Rapid development of translation tools

MT Summit VII Sept.1999 Rapid Development of Translation Tools Jan W. Amtrup Karine Megerdoomian Computing Research Laboratory Computing Research Laboratory New Mexico State University New Mexico State University email: jamtrup@crl.nmsu.edu email: karine@crl.nmsu.edu Rémi Zajac Computing Research Laboratory New Mexico State University email: zajac@crl.nmsu.edu Abstract The Computing Research Laboratory is cur- rently developing technologies that allow rapid deployment of automatic translation capabilities. These technologies are de- signed to handle low-density languages for which resources, be that human informants or data in electronically readable form, are scarce. All tools are built in an incremen- tal fashion, such that some simple tools (a bilingual dictionary or a glosser) can be de- livered early in the development to support initial analysis tasks. More complex appli- cations can be fielded in successive func- tional versions. The technology we demon- strate has first been applied to Persian- English machine translation within the Shi- raz project and is currently extended to cover languages such as Arabic, Japanese, Korean and others. 1 Introduction One major goal that is pursued at the Computing Re- search Laboratory (CRL) is to create a machine trans- laiion environment that allows the rapid development and deployment of translation systems for low-density languages. The main problem lies with the usually low amount of machine-readable data for the development of knowledge sources (machine readable dictionaries, large corpora) and the difficulty to acquire such knowl- edge manually. In order to support the creation of ma- chine translation systems for such languages, we de- veloped techniques that allow the incremental installa- tion of successively more complex translation systems and tools. The framework which is used at CRL con- sists of the following tools, presented here in increasing order of processing complexity (and, consequently, in increasing order of acquisition complexity): • A bilingual dictionary lookup tool for human translators. • A dictionary editor which allows modifying ex- isting entries in a dictionary as well as adding new words to a dictionary. • A preprocessor which combines spell-checking of a source text and augmentation of the underly- ing dictionary with previously unknown words. • A glosser that performs dictionary lookup for a complete text and allows viewing of translations in context. • A translation system that produces automatic translations of source texts on various levels of quality. In this paper, we will demonstrate several of these components as applied to Persian-English machine trans- lation in the Shiraz project1. Knowledge sources and tools for other languages are under development, including Arabic, Japanese, Korean, Russian, Serbo- Croatian and Spanish. 2 The Shiraz System The aim of the Shiraz project is to develop a machine translation system that translates Persian text into English. We target news material and various mis- cellaneous texts. The source texts are usually taken from on-line sources (e.g., web pages), although plain text can be handled as well. The dictionary, the pri- mary knowledge source in the system, contains ap- proximately 50,000 entries. It has been built by Per- 1 http://crl.nmsu.edu/Research/Projects/shiraz - 385- MT Summit VII _______________________________________________________________ Sept. …

Content truncated to 500 words