My dissertation will address a critical gap in SLA by providing the first comprehensive quantitative synthesis integrating explicit aptitude, implicit aptitude, and working memory research. Despite decades of investigation into how cognitive abilities predict language learning success, the field suffers from definitional confusion, measurement inconsistencies, and reliability concerns—particularly for implicit measures. Through systematic meta-analysis, I examine how these cognitive constructs predict learning outcomes across different knowledge types (implicit vs. explicit) and instructional conditions. This work tackles fundamental methodological challenges, including circular validation problems and the absence of cross-validation between aptitude batteries. By documenting reliability reporting practices and construct operationalizations, the dissertation will reveal how measurement issues may distort our empirical understanding. The goal is to provide empirical grounding to validate or challenge theoretical claims about differentiated aptitude profiles and aptitude-treatment interactions (ATIs), moving the field toward greater methodological rigor and theoretical coherence.
Generative AI tools like ChatGPT offer unprecedented opportunities for language learners and educators, even though research has focused on the perception of these tools over their uses and effectiveness. Working with Frederick Poole, we synthesize current research and provides a comprehensive framework for understanding GenAI's role in language education, positioning AI literacy as a crucial component of digital literacies. We argue that unlike earlier chatbots, GenAI can handle open-ended dialogue and adapt to learner needs, though it should enhance rather than replace learning. Our empirical work with Catalan learners demonstrates that task design matters more than proficiency prompting, and reveals linguistic alignment between learners and ChatGPT. This work has important implications for LCTL pedagogy and the ethical integration of AI in language education. Upcoming work focuses on the literacy and uses of AI by language educators in the field of foreign, additional, and second language education.
Work derived from this research has been made available in publications with Frederick Poole in CALICO Journal and Technology and Instructed Second Language Acquisition: Connecting Research and Pedagogy (John Benjamins), as well as conference presentations at EuroSLA, the Artificial Intelligence Research in Applied Linguistics, and the Task-Based Language Teaching Conference.
If you have insights that you would like to share, please contact me at todacosi@umd.edu!
Catalan represents a unique case in applied linguistics as a less commonly taught language with political and cultural significance. My teaching experience and research in this area focus on the sequences followed by learners of Catalan and the origins of errors in their interlanguage. I investigate how innovative technologies, including AI, can support Catalan language learning, particularly to facilitate independence in the learning process when traditional immersion opportunities are limited. Working with Frederick Poole, I examined ChatGPT's effectiveness as a conversational partner for Catalan learners, exploring proficiency-level adaptation and ecological validity. Our findings suggest that while task design matters more than proficiency prompting, ChatGPT demonstrates linguistic alignment with learners and is perceived as a realistic conversation partner. This work has important implications for LCTL pedagogy and AI literacy in language education.
Work derived from this research has been made available in some of my publications on AI with Frederick Poole, as well as conference presentations at EuroSLA, the Artificial Intelligence Research in Applied Linguistics, and the Task-Based Language Teaching Conference.
If you are a Catalan learner, contact me at todacosi@umd.edu, you may be able to participate in some of my projects!
How do learners acquire grammar without explicit instruction? My research examines incidental learning of morphosyntactic features, with a focus on Spanish grammatical gender among intermediate learners. I investigate whether enhanced linguistic context (such as adjective-noun pairings) facilitates learning, how different types of knowledge (implicit vs. explicit) develop under various instructional conditions, and what role individual differences play in these processes. Using measures like the Visual Word Monitoring Task for implicit knowledge and cloze tasks for explicit knowledge, alongside assessments of statistical learning ability (Alternating Serial Reaction Time) and working memory (Running Memory Span), this work reveals the complex interplay between instructional manipulation, linguistic properties, and learner cognition. A key finding is that item-level information matters less than total scores when predicting learning outcomes, highlighting the need for nuanced measurement approaches.
Work derived from this research has been made available at several conferences, such as the Interdisciplinary Advances in Statistical Learning, Vocab @Vic, the American Association of Applied Linguistics, and the Annual Graduate Portuguese and Hispanic Symposium (where it was awarded best paper). This research has received funding from National Science Foundation Research Traineeship through the Language Science Center at the University of Maryland in 2022, as well as the Graduate Program in Second Language Acquisition at the University of Maryland, College Park.
Accurate proficiency assessment is fundamental to SLA research, yet traditional measures often fail to capture the multidimensional nature of language ability. My work on proficiency measurement spans multiple projects and languages. I developed an Elicited Imitation Task (EIT) to distinguish intermediate-high from superior-level Spanish learners, facilitating the identification of highly capable speakers. This research reveals that proficiency is not unidimensional—different measures tap into distinct constructs, with some reflecting reading/writing skills while others capture speaking/listening abilities based on factor analyses to understand how different proficiency measures (DELE, LexTALE, EIT, OPI, self-ratings) relate to one another. Or that we should exercise additional caution when assuming the comparability of findings across tests. These findings have important implications for how we screen participants, interpret research results, and design studies in SLA.
I also created a c-test to assess Catalan proficiency—if you are interested in helping me pilot it and develop a valid and free measure for Catalan proficiency online or accessing the modified EIT, please email me at todacosi@umd.edu!
Work derived from this research has been presented at the East Coast Organization of Language Testers, and has received funding from National Science Foundation Research Traineeship through the Language Science Center at the University of Maryland in 2022.