-hoc Meaning Substitution Causes Shape and Material Biases: Computational Explanation for Emergence of Word Learning Biases

We illustrated a computational mechanism of shape bias and material bias with “ad-hoc meaning substitution (AMS)” hypothesis and verified it by computer simulations. AMS represents that when given a novel word and a instance object/substance, children substitute a known noun meaning nearest to the instance and the instance itself as an ad-hoc template of the novel noun meaning. The substitution enables fast mapping and subsequent vocabulary spurt. To describe internal process of AMS, we introduced “word distributional prototype (WDP)” as an explicit representation of word meaning with an inductive learning function. Simulation 1 revealed that when neural networks with WDP and AMS were given biased vocabularies reflecting those of young children, it demonstrated shape, material, and overgeneralized shape biases which means wrong shape bias over material bias. This result suggests that the triad of word meaning induction, AMS, and early biased vocabulary is essential for emergence of the biases. Simulation 2 introduced a notion of maturity that denote a degree of learning convergence for each word meaning, and then networks showed neither shape nor material bias during early small vocabulary. This result indicates that the age of bias emerges is decided by the maturity. These results suggest that phenomena concerning shape and material biases are explicable with the simple ad-hoc learning mechanism instead of meta learning or built-in language-specific ones.