Some issues and problems in text tagging using neural networks

ABSTRACT

This paper reports the results of several experiments conducted on automatic text tagging using neural networks. Error backpropagation networks were tested to see how effective they would be at correctly predicting the syntactic and semantic classification ("tag") of a word in a sentence, given some or no contextual information. The following contexts were examined: (1) the ending (last three letters) of the word alone, and (2) an encoded representation of the word, preceded by the representations of the three previous words in the sentence. Although each study suffered from some interesting problems with data representation, the results seemed promising and suggest that further investigation is warranted.

REFERENCES

• 1: Boggess, L.,Agarwal, R. & Davis, R. Disambiguation of propositional phrases in automatically labeled technical text. Proceedings ninth national conference on artificial intelligence, Volume 1, Menlo Park, MIT Press, 1991, 155-159.
• 2: Kenneth Ward Church, A stochastic parts program and noun phrase parser for unrestricted text, Proceedings of the second conference on Applied natural language processing, February 09-12, 1988, Austin, Texas [doi>10.3115/974235.974260]
• 3: Elenius, Kjell. Comparing a connectionist and a rule-based model for assigning parts of speech. ICASSP 90, 1990, Sll.9, 597 - 600.
• 4: Nakamura, M. & Shikano, K. A study of English word category prediction based on neural networks. ICASSP, 1989, S13.10, 731- 734.