Please use this identifier to cite or link to this item:
|Title:||TRANSLATION MODELS FOR GRAMMATICAL ERROR CORRECTION||Authors:||SHAMIL CHOLLAMPATT MUHAMMED ASHRAF||Keywords:||grammatical error correction, deep learning, convolutional neural networks, machine translation, quality estimation, natural language processing||Issue Date:||30-Sep-2018||Citation:||SHAMIL CHOLLAMPATT MUHAMMED ASHRAF (2018-09-30). TRANSLATION MODELS FOR GRAMMATICAL ERROR CORRECTION. ScholarBank@NUS Repository.||Abstract:||Grammatical error correction (GEC) automatically corrects various kinds of errors in writing, including spelling, punctuation, grammatical, and word choice errors. This thesis explores a machine translation (MT) approach which treats GEC as a translation task from “bad English” to “good English”. Apart from investigating several features for the statistical MT (SMT) approach, an SMT component for spelling error correction (Spell-MT) employing a character-level model is proposed. A domain-adapted neural network joint model is incorporated in the SMT framework to improve its generalization. Further, a multilayer convolutional encoder-decoder network is proposed for GEC, employing several techniques such as transfer learning and feature-based re-ranking and incorporating Spell-MT. The proposed approach achieves the highest published F0.5 score on the benchmark CoNLL-2014 test data, based on training on non-proprietary training data. Neural quality estimation (QE) models are proposed to assess the quality of real-time GEC outputs and to improve downstream GEC.||URI:||http://scholarbank.nus.edu.sg/handle/10635/151890|
|Appears in Collections:||Ph.D Theses (Open)|
Show full item record
Files in This Item:
|ChollampattS.pdf||1.08 MB||Adobe PDF|
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.