• Linguistics 2026 Paris

    Geoffrey T. LaFlair

    Duolingo

    Geoffrey T. LaFlair

    Research Areas:

    • Second language acquisition
    • AI in language learning
    • Language assessment
    • Learning analytics and large-scale learner data
    • Data-driven and personalised learning systems
    Publications

    Bio

    Geoff LaFlair is a Principal Assessment Scientist at Duolingo where he co-leads Assessment Research and Development for the Duolingo English Test. He holds a PhD in Applied Linguistics from Northern Arizona University. Prior to joining Duolingo, he was an Assistant Professor in the Department of Second Language Studies at the University of Hawaiโ€™i at Mฤnoa and the Director of Assessment in the Center for ESL at the University of Kentucky. His research interests are situated at the intersection of language assessment, psychometrics, and AI, focusing on the application of research from these fields in conducting research on and developing operational language assessments. He has co-authored research that has been published in Language Testing, Applied Linguistics, The Modern Language Journal, Language Learning, the Transactions of the Association for Computational Linguistics, Journal of Computer Assisted Learning, Frontiers in Artificial Intelligence, Empirical Methods in Natural Language Processing, and numerous conference proceedings.

    Applications of Linguistic and Psychometric Methods in AI-driven Language Assessments

    Outline


    Recent advances in AI have transformed digital-first language assessment development and administration (LADA). However, the development process of modern, large-scale, high-stakes language assessments requires more than “just AI”. The development process should be rooted in an interdisciplinary, human-centered approach that leverages AI as a tool, the outputs of which are evaluated. In this presentation, I will discuss applications of empirical linguistic and psychometric methods to three key challenges in large-scale language assessment: content generation, item parameter estimation, and automated scoring.

    Content generation (stimuli and items) is often a bottleneck for large-scale LADA systems, which require very large item banks. Large item banks are especially important for assessments that adapt the difficulty of questions to test taker ability in order to ensure no two tests are the same and that the test can measure a wide range of abilities. Artificial intelligence can generate content en masse. The challenge for parameter estimation (e.g., item difficulty) is that traditional methods require extensive pilot testing, which is time-consuming, costly, and poses security risks. Artificial intelligence coupled with psychometric methods, or computational psychometrics (von Davier, 2022), offers solutions for incorporating features of the stimuli in addition to response data to create high-quality and stable “synthetic” parameter estimates. Automated scoring systems have a long history of operational use in language assessment (e.g., e-rater, Attali & Burstein, 2006). However, some subconstructs of writing and speaking have been difficult to capture using traditional approaches to feature engineering. Artificial intelligence now offers opportunities to better measure these subconstructs, such as cohesion (Naismith et al., 2023) and intelligibility (Cai et al., 2025).

    However, with these opportunities comes the responsibility to ensure the systems are working as intended. One of the primary assumptions of AI-generated content is the ability to automatically generate text that is linguistically appropriate for the purposes of measurement. I will discuss how register-based corpus linguistics (Biber & Conrad, 2019) offers approaches for shaping prompts, or test specifications, and for evaluating the linguistic features of the generated content. An assumption of synthetic parameter estimates is that they are stable, accurate, and capture difficulty in a way that is construct-relevant. I will show how synthetic parameter estimates can be evaluated empirically through natural language processing, corpus-linguistic and psychometric methods, such as examining feature importance as well as their effects on the reliability of test scores. Additionally, I will demonstrate how salient quantifiable aspects of spoken language, such as intelligibility, can be included as features in automated scoring systems in order to improve construct coverage. Through the lens of responsible AI (Burstein, 2026), I will present systems and methods (built on linguistics and psychometrics) for ensuring that test scores from digital-first assessments remain valid, reliable, and fair.

  • Linguistics 2026 Paris

    Monica Melby-Lervรฅg

    University of Oslo

    Monica Melby-Lervรฅg

    Research Areas:

    • Psycholinguistics and language development
    • Dyslexia and language impairments
    • Reading and literacy acquisition
    • Cognitive mechanisms in language processing
    Publications

    Bio

    Monica Melby-Lervรฅg is a Professor at the Department of Special Needs Education at the University of Oslo. Her research focuses on children’s language, literacy and cognitive development, with a particular interest in understanding what shapes developmental trajectories and whether they can be changed through targeted intervention. She has a long-standing interest in intervention research and the use of meta-analysis to evaluate the effectiveness of educational programmes.

    Academic Interests

    My research is mainly focused on language and reading development in children with dyslexia, specific language difficulties and children with minority language.

    Area of the particular interest includes educational psychological counseling and evaluation in relation to the aforementioned group, which is reflected in my research work.

    I also have a special interest in quantitative methods where meta-analyses and experiments are my mainย fields.

    From First Words to Fluency: Understanding and Influencing Language Development

    Outline


    Language is one of the most remarkable things a human being ever does, and understanding why some trajectories diverge and whether they can be redirected is the core of this talk.

    This keynote draws on research into how language develops in young children. The first part tackles a deceptively fundamental question: what is language, developmentally speaking? In our longitudinal studies, we examine its dimensionality throughout development: how components such as vocabulary, grammar, and phonological awareness relate to each other, and whether that structure shifts as children grow. We then look at what drives variation in developmental trajectories, drawing on data that separates the contributions of genetic and environmental factors and how these interact over time.

    These questions set the stage for the talk’s central challenge: malleability. Our intervention studies involving intensive, targeted training programs suggest that some aspects of language are more responsive to intervention than others. Specific skills can be moved; broader language competence is harder to shift. The distinction matters enormously for how we design.

    The talk closes by outlining what this body of work leaves open and the directions future research needs to take โ€” from longer follow-up windows and stronger causal designs to a sharper framework for what it means to support language development that genuinely lasts.

  • Linguistics 2026 Paris

    Emma Marsden

    University of York

    Emma Marsden

    Research Areas:

    • Grammar & Morphosyntax
    • Vocabulary & Lexical Knowledge
    • Input Processing & Learner Cognition
    • Open Science, Accessibility & Transparency
    • Explicit Instruction in Morphosyntax
    • Components of Language Proficiency
    • MultilingProfiler Development
    • Open Science Advocacy
    • Applied Linguistics
    Find out more

    Overview

    Emma Marsden is Professor of Second Language Education at the University of York, England. She has published widely on second language learning, processing, teaching, and testing. Her initial experiences as a French, Spanish, and English teacher in schools have inspired much of her research. With colleagues and students, she has co-authored over 80 peer-reviewed publications in the areas of grammar and vocabulary learning and teaching, processing, attention, implicit learning, priming, research methods, and open scholarship. She is co-author of the book ‘Second Language Learning Theories‘ with Ros Mitchell and Florence Myles. Emma was Associate Editor and Journal Editor of Language Learning between 2015 and 2022 and she (co-)directs several international open scholarship initiatives: for materials and data (IRIS), accessible summaries (OASIS), vocabulary profiling (MultilingProfiler), and teaching and professional development resources (Language-Driven Pedagogy). Between 2018-2023, she directed the National Centre for Excellence for Language Pedagogy, funded by the governmentโ€™s Department for Education.

    Outline:

    Modern Language Education in School Classrooms in England

    Ongoing projects

    In a world where English(es) constitute global lingua franca, what is happening to language education in countries where most of the population already speak English as their firstโ€”and often their only functioningโ€”language? I focus on England, where every year about ยผ million 16-year-olds take a high-stakes national exam in French, German, or Spanish, equating to about 45% of the cohort. Despite these numbers, there is a shortage of language students at more advanced levels and, therefore, of language expertise among teachers and other professions where language is needed. In this talk, I will consider a small but powerful cluster of factors involved in the complex picture underlying this situation: I focus on the curriculum, pedagogy, and assessment that are experienced by learners in the first 450 hours of their exposure to the language. These are tangible factors, over which educators, policymakers, and materials developers have control.

    First, I trace foreign language education over the last fifty years in England. I start at a time where foreign language study was undertaken by an elite in selective schools, with pedagogies and assessments drawing largely on methods inspired by the teaching of ancient languages. I then trace the rise of โ€˜Communicativeโ€™ approaches as they have been interpreted and operationalised in classrooms, textbooks, and governmentsโ€™ assessment regimesโ€”all with an ambition to open up languages to (almost) the whole population. I argue that largely due to the nature of the high-stakes assessments, which have needed to cater for a wide-range of student aptitudes and motivations, a โ€˜washbackโ€™ effect has led to narrow interpretations of Communicative Language Teaching, foregrounding memorisation of (often) unanalysed phrases of language through heavily topic-constrained curricula.

    Next, I review four dominant applied linguistics research agendas in this context, spanning the 1990s to early 2020s, which had to, by necessity, focus on the learning and pedagogy occurring in the educational status-quoโ€”in which the government-defined curriculum only very loosely identified the linguistic content (vocabulary, grammar, and sound-writing relations). In such a context, researchers could: (1) document the development of learner language, including the potential role of formulaic chunks of language; (2) observe the motivational forces at play among students and teachers; (3) evaluate interventions that boosted learnersโ€™ confidence and equipped them with strategies to deal with unpredictable and complex language; and (4) advocate alternative visions, such as Content and Language Integrated Learning or greater emphasis on inter-cultural competencies, running somewhat in tension with the assessment regimes. In a world where the linguistic content was not mapped out, these were appropriate, useful, and theoretically and practically insightful research agendas.

    In the third part of my talk, I describe policy-changes (2022-2026) that identified the linguistic content for teachers and students, and I draw parallels with some other jurisdictions. I present research that we conducted during and just after these policy-changes that showed concerning misalignment between curriculum and assessment, likely adversely affecting student and teacher motivation. Despite controversy surrounding these policy changes, I will argue that โ€˜the baby has not been thrown out with the bathwaterโ€™; that the high-level educational rationalesโ€”of promoting communication and nurturing an appreciation of other culturesโ€”do not have to be  forsaken by the existence of a defined linguistic content. However, these policy-changes were relatively minor. I end with alternative and potentially โ€˜disruptiveโ€™ visions of assessments and of language education that may provide more of the population in majority-Anglophone countries with an appreciation of languages and language learning. I also advocate for robust, open scholarship to facilitate a collaborative ethos between researchers, policymakers, and educators.