Each of these possible interpretations rely on a different set of concepts being utilized from within the ontology of my theory. Although all are possible, some may be more probable than others in a given context. How is a reasoner to select the ``best'' interpretation given all other information available? A reader could simply ``skip'' this sentence and assume that later sentences will disambiguate the confusion. Unfortunately, this simply pushes the problem back a level--at some point, an interpretation must be created (given the problems of keeping all possible options simultaneously active in memory). Also, human readers are able to make a choice between the alternatives if you stop the reading process and query them. Some mechanism must be allowing the selection of a ``best'' choice at some point in the reading process.
One option is simply to allow an arbitrary choice. On the other end of possibilities, a designer of a reading system could encode a set of rules which would allow the system to select one choice over the other. Unfortunately, the former of these options leads to potentially bizarre results occurring (how often, after all, is a story beginning with the example sentence going to be about were-bears?); the latter option leads to a huge knowledge engineering problem and the potential to overlook possibilities which would then cause less than optimal performance. I follow a more general approach of allowing the basic ontology to constrain the problem. This allows the creative understanding of concepts to occur, while preventing bizarre interpretations from being formed.
Figure 12 depicts a portion of the ``generic'' human; Figure 13 depicts a similar portion of the ``generic'' bear. One possible interpretation of the sentence ``John was a bear'' is that the human named John is acting like a bear. One possibility is that John is as strong as a bear. ISAAC can capture this interpretation as shown, in part, in Figure 14. Or, John might be as angry as a bear (Figure 15). In either case, John is a human with a characteristic that can be described as ``bear-like.'' The bear-like states are generated by appealing to primary attributes of the bear concept.
Another alternative is that a human has been changed into a bear. This is seen in Figure 16. Notice that it is currently a very sparse representation. Nothing is known about the transformation, only that one occurred. Another possible interpretation is that John is not a true human but that he is a were-bear--a magical creature capable of transforming from a human form to a bear form. This is seen shown in Figure 17. Finally, the simplest interpretation is shown in Figure 18. Bears possess names just as humans do, John is a name, so John might simply be the name attached to a particular bear.
As Chapter 6 will describe, the creative understanding process uses the ontology described here to limit the amount of alteration which is allowed to occur while trying to understand a concept. Any of the possible meanings for the sentence John was a bear is possible; however, some require more alterations if the necessary concepts are not already in memory. For example, of a person does not know about were-creatures, the possible meaning shown in Figure 17 will not be considered, solely on the basis of this one sentence. Similarly, if a person does not realize that one can describe humans as having characteristics of other animals, John being bearlike will not be an interpretation considered. The knowledge representation must allow all of these possibilities to be representable (thereby permitting potential novel concepts to be handled); the specific interpretation which is produced and reasoned about depends on what background knowledge the reasoner already possesses. Some of the interpretations require that the reasoner do more work manipulating existing knowledge to make the new concept ``fit'' into what is already known. If an interpretation can be found which requires less of this alteration, it will be preferred.
This means that the ISAAC system can be a better understander if it contains a wide-range of concepts, as well as being able to develop new ones as needed. At startup time in ISAAC, there are approximately 1200 concepts internally represented; these fit into the twenty ontological categories I have described. The concepts include general ones which are useful in a variety of story comprehension situations, as well as more specific ones which are ties to particular texts. Figure 19 through Figure 22 show a subset of the concepts which the instantiated model utilizes.
Although there are dozens of examples of any particular type and domain, I have not shown all of them; the difference in numbers I did elect to show are somewhat indicative of the overall proportions which exist in the knowledge system. In other words, to understand stories which are set in the physical world, a great number of physical entities are needed. On the other hand, few temporal agents are needed as that class of entity is extremely rare.
The other element to note is that I have not shown the hierarchical organization; this was an intentional ``oversight'' on my part. One reason is that the large size of the conceptual framework is somewhat prohibitive to inclusion en toto. Another, more theoretically grounded reason is related to the dynamic nature of my hierarchy. Since it can be shuffled when necessary, presenting a hierarchical breakdown might bias the reader into thinking that that was the primary organization. While a pre-reading organization does exist, it is an implementational detail rather than a theoretical commitment. Tied in to this point is the fact that the hierarchies tend to be very broad rather than deep. The primary importance is the ontological category which the concept belongs to, and I have shown that in the four figures.
