本稿は,日本語名詞の意味を連続的かつ可視的に表現する意味マップを神経回路網モデルSOMによる自己組織化によって自動構築する手法を提案する.共起する連体修飾要素の観点から,まず,意味マップの自己組織化に有効と思われる,連体修飾要素が名詞の具体的な内容を表すような名詞句を新聞から人手で収集し,その名詞句を用いた意味マップの構築を試みる.そして,大規模意味マップの構築にはデータ収集の自動化が不可欠という観点から,新聞から名詞およびそれと共起する形容詞と形容動詞を共起頻度の高いものから自動的に取り出して構成される名詞句を用いた意味マップの構築を試みる.計算機実験で得られた意味マップはまず実際に用いた学習データを用いて検討し,意味マップ上の名詞は全般的に学習データが示唆する意味で配置されていることを確かめる.そして,分類結果に可視性や連続性のない階層型クラスタリング手法との比較を行い,本手法の分類能力を評価する.さらに,可視化能力を有す多変量解析手法が本タスクにうまく適用できないことを主成分の寄与率分析および計算機実験を通じて明らかにし,提案手法の必要性を補強する.A method is described for automatically constructing a semantic map,a visible and continuous representation in which Japanese nouns with similar meanings are placed at the same or neighboring points so that the distance between them represents semantic similarity.This is done by using the self-organizing neural network, SOM.From the point of view of common adnominal constituents,we first manually gather noun phrases whose adnominal constituents concretely describe the contents of head nouns from newspapers and construct a semantic map of the nouns using these noun phrases.Such types of noun phrases are thought to be effective for self-organizing a semantic map.Because it is indispensable to gather data automatically for constructing a large semantic map,we then construct a semantic map of the nouns using the noun phrases that consist of nouns and their co-occuring adjectives and nominal adjectivals.They are gathered automatically from newspapers in the order of the frequency of their co-occurrent words.Examination of semantic maps obtained in computer experiments showed that the nouns were mapped to the points corresponding to the training data.And, to objectively evaluate the SOM's ability in semantic classification,the semantic maps are compared to the results of classification by hierarchical clustering,which cannot give results with visible and continuous representation.Further, it is clarified that the multivariate statistical analysis such as principle component analysis and factor analysis cannot be used to construct semantic maps which reinforces the necessity of the proposed method for this task.