PADS Restoration and Its Importance in Reading Comprehension and Meaning Representation

摘要

Unlike competent human readers capable of inferring, tracing, and filling out gaps or hurdles left behind by authors' use of transformations in their writing such as permutation, addition, deletion, and substitution (PADS), these operations are challenging to computer readers and new foreign language learners. This paper reports a parser's use of a suite of NLP technologies - clause boundary detection, resolution of different anaphors, inter-event relation finding, and case frame building - to fill out PADS gaps and output a much more explicit kernel-like meaning representation that includes case relation tuples of "Who Did What to Whom" and the inter-event relations based on conjoining, embedding, branching, insertion and apposition. According to Halliday and Hasan (1976), those gaps serve as cohesive devices to achieve better texture of the text organization. The transformations are ruled-based and they are important part of native speakers' competence. Though the rule-based parser is still short of perfection, the necessary design is in place and it has quite a few encouraging results. This report will also show the usefulness of PADS restoration technology in CALL and information extraction.