A Language Model for Improving the Graph-Based Transcription Approach for Historical Documents

Language Models (LMs) capture the contextual dependencies of a language and assign higher probabilities to well-formed sequences of words. For that reason, LMs have been commonly used in generic handwriting recognition, improving recognition results. In this paper, we present the integration of a Language Model along with a dictionary into a graph-based recognizer, which aims at transcribing handwritten historical documents. The results of such integration show a significant improvement on word accuracy when applied to our corpora.