In the practical development of effective machine translation systems the direction of translation is fundamental. Import translation from foreign languages into the user's language should be contrasted with export translation from the user's language into foreign languages. This difference in direction determines what kind of MT system design is most effective.
In import translation the source text is normally in only one foreign language, but in export translation the same text often needs to be translated into several target languages. For example, a company which sells products throughout the European Community needs to translate the same product documentation into all the EC languages.
If the original documentation is in Japanese, and is to be translated into the EC languages, there is a major linguistic difference in the direction of translation. Translation into European languages from Japanese requires decisions about singular/plural and definite/indefinite for every noun phrase in every sentence, and decisions about pronouns to fill missing subjects and objects in nearly every sentence. These decisions require both text-wide linguistic analysis of the source text and factual knowledge about the real-world domain. However, most of the information required for these decisions is the same for all the EC languages.
Another difference is that import translation always starts with a source text, but export translation sometimes has no source text. For example, there are 2 approaches for a Japanese company needing product documentation for the EC. One approach is to write the documentation in Japanese first, then translate from the Japanese into the 9 EC languages. The other approach is to write the documentation in both Japanese and European languages at the same time. These alternatives are discussed further in Section 4.3.
Another difference is that in export translation it is usually not so difficult to get further information about the subject matter to be translated. The company which wants the export translation is usually the same company which makes the products. So the people who do the export translation can ask the people who designed the products for extra information and real-world knowledge.
Above all, there is a crucial difference in the requirement for post-editing. The quality of the raw output of MT has reached a level at which, in import translation, the post-editing stage is no longer always necessary for comprehension. The user, without needing a knowledge of the source language, can read the rough translation in his own language, and may find it sufficiently useful as it is.
For export translation, however, the target text is typically intended for customers, and the raw output of current systems falls short of the required standard. So post-editing by a target language expert is essential. This extreme dependence on the human post-editor means that current systems are restricted, for export translation, to use by professional translation agencies.