Chinese Grammatical Error Diagnosis Using Ensemble Learning

Abstract

Automatic grammatical error detection for Chinese has been a big challenge for NLP researchers for a long time, mostly due to the flexible and irregular ways in the expressing of this language. Strictly speaking, there is no evidence of a series of formal and strict grammar rules for Chinese, especially for the spoken Chinese, making it hard for foreigners to master this language. The CFL shared task provides a platform for the researchers to develop automatic engines to detect grammatical errors based on a number of manually annotated Chinese spoken sentences. This paper introduces HITSZ's system for this year's Chinese grammatical error diagnosis (CGED) task. Similar to the last year's task, we put our emphasis mostly on the error detection level and error type identification level but did little for the position level. For all our models, we simply use supervised machine learning methods constrained to the given training corpus, with neither any heuristic rules nor any other referenced materials (except for the last years' data). Among the three runs of results we submitted, the one using the ensemble classifier Random Feature Subspace (HITSZ_Run1) gained the best performance, with an optimal F1 of 0.6648 for the detection level and 0.2675 for the identification level.

Arabic Dialect Identification Using iVectors and ASR Transcripts

“…In light of this, we decided to test two ensemble methods. Classifier ensembles have also proven to be an efficient and robust alternative in other text classification tasks such as language identification (Malmasi and Dras, 2015a), grammatical error detection (Xiang et al, 2015), and complex word identification (Malmasi et al, 2016a We follow the methodology described by Malmasi and Dras (2015a): we extract a number of different feature types and train a single linear model using each feature type. Our ensemble was created using linear Support Vector Machine classifiers.…”

Proceedings of the 12th Workshop on Innovative Use of NLP for Building Educational Applications

“…Such systems have proved successful not only in NLI and dialect identification, as evidenced in the previous sections, but also in numerous text classification tasks, among which are complex word identification (Malmasi et al, 2016a) and grammatical error diagnosis (Xiang et al, 2015). The classifiers can differ in a wide range of aspects, such as algorithms, training data, features or parameters.…”

“…In most of the previous studies (Mihalcea and Strapparava, 2005;Purandare and Litman, 2006;Yang et al, 2015), humor recognition was modeled as a binary classification task.…”

“…Three humor-specific stylistic features, including alliteration, antonymy, and adult slang were utilized together with content-based features to build classifiers. In a recent work (Yang et al, 2015), a new corpus was constructed from the Pun of the Day website. Yang et al (2015) explained and computed stylistic features based on the following four aspects: (a) Incongruity, (b) Ambiguity, (c) Interpersonal Effect, and (d) Phonetic Style.…”

MAZA at SemEval-2016 Task 11: Detecting Lexical Complexity Using a Decision Stump Meta-Classifier

“…Classifiers ensembles have proved to an efficient and robust alternative in other text classification tasks such as language identification (Malmasi and Dras, 2015a) and grammatical error detection (Xiang et al, 2015). This motivated us to try this approach in the CWI SemEval task.…”