On using syntactic constraints in text recognition

The human reading process makes use of a number of knowledge sources which communicate with each other in arriving at correct interpretation. In order to obtain human competing performance in machines, this knowledge integration capability has to be built into the machines. A strategy for integration of the natural language syntactic knowledge in the text recognition task is presented. The syntactic knowledge layer is helpful only in those cases where the bottom-up hypothesized word alternatives belong to different syntactic categories. However, in the cases of unresolved ambiguities, the system comes with character pairs/groups over which more computationally-intensive low level processing is performed to ascertain its/their correct identity(/ies)