Unifying Natural Logic and Probabilistic Logic for Recognizing Textual Relations

主要研究人员建立了一种识别文本关系的新的概率模型，该模型将自然逻辑、自然语言上的逻辑和统计概率模型统一起来。通过将文本关系的推理过程建模为句子中单元之间的对齐和因子图上的推理，可以精确地处理自然语言中的几种语言现象，并通过从一组数据集中自动学习语言关系来提供稳健性。对多个数据集进行了评价，证明了模型的有效性。

The principal investigator established a novel probabilistic model for recognizing textual relations that unifies Natural Logic, a logic over natural language, and a statistical probabilistic model. By modeling an inference process of a textual relation as alignment between units in sentences and inference over a factor graph, it makes possible to precisely handle several linguistic phenomena in natural languages, and provide robustness by automatically learning linguistic relations from a set of datasets. Evaluation over multiple datasets had been conducted, and the effectiveness of the model had been proved.