Human terminology and translation research uses source- and target-language parallel texts and background documents that exist in a variety of available machine-readable corpora. These corpora are, for the most part, under used and have not been effectively integrated into translation technology or into the internationalization strategies of most companies. Only some of the results of translation and terminology research, the translated sentence or terminological equivalent, have been stored and reused. Other textual and linguistic objects that could be derived from full texts, such as term contexts, changes made at the suprasentential level by translators or editors, as well as potentially reusable elements that are not translation memory segments, are not retained in translation databases. Although terminology management and translation memories have provided computer assistance for recording some of the results of translation research, there has been precious little assistance for more extensive translation research and resource discovery processes. This assistance is impossible to provide without exploiting the resources of document corpora and the analytic tools of corpus linguistics.

The language industry needs new approaches to internationalization that exploit available corpora to enable automation of some of the laborious human activity involved in supporting translation decisions and populating translation-oriented databases and memories. These approaches should address some of the shortcomings of translation memories and terminology managers by discovering and retaining linguistic, semantic, and textual objects of value to translation in addition to translated sentences and terminology equivalents. Integrating corpora more explicitly in internationalization strategy also means recognizing that human-populated terminology glossaries and translation memories are only the initial applications of translation technology in the language industry.

If relevant corpora could be discovered or constructed and then processed by computational tools, as for example by automatic term extractors, then the translator could be presented with many more terms and term equivalents - in context - than purely human research would allow. Many more sources of parallel and background texts could be identified and consulted, and many more candidates for term and translation equivalence considered for selection. There would be greater coverage of research materials during the translation process. The use of computational tools would remove restrictions on the range of possible research results exerted by the pressure of project deadlines. Corpus-based methods could substantially shorten the time it takes to populate or fill terminology and translation databases with translation equivalents. All available corpora would need to be leveraged, including the document corpora already owned by organizations, to improve the speed and quality of translation. Where appropriate corpora or appropriately structured corpora don't exist, mechanisms must be developed for creating them.