One of the major objective in story understanding is to discover the
causal reasoning behind characters' actions and to link these into a overall
picture of the characters' motivations and actions. Thus the main aim
when processing a sentence is to discover a character's goal in which this
sentence can be considered as a step towards its achievement. The above
process uses abductive reasoning in drawing its inferences and as a
consequence of this any facts that are derived from a sentence might be
invalid, causing a number of facts to be generated that are inconsistent
with the knowledge base. A further complication to story understanding is
that much of the information that is necessary for understanding to occur
can only be obtained using default reasoning. Any such default fact remain
valid unless a further statement proves that this is not the case.
As a consequence of the above any new statements must be check
against the rest of the knowledge base to make sure there are no
inconsistencies and a list of supporting statements must be held so that
any inconsistency found can be resolved and erased. An alternative to
erasing these inconsistent statements within the knowledge base is to
maintain a number of consistent environments using an assumption
based truth maintenance system to enforce consistency. This has the
advantage that more than one environment may be worked on at once
and environments can be compared.
The thesis discusses the maintenance of more than one environment
and proposes a blackboard system, along with an assumption based truth
maintenance system, as an ideal architecture to support the requirements
of a story understanding program. The thesis also describes the knowledge
sources, such as syntax and semantics, that are necessary for story
understanding and how their operation should be controlled using a
dynamic scheduling system.
A Doctoral Thesis. Submitted in partial fulfillment of the requirements for the award of Doctor of Philosophy of Loughborough University.