Learning Grammar Patterns for Assisting Learners in Writing

• Main Authors: Chang, Jin, 張竟
• Other Authors: Chen, Hwann-Tzong
• Format: Others
• Language: en_US
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/55533848007974997024

碩士 === 國立清華大學 === 資訊工程學系 === 103 === Grammatical errors related to articles, prepositions, noun number, verb form, and complementation is the most frequent and problematic writing errors produced by learners of English as a Second Language (ESL). We introduce a method for extracting grammar patterns from a corpus for the purpose of assisting ESL learners in writing fluently and avoiding errors. In our approach, phrase chunks are transformed into grammar patterns for statistical analysis and filtering. The method involves extracting grammar patterns using templates, and selecting characteristic patterns for each content words. We present an interactive writing environment, WriteAhead, that automatically displays extracted pattern and examples to prompt the users as they write away, or mouse around a draft during self-editing. We also extend the method to apply to error-annotated corpus and for extracting synchronous grammar patterns (SGP). These extracted SGPs can be used to provide more informative feedback to the user with a given sentence with potential errors for the purpose of assisted editing. In our approach, grammar patterns are extracted from training data to generate feedback with corrections and explanation. The method involves identifying patterns in pairs of erroneous and edited sentences covering an edit, and validate patterns extracted from the edited sentences, and counting the frequency of distinct SGPs. We present a new system, WriteAhead, in writing mode, proactively provides writing suggestions using extracted grammar patterns and examples to assist student writers, as they write away. Additionally, in editing mode, WriteAhead displays common edit patterns as corrective feedback for potential errors near the word under the cursor. Preliminary experiments and evaluation show that WriteAhead with the acquired patterns and examples give reasonably effective writing suggestion and corrective feedback, and has potentials of promoting fluent and correct writing, based on evaluation on publicly available datasets.