Monte Carlo Semantics: McPIET at RTE4

摘要

The Monte Carlo Pseudo Inference Engine for Text (McPIET) addresses the RTE problem within a new theoretic framework for robust inference and logical pattern processing based on integrated deep and shallow semantics. In this report we outline, in some detail, this new theoretic framework, and we will use it to shed some light on the informativity and robustness characteristics for the extreme cases of deep and shallow processing. Unsurprisingly, it will turn out that there is a tradeoff between informativity and robustness. We will be able to characterize an important new notion of a degree of validity, and provide some evidence to suggest that this concept plays a crucial role in the robustness of shallow inference. At the same time our framework still supports informationally rich semantic representations and background theories, which play the central role in the informativity of deep inference. Within our new theory we can then pose, from a completely new perspective, the problem of deep/shallow integration, and also propose a solution to it, which we will call Monte Carlo Semantics.