Застосування методів глибого навчання в задачах NPL: граматична корекція текстів

Abstract

. Modern approaches to solving the problem of text Grammatical Error Correction (GEC), based on deep learning methods, are considered. The main algorithmic methods, including rule-based, dictionary-based and more recent neural network-based methods, are considered. The problem of regeneration of whole input sequence in sequence-to-sequence models is described. The sequence-to-edits method with an example of such architecture is described as a solution to the regeneration problem. A perspective approach to combining algorithmic and neural network methods is proposed.