A Recurrent Neural Network for Multiple Language Acquisition: Starting with English and French

Abstract : How humans acquire language, and in particular two or more different languages with the same neural computing substrate, is still an open issue. To address this issue we suggest to build models that are able to process any language from the very beginning. Here we propose a developmental and neuro-inspired approach that processes sentences word by word with no prior knowledge of the semantics of the words. Our model has no "pre-wired" structure but only random and learned connections: it is based on Reservoir Computing. Our previous model has been implemented in the context of robotic platforms where users could teach basics of the English language to instruct a robot to perform actions. In this paper, we add the ability to process infrequent words, so we could keep our vocabulary size very small while processing natural language sentences. Moreover, we extend this approach to the French language and demonstrate that the network can learn both languages at the same time. Even with small corpora the model is able to learn and generalize in monolingual and bilingual conditions. This approach promises to be a more practical alternative for small corpora of different languages than other supervised learning methods relying on big data sets or more hand-crafted parsers requiring more manual encoding effort.