Plagiarism Detection across Distant Language Pairs

Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010), pages 37–45, Beijing, August 2010 Plagiarism Detection across Distant Language Pairs Alberto Barr´on-Cede˜no Paolo Rosso Natural Language Engineering Lab. - ELiRF Universidad Polit´ecnica de Valencia {lbarron, prosso}@dsic.upv.es Eneko Agirre Gorka Labaka IXA NLP Group Basque Country University {e.agirre, gorka.labaka}@ehu.es Abstract Plagiarism, the unacknowledged reuse of text, does not end at language boundaries. Cross-language plagiarism occurs if a text is translated from a fragment written in a different language and no proper citation is provided. Regardless of the change of language, the contents and, in particular, the ideas remain the same. Whereas dif- ferent methods for the detection of mono- lingual plagiarism have been developed, less attention has been paid to the cross- language case. In this paper we compare two recently proposed cross-language plagiarism de- tection methods (CL-CNG, based on char- acter n-grams and CL-ASA, based on sta- tistical translation), to a novel approach to this problem, based on machine trans- lation and monolingual similarity analy- sis (T+MA). We explore the effectiveness of the three approaches for less related languages. CL-CNG shows not be ap- propriate for this kind of language pairs, whereas T+MA performs better than the previously proposed models. 1 Introduction Plagiarism is a problem in many scientific and cul- tural fields. Text plagiarism may imply differ- ent operations: from a simple cut-and-paste, to the insertion, deletion and substitution of words, up to an entire process of paraphrasing. Differ- ent models approach the detection of monolin- gual plagiarism (Shivakumar and Garc´ıa-Molina, 1995; Hoad and Zobel, 2003; Maurer et al., 2006). Each of these models is appropriate only in those cases where all the implied documents are written in the same language. Nevertheless, the problem does not end at lan- guage boundaries. Plagiarism is also committed if the reused text is translated from a fragment writ- ten in a different language and no citation is pro- vided. When plagiarism is generated by a transla- tion process, it is known as cross-language plagia- rism (CLP). Less attention has been paid to the detection of this kind of plagiarism due to its enhanced diffi- culty (Ceska et al., 2008; Barr´on-Cede˜no et al., 2008; Potthast et al., 2010). In fact, in the recently held 1st International Competition on Plagiarism Detection (Potthast et al., 2009), no participants tried to approach it. In order to describe the prototypical process of automatic plagiarism detection, we establish the following notation. Let dq be a plagiarism suspect document. Let D be a representative collection of reference documents. D presumably includes the source of the potentially plagiarised fragments in dq. Stein et al., (2007) divide the process into three stages1: 1. heuristic retrieval of potential source doc- uments: given dq, retrieving an appropri- ate number of its potential source documents D∗ ∈ D such that |D∗| ≪ |D|; 2. exhaustive comparison of texts: comparing the text from dq and d ∈ D∗ in order to identify reused fragments and their potential 1This schema was formerly proposed for …

Content truncated to 500 words