To use Gerhard Kempen's words (cf. [Kempen1989], page 15)
Nevertheless, the addition of a monitor may contribute to the solution of practical and theoretical problems significantly. Take for example the above issue of one-way versus two-way traffic between strategic and tactical components.Suppose the monitor can intercept the linguistic output from the tactical component (preferably before the point of speech) and feed it into a parser/understander. The latter evaluates the generator's utterance from relevant viewpoints and informs (via the monitor) the strategic component of its diagnosis. This would establish the line of communication postulated by Danlos and others without complicating the generator's design -- the parser is needed anyway.
In a similar vein, Mann Mann:89, page 207 suggests that
More substantial sharing occurs in the area of knowledge representation and inference. Here the problems and solution, not just the recognition of phenomena, are shared. There is hope for convergence, for one all-sufficient underlying representational form, and for a non-directional view of language. It is often suggested that an adequate text generator must have an understander inside to check its work. Still, the research activity is dominated by the differences rather than the shared elements.
Putting both citations together our approach can be seen as an important step in that direction. We have to incorporate into language interfaces the same kind of sensitivity to later audience reactions that we have ourselves. If we are able to do this, then this will lead to more flexible natural language systems. Clearly, in order to get a real-time behaviour of our systems, the amount of feedback between the conceptual and linguistic system should be reduced as far as possible. If the evaluation of some decision points cannot be performed deterministically with information of the input the system should be free to choose according some preferences (for example `the first alternative you can get' or `take that with the highest priority'). However, if we are in a particular situation, then the conceptual component especially the monitor should be able to delegate the appropriate goals to obtain a more carefully realization to the linguistic component. This component is now responsible for doing its best. By means of monitoring the conceptual component can be part of this process in that it can observe the input/output behaviour of the linguistic system and evaluates the results with respect to the specified goals.
This kind of modelling means for the design of a whole system, that the particular goals that have to be fulfilled in order to obtain adequate realization can be formulated in a discourse independent abstract way. Because during run-time the conceptual and linguistic component are able to solve the goals in a cooperative way, the overall system is able to react on particular situations more flexible during run-time. If we would deny the importance of such mechanisms for the investigation of NLS then we have to foresee in a system the creative aspect of possible situations and hence, we are forced to specify the flexibility by hand.