From Lexical Entailment to Recognizing Textual Entailment Using Linguistic Resources

From Lexical Entailment to Recognizing Textual Entailment Using Linguistic Resources Bahadorreza Ofoghi Centre for Informatics & Applied Optimization, University of Ballarat Victoria 3350, Australia b.ofoghi@ballarat.edu.au John Yearwood Centre for Informatics & Applied Optimization, University of Ballarat Victoria 3350, Australia j.yearwood@ballarat.edu.au Abstract In this paper, we introduce our Recognizing Textual Entailment (RTE) system developed on the basis of Lexical Entailment between two text excerpts, namely the hypothesis and the text. To extract atomic parts of hypotheses and texts, we carry out syntactic parsing on the sentences. We then utilize WordNet and FrameNet lexical resources for estimating lex- ical coverage of the text on the hypothesis. We report the results of our RTE runs on the Text Analysis Conference RTE datasets. Using a failure analysis process, we also show that the main difficulty of our RTE system relates to the underlying difficulty of syntactic analysis of sentences. 1 Introduction Success in many automated natural language ap- plications implies an accurate understanding of the meaning (semantics) of texts underlying the surface structures (syntax) by machines. This becomes chal- lenging with different syntactic forms and dissimi- lar terms and phrases expressing the same seman- tics. Automated natural language applications make extensive use of fine-grained text processing mod- ules that enable them in more effective dealings with structurally complicated texts. One of the current text processing tasks is con- cerned with inferring the meaning of a piece of text from that of another potentially larger text excerpt. This has now become a direction of study for the members of the natural language processing com- munity and is known as Recognizing Textual Entail- ment (RTE). The problem of RTE is formally de- scribed as recognizing the relationship between a pair of texts referred to as hypothesis and text. The hypothesis (H) is a succinct piece of text and the text (T) includes a few sentences the meaning of which may or may not entail the meaning of the hypothesis. If the meaning of H can be inferred from that of T, then the relationship is denoted by T → H. For instance, given H=“UN peacekeepers abuse children.” and T=“Children as young as six are being sexually abused by UN peacekeepers and aid workers, says a leading UK charity.” the relation T → H holds true. The classification of the relationship between the hypothesis and the text can be either a 3-way clas- sification or a 2-way classification task. The 3-way classes are: • Entailment: where T → H. • Contradiction: where T → ¬H. • Unknown: where there is not enough evidence available in the text to decide whether T → H or T → ¬H. In the 2-way classification method, the Contradic- tion and Unknown relations are unified into a single class called No Entailment. Our RTE system only considers the 2-way classification task. 2 Related work A few approaches to RTE have been developed dur- ing recent years. This includes the following. Term-based approach – Most of the systems that take this approach consider …

Content truncated to 500 words