Use of Language Models in Handwritten Sentence/Phrase Recognition
Rohini Srihari Principal Investigator
MetadataShow full item record
Srihari 9315006 This is a standard award in the NSF/ARPA Human Language Technology for cooperative research on how to utilize new and improved approaches to the recognition of handwritten text with the support of language models based on linguistic knowledge, involving researchers at the State University of New York at Buffalo and the Eastman Kodak Company. The approach concerns the use of contextual constraints to improve the recognition and, at times, override the results of classical pattern recognition. This includes the use of lexicons and holistic filtering, of dictionary statistics, and of high-level syntactic and semantic post-processing. Although the research focuses on models of the English language, with emphasis on the linguistic component rather than on the pattern recognition component, and utilizes statistical training with representative corpora, it can be extended to other European languages and to the recognition of printed text.