ABSTRACT
The mechanical generation of natural-language sentences or texts from internal representations (semantic representations) encompasses the phase of planning the content (‘what to say’) as well as the form (‘how to say’) of assertions. This division of labor is reflected in the system architecture in strategic and in tactical components, respectively. However, language can only be adequately generated if both components interact. The informational resources required for generation and production clearly overlap, but it is an open question as to what degree. In particular, it is unclear to what extent comprehension and production can be seen as inverse processes.
