For many decades, language sciences sought to ignore variability—from variation across people, to variation in the input. Early bilingualism research was no exception. This was due in part to the bilingualism theories being built on the monolingual comparisons and in part to methodological limitations. Theoretically, early bilingualism speech perception research adopted Categorical Perception—a theory which sought to ignore acoustic variation to rapidly extract categories. This led to the twin assumptions that the early acquisition of a language is key for creating speech categories and that listeners’ goal were to ignore the acoustic/phonetic variation. These theoretical assumptions were bound to methodological limitations, where tasks that were used were geared towards ignoring variation.

Recently, diverse bilingual communities have been spotlighted as examples of theoretical and methodological challenges in bilingualism research. Here, I will provide data from multiple studies that centralize diversity with cutting-edge statistical, social network, and eye-tracking methods. Doing so, I show how diversity contributes to stronger theory building and advancements in ecological methodologies.

First, multi-site experiments show the importance of variation in listeners' social networks such that listeners do not ignore variation in speech in locations where language ideologies are not against multilingualism.

Next, I start with work on speech categorization that uses a novel task designed to elicit subjects’ gradient sensitivity to variation; this is coupled to a social network analysis and Bayesian Hierarchical Modelling, which allows us to parametrize speech categorization performance in relation to one's social network diversity.

Finally, I present data from a study of heritage speaker children in the USA using eye-tracking methods to understand real-time word recognition in language diverse individuals across different age groups. I show how using Principal Component Analysis on the processing measures, can identify different profiles that emerge as a function of language use in social networks.