Extrinsic evaluation of web-crawlers in machine translation: a study on Croatian-English for the tourism domain

Extrinsic Evaluation of Web-Crawlers in Machine Translation: a Case Study on Croatian–English for the Tourism Domain∗ Antonio Toral†, Raphael Rubino⋆, Miquel Espl`a-Gomis‡, Tommi Pirinen†, Andy Way†, Gema Ram´ırez-S´anchez⋆ † NCLT, School of Computing, Dublin City University, Ireland {atoral,tpirinen,away}@computing.dcu.ie ⋆ Prompsit Language Engineering, S.L., Elche, Spain {rrubino,gramirez}@prompsit.com ‡ Dep. Llenguatges i Sistemes Inform`atics, Universitat d’Alacant, Spain mespla@dlsi.ua.es Abstract We present an extrinsic evaluation of crawlers of parallel corpora from multi- lingual web sites in machine translation (MT). Our case study is on Croatian to English translation in the tourism domain. Given two crawlers, we build phrase-based statistical MT systems on the datasets pro- duced by each crawler using different set- tings. We also combine the best datasets produced by each crawler (union and in- tersection) to build additional MT systems. Finally we combine the best of the previ- ous systems (union) with general-domain data. This last system outperforms all the previous systems built on crawled data as well as two baselines (a system built on general-domain data and a well known on- line MT system). 1 Introduction Along with the addition of new member states to the European Union (EU), the commitment with multilingualism in the EU is strengthened to give support to new languages. This is the case of Croa- tia, the last member to join the EU in July 2013, and of the Croatian language, which became then an official language of the EU. Croatian is the third official South Slavic lan- guage in the EU along with Bulgarian and Slovene. Other surrounding languages (e.g. Serbian and ∗The research leading to these results has received fund- ing from the European Union Seventh Framework Pro- gramme FP7/2007-2013 under grant agreement PIAP-GA- 2012-324414 (Abu-MaTran). ∗c⃝ 2014 The authors. This article is licensed under a Creative Commons 3.0 licence, no derivative works, attribution, CC- BY-ND. Bosnian), although still not official in the EU, be- long also to the same language family and are the official languages of candidate member states, thus being also of strategic interest for the EU. We focus on providing machine translation (MT) support for Croatian and other South Slavic languages using and producing publicly available resources. Following our objectives, we developed a general-domain MT system for Croatian–English and made it available online on the day Croatia joined the EU. It is, to the best of our knowledge, the first available MT system for this language pair based on free/open-source technologies. New languages in the EU like Croatian can ben- efit from MT to speed up the flow of information from and into other EU languages. While this is the case for most types of content it is especially true for official documentation and for content in particular strategic sectors. Tourism is one of the most important economic sectors in Croatia. It represented 15.4% of Croa- tia’s gross domestic product in 2012 (up from 14.4% in 2011).1 With almost 12 million foreign tourists visiting Croatia annually, the tourism sec- tor results in income of 6.8 billion euro. The increasing …

Content truncated to 500 words