HPSG 2013

Frank Richter: 87, 94, 14? The Nature of Linguistic Theory

In the early days of HPSG, there was considerable emphasis on its formal foundations and on the 'nature of linguistic theory' that would go with a particular formalization. In fact, HPSG came with at least two significantly different sets of foundational assumptions: A logic of partial information in 1987, and a theory of linguistic types in 1994, soon to be followed by a number of modifications as the mathematical details were spelled out and a few philosophical disagreements over their interpretation were settled. In the meantime, the way in which linguistic research is conducted has changed: Controlled psycholinguistic experiments provide a much more fine-grained picture of processing facts and the gradedness of human acceptability judgments, statistical language modeling has become the standard in computational linguistics, and large-scale corpora allow insights into actual language use that were not available before. Closer to home for HPSG, there is now a considerable number of computational grammar implementations which show how HPSG fares with respect to grammar fragments in different languages, and how these grammars behave computationally. In this talk, I would like to reconsider the nature of HPSG-style grammar theory and its perspective on language in light of the empirical turn in linguistics. Questions to be touched upon include how suitable its foundational assumptions are for the description of data gained with any of the empirical methods above, what these data can or should tell us about linguistic theory, and which of them should be considered subject to what kind of linguistic theory. We know HPSG in 1987 and in 1994. My modest aim is to present HPSG 2014.