Lexicon-coding workflow at CLS Communication

EAMT 2005 Conference Proceedings 255 Lexicon-Coding Workflow at CLS Communication Hans-Udo Stadler CLS Communication AG Elisabethenanlage 11, CH-4051 Basel hans-udo.stadler@cls-communication.com Abstract. Offering Machine Translation (MT) as a commercial service to end users means having to build up and regularly update large lexicons that can handle both general lan- guage and client specific terminology. This paper describes the processes developed at CLS Communication to cover the demand of users from several fields. Over a period of four years, processes have been set up and constantly refined in order to tackle the issue of im- provement of MT output incorporating the resources at disposal. Different approaches have been tested, that have led to the implementation of three main coding streams. The coding of „unknown words”, translation analysis and exports from terminology databases enable the MT team at CLS to react to changing external and internal needs. Accuracy and expenditure of time have been evaluated, as have the applications of automation and human skills. 1. Introduction The initial idea for CLS Communication’s MT service came from the clients themselves, that is from a Swiss Bank whose employees needed a solution for fast and inexpensive translations. They were looking for a way to translate mainly internal communications to and from several languages within a secure environment. CLS Communication had already provided them with human translations and other language services and thus had the competency and synergy po- tential to offer them the requested service. 2. Implementation Building an MT system from scratch would have meant years of development apart from having to recruit the necessary experts from fields which did not belong to the company’s core competencies. CLS Communication there- fore decided to acquire a license to use and adapt an existing MT system. In 2000 and 2001 several commercial systems were evaluated with regard to translation quality and scalabil- ity, among other aspects (Maier and Clarke, 2001). Finally, the transfer system by Com- prendium Lingua (formerly SailLabs) was cho- sen and installed in a server as well as a desktop version. One of the main advantages of this par- ticular software was the lexicon development component called LexShop, which allowed CLS to not only add new complex terminology but also edit all entries, including the very sophisti- cated linguistic information needed for the trans- lation software’s rules. After all, specialized lexi- cons resulting in quality improvement are the main added value or, from a clients’ perspec- tive, the main reason to choose CLS Machine Translation. 3. Target Groups of CLS MT Most of CLS’ clients are companies from the finance, telecommunications and pharmaceutical sectors. MT is offered to them online so they can perform the translations themselves, or they can opt for our post-editing service, where the MT output is revised by a human translator who removes the most serious errors and selects the correct alternatives. Those translators can be regarded as internal clients for the MT team, since they proved to have their own wishes and needs with respect to our MT …

Content truncated to 500 words