Distributional semantic models: Deepening the methodology of cognitive modeling and exploring cognitive processes in human semantic memory

2018 Fiscal Year Final Research Report

• Project/Area Number: 15H02713
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Single-year Grants
• Section: 一般
• Research Field: Cognitive science
• Research Institution: The University of Electro-Communications
• Principal Investigator: Utsumi Akira 電気通信大学, 大学院情報理工学研究科, 教授 (30251664)
• Co-Investigator(Kenkyū-buntansha): 猪原 敬介 くらしき作陽大学, 子ども教育学部, 講師 (10733967)
• Project Period (FY): 2015-04-01 – 2019-03-31
• Keywords: ベクトル空間モデル / 単語の意味 / 記号接地 / 意味記憶 / 抽象語 / ベクトル意味論 / 意味ネットワーク

Outline of Final Research Achievements

In this research project, using the technique of network analysis, we demonstrated that a vector space model or a distributional semantic model, in which the meaning of words is represented as a multi-dimensional vector, is plausible as a cognitive model of human semantic memory or mental lexicon. In addition, it was found from a simulation experiment of network growth that human lexical acquisition process can be better explained by the following two processes: semantic differentiation (i.e., a process of adding some kind of variation on the meaning of existing words by a new word) and experiential correlation (i.e., a process of relating a new word to existing words by experiential correlation). Furthermore, we proposed a novel multimodal distributional semantic model in which abstract words are represented indirectly through grounded representations of their semantically related concrete words, and provided computational evidence for the indirect grounding view of abstract words.

Free Research Field

認知科学

Academic Significance and Societal Importance of the Research Achievements

単語の意味とは何か，それはどのように表現・処理・獲得されるのか，といった言語の意味に関する諸問題は，人間の知の解明を目指す認知科学などの学術分野の中心課題であり，本研究で得られた結果はそのさらなる解明に向けて，ベクトル空間モデルが認知的に妥当な計算モデリング手法であることを実証した．ベクトル空間モデルによるシミュレーションを行うことで，人間の意味記憶に関する仮説検証が可能になる点で学術的に意義がある．さらに，近年の人工知能の基盤である深層学習では，基盤技術として単語ベクトルが用いられているため，深層学習の内部構造の解明や学習手法の発展にも寄与する点で社会的意義も大きい．