An L2 Neural Language Model of Adaptation to Dative Alternation in English

Abstract

Neural(-network) language models (NLMs) have recently been shown to adapt not only to lexical items but also to abstract syntactic structures. In this study, we provide further evidence for this thesis by showing that the syntactic priming paradigm on an L2 LSTM (Long Short-Term Memory) language model (LM) enhances the ability for it to track abstract properties of sentences compared to the non-cumulative priming paradigm. Furthermore, we investigate the effect of the learning rate on adaptation. In so doing, we probe how much enhancement is due to adapting such an L2 NLM’s syntactic representations. We report the performances of the L2 LSTM LM in the adaptation experiment focusing on dative alternation in English and confirm that they adapt both lexical items and syntactic structures, just as L1 NLMs do.