Episodic grammar: a computational model of the interaction between episodic and semantic memory in language processing

We present a model of the interaction of semantic and episodic memory in language processing. Our work shows how language processing can be understood in terms of memory retrieval. We point out that the perceived dichotomy between
rule-based versus exemplar-based language modelling can be interpreted in a neuro-biological perspective in terms of the interaction between a semantic memory system that encodes linguistic knowledge in the form of abstract rules, and an episodic memory that stores concrete linguistic events. We implement the idea of a semantic-episodic memory integration in a probabilistic grammar, and evaluate its performance as a syntactic parser on corpora of natural language. Our labeled precision and recall results are competitive with state-of-the-art syntactic parsers, with F-scores up to 90.68 on section 22 of the Penn WSJ corpus.