Langnage-Level Syntactic and Semantic Constraints Applied to Visnal Word Recognition

VariOllS aspects of using language-level syntactic and semantic cOUM straints to improve the performance of word recognition algorithms are dis~ cussed, Following a brief presentation of a hypothesis generation model for handwritten word recognition, various types of languageMlevel constraints are reviewed, Methods that exploit these characteristics are discussed including intraMdocument word correlation, common vocabularies, partMof-speech tag COM occurrence, structural parsing with a chart data structure. and semantic biasing with a thesaurus.