A Model Of Fuzzy Sense Inference Of English Modal Verb Can By ANFIS

Fuzziness is one of the intrinsic features of natural language and a widely-existing phenomenon in every aspect of human society. Since Fuzzy Set Theory was first proposed by L. A. Zadeh in 1965, a number of researches on fuzziness have been carried out by means of mathematical methods. Fuzzy linguistics applies fuzzy set theory into linguistic studies. Studies and researches on fuzziness in linguistics have begun in 1970s at abroad, while Professor Wu Tieping first introduced fuzzy linguistics to China. Fuzzy semantics, as an interdiscipline, needs to find its closer connection with other adjacent discipline, like math, computer and cognitive science, etc. Currently in this field, much effort has been devoted to systematic studies and researches on semantic and grammatical level. However, few researches on fuzzy sense inference by integrating Fuzzy Set Theory and modern computational techniques have been done so far.Since modality is the mental representation of the speaker's attitude, the English modal verbs are remarkably indeterminate in their meanings. Gradience is one important feature of the indeterminacy in modal meanings. In this research, the quantification of the grades of membership of the three interrelated meanings of modal verb can is realized and a group of semantic and syntactic features that potentially influence the gradience are selected. This paper aims at establishing a highly accurate model of fuzzy sense inference by means of Adaptive Neuro-Fuzzy Inference System. To realize the quantification of the linguistic features, Woncord software is utilized to do statistics and Mutual Information of related linguistic information is calculated. Thereafter, Adaptive Neuro-Fuzzy Inference System is adopted to set up models among which the optimal one is selected. The experimental results reveal a sequence of influences of different semantic and syntactic features on the meanings of can. This result is a valuable reference for the semantic studies on modal can and a better comprehension of English learners on the semantic gradience of modality and its influencing factors. Meanwhile, the establishment of this inference model does not only realize the computational inference of modal meanings, but also free linguists from massive corpus tagging work for sense inference and promote the intellectualization of effective man-machine exchange.