Random Graph Model Simulations of Semantic Networks for Associative Concept Dictionaries

Word association data in dictionary form can be simulated through the combination of three components: a bipartite graph with an imbalance in set sizes; a scale-free graph of the Barabási-Albert model; and a normal distribution connecting the two graphs. Such a model makes it possible to simulate the complex features in degree distributions and the interesting graph clustering results that are typically observed for real data. 1 Modeling background Associative Concept Dictionaries (ACDs) consist of word pair data based on psychological experiments where the participants are typically asked to provide the semantically-related response word that comes to mind on presentation of a stimulus word. Two well-known ACDs for English are the University of South Florida word association, rhyme and word fragment norms (Nelson et al., 1998) and the Edinburgh Word Association Thesaurus of English (EAT; Kiss et al., 1973). Two ACDs for Japanese are Ishizaki’s Associative Concept Dictionary (IACD) (Okamoto and Ishizaki, 2001) and the Japanese Word Association Database (JWAD) (Joyce, 2005, 2006, 2007). While there are a number of practical applications for ACDs, three are singled out for mention © 2008. Licensed under the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported license (http://creativecommons.org/licenses/by-ncsa/3.0/). Some rights reserved. here. The first is in the area of artificial intelligence, where ACDs can contribute to the development of intelligent information retrieval systems for societies requiring increasingly sophisticated navigation methods. A second application is in the field of medicine, where ACDs could be used in developing systems that seek to prevent dementia by checking higher brain functions with a brain dock. Finally, within educational settings, ACDs can greatly facilitate language learning through the manifestation of inherent cultural modes of thinking. The typical format of an ACD is to list the stimulus words (cue words) and their response words together with some statistics relating to the word pairing. The stimulus words are generally basic words determined in advance by the experimenter, while the response words are semantically associated words provided by respondents on presentation of the stimulus word. The statistics for the word pairing include, for example, measured or calculated indices of distance or perhaps some classification of the semantic relationship between the pair of words. In order to mathematically analyze the structure of ACDs, the raw association data is often transformed into some form of graph or complex network representation, where the vertices stand for words and the edges indicate an associative relationship (Joyce and Miyake, 2007). However, to our knowledge, there have been no attempts at mathematically simulating an ACD as a way of determining in advance the architectural design of a dictionary. One reason is that it is a major challenge to compute maximum likelihood estimations (MLEs) or Monte-Carlo simulations for graph data (Snijder, 2005). Thus, it is extremely difficult to predict dependences for unknown