Linguistic Approaches to Bilingualism - Volume 15, Issue 4, 2025

While recent years have seen a shift towards random effects modelling, particularly in areas of linguistics in which nested structure is the norm (e.g., trial repetitions nested within participants), an over-reliance on standard linear modelling prevails, particularly in the cases of dynamic phenomena that may not constitute a linear relationship, e.g., vowel trajectories, pitch contours, acquisition processes, etc. Generalised Additive (Mixed) Models (GAMMs) are now commonly employed in phonetic research (given the naturally dynamic nature of speech data) and this is reflected by the availability of several tutorials which focus on phonetic data. This tutorial aims at making GAMMs accessible to researchers from other fields within linguistics. In particular, this tutorial is written for researchers in bilingualism and multilingualism who wish to be able to start using GAMMs for non-linear data, which is very common in developmental and learning phenomena. While only the basics will be covered here, we hope that researchers will get the necessary foundations to be able to learn GAMMs from existing resources.

Bayesian analyses have been increasingly adopted in psychology and linguistics as an addition (or replacement) to traditional frequentist methods. However, Bayesian methods are not yet widely applied in bilingualism research, possibly because existing introductions and tutorials have not been directed specifically at our field. The current paper highlights the advantages of Bayesian statistics to the bilingualism researcher, by providing both an introduction to its foundational principles and a practical tutorial on estimation and hypothesis testing using the brms R package. The examples build up from simple linear regression to more advanced mixed-effects models and showcase different aspects of a Bayesian workflow. All data, code, models, and supplementary materials are publicly available at https://osf.io/n3jgm/.

This research addresses two major challenges in studying second language acquisition and bilingualism: reducing overlap in predictor variables and correctly classifying participants into language proficiency levels. Too many relevant predictors can harm statistical analysis due to an increased chance of overlap, known as multicollinearity. To tackle this, we use Principal Component Analysis (PCA) on selected predictors to identify proficiency indicators, combining the length of stay in the UK and language test scores. Additionally, traditional methods, especially IELTS-based proficiency classifications, often miss subtle differences in language skills, particularly when they fail to consider how long participants have been exposed to the target language. We counter this by using non-hierarchical Cluster Analysis (NCA) for a grounded, data-driven way of detecting distinct language proficiency groups. This new approach is demonstrated on a dataset of eye movements from reading tasks, collected from Chinese–English bilinguals in the UK.

The time it takes an individual to respond to a probe (e.g., a word, picture, or question) or to read a word or phrase provides useful insights into cognitive processes. Consequently, timed measures are a staple in bilingualism research. However, timed measures usually violate assumptions of linear models, one being normal distribution of the residuals. Power transformations are a common solution but which of the many possible transformations to apply is often guesswork. Box and Cox (1964) developed a procedure to estimate the best-fitting normalizing transformation, coefficient lambda (λ), that is easy to run using standard R packages. This practical primer demonstrates how to perform the Box-Cox transformation in R using as a testbed the distractor items from a recent eye-tracking study on sentence reading in speakers of Spanish as a majority and a heritage language. The analyses show (a) that the exponents selected via the Box-Cox procedure reduce positive skewness as well as or better than the natural log; (b) that the best-fitting value of λ varies based on factors such as group and, in the case of eye-movement data, the measure of interest; and (c) that the choice of transformation sometimes impacts p values for model estimates.

Web-based eye-tracking is more accessible than ever. Researchers can now carry out visual world paradigm studies remotely and access never before tested, multilingual populations via the internet all without the need for an expensive eye-tracker. Web-based eye-tracking, however, requires careful experimental design and extensive data wrangling skills. In this paper, we provide a framework for reproducible, open science visual world paradigm studies using online experiments. We provide step-by-step instructions to building a typical visual world paradigm psycholinguistics study, and walk the reader through a series of data wrangling steps needed to prepare the data for visualization and analysis using the open-source software environment, R. Importantly, we highlight the key decisions researchers need to make and report in order to reproduce an analysis. We demonstrate our approach by carrying out a single change replication of an in-person eye-tracking study by Porretta et al. (2020). We conclude with best practices and recommendations for researchers carrying out bi-/multilingualism web-based visual world paradigm studies.

Experimental research has yielded many important psycholinguistic findings in bi-/multilingualism. However, cognitive computational models of sentence processing were limited to the single-language case until recently. In this methodological review, we discuss cognitive modelling work that uses the Bilingual Dual-path model to simulate experimental research on bilingual sentence production, processing, and development. As a detailed example of such work, we then report on new simulations conducted with the model. Finally, we suggest directions for future cognitive modelling research using this model.