A neural lexical post-processor for improved neural predictive word recognition

Abstract

This work presents a neural post-processor introducing lexical knowledge in a neural predictive system for on-line word recognition [4]. Each word is modeled by the natural concatenation of letter-models corresponding to the letters composing it. Successive parts of a word trajectory are this way modeled by different Neural Networks. A dynamical segmentation allows to adjust letter-models to the great variability of handwriting encountered in the words. Our system combines Multilayer Neural Networks and Dynamic Programming with an underlying Left-Right Hidden Markov Model (HMM). Training was performed on 7000 words from 9 writers, leading to already good results in the letter-labelling process. These results are significantly improved, at the word level, thanks to the use of the postprocessor.