Textual Entailment Recognition And Its Application In Information Retrieval

In recent years, the rise of interest in textual entailment recognitiondoes not come with correspond improvement of its accuracy. An importantfactor of the inaccuracy is caused by the lack of background information orbackground knowledge. On the other hand, the background knowledge isrestricted in many areas of expertise, which is relative easy to model andput into use. With the help of domain knowledge, the promotion ofaccuracy of entailment recognition becomes within reach. Using highaccurate domain entailment recognition into search helps use to identifythe required information more quickly, since a general assessment of thedocument is provided by the result of entailment recognition.In our work, we introduce domain ontology into the system to modeldomain knowledge and use natural logic to boost the process of entailmentinference, which transforms the text only halfway into logical form. Bydoing this, the texts are equipped with inference power and the complexityas well as fragility is avoided.The segmentation of the process of recognition inference makes theresult of each step human-understandable, which provides a window ofobservation for further analysis and improvement.The new inference algorithm is designed according to the specialty ofdomain search and minimum human labor from the user is also put intoconsideration, which leverages the power of entailment recognition tofacilitate in-domain information retrieval.