With false hope of a firm foundation gone, with the world displaced by words that are but versions, with substance dissolved into function, and with the given acknowledged as taken, we face the questions of how worlds are made, tested, and known.-Nelson Goodman Ways of Worldmaking
This approach requires . . . a fundamental change in perspective, such that the contingence of action on a complex world of objects, artifacts, and other actors, located in space and time, is no longer treated as an extraneous problem with which the individual actor must contend, but rather is seen as the essential resource that makes knowledge possible and gives action its sense.-Lucy Suchman Plans and Situated Actions
We have seen that on-line meaning construction is a non-trivial process in which a speaker assembles utterance meaning in response to linguistic clues. Because meaning is considerably underdetermined by the overt structure of language, it often requires the creative application of background knowledge. The importance of background knowledge is especially obvious when the objects of analysis are stretches of connected discourse. Schank & Abelson (1977), for example, point to the somewhat surprising difficulty of constructing a computational model capable of understanding simple stories like this one:
Seana went to a restaurant.
She ordered chicken.
She left a large tip.
Although one might conceivably build a model that could construct meanings for each individual sentence, Schank & Abelson argued that such a model would fail to compute a number of things human readers would naturally assume to have transpired.
Seana went to a restaurant.
(She sat down.)
She ordered chicken.
(She ate the chicken. )
(She paid the check.)
She left a large tip.
(She left the restaurant.)
The problem by now is a familiar one: the interpretation of the combined utterances is much richer than that one might derive from formal composition of its components. Moreover, it arises because the literal content of the sentences underspecify Seana's adventures in the restaurant. Because understanding what happened to Seana in this story seems to require a more general understanding of what goes on in restaurants, Schank & Abelson reasoned that representations of stereotyped situations might be useful for natural language processing systems. Consequently, in the course of building a system that could understand stories like Seana's, Schank & Abelson (1977) utilized scripts, data structures they proposed as being analogous to Minsky's frames. Scripts represent stereotyped sequences of common events, and provide the information the system needs to fill in gaps in the story.
But really, how often do we encounter stories like Seana's? She went to a restaurant, ordered chicken, and left a large tip? That's about as interesting as a story about a guy watching paint dry. In fact, the latter would probably be more interesting since one could at least hope for the story to provide hints about the protagonist's motives and explain his unusually subdued behavior. Consider instead the following story about Seana and Rodney, a young couple who meet for dinner shortly after the Christmas holidays.