Application of the Infinite Relational Model combined with the Bayesian Model of Generalization for effective cross-cultural knowledge transfer

Abstract

This paper first introduce a unique approach to convey meanings of culturally-specific concepts (CSCs) existing in one culture to a person coming from another culture via English as an intermediate language. The core algorithm employed is the Bayesian Model of Generalization (Tenenbaum & Griffiths, 2001). This model enables one to not only link cross-cultural CSCs but also estimate probabilities of how an information receiver generalizes a source concept in question from a given stimulus. The application of this model requires highly appropriate datasets consisting of concepts and their definitional features. In (Glückstad & Mørup, under review), an empirical study was performed with datasets obtained from a semi-automatic ontology construction method known as Terminological Ontology (TO) proposed by (Madsen et al. 2004). The results from that study indicated that particularly strict rules for constructing TOs may risk causing the elimination of important features. It means that the original TO-approach may require a more flexible taxonomic organization of feature structures. Hence in this work, we investigate how the Infinite Relational Model (Kemp et al., 2006), a novel unsupervised machine learning method, is applied for generating more flexible feature structures and combined with the aforementioned algorithm, Bayesian Model of Generalization.