Semantic and Pragmatic computing with GETARUNS

Abstract

We present a system for text understanding called GETARUNS, in its deep version applicable only to Closed Domains. We will present the low level component organized according to LFG theory. The system also does pronominal binding, quantifier raising and temporal interpretation. Then we will introduce the high level component where the Discourse Model is created from a text. Texts belonging to closed domains are characterized by the fact that their semantics is controlled or under command of the system; and most importantly, sentences making up the texts are fully parsed without failures. In practice, these texts are short and sentences are also below a certain threshold, typically less than 25 words. For longer sentences the system switches from the topdown to the bottomup system. In case of failure it will backoff to the partial system which produces a very lean and shallow semantics with no inference rules. The small text we will present contains what is called a "psychological statement" sentence which contributes an important bias as to the linking of the free pronominal expression contained in the last sentence.