Although improvements in post-editing facilities for professional translators are very desirable, the really significant challenge is to find ways to enable ordinary people to achieve successful personal export translation without professional post-editing. Some approaches to this challenge were described in Sections 4.2.1 and 4.2.2.
The approach taken in the demonstration prototype directly addresses this challenge. The linguistic knowledge of the target language is the responsibility of the system. The knowledge of the subject matter to be communicated is the responsibility of the user. The process of generating the target text is shared between the system and the user, as the system initiates inquiries when further information is necessary for linguistic decisions, and the user initiates revisions when the generated text is unsatisfactory. The user can stop worrying about grammatical correctness, which is guaranteed by the control imposed by the grammar, and can concentrate on the correctness of the choices required.
In terms of the three problems listed in the Introduction, following the framework of [Tsujii 1986], it is the third problem which is crucial. The knowledge of the target language functional systems (and their internal relationships) is built into the systemic grammar network. The knowledge of target language structures (and their relationships to the functional systems) is built into the realization rules. The hard part is how to help the monolingual user to understand and make those functional distinctions which are not made in the user's own language.
The prototype shows the basic approach of displaying felicity conditions, in Japanese, for the English choices. It would be best if the felicity conditions were initially provided by the grammar writers, and then tested and modified in practical use with Japanese users. In the development of the prototype, suggestions from Japanese colleagues were adopted in the felicity conditions, with the result that the English and Japanese versions are not necessarily equivalent.
Further developments could provide more feedback to the user. It would be possible to add ``help texts'' with further explanations of the distinctions. It would be possible to add ``back generation'' from the set of choices into Japanese, to confirm the choices made. It would be possible to add ``back translation'' of the English surface text into Japanese, to check for misunderstandings. These further developments were outside the scope of the implemented prototype.