AGILE Review meeting - April 2000
The AGILE IMD text planner
®Implementation in KPML
®Organization of textual content
®Interpretation of the structure of the content specification in terms of text plan elements that build up a text plan
®Relating text plan elements to the pieces of content they plan.
®“Unlimited” recursive depth of the content specification can be planned.
®Incremental planning of complexity
®Multilinguality
®Common text plan but –possibly- language-specific constraints on the realization of specific text plan elements (e.g. titles, style)
®Coreference in the content spec.and basic discourse model
®Layout formatting
®Formatting of the layout of the text using HTML annotation
The text planner is implemented in KPML. This narrows substantially the (in the literature noted) “generation gap” between strategic generation and tactical generation: There exists now the possibility for the text planner to guide realization issues sentence planning and lexico-grammatical generation. Strategic and tactical generation no longer operate in total ignorance of one another.

The text planner consists mainly of a region that specifies how the text plan is build up as an interpretation of the structure of the content specification. Thus, we are using the same familiar means as employed in the development of lexico-grammars. The basic text plan elements are Task-Title, Tasks, Constraints, Preconditions, Side-effects.

The text planner can handle the “unlimited” recursive depth made possible by the (intertwined) recursivity built into the domain model. This presents a significant generalization not only over the initial AGILE prototype, but also over previous comparable systems like DRAFTER.

The text planner has been developed such that it constructs a text plan in an “incremental” way, i.e. in a way that is sensitive to the complexity of the content specification. For example the text plan can consist just of one or two “sentences”, or can be more elaborate and include a title, steps, etc. When the content specification is fairly minimal and would specify the meaning that could be realized by a single sentence, we do not plan that content as a document title but as a simple task/step.

The text planner is fully “multilingual” in that we can specify language-specific constraints where desired, or share resources among languages when there are no differences. For example, the text plans produced for the three languages are –so far- identical, but may contain language-specific constraints on how particular text plan elements should be realized (like titles, tasks in a particular style, etc.).

The text planner maintains a basic discourse model (in the spirit of for example the Praguian Stock of Shared Knowledge). This discourse model keeps track of entities that have already been introduced in the context. Such relations/reuse can be indicated in the content specification through the use of identical identifiers. The sentence planner makes use of the discourse model in order to plan appropriate forms of (basic) contextual reference – for example, in Czech to result in a contextually appropriate word order.

The text planner also annotates the text plan elements with HTML tags, so as to give rise to a proper layout of the content.