A Lexico-Syntactic-Semantic Approach to Recognizing Textual Entailment

Introduction

The task of recognizing textual entailment (RTE) takes a pair of text fragments as input and determines whether the meaning of one fragment (called hypothesis, H) can be derived from the meaning of the other fragment (called text, T), that is, whether T logically entails H. Due to huge impact of the RTE task in many application areas of natural language processing (NLP), such as information extraction, information retrieval, question answering, summarization, paraphrase acquisition, machine translation and reading comprehension, the PASCAL Network of Excellence funded by the European Union has organized a number of RTE challenges and, in particular, has released a number of RTE datasets, such as RTE1 to RTE4 (Dagan et al., 2005; Bar-Haim et al., 2006; Giampiccolo et al., 2007; Giampiccolo et al., 2008).

Textual entailment (TE) is a directional relationship between T–H pairs: the entailment relation may hold from T to H but not from H to T, as in the following example from the RTE1 dataset:

• Example 1 (Entailment).

• T: The Daily Telegraph, most prized asset in Lord Conrad Black’s crumbling media empire, has been sold to Britain’s Barclay twins.

• H: Daily telegraph is sold.

Here, the meaning of H can be derived from that of T. However, T contains additional information other than that contained in H and thus cannot be inferred from H.

Typically, RTE is a 2-way classification task, where T–H pairs are labeled as YES or NO entailment. This 2-way classification task has been extended to a 3-way task in the fourth PASCAL RTE challenge, where the no-entailment cases were further classified into mutually contradicting or not. In the RTE4 dataset, T–H pairs are labeled as Entailment, Contradiction or Unknown:

• Example 2 (Contradiction).

• T: As German voters go to the polls on Sunday, unemployment will be a key issue. Despite tough labour market reforms, the number of unemployed has risen to 5m. And Germany's jobless are getting despondent.

• H: Germany's jobless rate decreases.

Here, not only no entailment holds from T to H, but T and H expresses contradicting information.

• Example 3 (Unknown).

• T: Proposals to extend the Dubai Metro to neighbouring Ajman are currently being discussed. The plans, still in the early stages, would be welcome news for investors who own properties in Ajman.

• H: Dubai Metro will be expanded.

Here, T only represents a possibility of an event (expansion of Dubai Metro) but not asserts this event. Therefore, H is not entailed by T but does not contradict it.