Open Subtitles Paraphrase Corpus for Six Languages

Open Subtitles Paraphrase Corpus for Six Languages Mathias Creutz Department of Digital Humanities, Faculty of Arts, University of Helsinki, Finland mathias.creutz@helsinki.fi Abstract This paper accompanies the release of Opusparcus, a new paraphrase corpus for six European languages: German, English, Finnish, French, Russian, and Swedish. The corpus consists of paraphrases, that is, pairs of sentences in the same language that mean approximately the same thing. The paraphrases are extracted from the OpenSubtitles2016 corpus, which contains subtitles from movies and TV shows. The informal and colloquial genre that occurs in subtitles makes such data a very interesting language resource, for instance, from the perspective of computer assisted language learning. For each target language, the Opusparcus data have been partitioned into three types of data sets: training, development and test sets. The training sets are large, consisting of millions of sentence pairs, and have been compiled automatically, with the help of probabilistic ranking functions. The development and test sets consist of sentence pairs that have been checked manually; each set contains approximately 1000 sentence pairs that have been verified to be acceptable paraphrases by two annotators. Keywords: paraphrase, subtitle, colloquial language, annotation, ranking, German, English, Finnish, French, Russian, Swedish 1. Introduction This paper introduces the first release of the Opusparcus multilingual corpus of paraphrases (Creutz, 2018). Para- phrases are pairs of phrases in the same language that es- sentially convey the same meaning, such as “Have a seat.” versus “Sit down.”. Paraphrase resources have been pub- lished earlier, for instance by Quirk et al. (2004), Dolan et al. (2004), Dolan and Brockett (2005), Ganitkevitch et al. (2013), Ganitkevitch and Callison-Burch (2014), and Pavlick et al. (2015). However, Opusparcus has a few dis- tinctive characteristics. Firstly, and most importantly, all paraphrases in Opus- parcus (OpenSubtitlesParaphraseCorpus) consist of movie and TV subtitles extracted from the OpenSubtitles2016 col- lection of parallel corpora (Lison and Tiedemann, 2016). Previous paraphrase collections mostly contain fairly for- mal language in the form of news text and transcripts of parliamentary proceedings. The more colloquial language used in subtitles can be a valuable addition, for instance, in computer assisted language learning, to help learners find natural and idiomatic expressions in real-life situations. Secondly, in this work the pivot language technique intro- duced by Bannard and Callison-Burch (2005) is applied us- ing multiple pivot languages rather than just one or a few. The technique consists in finding paraphrases in one tar- get language by translating to another, so-called pivot lan- guage and then translating back. For example, English “Have a seat.” can be translated to French “Asseyez- vous.”, which can be translated back to “Sit down.”. Now, a well known fact is that different languages make dif- ferent distinctions; for instance, the English pronoun you corresponds to French toi or vous, depending on number and degree of politeness. If French paraphrases are ex- tracted using English as a pivot, then the toi/vous distinc- tion will typically disappear, such that “Asseyez-vous.” and “Assieds-toi.” emerge as paraphrases, because they can both be translated as “Sit …

Content truncated to 500 words