A study of cognitive processes of natural language understanding 自然言語理解における認知プロセスに関する研究



    • 森, 辰則 モリ, タツノリ



A study of cognitive processes of natural language understanding




森, 辰則


モリ, タツノリ











From the viewpoint of computational linguistics, of which main purpose is to formalize human processes of language understanding as computational processes, this dissertation proposes three models which formalize three different aspects of natural language understanding respectively. First, we propose a model of metaphor understanding in which contexts play an important role. The model of metaphor understanding proposed by this dissertation deals not only with understanding an isolated metaphorical sentence, but also with understanding metaphorical expressions in a discourse. We adopt the formalization based on situation semantics, in which by dealing with the interpretation of a metaphorical expression and the information in the expression separately the metaphor understanding is done according to hearer’s resource. Secondly, we propose a human parsing model of natural language from the point of view of performance, or actual usage of language. Several researchers have paid attention to some remarkable phenomena in the human parsing process, since these phenomena may be a clue to the explication of the high efficiency on the human parsing process. On the hypothesis that these phenomena should be due to delicate aspects of performance, we propose a parsing model based on connectionist model, which is founded on interaction among a large number of simple computing element. Finally, we propose a model of common sense reasoning, which is used for inference by world knowledge, in natural language understanding. Some mechanism of common sense reasoning, such as circumscription, is necessary in order to make a natural language understanding system be able to comprehend some natural language expression more deeply, that is, to make the system be able to obtain some useful information, which human extracts from some expression by using his/her own background knowledge, especially common sense. While circumscription is a rule for the first order logic, we propose an application of circumscription to logic programming languages, such as Prolog. We also present an application of circumscription to the programming language Uranus which is an extended logic programming language with the multiple world mechanism. The application of circumscription to Uranus is useful to formalize a common sense reasoning on some frame-style knowledge base.

横浜国立大学, 平成3年3月27日, 工学博士, 工博甲第58号


  1. Contents / p5 (0005.jp2)
  2. Abstract / p3 (0004.jp2)
  3. Acknowledgments / p11 (0008.jp2)
  4. 1 Introduction / p1 (0009.jp2)
  5. 2 A Formalization of Metaphor Understanding in Situation Semantics / p7 (0012.jp2)
  6. 2.1 Introduction / p7 (0012.jp2)
  7. 2.2 The Outline of Metaphor Understanding in Situation Semantics / p9 (0013.jp2)
  8. 2.3 The Details of Each Step / p11 (0014.jp2)
  9. 2.4 Conclusion / p31 (0024.jp2)
  10. 3 A Parsing Model based on Connectionist Model / p33 (0025.jp2)
  11. 3.1 Introduction / p33 (0025.jp2)
  12. 3.2 Survey of Connectionist Model / p35 (0026.jp2)
  13. 3.3 Parsing Mechanism / p36 (0027.jp2)
  14. 3.4 Analysis of Several Linguistic Phenomena by the CM Parser / p51 (0034.jp2)
  15. 3.5 Conclusion / p61 (0039.jp2)
  16. 4 A Model of Common Sense Reasoning Based on Circumscription / p63 (0040.jp2)
  17. 4.1 Introduction / p63 (0040.jp2)
  18. 4.2 Circumscription / p65 (0041.jp2)
  19. 4.3 Introducing Abnormality Predicates by Finding Inconsistency / p69 (0043.jp2)
  20. 4.4 Circumscription of Hierarchical Knowledge with Inheritance / p70 (0044.jp2)
  21. 4.5 Control of Splitting Hierarchy / p79 (0048.jp2)
  22. 4.6 Conclusion / p83 (0050.jp2)
  23. 5 Conclusion / p85 (0051.jp2)
  24. Published Papers / p87 (0052.jp2)
  25. Bibliography / p91 (0054.jp2)
  26. 3.5 The connectionist model parser / p45 (0031.jp2)
  27. 3.7 The parse tree just after the CM parser reads the word 'chased' / p53 (0035.jp2)


    • 8000000076386
  • DOI(NDL)
  • 本文言語コード
    • eng
  • NDL書誌ID
    • 000000240497
  • データ提供元
    • 機関リポジトリ
    • NDLデジタルコレクション