Research on intelligent word completion
A word-completion utility facilitates the typing of text by a user who has physical or cognitive disabilities or is working on a keyboard with few keys, such as a telephone keypad. As the user enters each keystroke, the program displays a list of the most likely completions of the partially typed word. As the user continues to enter letters, the program updates the suggestion list accordingly. If the intended word is in the list, the user can select it with a single keystroke or mouse-click. For a user with physical disabilities, for whom each keystroke is an effort, this saves time and energy; for a user with cognitive disabilities, this can assist in the composition of well-formed text. A number of word-completion utilities are available commercially; but their suggestions, which are based on n-gram frequencies, are often syntactically or semantically implausible, excluding more-plausible but lower-frequency possibilities from the list. This can be particularly problematic for users with certain cognitive disabilities, such as dyslexia, who often are easily confused by inappropriate suggestions.
Our research aims to improve these programs by adding linguistic intelligence to them. Afsaneh Fazly's thesis explores the addition of syntactic information to word completion, developing new algorithms in which the part-of-speech tags of words are used in addition to the words themselves to improve the accuracy of the suggestions. The results show a small but statistically significant improvement in keystroke savings over baselines that use only word n-grams. Jane (Jianhua) Li's thesis adds semantic knowledge to this idea. Knowledge of semantic relationships is used to measure the semantic association of completion candidates with the context. Those that are semantically appropriate to the context are promoted to the top positions in prediction lists due to their high association with context. Experimental results show a performance improvement when using the integrated model for the completion of nouns.
This research is carried out in collaboration with Dr Fraser Shein of the Bloorview Research Institute at Bloorview Kids Rehab, developers of WordQ writing software.
This research is related to our work on semantic distance and on lexical chains and threads of meaning in documents.
References:
-
Li, Jianhua. Modelling Semantic Knowledge for a Word
Completion Task.
M.Sc. thesis, Department of Computer Science, University of Toronto, October 2006. Abstract | PDF. - Li, Jianhua and Hirst, Graeme. ``Semantic knowledge in a word completion task.'' Proceedings, 7th International ACM SIGACCESS Conference on Computers and Accessibility, Baltimore, October 2005. Abstract | PDF
- Fazly, Afsaneh and Hirst, Graeme. ``Testing the efficacy of part-of-speech information in word completion.'' Workshop on Language Modeling for Text Entry Methods, 11th Conference of the European Chapter of the Association for Computational Linguistics, Budapest, April 2003. Abstract | PDF | PostScript
- Fazly, Afsaneh. The use of syntax in word completion utilities. MSc thesis, Department of Computer Science, University of Toronto, January 2002. Abstract | gzipped PostScript (448 Kb) | PostScript (2.4 Mb) | PDF (532 Kb).