Simulating Food Evacuation Behavior : An application of Quasi-isomorphic Knowledge Model

2004 Fiscal Year Final Research Report Summary

• Project/Area Number: 14390021
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Single-year Grants
• Section: 一般
• Research Field: 広領域
• Research Institution: TOKYO INSTITUTE OF TECHNOLOGY
• Principal Investigator: SAKANO Tatsuro Tokyo Institute of Technology, School of Decision Science, Associate Professor, 大学院・社会理工学研究科, 助教授 (40196077)
• Project Period (FY): 2002 – 2004
• Keywords: Flood Evacuation / Protocol Analysis / Implicature

Research Abstract

The purpose of this study is to build a simulation model to emulate decision process of individual citizens' evacuation behaviors in the face of flood emergency. The study consists of three parts : 1)protocol experiment for collecting think aloud protocol, 2)detecting factual-inference rules and behavioral rules from the recorded protocol, and 3)simulate the decision process.
In the first year, an experimental setting was constructed by using video image to give virtual flood evacuation experiences to the 20 subjects, and recorded the think aloud protocol to reveal what are their decisions and how they come about such decisions. From the protocol, the propositions related to evacuation decision are extracted. Most of think aloud protocols are ill-formed sentences similar to spoken language. Besides that, various modality expressions are attached to proposition in Japanese. To deal with these two causes of ambiguity, we developed a method to separate modality and proposition and a method to transform ill-formed sentence to well-formed sentence.
In the second year, behavioral propositions and factual propositions are separated using the functional classification of sentence. Then four categories of evacuation behaviors, information gathering, preparation of evacuation, communication, and evacuation, are detected. It is shown that the meaning of transitive verves can be analyzed using case restriction, thus enabling to correspond transitive to the four categories detected. Against the expectation, the meaning analysis does not require pragmatic interpretation, probably because the decision context is fixed like in this study.
In the final, third year, factual propositions are analyzed. Three different aspects of propositions are identified : 1)time frame, 2)case relatedness of objects, and 3)descriptive forms. Finally, the factual rules and behavioral rules detected above are modeled into a simulator and can emulate successfully the difference between the experienced subjects and the non-experienced subjects.