Definite Noun Phrases in Statistical Machine Translation into Scandinavian Languages

Definite Noun Phrases in Statistical Machine Translation into Scandinavian Languages Sara Stymne Link¨oping University, Link¨oping, Sweden Xerox Research Centre Europe, Grenoble, France sara.stymne@liu.se Abstract The Scandinavian languages have an un- usual structure of definite noun phrases (NPs), with a noun suffix as one possibility of expressing definiteness, which is prob- lematic for statistical machine translation from languages with different NP struc- tures. We show that translation can be im- proved by simple source side transforma- tions of definite NPs, for translation from English and Italian, into Danish, Swedish, and Norwegian, with small adjustments of the preprocessing strategy, depending on the language pair. We also explored target side transformations, with mixed results. 1 Introduction One problem for statistical machine translation is when the source language has a different structure in some respect than the target language. One such issue is the unusual realization of definite noun phrases in Scandinavian languages. Definiteness can be expressed in two ways in Scandinavian lan- guages, either by a definite article or by a suffix on the head noun. This is problematic for translation from languages that only use definite articles, such as English or Italian, leading to problems such as wrong noun forms and spurious definite articles in the translation output. It has previously been shown that definite noun phrases can successfully be handled by a prepro- cessing step for translation between English and Danish (Stymne, 2009b). In this study, source lan- guage noun phrases were transformed to be sim- ilar in structure to target language NPs. In this paper we show that preprocessing of definiteness c⃝ 2011 European Association for Machine Translation. also can be successful for translation from English into Swedish and Norwegian, and from Italian to Danish, using the same basic strategy as in Stymne (2009b). However, some small careful modifica- tions to the original English-Danish preprocessing strategy were necessary. 2 Definiteness In the Scandinavian languages there are two mech- anisms for expressing definiteness, by using a def- inite article or by using a suffix on the head noun. These mechanisms can also be used in combina- tion, so called double definiteness. The distri- bution rules for these two mechanisms are quite strict in all Scandinavian languages, but they dif- fer somewhat between them. The noun phrase re- alization is different in NPs with a pre-modifier such as an adjective or numeral and in NPs with- out pre-modifiers. Table 1 shows the allowed and disallowed combinations in Swedish, Norwegian Bokm˚al1 and Danish, which are the target lan- guages we focus on in this paper, and compares it to English and Italian, the source languages we focus on. There are similar phenomena in the other Scandinavian languages – Norwegian Nynorsk, Icelandic, and Faroese – as well. As can be seen in Table 1 there is a difference in pre-modified noun phrases, where the definite arti- cle is used, and in simple noun-phrases where the suffix is used. In Swedish and Norwegian there is double definiteness in pre-modified noun phrases, something that never occurs …

Content truncated to 500 words