Development of an english spell checker

Abstract

Spelling correction is an important part of modern word processing systems, helping to accurately interpret user input. This thesis explores the development of a contextual, edit distance corrector built on foundational computational linguistics techniques. The work is organized around the noise channel model, a complex framework that corrects typographical errors by leveraging the likelihood of specific character permutations alongside prior word probabilities. Initially, we define practices and apply them to perform possible corrections for deletion, insertion, substitution errors, and transposition errors. This function is validated using English vocabulary definitions obtained from the NLTK corpus to ensure that candidate disciplines are properly English words. At the core of our spell corrector is a noise channel model, which combines possible typos with unigram probabilities derived from a large text corpus This model is enhanced by the inclusion of a bigram language model, which considers context around typos in a sentence to improve the accuracy of correction. Our approach includes constructing two grams of frequency cells and using Laplace smoothing to compute conditional probabilities. The final spelling correction task combines these possibilities to determine the most likely correction for a given typo by considering the noise channel and context. The analysis highlights the effectiveness of the system in correcting isolated words and spelling errors in sentences. However, it also highlights limitations, such as the need for extensive corpus data, the potential for contextual confusion in ambiguity. This thesis makes assumptions about of the development process and concludes with recommendations for future improvements to improve the accuracy and usability of the maintenance script This review contributes to a foundational understanding of the spelling correction tool and provides a practical application that balances simplicity and efficiency, and paves the way for advanced text processing and recognition tools the relevant information.