Teaching, coding, learning: Using AI to unlock patterns in learner language [r-libre/4261]

Couture-Matte, Robin, Cruz Enríquez, Maura et Pichette, François (2026). Teaching, coding, learning: Using AI to unlock patterns in learner language. Dans Book of abstracts on the Second Language Research Forum (SLRF) (p. 214).

Fichier(s) associé(s) à ce document :

[thumbnail of SLRF 2026 Book of abstracts.pdf]

PDF - SLRF 2026 Book of abstracts.pdf
Contenu du fichier : Mise à jour de la version publiée

Télécharger

Catégorie de document :	Communications dans des actes de congrès/colloques
Évaluation par un comité de lecture :	Oui
Étape de publication :	Publié
Résumé :	In second language acquisition (SLA), analyzing learner production data provides valuable insights into language development and informs pedagogical practices. However, processing large corpora of student assignments can be time-consuming and technically demanding. Although advances in natural language processing have made automated analysis more accessible, programming expertise remains a barrier for many applied linguistics professionals. In this context, generative artificial intelligence (AI) offers a promising solution for researchers with limited coding experience (Kim, 2023). This study examined how AI tools can support linguists and language teachers in automating corpus analysis. Specifically, we explored whether ChatGPT and Copilot could be integrated across the research workflow to assist researchers with limited coding knowledge with data manipulation, library selection, code generation, troubleshooting, and output verification. The dataset consisted of a Spanish corpus of more than 350,000 words drawn from the assignments of 219 learners (A1–A2: 77; B1–B2: 62; C1–C2: 80) and an English corpus of over 45,000 words from 98 learners (A2: 50; B1: 49). All participants were adult francophone learners at a Canadian university. The analysis examined patterns in learner writing, focusing on gender assignment in Spanish and the use of gerunds and participles in English. For the Spanish corpus, AI-assisted procedures enabled data processing in Python (Python Software Foundation, 2025) using the spaCy library (Honnibal et al., 2020), followed by Bayesian mixed-effects logistic regression in R (brms package). Results identified key predictors of gender assignment, including proficiency level (β = 1.96, 95% CrI [1.62, 2.32]), non-prototypical gender markers (β = –2.47, 95% CrI [–3.25, –1.71]), and cross-linguistic gender mismatches (β = -1.45, 95% CrI [-1.81, -1.10]). For the English corpus, AI-supported analyses using spaCy and SciPy revealed significantly greater complexity in gerund and participle use at higher proficiency levels (χ² (15) = 26.85, p = .043). Findings aligned with previous research and expert evaluations, demonstrating that AI can effectively support corpus analysis workflows and generate functional code. This study highlights the potential of generative AI to make corpus linguistics more accessible, while also underscoring the value of medium-scale learner corpora for identifying systematic patterns in L2 development.
Adresse de la version officielle :	https://9d42e4d3-e9c0-4687-aa50-840fb89a0f9b.files...
Déposant:	Cruz Enríquez, Maura
Responsable :	Maura Cruz Enríquez
Dépôt :	08 oct. 2026 12:43
Dernière modification :	08 oct. 2026 12:43

Actions (connexion requise)

RÉVISER