Spell Checker for Punjabi Language Using Deep Neural Network

Abstract

Spell Checker is an important part of language specific text processing applications. In this paper, we propose a novel hybrid approach for spell checker for Punjabi language. Available spell checkers use traditional method for error detection such as dictionary lookup technique and minimum edit distance for error correction. The proposed spell checker aims to improve performance as well accuracy. In this paper, we present the use of trie data structure to store Punjabi words dictionary and then use tree based algorithm along with n-gram analysis to detect misspelled words. To correct misspelled words, best possible suggestion is listed using Long-short term memory (LSTM) recurrent neural network along with rule based approach and minimum edit distance. Error detection using trie-based dictionary improves the performance and Error correction using LSTM improves the accuracy of proposed spell checker. In addition to error detection techniques and error correction techniques, proposed spell checker uses handcrafted rules, language syntax rules and rules regarding tokenization.