Please use this identifier to cite or link to this item:
Title: Deep learning for English grammatical error correction
Authors: Luo, Jingying
Keywords: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
Engineering::Electrical and electronic engineering
Issue Date: 2021
Publisher: Nanyang Technological University
Source: Luo, J. (2021). Deep learning for English grammatical error correction. Final Year Project (FYP), Nanyang Technological University, Singapore.
Project: B3310-201
Abstract: Rule-based approach and deep learning approach are two most popular approaches while dealing with Grammatical Error Correction (GEC) task. The rule-based approach is strict, fast and precise but unable to deal with complex errors. The deep learning approach is more powerful with the ability to deal with complex or semantic errors, but minor errors are sometimes ignored due to the complexity of neural networks. This Final Year Project report has investigated the Encoder-Decoder based Sequence-to-Sequence deep learning model on GEC task and incorporated it with the rule-based pre-processing approach. The Deep Dynamic BERT (Bidirectional Encoder Representation from Transformer) -fused model is proposed with GLEU score result of 61.0 on JFLEG system. By incorporating rule-based pre-processing into the model, the system is able to deal with more detailed grammatical errors. The performance was improved especially on the errors at beginner’s level. What is more, a web application prototype with the ability to automatically generate suggestions for grammatical error correction is also built to demonstrate the capability of the model.
Fulltext Permission: restricted
Fulltext Availability: With Fulltext
Appears in Collections:EEE Student Reports (FYP/IA/PA/PI)

Files in This Item:
File Description SizeFormat 
  Restricted Access
3.6 MBAdobe PDFView/Open

Page view(s)

Updated on Jul 4, 2022


Updated on Jul 4, 2022

Google ScholarTM


Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.