Incremental Derivations in CCG

This paper presents a research note on the degree to which strictly incremental derivations (that is derivations which are fully connected at each point in time) are possible in Combinatory Categorial Grammar (CCG). There has been a recent surge of interest in incremental parsing both from the psycholinguistic community in a bid to build psycholinguistically plausible models of language comprehension, and from the NLP community for building systems that process language greedily in order to achieve shorter response times in spoken dialogue systems, for speech recognition and machine translation. CCG allows for a variety of different derivations, including derivations that are almost fully incremental. This paper explores the syntactic constructions for which full incrementality is not possible in standard CCG, a point that recent work on incremental CCG parsing has glossed over.