Атты І халықаралық конференция ЕҢбектері
жүктеу/скачать 8,57 Mb. Pdf көрінісі
|
Болатбек М. (1)
|
A. SUNDETOVA 1 , M.L.FORCADA 2 , A. SHORMAKOVA 1 , A. AITKULOVA 1 . 1 Information Systems Chair, Al-Farabi Kazakh National University, Al-Farabi av., 71, 050040 Almaty, Kazakhstan, and 2 Departament de Llenguatges i Sistemes Informàtics, Universitat d'Alacant, E-03071 Alacant, Spain STRUCTURAL TRANSFER RULES FOR ENGLISH-TO-KAZAKH MACHINE TRANSLATION IN THE FREE/OPEN-SOURCE PLATFORM APERTIUM Introduction Translating natural text from English to a Turkic language such as Kazakh faces important challenges: On the one hand, the complex agglutinative morphology of Turkic languages is very different from that of a fusional, morphologically not too complex language like English; an immediate effect is the fact that correspondences can seldom be modelled as word-for-word translations. Even if Turkic language morphology shows clear morphotactics (ordering of morphemes), its morphophonology shows complex phonological changes to due to interactions between neighboring morphemes (vowel harmony, sonorization, etc.) many of which are explicitly represented in writing. On the other hand, there are many differences between the syntax of Turkic languages and English. Just to name a few: subject–object–verb order (compare subject–verb–object in English), use of postpositions (compare prepositions in English), head-final syntax with modifiers and specifiers always preceding the modified/specified (normally following in English), overt case marking allowing for a rather free ordering of arguments (versus a more fixed order in English), lack of definite articles (extensively used in English), verbal-noun-centered structures where English uses modal verbs (must, have to, want to) or verbal-noun or verbal-adjective-centered constructions where English has subordinate clauses using finite verbs with relatives or subordinating conjunctions (the book which I read, the place where I saw him, before he came), lack of a parallel of the English verb have, as used for possession, etc. For an account (in Russian) of syntax differences between English and Kazakh, see Печерских & Амангельдина (2012). When sufficiently large sentence-aligned parallel corpora are available (for instance, as in the case of English to Turkish, see, for example, Tyers and Alperen 2010), statistical machine translation (Koehn 2010) may be used to attempt translation from English into a Turkic language 323 (in fact, statistical machine translation is currently offered by Google for two Turkic languages, Azeri and Turkish). However, in the case of Kazakh, it would be very hard to put together the necessary amount of sentence-aligned parallel text, and rule-based machine translation, in which experts write up dictionaries and grammatical rules that are applied by an engine, emerges as a clear solution; in fact, existing commercial systems for English to Kazakh (Sanasoft 7 , Trident 8 ) all appear to be rule-based. We are currently engaged in building a free/open-source rule-based machine translation system from English to Kazakh, and we are using the Apertium free/open-source machine translation platform (Forcada et al. 2011, http://www.apertium.org) for various reasons. On the one hand, the platform already contains free/open-source English morphological dictionaries and, what is more important, Kazakh morphological dictionaries (Salimzyanov et al. 2013) which take care of all of the morphotactics and morphophonology and provide a basic vocabulary; this allows us to concentrate our work in two fronts: building the lexical transfer part, that is, a bilingual dictionary (already underway) and building structural transfer rules (grammatical rules for translation), which will be the subject of this paper. On the other hand, building free/open-source dictionaries and rules for English to Kazakh means that they will be freely available, 9 for instance, to build translation systems for other Turkic languages; this gives a strategic value to our work, as most of the structural transfer rules will be ready for use with other Turkic languages with little modification or no modification at all. 10 The paper, which describes work in progress in the Apertium English-to-Kazakh structural transfer, is organized as follows: Section 323 describes the free/open-source rule-based machine translation platform, focusing on structural transfer. Section 0 describes the structural transfer rules currently available to tackle the main syntactic divergences between English and Kazakh; section 0 describes some successful structural translations and some limitations, and, finally, section 0 gives concluding remarks and outlines future work. жүктеу/скачать 8,57 Mb. Достарыңызбен бөлісу:
|