Rapid rule-based machine translation between Dutch and Afrikaans

Rapid rule-based machine translation between Dutch and Afrikaans Pim Otte Mendelcollege Pim Mulierlaan 4 2024 BT Haarlem 5666@mendelcollege.nl Francis M. Tyers Dept. Lleng. i Sist. Inform. Universitat d’Alacant E-03070 Alacant ftyers@dlsi.ua.es Abstract This paper describes the design, develop- ment and evaluation of a machine transla- tion system between Dutch and Afrikaans developed over a period of around a month and a half. The system relies heavily on the re-use of existing publically available resources such as Wiktionary, Wikipedia and the Apertium machine translation plat- form. A method of translating compound words between the languages by means of left-to-right longest match lookup is also introduced and evaluated. 1 Introduction Dutch is a West-Germanic language spoken by nearly 23 million people, mostly from the Nether- lands and Flanders, the Dutch-speaking part of Belgium, and a minority living in former colonies of the Netherlands, such as Suriname, Aruba and the Netherlands Antilles. Dutch, as it is today, started developing in the 16th century in the ma- jor trade cities, such as Amsterdam and Antwerp (Shetter and Ham, 2002). Afrikaans is spoken by at least 5 million people, mainly in South Africa but also in Namibia. Afrikaans is a vari- ety of Dutch that originates from that spoken by the Dutch colonists of the Cape Colony. In 1925 Afrikaans replaced Dutch as an official language in South Africa, to be the joint official language to- gether with English (Donaldson, 1993). Currently, Afrikaans is one of the eleven national languages. In this paper we will describe the development of apertium-af-nl, a bi-directional Afrikaans and Dutch machine-translation system based on the Apertium platform. As Afrikaans and Dutch c⃝ 2011 European Association for Machine Translation. are largely mutually intelligible, this machine translation system focuses on dissemination, the translation of text for the purpose of being post- edited and then being published. This is not the first system to work with this lan- guage pair, van Huyssteen and Pilon (2009) de- scribe a rule-based system to convert in a single direction from Dutch to Afrikaans. The reason we have chosen to work with a rule-based approach, instead of the ubiquitous corpus-based/statistical approach, is that the latter needs parallel corpora for the two languages. The only freely avail- able Afrikaans–Dutch corpus, is KDE41, which is translated via English and domain specific. We feel that these corpora do not approach the quality required for the statistical approach, which makes the rule-based approach favourable. The paper is laid out as follows: firstly, we will describe the reuse and creations of resources. We will then discuss several grammatical features of Afrikaans and Dutch and how these were treated in the machine-translation system. We will then present a section in which the system is evaluated. Finally, we will discuss the system and future work that could be done. 2 Method The system is based on the Apertium machine translation platform.2 The platform was originally aimed at the Romance languages of the Iberian peninsula, but has also been adapted for other, more distantly related, …

Content truncated to 500 words