Integrating Text Planning and Production in Generation

While the task of language generation seems to separate quite naturally into the two aspects of language generation (text planning and text production), it is necessary to have the planning and the production interact at generator decision points in such a way that the former need not contain explicit syntactic knowledge, and that the latter need not contain explicit goal-related information. This paper describes the decision points, the types of plans that are used in making the decisions, and a process that performs the task. These ideas are embodied in a program. 1 I n t r o d u c t i o n : the P rob lem Our current understanding of language generation includes text planning and text production. In generation work of a decade ago (Simmons & Slocum 72; Goldman 75), no text planning phase ever appeared. In the last few years, much work has been done developing text planners. The issue of interaction between planning and production phases was addressed in various ways. This paper suggests a better way to achieve the necessary interaction. In the simplest systems, planners make only very highlevel decisions, such as selecting appropriate speech acts (Cohen 78; Jacobs 85), and play no further role in text expansion. In systems with more elaborate plans, the text can be planned out in considerable detail before actual production is started. In this approach, there is a one-way flow of information from the planner to the generator (McDonald (personal communication); McDonald 80; Appelt 81). The production process requires this information whenever it must decide how to expand a generator instruction into a series of more detailed instructions. If the decision criteria are based purely on syntactic and rhetorical grounds (using notions such as sentence focus and stress); it is comparatively easy to This work was supported in part by the Advanced Research Projects Agency monitored by the Office of Naval Research under contract N00014-82-K-0M9 build them into a pre-expansion planner, since they are relatively simple and only impinge on expansion at a small number of points (such as subject choice and clause content). However, when you want the decisions to take into account pragmatic considerations such as speaker intentions, conversational setting, and hearer characteristics, it is much more difficult to plan all the decisions before commencing actual expansion. For example, suppose the generator wants to create in the hearer sympathy for a 65-year old beggar. In the sentence ■the [say-age AGE-INSTANCE-23] woman is homeless", say-age should return "old or even "ancient" rather than "65-year old". For the planner to precompute this decision, it will have to compute all the decisions (via say-sentence and say-subject, etc.), such as selecting a subject, a head noun, and adjectives, before it will be in a position firstly to realize that AGE is to be said as an adjective, and secondly to determine what the options are in this case. In order to do this computation, the planner will have to have access to information which one would like to claim is properly the exclusive concern of expansion, such as syntactic and lexical knowledge. (For instance, Appelt's planner contains grammatical knowledge spread throughout. Appelt alludes to the problems that this causes in (Appelt 81).) Furthermore, if the planner is going to do this computation down to the level of individual words, it may as well do the generation simultaneously. One can try to get around this problem by having the planner assemble a set of injunctions upon which the expansion can base its decisions. To do so, the injunctions would have to span the space of possible locutions arising from the representation; assembling them would be a very large task. A better solution is to perform planning only when necessitated by the expansion. This approach is characterized by a two-way communication at decision points. As the example in section 3 shows, five decision points enable a generator to produce flexible yet good text. These decision points are: topic choice; sentence content; sentence organization; clause organization; and word choice.