Annotating by Proving using SemAnTE

Abstract

We present SemAnTE, a platform for marking and substantiating a semantic annotation scheme of textual entailment according to a formal model. The platform introduces a novel approach to annotation by providing annotators immediate feedback whether the data they mark are substantiated: for positive entailment pairs, the system uses the annotations to search for a formal logical proof that validates the entailment relation; for negative pairs, the system verifies that a countermodel can be constructed. By integrating a web-based user-interface, a formal lexicon, a lambda-calculus engine and an offthe- shelf theorem prover, this platform facilitates the creation of annotated corpora of textual entailment. A corpus of several hundred annotated entailments is currently in preparation using the platform and will be available for the research community.