口語形式的資訊在人類日常溝通中佔極重要角色。最近的研究指出人類認知在處理口語形式的資訊與數值形式的資訊並無顯著差異。然而目前尚無適當的模式可用以描述認知之口語資訊處理。因此，本論文探討人類認知與規範性方法(包括數值方法與模糊集合方法)在處理口語資訊上的差異、並分析人類認知之共通決策法則，以藉此對口語運算之描述性模式提出建議。所探討之口語運算包括口語統計值之位置、均值、及變異。
在口語位置研究中，本論文探討口語五等級評估尺度之數值表示法、模糊表示法、及認知表示法之差異。其中，認知表示法是以所提之區間估計方式獲取。研究結果發現認知之口語表達與數值表示法的等距(equal interval)，或模糊表示法之對稱、等區間(equal space)的假設不符。
在均值運算方面，本研究首先探討數值方法、模糊方法、及認知方法三者在綜合口語評估詞之差異。研究結果顯示數值運算表現最差，與人類認知運算有顯著差異。模糊集合運算之表現亦普遍不佳；此意謂著以模糊集合運算充當描述性運算仍有一段距離。另外，以所提之認知表示法進行模糊數運算，可得到較接近人類認知之口語綜合(相符比由0.62提昇為0.77)。為瞭解決策者綜合口語資訊所採用之決策法則，本研究並以多維尺度法(Multi-Dimensional Scaling)分析三個實驗的資料。分析結果指出受試者綜合口語資訊時，除了受口語詞的數值均值影響外，尚會考慮口語詞的「極值」與「極性」。
在變異運算方面，本研究以基準比較法獲取受試者的主觀判斷，並以因子實驗探討受試者之口語變異估計受那些因素影響。其結果發現受試者之口語變異估計會受口語詞的「數值變異」、「熵值」、「極性」、「數值變異與極性的交互作用」、「熵值與極性的交互作用」、及「數值變異、熵值、與極性的交互作用」所影響。而且「熵值」比「數值變異」更能描述人類認知之口語變異估計。
本論文之結果可用以輔助現有數量方法處理定性資料。論文最後並說明這些研究發現的可能應用。

關鍵詞：口語資訊、認知直覺運算、口語表達、口語意見綜合、口語共識判斷。

Verbal information plays a pivot role in human daily communication. Recent research has pointed out that the performance of human cognition in processing verbal information has no significant difference from that in processing numerical information. However, no proper model is available to describe human cognition in processing of verbal information. Therefore, this dissertation explores the difference between human cognition and normative models in processing verbal terms, and further analyzes the decision rules employed by decision-makers to illustrate the proper form of a descriptive model. The explored verbal operations include the following statistics: representation, mean, and variance.
In the study of verbal representation, the differences among numerical representation, fuzzy representation, and cognitive representation of Likert verbal evaluations are revealed. This cognitive representation is obtained by the proposed interval estimation method. The proposed method can simultaneously construct the verbal categories in a Likert scale. The result shows that the cognitive representation is inconsistent with the assumption of equal interval in numerical representation, and those of symmetry and equal space in fuzzy representation.
In the study of verbal mean operation, the research first investigated the differences among numerical, fuzzy, and cognitive methods in aggregating verbal terms by conducting three experiments. The results reveal that the numerical operation deviates much from actually decision making. The performances of fuzzy aggregations are also poor. This fact shows that fuzzy aggregations are still not qualified as descriptive operators. However, using cognitive representation to conduct fuzzy number operations can obtain a higher match-rate with the human decision (from 0.62 to 0.77). To understand the decision rules underlying human cognition, the research conduct a Multi-Dimensional Scaling (MDS) analysis. The results show that, other than numerical mean, subjects use two intuitive rules to aggregate opinions, namely, extreme-value and polarity.
In the study of verbal variance operation, the research obtained the subjective judgments by a paired-comparison procedure. Furthermore, a factorial experiment is conducted to investigate the factors that might influence subjects’ verbal consensus judgment. The results show that subjects’ verbal consensus judgment is related to numerical variance, entropy, polarity, the interaction between numerical variance and polarity, the interaction between entropy and polarity, and the interaction among numerical variance, entropy, and polarity. Above all, entropy is a more significant descriptive operator than numerical variance.
The results of the dissertation could complement the current numerical methods in processing qualitative data. Possible applications of the research findings are also discussed.

Keywords: verbal information, cognitive operation, verbal representation, aggregation of verbal opinions, and consensus judgment of verbal opinions.

第 1 章 緒論 1-1
1.1. 口語直覺運算之重要性 1-1
1.2. 與其它研究之關聯 1-3
1.3. 研究流程與目的 1-4
1.3.1. 口語表達 1-5
1.3.2. 規範模式與人類認知在綜合口語詞之差異 1-5
1.3.3. 口語綜合決策型態 1-6
1.3.4. 口語變異估計決策型態 1-6
1.4. 論文架構 1-7
第 2 章 文獻探討 2-1
2.1. 人類認知與規範性模式在數值運算上之差異 2-1
2.1.1. 規範性方法與描述性方法之比較 2-1
2.1.2. 人類決策與規範性理論差異之解釋 2-2
2.2. 口語表達 2-7
2.2.1. 數值表達方法 2-8
2.2.2. 模糊表達方法 2-9
2.3. 口語均值運算 2-14
2.3.1. 數值均值運算 2-14
2.3.2. 模糊均值運算 2-15
2.4. 口語變異運算 2-20
2.4.1. 意見共識量測 2-20
2.4.2. 熵值(entropy)運算 2-23
2.4.3. 極性的影響 2-24
2.4.4. 量測程序與誤差間之關係推論 2-25
2.5. 心理量度相關方法 2-27
2.5.1. 單維尺度之區間尺度的獲取 2-27
2.5.2. 多維尺度之區間尺度的獲取 2-30
第 3 章 研究方法 3-1
3.1. 口語歸屬函數之實驗 3-1
實驗步驟及受試者 3-4
3.2. 口語均值比較實驗 3-5
3.2.1. 目的 3-5
3.2.2. 實驗設計 3-6
3.2.3. 以MDS分析口語綜合之決策策略 3-12
3.3. 口語變異估計實驗 3-13
3.3.1. 研究假說(Research Hypotheses) 3-13
3.3.2. 實驗設計 3-14
3.3.3. 實驗程序 3-16
3.3.4. 受試者、實驗問題與情境 3-17
第 4 章 結果分析與討論 4-1
4.1. 口語評估尺度之歸屬函數 4-1
討論 4-3
4.2. 口語均值比較實驗結果 4-3
4.2.1. 認知排名 4-4
4.2.2. 模擬 4-5
4.2.3. 數值、模糊、認知排名的聯合比較 4-7
4.2.4. 影響認知排名的因素 4-9
4.2.5. 小結 4-13
4.3. 口語均值比較決策策略 4-14
4.3.1. 實驗結果 4-14
4.3.2. 多維尺度(MDS)分析 4-16
4.3.3. 討論 4-21
4.4. 口語變異估計實驗結果 4-22
4.4.1. 三個口語意見之共識 4-23
4.4.2. 五個口語意見之共識 4-25
4.4.3. 討論 4-29
4.5. 口語均值與變異於決策中的應用

[1] 林信惠與劉永森(1993)，”層級分析(AHP)之區間判斷及綜合函數研究，”管理科學學報，第十卷第二期，第285-303頁。
[2] 張新華(1991)，資訊學概論。台北市：臺灣商務。
[3] Aczel, J. and Alsina, C. (1986). “On synthesis of judgment,” Socio-Economic Planning Sciences, Vol. 20(6), pp. 333-339.
[4] Agresti, A. (1996). An Introduction to Categorical Data Analysis, John Wiley & Sons, Inc.
[5] Agresti, A. and Finlay, B. (1997). Statistical Methods for the Social Sciences 3rd edition, Prentice-Hall.
[6] Allais, M. (1953). “Le comportement de l’homme rationnel devant le risque: Critique des postulates et axioms de l’ecole americaine,” Econometrica, Vol. 21, pp. 503-546.
[7] Anderson, J.R. (1990). Cognitive Psychology and Its Implications (3rd edition). New York:W.H. Freeman and Company.
[8] Anderson, N.H. (ed.)(1990). Contributions to Information Integration Theory Volume I: Cognition. Lawrence Erlbaum Associates, Publishers, Hillsdale.
[9] Aster, R. and Seidman, R. (1997). Professional SAS Programming Secrets