A Hybrid System for Patent Translation

A Hybrid System for Patent Translation Ramona Enache Cristina Espa˜na-Bonet Aarne Ranta Llu´ıs M`arquez Dept. of Computer Science and Engineering TALP Research Center Chalmers University of Technology LSI Department University of Gothenburg Universitat Polit`ecnica de Catalunya {enache,aarne}@chalmers.se {cristinae,lluism}@lsi.upc.edu Abstract This work presents a HMT system for patent translation. The system exploits the high coverage of SMT and the high preci- sion of an RBMT system based on GF to deal with specific issues of the language. The translator is specifically developed to translate patents and it is evaluated in the English-French language pair. Although the number of issues tackled by the gram- mar are not extremely numerous yet, both manual and automatic evaluations consis- tently show their preference for the hybrid system in front of the two individual trans- lators. 1 Introduction The predominant core of machine translation (MT) systems has been changing through the years. From the very beginnings in the 50s where only dictionary-based MT systems existed, the technol- ogy evolved towards rule-based systems (RBMT). Later in the 90s the everyday more powerful com- puters allowed to develop empirical translation systems. Recently a type of empirical system, the statistical one (SMT), has become a widely used standard for translation. At this point the two main paradigms, RBMT and SMT, coexist with their strengths and weaknesses. Luckily these strengths and weaknesses are complementary and current ef- forts are being made to hybridise both of them and develop new technologies. A classification and description of hybrid translation can be found in (Thurmair, 2009). In general RBMT provides high precision, due to an analysis of the text, but has limited coverage c⃝ 2012 European Association for Machine Translation. and a considerable amount of effort and linguistic knowledge is required in order to build such a sys- tem. On the other hand, SMT can achieve a huge coverage and is good at lexical selection and flu- ency but has problems in building structurally and grammatically correct translations. Hybrid MT (HMT) is an emerging and chal- lenging area of machine translation, which aims at combining the known techniques into systems that retain the best features of their components, and reduce the disadvantages displayed by each of the methods when used individually. This work presents a hybrid translation system specifically designed to deal with the translation of patents. The language of patents follows a formal style adequate to be analysed with a grammar, but at the same time uses a rich and particular vocabu- lary adequate to be gathered statistically. We focus on the English-French language pair so that the ef- fects of translating into a morphologically rich lan- guage can be studied. With respect to the engine, a grammar-based translator is developed to assure grammatically correct translations. We extend GF (Grammati- cal Framework, Ranta (2011)) and write a new grammar for patent translation. The SMT sys- tem that complements the RBMT is based on Moses (Koehn et al., 2007). This system works on two different levels. First, it is used to build the …

Content truncated to 500 words