Rapid development of RBMT systems for related languages

The article describes a new way of constructing rule-based machine translation systems (RBMT). RBMT systems are currently among the best performing machine translation systems. Most of the "big named" machine translation systems (Systran, 2007)(Promt, 2007) belong to this category, but these systems have a big drawback; construction of such systems demands a great amount of time and resources, thus resulting very expensive. The article describes methods that automate parts of the construction process. The methods were evaluated on a case study: construction of a fully functional machine translation system of closely related language pair Slovene Serb. Slovene and Serbian language belong to the group of southern Slavic languages that were spoken mostly in former Yugoslavia. Slovenian language is mostly spoken in Slovenia, Serbian language is mostly spoken in Serbia and in Montenegro. The languages share common roots and even more importantly they share common recent historical environment, these languages were spoken in the same country, even taught in schools as languages of the surroundings. Economies of all three states are closely connected and younger generations, the post-yugoslavia breakage generations, have difficulties in mutual communication, so there is quite big interest in construction of such translation system. The system is based on Apertium (Armentano-Oller et al., 2007) (Corbí-Bellot et al., 2005), an opensource RBMT toolkit. Apertium uses a shallow-transfer machine translation engine which processes the input text in stages, as in an assembly line: de-formatting, morphological analysis, part-of-speech disambiguation, shallow structural transfer, lexical transfer, morphological generation, and re-formatting. The data needed by the presented stages can be grouped into three categories: monolingual dictionaries used by morphological analysis and morphological generation, bilingual dictionaries used in lexical transfer and structural transfer rules used in structural transfer. Each group's data creation was addressed by a particular method; monolingual dictionaries were constructed using bilingual dictionary data and applying automatic paradigm tagging techniques; bilingual dictionary was constructed using available bilingual word-list but a few methods for automatic bilingual dictionary construction were investigated; a method for automatic structural shallow-transfer rule construction (Sánchez-Martínez et al., 2006) was used to construct a set of structural transfer rules.