I present two studies exemplifying the use of network approaches to uncover emergent properties of language in mind and society.

Study 1: Linguistic relativity emerges from distributional semantics.

The study investigated whether grammatical gender assignment influences how speakers of different languages conceptualize inanimate nouns. I analyzed 31k word meanings derived from training neural networks on millions of texts in 17 gendered languages, and compared 136 language pair combinations. The results suggest that grammatical gender affects the distributional semantics of words, by way of both sub-lexical and contextual similarities. These gender-related biases in word meanings were observed in language statistics that do not directly involve sensory experience, suggesting that much of the embodiment in language processing might become statistically embedded in how words are used.

Study 2: Social structure and multilingual coexistence.

Which social structures are optimal for maintaining minority languages, and which ones accelerate language shift? I test specific conditions of maintenance of bilingual behavior as emergent in constructed human social networks, by experimentally varying the topological properties of such networks - how individuals are made to interact, and who says what to whom. Then I apply quantitative analyses from social network analysis to assess which social connectivity structures promote or hinder the spread of language behavior in a community. This network science approach coupled with experimental paradigms can probe the effects of individual-to-individual interactions while also revealing how those interactions lead to larger-scale emergent patterns of bilingual behavior at the community level.

Luca Onnis

About the Higher seminar in Research on Bilingualism