Quantifying Subjectivity of Meaning

When human communication as well as computational modeling of knowledge and language is considered, it is usually taken granted that the meaning of all symbols used in the communication or representation of knowledge is shared by all human and/or artificial agents. It is, however, quite straightforward to show empirically that this is not the case. Practical and theoretical limitations of traditional knowledge representation were already highlighted in an early project the objective of which was to developed a natural language database interface [10]. It started to be obvious that meaning needs to be defined contextually and also the subjective aspects is relevant. The word 'red' has different interpretations in different contexts such as "red shirt", "red skin", or "red wine". Subjectivity is particularly notable when abstract complex concepts such as 'computation', 'democracy', or 'sustainability' are considered.

In this presentation, grounded in earlier  research in this area [6,7,8,9], a general framework for modeling epistemological subjectivity is outlined [5],  and two novel methods for analyzing differences in meaning and understanding are introduced: 1) GICA (Grounded Intersubjective Concept Analysis) method [1,4] and a method for assessing user-specific difficulty of documents [3].

In human communication, it is the occasional clear failure that allows us to see that understanding language is often difficult. In making the connection between a word and its typical and appropriate use, we humans rely on a long learning process.  The process is made possible and guided by our genetic make-up, but its success essentially requires extensive immersion to a culture and contexts of using words and expressions. To the extent that these contexts are shared among individual language speakers, we are then able to understand each other.  When our learning contexts differ, however, differences in  understanding the concepts themselves arise and subsequent communication failures begin to take place. Two main failure types can be detected. The first type is false agreement, where on the surface it looks as if we agree, but in fact our conceptual difference hides the underlying difference in opinions or world views. The second type of problem caused by undiscovered meaning differences is false disagreement. If we are raised (linguistically speaking) in different sub-cultures, we might come to share ideas and views, but might have learned to use different expressions to describe them. [1]

We have recently introduced a novel method to analyze and make visible differences among people regarding how they conceptualize the world [1,5].  The Grounded Intersubjective Concept Analysis (GICA) method first employs either a conceptual survey or a text mining step to elicit particular ways in which terms and associated concepts are used among individuals [1].  The subsequent analysis and visualization reveals potential underlying groupings of people, objects and contexts.  The GICA method extends the basic idea of the traditional term-document matrix analysis to include a third dimension of different individuals.  This leads to a formation of a third-order tensor of Subjects x Objects x Contexts. Through flattening, these Subject-Object-Context (SOC) tensors can be analyzed using various computational methods [2].

We have also recently introduced a novel approach for assessing the difficulty level of a document: our language-independent method assesses difficulty of a document for each user separately. The method enables, for instance, offering information in a personalized manner based on the user’s knowledge of different domains. The method is based on the comparison of terms appearing in a document and terms known by the user. We present two ways to collect information about the terminology the user knows: by directly asking the users the difficulty of terms or, as a novel automatic approach, indirectly by analyzing texts written by the users. [3]

References

[1] Timo Honkela, Juha Raitio, Krista Lagus, Ilari T. Nieminen, Nina Honkela, and Mika Pantzar (2012). Subjects, Objects and Contexts: Using GICA Method to Quantify Epistemological Subjectivity. In Proceedings of IJCNN 2012, to appear.

[2] Juha Raitio, Tapani Raiko, and Timo Honkela. Analysis of Subject-Object-Context Tensors. In preparation.

[3] Mari-Sanna Paukkeri, Marja Ollikainen, and Timo Honkela  Assessing user-specific difficulty of documents. Submitted.

[4] Timo Honkela, Nina Janasik, Krista Lagus, Tiina Lindh-Knuutila, Mika Pantzar, and Juha Raitio (2010). GICA: Grounded intersubjective concept analysis - a method for enhancing mutual understanding and participation. Technical Report TKK-ICS-R41, Aalto-ICS, Espoo, December 2010.

[5] Timo Honkela, Ville Könönen, Tiina Lindh-Knuutila, and Mari-Sanna Paukkeri (2008). Simulating processes of concept formation and communication. Journal of Economic Methodology, 15(3):245-259.

[6] Tiina Lindh-Knuutila, Timo Honkela, and Krista Lagus (2006). Simulating meaning negotiation using observational language games. In Symbol Grounding and Beyond: Proceedings of the Third International Workshop on the Emergence and Evolution of Linguistic Communication, pages 168–179, Rome, Italy, 2006. Springer, Berlin/Heidelberg.

[7] Juha Raitio, Ricardo Vigário, Jaakko Särelä, and Timo Honkela (2004). Assessing similarity of emergent representations based on unsupervised learning. In Proceedings of IJCNN 2004, pages 597-602, Budapest, Hungary.

[8] Timo Honkela, Kevin I. Hynnä, and Tarja Knuuttila (2003). Framework for modeling partial conceptual autonomy of adaptive and communicating agents. In Proceedings of Cognitive Science Conference, Boston, MA.

[9] Timo Honkela (1993). Neural nets that discuss: a general model of communication based on self-organizing maps. In Proceedings of ICANN'93, pages 408-411, Springer-Verlag, London.

[10] Harri Jäppinen, Timo Honkela, Heikki Hyötyniemi, and Aarno Lehtola (1998). A multilevel natural language processing model. Nordic Journal of Linguistics, 11:69-82.

Bio:

Prof. Timo Honkela is currently a chief research scientist at the Department of Information and Computer Science, Aalto University School of Science. He is the head of the Computational Cognitive Systems research group. Earlier he has served as a professor (pro tem) at the laboratory of computer and information science at TKK and as a professor at the Media Lab of University of Art and Design Helsinki. Honkela has published more than one hundred scientific publications. He has conducted research on several areas related to artificial and computational intelligence, cognitive modeling, natural language processing and text mining including a central role in the development of the Websom method for visual information retrieval and text mining. Honkela is a former chairman of the Finnish Artificial Intelligence Society and current vice-chair of Finnish Cognitive Linguistics Association. He is also the chair of the IFIP working group on knowledge representation and reasoning (WG 12.1) in which position he has initiated and chaired the AKRR conference series on adaptive knowledge representation and reasoning. He is a member of the executive committee of ENNS.