Supervised Word Sense Disambiguation with Frame-based Constructional Features : a pilot study of fán “to annoy/be annoying/be annoyed”

Research output: Journal Publications and ReviewsRGC 21 - Publication in refereed journalpeer-review

View graph of relations


Original languageEnglish
Pages (from-to)33-46
Journal / PublicationInternational Journal of Knowledge and Language Processing
Issue number2
Publication statusPublished - Oct 2018


Studies of Word Sense Disambiguation have made extensive use of unstructured contextual features for disambiguating polysemous words, while this paper aims to testify the validity of adopting linguistically-motivated and semantically-encoded features for classifying word senses. It conducts an innovative pilot study on supervised Word Sense Disambiguation of a Chinese polysemous verb fán 煩 “to annoy/be annoying/be annoyed” via the use of frame-based constructional (FC) features. Experimental results have shown significant improvement of using FC over the baseline sets, with the weight average FΔmax attains to 0.312. Besides, the noun phrase feature set also shows impressive performance compared to uni-grams and bi-grams, which tends to imply a close relation between verb meaning and core arguments. The promising results have proved the great discriminativeness of FC for sense disambiguation and suggest a possible alternation of employing/combining deep linguistic resources for Natural Language Processing applications in future.

Research Area(s)

  • Word Sense Disambiguation, Frame-based Constructional Features, Supervised Machine Learning, Polysemous Emotion Verb fán, Lexicalization Pattern