Recognizing Textual Entailment Using Inference Phenomenon

Inference phenomena refer to inference relations in local fragments between two texts. Current research on inference phenomenon focuses on the construction of data annotation, whereas there are few research on how to identify those inference phenomena in texts, which will contributes to improving the performance of recognizing textual entailment. This paper proposes an approach, which uses inference phenomena to recognize entailment in texts. In the approach, the task of recognizing textual entailment is formalized as two problems, that is, inference phenomenon identification and entailment judgment, then a joint model is employed to combine such two related subtasks, which is helpful to avoid error propagation. Experimental results show that the approach performs efficiently for identifying inference phenomena and recognizing entailment at the same time.