OBLs Hobble Computations

One of the wonders about Natural Language is the way it manages to repurpose limited resources. Unfortunately for linguistic computations the way it does this is by relying heavily on non-linguistic context. This means that given the incomplete information that we have when we try to analyze text relying only on syntactic structure and limited lexical resources, our analyses are most of the time ambiguous in multiple ways. This problem can be compounded by the incompleteness of our understanding of syntax and of the structure of the lexicon. In this paper we discuss a case where ambiguities are created by incomplete syntactic and contextual knowledge and where attempts to remedy the explosion of ambiguities that they allow in fact make the problem worse. The problem we focus on is how the interaction between some obliques and adjuncts in XLE and the interaction between XLE subcategorization and VerbNet information together create computational bottlenecks. We propose to solve part of these problems by assimilating one type of obliques to adjuncts.