Assignment of gender to French nouns in primary and secondary language : a connectionist model

Abstract

In French, grammatical gender is often represented phonologically and/or morphologically. Thus, a language learner's competence for gender iden tification might in part reflect the ability to recognize patterns in noun phonology and morphology. We herein describe a computer-based connec tionist-type network model which learned to identify correctly the gender of a set of French nouns. Subsequently, this model was able to generalize from that learning experience and assign gender to previously unstudied nouns with a high degree of reliability. This gender assignment was accomplished by relying solely upon information inherent in the structure of the nouns themselves, and it occurred in the absence of explicit rules for the evaluation of nouns. Instead, the model discovered criterial gender-specific features when shown examples of masculine and feminine nouns during its initial training period. The model's ability to learn these gender-specific features was found to be related both to its initial connectivity state and to a variable learning-rate parameter. These latter results are discussed with respect to their general implications for second language learning.