Europarl Datasets with Demographic Speaker Information

Rule-based machine translation from Kazakh to Turkish Sevilay Bayatli Dept. Elec. and Computer Engineering Altınbaş Üniversitesi sevilaybayatli@gmail.com Sefer Kurnaz Dept. Elec. and Computer Engineering Altınbaş Üniversitesi sefer.kurnaz@altinbas.edu.tr Ilnar Salimzianov School of Science and Technology Nazarbayev University ilnar@selimcan.org Jonathan North Washington Linguistics Deptartment Swarthmore College jonathan.washington@swarthmore.edu Francis M. Tyers School of Linguistics Higher School of Economics ftyers@hse.ru Abstract This paper presents a shallow-transfer ma- chine translation (MT) system for translat- ing from Kazakh to Turkish. Background on the differences between the languages is presented, followed by how the system was designed to handle some of these differ- ences. The system is based on the Apertium free/open-source machine translation plat- form. The structure of the system and how it works is described, along with an evaluation against two competing systems. Linguis- tic components were developed, including a Kazakh-Turkish bilingual dictionary, Con- straint Grammar disambiguation rules, lex- ical selection rules, and structural transfer rules. With many known issues yet to be ad- dressed, our RBMT system has reached per- formance comparable to publicly-available corpus-based MT systems between the lan- guages. 1 Introduction In this paper we present a prototype shallow- transfer rule-based machine translation system using the Apertium free/open-source machine translation platform (Forcada et al., 2011) for translating from Kazakh to Turkish. One of the most common criticisms towards Rule-Based Machine Translation (RBMT) regards the amount of work necessary to build a system for a new language pair (Arnold, 2003). In fact, in a traditional scenario, linguists with expertise in the source and target language need to manually build © 2018 The authors. This article is licensed under a Creative Commons 3.0 licence, no derivative works, attribution, CC- BY-ND. all the dictionary entries and transfer rules. Con- versely, in a corpus-based/statistical MT approach (Koehn, 2010), no such effort is required as the sys- tem can be automatically built from parallel corpora providing they exist. If parallel corpora do not ex- ist, then we see two options that remain. The first is to create a new parallel corpus, either by translat- ing millions of words from scratch (requiring effort from translators),¹ or by finding parallel text online and processing it (requiring effort from program- mers). The second option is to build a rule-based machine translation system (requiring effort from linguists). The most labour-intensive of these ap- proaches is to translate the data from scratch, al- though this might be practical in certain situations. Building a rule-based machine translation system and finding and processing parallel texts from the internet are, given equal available expertise, around equally time consuming. As large, freely available parallel corpora are not known to exist between Kazakh and Turkish and we were interested in struc- tural differences between these two languages — and producing new linguistic resources, we chose to build a rule-based MT system. Kazakh and Turk- ish are different enough that native speakers are not able to make sense of the other language, but also share similar enough structure that an RBMT system is feasible with some level of …

Content truncated to 500 words