Language is filled with uncertainties, and people who know multiple languages must manage multiple uncertainties across linguistic systems. I investigate the neural, cognitive, and linguistic mechanisms that allow people to manage language-related uncertainties in the environment and during language processing at multiple levels. I employ multiple methodologies (including neuroimaging, questionnaires, behavioral responses, and eye tracking), which are optimal for revealing associations between real-world language experience, language processing, executive control, and brain dynamics. In my recent work, I use machine learning, cross-validation, and network science to build and test neurocognitive models of real-world language use and task performance related to executive control.
My research has been funded from multiple sources, including the National Science Foundation (USA), the National Institutes of Health (USA), and the Social Sciences and Humanities Research Council (Canada).
Below you can find some more information about my main research questions, including some short presentations.
I developed a methodological approach using entropy (a measure of uncertainty and diversity; Shannon, 1948) to estimate and model individual differences in the extent to which people experience language-related uncertainty in their real-world environments: i.e., language entropy. Language entropy can be computed from questionnaire data about the engagement of two or more languages across multiple domains (e.g., language use in the home vs. language use at work). People with high language entropy report using two or more languages to an equal degree in their domains, and they likely experience high degrees of language-related uncertainty. I have made an R package available to compute this measure on GitHub.
I found that language entropy predicts the functional organization of brain networks implicated in language and executive control (Gullifer et al., 2018; Gullifer & Titone, 2020a) and aspects of language proficiency (Gullifer et al., in press; Gullifer & Titone, 2020b), as predicted by theories of neurocognitive adaptation and control (Abutalebi & Green, 2016; Green & Abutalebi, 2013). Speculatively, high entropy bilinguals may attend to cues so that they can rapidly identity which language will come next, resolving language-related uncertainty in their environment. Future work will assess relationships between language entropy and neural oscillations with resting-state EEG and ERPs. Planned work will investigate the validity of this construct with objective measures.
Below, you can watch a four-minute presentation on this topic (presented at AMLaP 2020 and Psychonomics 2020). You can also check out the project on OSF.
I developed and applied machine learning models of proactive control to approximate human performance on an AX-CPT (Gullifer & Titone, 2020a). I found agreement between multiple model architectures (including cross-validated mixed-effects models and least absolute shrinkage selection models) that language entropy is the most important individual difference feature to obtain optimum model performance, beyond other commonly used individual differences in language experience such as the sheer amount of second language exposure or the at which it was acquired. This suggests that cognitive models of proactive control must adapt to meet the demands of the linguistic environments in which they are embedded, including those created by language-related uncertainties. In future work, I plan to extend these models by incorporating individual differences in brain connectivity.
Machine learning and related methods hold high value for the future of the psychological and language sciences because the data in these fields are evolving to become massively multivariate. Researchers often collect many measures across many tasks and participants, and they need ways to assess the impact of these variables rigorously.
Below, you can watch a 15-minute talk on this topic (presented at Psychonomics 2020). You can also check out the project on OSF.
Importantly, how people use language goes far beyond the joint engagement of different language systems. People discuss various conversational topics in diverse language environments among different contacts within their social networks, all of which may lead to varying amounts of language-related uncertainty. Thus, a rich quantification of these experiences will be critical in further modeling language proficiency, language control, and executive control. Fundamentally, many experiences with language can be modeled as interconnected networks. While network models of multilingual language usage have been constructed from sources of language usage online, like Twitter (e.g., Eleta & Golbeck, 2014), they have not, to my knowledge, been constructed from in-person language usage.
In recent work, funded by an SSHRC Insight Development Grant, we modeled data about individuals’ language engagement for various topics of conversation (e.g., politics, sports, moral issues, religious issues) as a way to elucidate differences among different domains of language usage (e.g., use in formal work settings vs. informal social settings; see figure below) and among the languages spoken by the individual (Tiv, Gullifer, et al., 2020). Ultimately, network science is an ideal tool for representing complex information, like individuals’ language practices within their broader communities.
You can also check out the project on OSF.
Language-related uncertainties occur when people, such as monolingual English speakers, encounter ambiguous words with multiple meanings, such as the word “bank” which could refer to the edge of land near a body of water or a financial institution. However, such uncertainties are greater for bilinguals or multilinguals because virtually every concept can minimally be ascribed to a word in each language and word forms can be ambiguous across languages. For example, in Spanish, un vaso is a drinking glass, but the word form looks strikingly like the English word vase. While these concepts are distinct, I use traditional psycholinguistic methods, including eye-tracking, to demonstrate that even highly proficient bilinguals experience momentary competition between conflicting meanings in the irrelevant language during spoken comprehension (Titone, Mercier, Sudarshan, Pivneva, Gullifer, & Baum, 2020), written comprehension (Gullifer, Kroll, & Dussias, 2013; Gullifer & Titone, 2019, and production (Dussias, Gullifer, & Poepsel, 2016; Gullifer et al., 2013).
I received my BA in Linguistics and Psychology from The University of Massachusetts Amherst in 2007. At UMass, I worked in the Clifton-Frazier Lab under the direction of Charles Clifton and Lyn Frazier.
Following my undergraduate studies, I joined The Pennsylvania State University community as a Research Technologist in the Center for Language Science. The CLS is an interdisciplinary group of linguists, psycholinguists, applied linguists, speech-language pathologists, speech scientists, and cognitive neuroscientists who share a common interest in language acquisition and bilingualism. From 2007 - 2009, I was the resident webmaster for the CLS and conducted various administrative duties.
In 2009 I became a graduate student in the Psychology Department at Penn State where I continued working with Judith Kroll and Paola Dussias. I received my MS in Psychology in 2011. I then received my dual-title PhD in Psychology and Language Science from The Pennsylvania State University in 2015.
In 2015, I moved to Montréal to join McGill University as an NIH-funded postdoctoral fellow in the McGill Language & Multilingualism Laboratory, under the direction of Debra Titone, and in the Language Experience and the Brain Laboratory, under the direction of Denise Klein.