|
|
We describe a computational framework for exploring the interaction among lexicon size, phonotactic probability (PP) and neighborhood density (ND) in phonological word form learning. We developed a feed-forward neural network model to examine the impact of PP and ND on word learning for a specified lexicon size. A unique feature of this model is that the previous knowledge of known and unknown words is stored in bias units, allowing us to generate and test predictions about the independent effects of lexicon size, PP and ND both on word learning generally and on the learning of specific words. We tested the frameworkâs learning of nonwords uncontrolled for PP but varying in ND in three lexicon sizes based on CVCs from the MCDI (Fenson, et al., 1994). On average, the number of cycles to learn an unknown word decreased as the lexicon increased. In addition, within each lexicon size, high ND nonwords were learned more quickly than low ND nonwords. |
|