Modeling Human Sentence Processing with Left-Corner Recurrent Neural Network Grammars

In computational linguistics, it has been shown that hierarchical structures make language models (LMs) more human-like. However, the previous literature agnostic about a parsing strategy of models. this paper, we investigated whether LMs human-like, and if so, which is most cognitively plausible. order to address question, evaluated three against human reading times in Japanese with head-final left-branching structures: Long Short-Term Memory (LSTM) as sequential model Recurrent Neural Network Grammars (RNNGs) top-down left-corner strategies Our modeling demonstrated RNNGs outperformed LSTM, suggesting architectures are plausible than or architectures. addition, relationships between cognitive plausibility (i) perplexity, (ii) parsing, (iii) beam size will also be discussed.