From Syntax to Semantics

Abstract

A language's syntax provides a set of grammar rules for properly organizing the words in a sentence to make it semantically interpretable. A sentence such as John gives a book to Mary, because of its construction using proper English grammar rules, is semantically interpretable, as opposed to an arbitrary sequence of words John Mary a gives to book which leaves us puzzled as to the underlying interpretation. An example of a grammar rule would state that a verb phrase (gives a book) can be constructed by combining a verb (gives) and a noun phrase (a book). Note that semantically interpretable does not infer semantically plausible. For example, the sentence The desk gives a sun to Zoe. is semantically interpretable, but within our current world, the interpretation of a desk being an animate object able of giving, and a sun being something the desk can give, is, to say the least, challenging. To rephrase this last statement using the terminology introduced in Chap. 9, we can say that the entities (desk, sun) in the sentence The desk gives a sun to Zoe. are quite difficult to ground to our mental grounding space. The purpose of this chapter is to understand this gap between syntax and semantics and develop NLP approaches which try to fill it. To do so, we start with an exploration of the syntactic side, then jump to an exploration of the semantic side, and then look at the in-between steps necessary to go from a syntactic interpretation of a sentence into a semantic interpretation of a sentence. On the syntactic side, this chapter introduces a common type of grammar used for syntactic processing in current NLP research: dependency grammars. Dependency parsers are different in their output than the parse trees generated by constituency parsers, which we presented in Chap. 8. Dependency parser results can be shown in predicate-like form, made of a grammatical relation, a source, and a target term. For example, nsubj(eats,John) would be obtained from the sentence John eats. and express explicitly the dependency between the verb eats and its subject John. Such representation will be familiar to the Semantic Web audience. Jumping to the semantic side, we look at a formalism for semantic sentence interpretation , called semantic frames. Semantic frames define typical events and the