Automatic Detection of Comma Splices

19126

Authors

nan

Abstract

In English text, independent clauses should be demarcated with full-stops (periods), or linked together with conjunctions. Non-native speak- ers are often prone to linking them improp- erly with commas instead of conjunctions, producing comma splices. This paper de- scribes a method to detect comma splices us- ing Conditional Random Fields (CRF), with features derived from parse tree patterns. In experiments, our model achieved an average of 0.91 precision and 0.28 recall in detect- ing comma splices, significantly outperform- ing both a baseline model using only local fea- tures and a widely used commercial grammar checker. 1

Keywords

nan

Full Content

PACLIC 28 !551 Automatic Detection of Comma Splices John Lee, Chak Yan Yeung Halliday Centre for Intelligent Applications of Language Studies Department of Linguistics and Translation City University of Hong Kong jsylee@cityu.edu.hk chak.yeung@my.cityu.edu.hk Martin Chodorow Department of Psychology Hunter College City University of New York martin.chodorow @hunter.cuny.edu Abstract In English text, independent clauses should be demarcated with full-stops (periods), or linked together with conjunctions. Non-native speak- ers are often prone to linking them improp- erly with commas instead of conjunctions, producing comma splices. This paper de- scribes a method to detect comma splices us- ing Conditional Random Fields (CRF), with features derived from parse tree patterns. In experiments, our model achieved an average of 0.91 precision and 0.28 recall in detect- ing comma splices, significantly outperform- ing both a baseline model using only local fea- tures and a widely used commercial grammar checker. 1 Introduction English text consists of a sequence of clauses linked and separated by punctuation and conjunctions. To separate two independent clauses, one uses a full- stop (period); to link together two related clauses, one typically uses a semicolon or a comma with an appropriate conjunction, which can be either co- ordinate (“and”, “but”, “or”) or subordinate (“be- cause”, “so”). For example, to link the two related clauses “it was raining” and “we stayed home”, one may use a comma and the conjunction “so”, yielding the complex sentence “It was raining, so we stayed home”. When a comma is used instead of a full- stop, or when it is used without a conjunction (e.g., “It was raining, we stayed home”), the result is a comma splice1. 1Note that a list of noun phrases with a missing conjunction (e.g., “I like apples, oranges.”) is not a comma splice. Our use of the term comma splice also includes improper linking of verb phrases. This occurs when a comma is used without a conjunction (e.g., “We stayed home, watched TV.”); without a relative pro- noun (e.g., “The boy chased after the rat, fled into the sewer”); or with the wrong verb form (e.g., “Wa- terborne pathogens are the pathogenic microorgan- isms, includes bacteria”). Comma splices are not only considered poor writing style, but they also compromise the readability of a text. Although native speakers have been found to commit a substantial number of common splice er- rors (Connors and Lunsford, 1988; Lunsford and Lunsford, 2008), non-native speakers appear to be especially prone to producing them, possibly due to interference from syntactic differences in L1 (Tseng and Liou, 2006; Bennui, 2008; Rahimi, 2009). This may be especially true for L1s where comma splices are frequently found and are not considered mis- takes, such as in Chinese (Lin, 2002). Comma splices are one of the errors addressed in the 2014 CoNLL Shared Task on Grammatical Error Cor- rection (Ng et al., 2014). They are annotated in many learner corpora, including the NUS Corpus of Learner English (Dahlmeier et al., 2013) and the EF- Cambridge Open Language Database (Geertzen et al., 2013). This …

Content truncated to 500 words

Labeled Text

In English text, independent clauses should be demarcated with full-stops (periods), or linked together with conjunctions. Non-native speakers are often prone to linking them improperly with commas instead of conjunctions, producing comma splices. This paper describes a method to detect comma splices using Conditional Random Fields (CRF), with features derived from parse tree patterns. In experiments, our model achieved an average of 0.91 precision and 0.28 recall in detecting comma splices, significantly outperforming both a baseline model using only local features and a widely used commercial grammar checker. 1

Publication Details

Paper ID:
19126
Year:
2014
Publication:
TIPSTER
Publication Type:
conf
DOI:
nan