Diseño de prompt y calidad del feedback correctivo automatizado de ChatGPT en un contexto de EAP.
Loading...
Date
2026
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad de Concepción
Abstract
En la presente tesis se tiene por objetivo analizar la relación entre el diseño del prompt y las características del feedback correctivo escrito (FCE) metalingüístico directo generado por ChatGPT en ensayos argumentativos académicos producidos por estudiantes de inglés como segunda lengua en un contexto universitario de inglés para propósitos académicos (EAP).
Para ello, se desarrolló un estudio cuantitativo descriptivo con alcance exploratorio, estructurado en dos etapas. En la primera, se generó FCE a partir de dos condiciones de prompt (elaborado y general) aplicadas a un corpus de 30 ensayos argumentativos escritos por estudiantes de nivel B2 del programa de pedagogía en inglés de la Universidad de Concepción, enmarcados en el proyecto FONDECYT 1230706. El prompt elaborado fue diseñado mediante un proceso iterativo fundamentado en la literatura sobre FCE, que incluyó la revisión del corpus, la identificación de brechas lingüísticas y el refinamiento progresivo de las instrucciones. En la segunda etapa, el feedback generado fue evaluado por una muestra de seis profesores de inglés en ejercicio en el sistema educativo chileno, mediante una rúbrica de siete ítems distribuidos en tres criterios: pertinencia, comprensibilidad y precisión. Los datos fueron analizados mediante tablas de frecuencias y la prueba Chi-cuadrado de Pearson.
Los resultados indican la existencia de una relación estadísticamente significativa entre el tipo de prompt y las valoraciones del FCE generado (χ² = 23,560, gl = 2, p < 0,05). El prompt elaborado concentró un 88,10% de valoraciones Logrado frente al 69,05% del prompt general, con diferencias más pronunciadas en los criterios de pertinencia y comprensibilidad que en el de precisión. Se concluye que el nivel de especificidad del prompt constituye una posible variable relevante en la calidad del FCE generado por ChatGPT, y que el diseño informado de prompts puede representar una competencia metodológica importante para la integración efectiva de herramientas de inteligencia artificial generativa en la enseñanza de la escritura académica en inglés como segunda lengua.
This thesis work aims to analyze the relationship between prompt design and the characteristics of direct metalinguistic written corrective feedback (WCF) generated by ChatGPT in academic argumentative essays produced by English as a second language students in a university-level English for Academic Purposes (EAP) context. A quantitative descriptive study with exploratory scope was developed, structured in two stages. In the first stage, WCF was generated using two prompt conditions — elaborated and general — applied to a corpus of 30 argumentative essays written by B2-level students from the English teacher education program at the Universidad de Concepción, within the framework of FONDECYT project 1230706. The elaborated prompt was designed through an iterative process grounded in WCF literature, which included corpus revision, identification of linguistic gaps, and progressive refinement of the instructions. In the second stage, the generated feedback was evaluated by a sample of six in-service English teachers from the Chilean educational system, using a seven-item rubric distributed across three criteria: pertinence, comprehensibility, and precision. Data were analyzed using frequency tables and Pearson's Chi-square test. Results indicate the existence of a statistically significant relationship between prompt type and WCF ratings (χ² = 23.560, df = 2, p < .05). The elaborated prompt concentrated 88.10% of Achieved ratings compared to 69.05% for the general prompt, with more pronounced differences in the pertinence and comprehensibility criteria than in precision. It is concluded that the level of prompt specificity could constitute a relevant variable in the quality of ChatGPT-generated WCF, and that informed prompt design might represent an important methodological competence for the effective integration of generative artificial intelligence tools in the teaching of academic writing in English as a second language.
This thesis work aims to analyze the relationship between prompt design and the characteristics of direct metalinguistic written corrective feedback (WCF) generated by ChatGPT in academic argumentative essays produced by English as a second language students in a university-level English for Academic Purposes (EAP) context. A quantitative descriptive study with exploratory scope was developed, structured in two stages. In the first stage, WCF was generated using two prompt conditions — elaborated and general — applied to a corpus of 30 argumentative essays written by B2-level students from the English teacher education program at the Universidad de Concepción, within the framework of FONDECYT project 1230706. The elaborated prompt was designed through an iterative process grounded in WCF literature, which included corpus revision, identification of linguistic gaps, and progressive refinement of the instructions. In the second stage, the generated feedback was evaluated by a sample of six in-service English teachers from the Chilean educational system, using a seven-item rubric distributed across three criteria: pertinence, comprehensibility, and precision. Data were analyzed using frequency tables and Pearson's Chi-square test. Results indicate the existence of a statistically significant relationship between prompt type and WCF ratings (χ² = 23.560, df = 2, p < .05). The elaborated prompt concentrated 88.10% of Achieved ratings compared to 69.05% for the general prompt, with more pronounced differences in the pertinence and comprehensibility criteria than in precision. It is concluded that the level of prompt specificity could constitute a relevant variable in the quality of ChatGPT-generated WCF, and that informed prompt design might represent an important methodological competence for the effective integration of generative artificial intelligence tools in the teaching of academic writing in English as a second language.
Description
Tesis presentada para optar al grado de Magíster en Lingüística Aplicada.
Keywords
Prompts (Education), ChatGPT, Inteligencia artificial generativa, Inglés Enseñanza Hablantes extranjeros