The EuroCALL Review - Reflective Practice Paper - Automatically Generated Subtitles and Closed Captions: ESP Students’ Perceptions of Artificial Intelligence Hallucinations

Oct 15, 2026

Uploaded by Ophelia

An exploratory study re-evaluates automatically generated subtitles and closed captions used inside and outside the classroom in an ESP context. Data come from questionnaire responses from 28 Chemical and Process Engineering students learning English for Specific Purposes. The research centers on AI errors in captions and subtitles, examining how learners perceive hallucinations and deriving three learner profiles from their answers. Results indicate substantial learner autonomy in caption use while clarifying students’ beliefs about generative AI.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages17

Reading Time43 min

AccessFree to download

Abstract

1. Introduction

2. The study

2.1. Data collection and additional contexts

2.1.1. First-year students

The EuroCALL Review Volume 32, No. 2, 2025, 140-156 ISSN: 1695-2618 [https://doi.org/10.4995/eurocall.2025.23497](https://doi.org/10.4995/eurocall.2025.23497) REFLECTIVE PRACTICE PAPER Automatically Generated Subtitles and Closed Captions: ESP Students’ Perceptions of Artificial Intelligence Hallucinations Ana Bumber *; Denyze Toffoli ** Université de Toulouse, France *[ana.bumber@iut-tlse3.fr](ana.bumber@iut-tlse3.fr |)[ |](ana.bumber@iut-tlse3.fr |) **[denyze.toffoli@iut-tlse3.fr](denyze.toffoli@iut-tlse3.fr) How to cite this article: Bumber A. & Toffoli D. (2025) . Automatically Generated Subtitles and Closed Captions: ESP Students’ Perceptions of Artificial Intelligence Hallucinations. The EuroCALL Review, 32(2), 140-156. [https://doi.org/10.4995/eurocall.2025.23497](https://doi.org/10.4995/eurocall.2025.23497) Abstract This article is an exploratory study which re-evaluates the use of automatically generated subtitles and closed captions inside and outside the classroom with the help of questionnaire responses from twenty-eight students of Chemical and Process Engineering who were learning English for Specific Purposes (ESP) as part of their compulsory classes. By focusing on AI errors in closed captions and subtitles this research explores an area that is presently absent from the literature. This article gives an insight into how students perceive the AI errors or hallucinations they encounter and establishes three learner profiles based on their answers. The findings show a large degree of autonomy in students’ use of captions and subtitles while at the same time giving insight into their beliefs about generative AI. Keywords Artificial Intelligence hallucinations; AI-generated captions; subtitle use; students’ habits; English for Specific Purposes; scientific and technical English. 1. Introduction Platforms such as YouTube provide teachers of English for Specific Purposes (ESP) with authentic English content for formal in-class teaching. In addition, students who already practice English in informal settings outside the classroom, are exposed to online videos in English to an even greater extent (Sockett & Toffoli, 2020) . While YouTube hosts varied and authentic content in English, the subtitles and the closed captions (CCs) in its videos are often automatically generated. Vanderplank (2016, 2020) and Teng (2021) have largely established the relevance of subtitles and CCs as language learning aids. Their studies and those they cite have primarily examined vocabulary acquisition. Watching authentic content with captions has also been shown to have a positive impact on students’ confidence (Teng, 2021, p. 44) . In his 2016 book Vanderplank excluded automatically generated CCs as efficient learning tools because they contain errors: “Atthe time of writing, YouTube’s automatically generated closed captions tend to be rather error-prone and more of a hindrance than a help, but undoubtedly they will improve with time and further development”(2016, 16) . Automatically generated subtitles and CCs are trained to recognise spoken utterance thanks to Automatic Speech Recognition (ASR) which relies on Deep Neural Networks. With Open AI and the democratisation of different AI tools, these errors in automatically generated CCs have become known as hallucinations, which stem from ASR limitations. Despite constant improvements, ASR is still, in 2025, not one hundred percent accurate. Extensive studies cited by Vanderplank (2016, 2020) and later Teng (2021) exclude AI-generated subtitles as being useful. Despite this, studies related to informal learning (Toffoli, Sockett & Kusyk, 2023) were an incentive to further investigate the impact these errors have on students who are learning English. In line with research on informal language learning, encouraging online video use in ESP contexts could foster language acquisition specifically through fiction based on professional contexts and autonomy supportive

FAQs

What is the focus of the study on automatically generated captions?

The study focuses on AI errors in automatically generated subtitles and closed captions, and on how ESP students perceive these errors as hallucinations.

Who participated in the research?

The study used questionnaire responses from 28 students of Chemical and Process Engineering learning English for Specific Purposes as part of compulsory classes.

What do the findings indicate about students’ use of captions?

The findings show a large degree of autonomy in students’ use of captions and subtitles, while also revealing their beliefs about generative AI.

Oct 15, 2026

Uploaded by Ophelia