Phonological Networks and Their Relationship with the Mental Lexicon

Abstract

Language networks seem to be a new way to picture the complexity and evolution of language. Thus, they have been divided primarily into semantic -the most widely studied- syntactic, and phonological. In fact, Phonological Networks (PN) may provide an explanation for the phenomenon of how the mental lexicon is built within the brain. They work under the principle of phonological similarities or phonological neighbours and it is suggested that those PN share a common set of properties even though they are different from other types of networks found in the literature [1]. Identifying the structure of these given networks may provide some insight into the mechanisms that might have influenced the development of the observed network. Besides, with the aid of Graph Theory, it would be plausible to have some insight into the mental lexicon [2].

Using the software for visualization networks Gephi, the graphs of three data sets will be compared. Those data sets will be organized into # of nodes (the words) and # of edges (the relationship between these words). The data sets are the followings: One random with 268 nodes and 938 edges, one from children lexicon with 268 nodes and 496 edges, and one from adults lexicon with 2588 nodes and 35582 edges. The exploratory software Pajek would be used. Properties such as Giant component, Small world property and the Degree distribution would be analysed. Besides, it is expected to learn about how these properties are connected to the composition of the mental lexicon.

Taking into account some studies carried out on the phonological similarities in the lexicon of English [2] and applied to other languages [1], the three main properties shared in the emergence of PN might differ from other complex networks. The Giant Component, which shows the number of nodes connected, is smaller than in other complex networks. The degree distribution which is normally defined by a power law, in PN would be driven by an exponential law due to constraints like the length of words or number of phonemes in them. Finally, the small world property might be defined by other components like the average path- length and the clustering coefficient. Thus, in the emergence of these three networks to analyse these components is still very abstract. The difference in nodes within the data sets and the also the edges might influence the results. Besides, it is necessary to go deeper into an analysis from a psycholinguistics perspective. To see how these results matched theories of the composition of the mental lexicon in the brain.

References

[1] S. Arbesman, S. Strogatz, and M. Vitevitch,“The structure of phonological networks across multiple languages”, International Journal of Bifurcation and Chaos, vol. 20 no. 3 pp. 679–685, 2010.

[2] M. S Vitevitch, “What can graph theory tell us about word learning and lexical retrieval?.” Journal of speech, language, and hearing research : JSLHR, 51(2), 408–422, 2008.

[3] M. Takac, A. Knott and S. Stokes, “What can Neighbourhood Density effects tell us about word learning? Insights from a connectionist model of vocabulary development”, Journal of Child Language, 44(2):346–379, 2017.