Report #2A Computational Model of On-line StoryUnderstandingSix Month Progress Report: 21 April 1997Student: Elliot SmithSupervisor: Peter HancoxOther members of thesis group:Ela ClaridgeMark TorranceOn-Line Story Understanding1AbstractModels of story understanding in artificial intelligence and cognitive psychology typicallyconcentrate on the construction of representations of story content. Construction of a‘complete’ representation (informally equivalent to an ‘understanding’) requiresinferences which establish coherence links between text statements. This report examineshow reality-based links (e.g. causal connections) and narrative-based links (e.g. thosethat establish rhetorical structure and delineate text segments) have been handled inprevious models. My suggestion is that both types of links are required for narrativeunderstanding, and that each is dependent on the other. I also indicate how the inferencesrequired to create both types of link may be uniformly represented and processed withina story grammar framework. This framework could also be used to model theories aboutrecall (hierarchical storage) and processing.On-Line Story Understanding2Contents1. Introduction.............................................................................................................. 31.1. Why Story Understanding?................................................................................. 31.2. Schemas ............................................................................................................. 41.3. Schema-Based Approaches in Artificial Intelligence............................................ 41.4. Schema-Based Approaches in Cognitive Psychology........................................... 52. Scripts as Models of Story Processing....................................................................... 52.1. Introduction........................................................................................................ 52.2. Scripts and Inferences......................................................................................... 52.3. Inferential Promiscuity........................................................................................ 62.4. How Scripts Help............................................................................................... 72.5. Limitations of Scripts.......................................................................................... 72.6. Summary............................................................................................................ 83. Story Grammars as Models of Story Memory............................................................ 83.1. Introduction........................................................................................................ 83.2. The Structure of Story Grammars....................................................................... 93.3. Evidence for Story Grammars........................................................................... 103.3.1. The Validity of Story-Level Constituents.................................................... 103.3.2. The Levels Effect in Recall......................................................................... 103.4. Similarities Between Scripts and Story Grammars............................................. 103.5. Problems with Story Grammars ........................................................................ 113.6. Summary.......................................................................................................... 124. Implications and Future Work................................................................................. 124.1. Representing Story Grammars Using Discourse Representation Theory ............ 124.2. Parsing with Story Grammars ........................................................................... 134.2.1. Control Strategies for Story Parsing........................................................... 134.2.2. Assigning Story Statements to Constituent Categories................................ 145. Timetable................................................................................................................ 156. Bibliography............................................................................................................ 16On-Line Story Understanding31. Introduction1.1. Why Story Understanding?Story understanding is a classic problem in Artificial Intelligence (AI) and is interestingfrom many perspectives:• Search, control and representationStory understanding involves three important aspects of AI: search, control, andrepresentation. My own interest arose from looking at the inferencing processduring story understanding, which corresponds to a search problem. The control ofsearch is a problem which occurs in many branches of AI, such as planning,problem solving, and theorem proving. Scripts are one means of implementing(inferential) control, and are specifically geared towards story understanding (seesection 2).• Cognitive modellingComputer-based and theoretical models of how people understand stories and textshave been developed in cognitive psychology (e.g. Thibadeau et al., 1982; Kieras,1982). These models are directly concerned with forming hypotheses to explainexperimental data. My main interest in these simulations is their attempt to account for story analysis,given a resource-bounded system. The limitations of memory imply that there mustbe a mechanism which can prevent a combinatorial explosion of inferences (thebane of many AI systems). These models attempt to show how a representation ofa story can be incrementally derived, given a limited-capacity working memory. Some AI systems have similar goals, as processing time and computer memorylimitations bound the operation of any program. However, many (story)understanding systems place no limits on the length of inference chaining (e.g.Hobbs et al., 1993).• Engineering solutionsI consider story understanding to be a subset of the field of text understanding.From an engineering point of view, text understanding is a profitable area, as it canproduce practical programs, e.g. for extracting information from texts (Mikheev,1996), or answering questions about them (Lehnert et al., 1983). Descriptions ofimplemented systems are a useful resource, as they describe how to deal with thereality of computer processing.In the next sections, I describe models that have been developed in AI
View Full Document