Automatic Spelling Recognition System:
See this
Snapshot, read the
paper, and play with an online
Java webstart prototype
This prototype system is designed to process messages corrupted by spelling errors, typos, or distorted scanned texts to cleverly recover the original message from the corruption of data. Variants of hidden Markov models (HMMs) are used to model the source of errors, and variants of bigram and trigram language models are used to model the context and how likely one word can appear after another. The system has to calculate the most likely original message based the corrupted message and the models mentioned above. We have already implemented the part of BioLinguistic to recover message from typing errors. Part of the prototype was used in the artificial intelligence (CSCI 440) class.
