Spelling Correction Using Encoder-Decoder and Damerau-Levenshtein Distance

A spell checker is a tool for detecting and correcting various spelling errors. Using memory and pattern recognition skills, humans find it easy to correct spelling errors. In contrast, for a machine to utilize these functions to perform a similar task, we need to train it to recognize patterns. One of the critical tasks of NLP has been to devise solutions for denoising and correcting misspelt words in a document and ease typing for typists by providing support infrastructure. In recent years, Long Short-term Memory (LSTM) has been utilized for similar operations because it aids the decision-making process by utilizing contextual information. To tackle this problem and to provide a new understanding of this problem, we propose a mixture model that uses an LSTM model to correct the spelling in the first go and then pass it to a probabilistic model that can then make edits to it. After getting the output from the LSTM model, which utilizes the encoder-decoder architecture, the probability model calculates the DL distance up to 2 to find the closest word in the English dictionary and replace the passed word.