NegPar: A parallel corpus annotated for negation

17473

Authors

nan

Abstract

Although the existence of English corpora annotated for negation has allowed for extensive work on monolingual negation detection, little is understood on how negation-related phenomena translate across languages. The current study fills this gap by presenting NegPar, the first English-Chinese parallel corpus annotated for negation in the narrative domain (a collection of stories from Conan Doyle’s Sherlock Holmes). While we followed the annotation guidelines in the CONANDOYLE-NEG corpus (Morante and Daelemans, 2012), we reannotated certain scope-related phenomena to ensure more consistent and interpretable semantic representation. To both ease the annotation process and analyze how similar negation is signaled in the two languages, we experimented with first projecting the annotations from English and then manually correcting the projection output in Chinese. Results show that projecting negation via word-alignment offers limited help to the annotation process, as negation can be rendered in different ways across languages. Keywords: negation, annotation projection, English, Chinese, parallel corpora, cross-lingual semantics 1.

Keywords

nan

Full Content

NegPar: a parallel corpus annotated for negation Qianchu Liu1 Federico Fancellu2 Bonnie Webber2 1 LTL,University of Cambridge 2 ILCC, School of Informatics, University of Edinburgh ql261@cam.ac.uk f.fancellu@sms.ed.ac.uk bonnie@inf.ed.ac.uk Abstract Although the existence of English corpora annotated for negation has allowed for extensive work on monolingual negation detection, little is understood on how negation-related phenomena translate across languages. The current study fills this gap by presenting NegPar, the first English-Chinese parallel corpus annotated for negation in the narrative domain (a collection of stories from Conan Doyle’s Sherlock Holmes). While we followed the annotation guidelines in the CONANDOYLE-NEG corpus (Morante and Daelemans, 2012), we reannotated certain scope-related phenomena to ensure more consistent and interpretable semantic representation. To both ease the annotation process and analyze how similar negation is signaled in the two languages, we experimented with first projecting the annotations from English and then manually correcting the projection output in Chinese. Results show that projecting negation via word-alignment offers limited help to the annotation process, as negation can be rendered in different ways across languages. Keywords: negation, annotation projection, English, Chinese, parallel corpora, cross-lingual semantics 1. Introduction Considering negation when extracting information from text is crucial for a range of natural language processing (NLP) tasks, especially when one is interested in whether an action or a state described in a statement is factual or counter-factual. To date, most work has focused on automatically identi- fying which tokens in a sentence constitute the three key elements of negation: cue, event and scope. As shown in (1), ‘not’ is marked as the presence of negation, the cue; ‘eat’ is a negated event; and [‘I’, ‘do’, ‘eat’, ‘pizza’] are part of the scope.1 If not otherwise stated, here and in the remainder of the paper, the cue is marked in bold with the scope underlined and the event marked in a box. (1) I do not eat pizza Corpora annotated for negation have been developed for both English and Chinese and for a variety of domains (§ 2.). However, these are monolingual corpora, and there is little empirical work on how negation is represented across languages. The current study aims at addressing this shortcoming by presenting the first parallel corpus annotated for negation. We created it by leveraging the CONANDOYLE-NEG cor- pus (Morante and Daelemans, 2012), a collection of four Sherlock Holmes stories. Before extending the English an- notations onto the sentence-aligned Chinese translation, we edited the English annotations to better capture phenomena related to negation scope. (§ 3.2.). To ease the annotation task and to investigate differences in the way negation is represented in the two languages, we experimented with first automatically projecting the anno- tations and then manually correcting them ( § 3.3. and § 4.). 1The negated event is usually considered as part of the scope of negation We experimented with projecting onto both Chinese char- acters and words via character or word alignment extracted from a large parallel corpus. In general, the performance of automatic projection yields relatively low results with …

Content truncated to 500 words

Labeled Text

Although the existence of English corpora annotated for negation has allowed for extensive work on monolingual negation detection, little is understood on how negation-related phenomena translate across languages. The current study fills this gap by presenting NegPar, the first English-Chinese parallel corpus annotated for negation in the narrative domain (a collection of stories from Conan Doyle’s Sherlock Holmes). While we followed the annotation guidelines in the CONANDOYLE-NEG corpus (Morante and Daelemans, 2012), we reannotated certain scope-related phenomena to ensure more consistent and interpretable semantic representation. To both ease the annotation process and analyze how similar negation is signaled in the two languages, we experimented with first projecting the annotations from English and then manually correcting the projection output in Chinese. Results show that projecting negation via word-alignment offers limited help to the annotation process, as negation can be rendered in different ways across languages. Keywords: negation, annotation projection, English, Chinese, parallel corpora, cross-lingual semantics 1.

Publication Details

Paper ID:
17473
Year:
2018
Publication:
TIPSTER
Publication Type:
conf
DOI:
nan