We introduce a new formal semantic model for annotating textual entailments that describes restrictive, intersective, and appositive modification. The model contains a formally defined interpreted lexicon, which specifies the inventory of symbols and the supported semantic operators, and an informally defined annotation scheme that instructs annotators in which way to bind words and constructions from a given pair of premise and hypothesis to the interpreted lexicon. We explore the applicability of the proposed model to the Recognizing Textual Entailment (RTE) 1a??4 corpora and describe a first-stage annotation scheme on which we based the manual annotation work. The constructions we annotated were found to occur in 80.65% of the entailments in RTE 1a??4 and were annotated with cross-annotator agreement of 68% on average. The annotated parts of the RTE corpora are publicly available for further research. ?? ??.
展开▼
