Fecha
2026-07-10
Derechos de acceso
info:eu-repo/semantics/openAccess
Título de la revista
ISSN de la revista
Título del volumen
Editorial
Resumen
A pesar de ser la lengua extranjera más estudiada en la educación superior estadounidense, la enseñanza del español no logra con frecuencia desarrollar la competencia oral funcional en el alumnado, debido a la escasez de práctica individualizada, la heterogeneidad del aula y los elevados niveles de ansiedad ante la lengua extranjera. Los avances recientes en modelos de lenguaje de gran escala (LLM) y en reconocimiento automático del habla (ASR) ofrecen una respuesta posible: sistemas de inteligencia artificial conversacional capaces de proporcionar práctica oral adaptativa en un entorno de baja presión afectiva a través de dispositivos móviles. Este estudio cuasiexperimental de métodos mixtos evaluó el impacto de la práctica oral autónoma con inteligencia artificial, realizada fuera del aula, sobre la competencia oral y la ansiedad lingüística de 129 estudiantes matriculados en ocho asignaturas universitarias de español en una institución de artes liberales en Pensilvania. El grupo experimental (n = 72) realizó sesiones semanales de práctica oral asistida por ChatGPT (versión gratuita, Voice Mode) en sus dispositivos móviles personales, con una duración mínima de treinta minutos por sesión fuera del horario de clase. La evidencia de práctica se documentó 4 mediante un protocolo dual de Canvas: reflexiones escritas semanales en los Foros y grabaciones de vídeo breves en las Tareas, canal este último que permite la descarga directa para la evaluación sistemática con rúbrica por parte de la investigadora principal. El grupo de control (n = 57) realizó tareas orales estructuradas equivalentes sin mediación de inteligencia artificial, documentadas a través de los mismos canales. La competencia oral se evaluó mediante rúbricas analíticas alineadas con los descriptores ACTFL, calificadas por la investigadora principal sobre el corpus total de las producciones, con salvaguardas explícitas frente al sesgo de expectativa (evaluación en orden aleatorio, no consulta de las puntuaciones del pretest durante la evaluación del postest e intervalo temporal entre sesiones de calificación). Con el propósito de triangular los hallazgos, el estudio integró medidas cuantitativas adicionales: la Escala de Ansiedad en el Aula de Lengua Extranjera (FLCAS), y un cuestionario ampliado del Modelo de Aceptación Tecnológica (TAM), administrado al grupo experimental al concluir la intervención. Los datos cualitativos se obtuvieron a partir de las publicaciones en los foros de Canvas, las grabaciones de vídeo enviadas mediante Tareas y entrevistas semiestructuradas con una submuestra intencional de doce participantes del grupo experimental (n = 12). Fundamentado en la teoría sociocultural de Vygotsky, la Hipótesis de la Interacción de Long, la Hipótesis del Output de Swain, la Hipótesis del Filtro Afectivo de Krashen y el Modelo de Aceptación Tecnológica, el estudio confirmó sus tres hipótesis. El grupo experimental obtuvo ganancias en competencia oral significativamente mayores que el grupo de control (d = 0,82, efecto grande), con las ganancias diferenciales más pronunciadas en comunicación interactiva y fluidez oral. La ansiedad ante la lengua extranjera se redujo un 18,3 % en el grupo experimental frente a un 3,4 % en el de control (d = 0,91), con las reducciones mayores entre los estudiantes con ansiedad inicial más alta. Las percepciones del TAM fueron uniformemente positivas (Utilidad Percibida M = 5,84/7; Facilidad de Uso Percibida M = 5,61/7), y la intensidad de práctica correlacionó con las ganancias en competencia oral (r = ,61 para la consistencia semanal de práctica). Entre las líneas de investigación futura se contemplan el análisis acústico mediante PRAAT, con el objeto de complementar la evaluación experta con métricas fonéticas objetivas, y la medición 5 sistemática de la Disposición a Comunicarse (Fathi, Rahimi y Derakhshan, 2024) en réplicas ampliadas.
Despite being the most widely studied foreign language in U.S. higher education, Spanish instruction often fails to produce functional oral proficiency due to limited individualized practice, high classroom heterogeneity, and elevated foreign language anxiety. Recent advancements in Large Language Models (LLMs) and Automatic Speech Recognition (ASR) offer a potential solution: conversational AI systems that provide adaptive, low-anxiety oral practice via mobile devices. This quasi-experimental, mixed-methods study evaluated the impact of student-driven, AI-mediated oral practice conducted outside the classroom on the oral proficiency and foreign language anxiety of 129 students across eight university-level Spanish courses at a liberal arts institution in Pennsylvania. The experimental group (n = 72) engaged in weekly AI-assisted oral practice using ChatGPT (free version, Voice Mode) on personal mobile devices, with a minimum of thirty minutes per session outside class hours. Practice was documented through a dual Canvas-based protocol: weekly written reflections submitted to Canvas Discussions and short video recordings submitted through Canvas Assignments, the latter channel enabling direct download for systematic rubric-based evaluation by the principal investigator. The control group (n = 57) performed equivalent structured oral tasks without AI mediation, documented through the same channels to maintain comparability. Oral proficiency was assessed through ACTFL-aligned analytic rubrics scored by the principal investigator over the full corpus of recordings, with explicit safeguards against expectancy bias (randomized scoring order, pretest scores withheld during posttest evaluation, and a time interval between scoring sessions). To triangulate findings, the study incorporated additional quantitative measures: the Foreign Language Classroom Anxiety Scale (FLCAS), and an expanded Technology Acceptance Model (TAM) questionnaire administered to the experimental group at the close of the intervention. Qualitative data were gathered through Canvas Discussion posts, learner video reflections submitted via Assignments, and semi-structured interviews with a purposive subsample of twelve experimental-group participants (n = 12). Grounded in Vygotsky's sociocultural theory, Long's Interaction Hypothesis, Swain's Output Hypothesis, Krashen's Affective Filter, and the Technology Acceptance Model (TAM), the study confirmed all three hypotheses. The experimental group achieved significantly greater gains in overall oral proficiency than the control group (d = 0.82, a large effect), with the most pronounced differential gains in interactive communication and oral fluency. Foreign language anxiety fell by 18.3% in the experimental group versus 3.4% in the control group (d = 0.91), with the largest reductions among initially high-anxiety students. TAM perceptions were consistently positive (Perceived Usefulness M = 5.84/7; Perceived Ease of Use M = 5.61/7), and practice intensity correlated positively with proficiency gains (r = .61 for weekly practice consistency). Future research directions include the integration of acoustic analysis via PRAAT to complement expert-rater assessments with objective phonetic metrics, and the systematic measurement of Willingness to Communicate (Fathi, Rahimi & Derakhshan, 2024) in extended replications.
Despite being the most widely studied foreign language in U.S. higher education, Spanish instruction often fails to produce functional oral proficiency due to limited individualized practice, high classroom heterogeneity, and elevated foreign language anxiety. Recent advancements in Large Language Models (LLMs) and Automatic Speech Recognition (ASR) offer a potential solution: conversational AI systems that provide adaptive, low-anxiety oral practice via mobile devices. This quasi-experimental, mixed-methods study evaluated the impact of student-driven, AI-mediated oral practice conducted outside the classroom on the oral proficiency and foreign language anxiety of 129 students across eight university-level Spanish courses at a liberal arts institution in Pennsylvania. The experimental group (n = 72) engaged in weekly AI-assisted oral practice using ChatGPT (free version, Voice Mode) on personal mobile devices, with a minimum of thirty minutes per session outside class hours. Practice was documented through a dual Canvas-based protocol: weekly written reflections submitted to Canvas Discussions and short video recordings submitted through Canvas Assignments, the latter channel enabling direct download for systematic rubric-based evaluation by the principal investigator. The control group (n = 57) performed equivalent structured oral tasks without AI mediation, documented through the same channels to maintain comparability. Oral proficiency was assessed through ACTFL-aligned analytic rubrics scored by the principal investigator over the full corpus of recordings, with explicit safeguards against expectancy bias (randomized scoring order, pretest scores withheld during posttest evaluation, and a time interval between scoring sessions). To triangulate findings, the study incorporated additional quantitative measures: the Foreign Language Classroom Anxiety Scale (FLCAS), and an expanded Technology Acceptance Model (TAM) questionnaire administered to the experimental group at the close of the intervention. Qualitative data were gathered through Canvas Discussion posts, learner video reflections submitted via Assignments, and semi-structured interviews with a purposive subsample of twelve experimental-group participants (n = 12). Grounded in Vygotsky's sociocultural theory, Long's Interaction Hypothesis, Swain's Output Hypothesis, Krashen's Affective Filter, and the Technology Acceptance Model (TAM), the study confirmed all three hypotheses. The experimental group achieved significantly greater gains in overall oral proficiency than the control group (d = 0.82, a large effect), with the most pronounced differential gains in interactive communication and oral fluency. Foreign language anxiety fell by 18.3% in the experimental group versus 3.4% in the control group (d = 0.91), with the largest reductions among initially high-anxiety students. TAM perceptions were consistently positive (Perceived Usefulness M = 5.84/7; Perceived Ease of Use M = 5.61/7), and practice intensity correlated positively with proficiency gains (r = .61 for weekly practice consistency). Future research directions include the integration of acoustic analysis via PRAAT to complement expert-rater assessments with objective phonetic metrics, and the systematic measurement of Willingness to Communicate (Fathi, Rahimi & Derakhshan, 2024) in extended replications.
Descripción
Categorías UNESCO
Palabras clave
inteligencia artificial, chatbots, español como lengua extranjera, competencia oral, ansiedad ante la lengua extranjera, CALL, aprendizaje de idiomas asistido por móvil, Canvas LMS, foros de Canvas, Tareas de Canvas, métodos mixtos, educación universitaria, artificial intelligence, oral proficiency, chatbots, Spanish as a Foreign Language, foreign language anxiety, CALL, mobile-assisted language learning, Canvas LMS, Canvas Discussions, Canvas Assignments, mixed methods, university education
Citación
Mula Garrido, Inés. Trabajo Fin de Máster: Eficacia de los chatbots de inteligencia artificial en el desarrollo de competencias orales del español como lengua extranjera: un estudio cuasiexperimental con estudiantes universitarios. Universidad Nacional de Educación a Distancia (España) 2026
Centro
Facultad de Filología
Departamento
Filologías Extranjeras y sus Lingüísticas



