Integration of lexical and syntactical knowledge in a handwriting-recognition system

Abstract

This article presents a research project carried out with the aim of investigating the improvements in recognition performances that result from the use of linguistic information in a handwriting-recognition system. The purpose of the study was to design a postprocessor that would enhance an existing handwriting-recognition system by identifying and correcting words it did not recognize initially. This was done by integrating linguistic information (both lexical and syntactical) into the system. Every sentence containing one or more incorrect words is parsed and all possible grammatical classes for each incorrect word are listed. Then, a lexical enquiry searches for words in the lexicon corresponding to the grammatical class of the word in question. Finally, a string-comparison algorithm selects only the words in the lexicon that are close to the incorrect word. The results of this experimentation show that such a system is more efficient in correcting words (even highly distorted ones) than conventional systems that only integrate lexical information. In conclusion, the integration of linguistic information to correct words not recognized by a handwritingrecognition system is shown to be an effective approach, and one that might be worth pursuing.