S01-02 09

Identifying abstracts written by ChatGPT: Scientific English students’ challenge


Deja tu comentario

Participa en esta ponencia enviádole tu pregunta o comentario a los autores

Añadir comentario


profile avatar
Natalia Mora LópezUniversidad Complutense de Madrid


1. Introduction

The advent of ChatGPT, a language model developed by OpenAI, has made it easier to create texts automatically by asking this artificial intelligence to write a specific text following some indications given in a prompt. Users must only think of the characteristics of the text and the topic and enter their prompt. Also, after a text is generated, changes can be requested if the product is not totally what they wanted.

This practice can be used in any domain, including the teaching and learning of a foreign language. Contrasting real and AI-generated texts is helpful for students to reflect on the different features a text should have to correctly follow genre conventions. Also, it may help students improve their own texts by finding differences and similarities between the generated texts and their non-native texts.

2. Objective and research questions

For the reasons presented above, this presentation aims to determine how accurately English students can distinguish between a scientific abstract written by ChatGPT or a human. Additionally, some research questions will be answered: (1) How many students identify the abstract written by ChatGPT?; (2) What are the main reasons students provide to support their choice?; (3) What differences do students find between the abstract generated by ChatGPT and their own abstracts?

3. Methodology

To achieve the objective and answer the research questions, 62 students who were taking the course Introduction to Scientific English (optional course in the first year of the degree in Optics and Optometry, at the Complutense University of Madrid) were presented two texts, one generated by ChatGPT and one belonging to a real journal published online, and asked to identify the one generated by the former and to support their choice with a justification based on the features of the text.

4. Results and discussion

The results showed that 76% of the students were able to identify the text generated by ChatGPT correctly. Some of the reasons given to support their correct choice regarding the authorship of ChatGPT were that it uses a more general language regarding the content of the study the abstract is supposed to summarise. However, some reasons given by those who thought that the same text was written by a human were that it followed language rules (there were no mistakes in the writing), it was well written regarding scientific style, and some inner structure could be observed.

5. Conclusions

In conclusion, the utilization of ChatGPT in generating texts presents a promising avenue for various domains, including language education. By contrasting AI-generated texts with those crafted by humans, students can gain valuable insights into genre conventions and refine their own writing skills. This study underscores the importance of student familiarity and acclimation to such AI tools to fully capitalize on their potential benefits. Further exploration into the pedagogical usefulness of AI-generated texts could enhance student learning experiences in scientific writing courses.

Preguntas y comentarios al autor/es

    • profile avatar

      Mateusz Pietraszek

      Comentó el 14/03/2024 a las 19:38:19

      Congratulations on your presentation! This is indeed a very compelling topic. You mention a task where students improve an existing text using AI. Have you got any ideas as to how this type of activity could be assessed? Thanks again!

      • profile avatar

        Natalia Mora López

        Comentó el 15/03/2024 a las 15:36:26

        Hi Mateusz,
        Thank you for your comment and your interest. To assess the activity, I think it's important to have a clear idea of the objective students must achieve. A rubric may be created including the aspects that the assessment might focus on, such as:
        - The improved text shows no/few/some/many changes with respect to the original text: in this case we're measuring how much/how well the tool has been used.
        - The improved text presents no/few/some/many mistakes that were or were not in the original text: in this case we're measuring the writing skill.
        Hope these ideas are useful.

    • profile avatar

      Iratxe Serna Bermejo

      Comentó el 14/03/2024 a las 15:36:12

      Hello Natalia,
      congratulations on your presentation. Lately I hear many teachers say that they no longer ask their students for written texts for fear of them using ChatGPT. Do you think that activities to work on written skills will decrease for this reason? What advice would you give to these teachers so that they do not completely rule out these types of activities?
      Thank you,
      Iratxe Serna Bermejo.

      • profile avatar

        Natalia Mora López

        Comentó el 14/03/2024 a las 17:24:58

        Hi Iratxe,
        Thank you for your words. I think writing is a basic skill that every human should master for communication purposes. Even if we can use tools to help us do so, humans should still acquire enough knowledge to judge whether what they receive from the tool is what they need, etc. or it's not. So even if you don't write a text by yourself, you must have the knowledge needed to verify that the text is correct before using it.

    • profile avatar

      Maria C. Bravo

      Comentó el 14/03/2024 a las 15:05:04

      Hi Natalia! This is a very interesting and current topic. I am curious at to whether ChatGPT generates/provides sources/references in an academic text such as an abstract.
      Thank you.

      • profile avatar

        Natalia Mora López

        Comentó el 14/03/2024 a las 15:15:54

        Hi Maria,
        Thank you for your question. ChatGPT definitely knows what references are in scientific writing and their different formats. And it may well introduce them in a text. I'll copy here a similar abstract to the one used in my presentation, after asking ChatGPT to include references in it using APA format.

        -------------ChatGPT's text-------------
        This paper thoroughly examines the multifaceted applications of Hyaluronic Acid (HA) in both ophthalmology and contact lenses, drawing on recent scientific literature. HA, renowned for its biocompatibility and viscoelastic properties, plays a pivotal role in ocular health and lens technology. Studies such as those by Dogru et al. (2014) and Liu et al. (2019) demonstrate HA's efficacy in promoting corneal wound healing and alleviating dry eye symptoms, respectively. Furthermore, research by Maïssa et al. (2019) highlights HA's impact on enhancing the comfort and lubrication of contact lenses. Additionally, investigations into HA's potential as a vehicle for sustained drug delivery in ocular therapies, as discussed in the work of Galante et al. (2018), further underscore its versatility and promise. By synthesizing these findings, this comprehensive review elucidates how Hyaluronic Acid serves as a vital component in advancing ophthalmic care and optimizing contact lens performance.

        Dogru, M., Katakami, C., Inoue, M., & Tear Film & Ocular Surface Society Dry Eye Workshop Subcommittee. (2018). Tear function and ocular surface changes in noninsulin-dependent diabetes mellitus. Ophthalmology, 105(6), 1004-1009.
        Liu, H., Begley, C., Chen, M., Bradley, A., Bonanno, J., McNamara, N. A., & Nelson, J. D. (2009). A link between tear instability and hyperosmolarity in dry eye. Investigative ophthalmology & visual science, 50(8), 3671-3679.
        Maïssa, C., Guillon, M., & Garofalo, R. J. (2019). Contact lens wear affects tear film evaporation. Eye & contact lens, 45(2), 100-107.
        Galante, J., Holgado, M. A., Arias, J. L., & Alonso, M. J. (2018). Polysaccharide hydrogels for modified release formulations. Journal of Controlled Release, 53(3), 235-259.

        Now the problem is that, even though some sources are correct, some others include incorrect information (e.g. although the title and the paper may exist, the author may be a different one). So, as usual, you can't take ChatGPT's answer straight away but it needs human supervision.

        • profile avatar

          Maria C. Bravo

          Comentó el 14/03/2024 a las 15:33:18

          Thank you, Natalia. That is indeed impressive. It provides the human researcher with a good starting point.

    • profile avatar

      Víctor Álvarez Rodríguez

      Comentó el 14/03/2024 a las 11:59:49

      Estimada Natalia,
      En primer lugar, felicidades por tu interesante ponencia. Me parece especialmente relevante la visión que realizas sobre el tema. Como profesor de comunicación, me encuentro ante una situación que me provoca muchas dudas ¿Cómo crees deberíamos afrontar el uso de las IAs por parte del alumnado en el aula? Por un lado, pienso que deben desarrollar sus destrezas como hasta ahora. Eso les hará competitivos y aportar criterio. Pero por otro, no podemos negar la realidad a la que se enfrentan. Muchas gracias.

      • profile avatar

        Natalia Mora López

        Comentó el 14/03/2024 a las 15:24:10

        Hola, Víctor:
        Muchas gracias por tu comentario y tus palabras. Para responder a tu pregunta, creo que ante todo lo importante es tener claro cuál es el objetivo que tenemos en nuestras clases: ¿qué queremos que aprendan los alumnos y que sean capaces de hacer cuando se marchen? Creo que si los objetivos de nuestras clases y el tipo de actividades que hacemos para que consigan esos objetivos están acordes a las necesidades sociales y laborales actuales y se fomenta el pensamiento crítico, no podrán utilizar las IA para que trabaje por ellos. Así que desde mi punto de vista, la forma de afrontar su uso sería tomando la IA como una herramienta más (como podría hasta ahora ser un motor de búsqueda o un libro donde viene información), y dedicar más tiempo a las tareas que requieran una transformación o adaptación del contenido a ciertos contextos o supuestos, o un acercamiento crítico al contenido en sí.

Deja tu comentario

Lo siento, debes estar conectado para publicar un comentario.


Egregius Ediciones


Configuración de Cookies

Utilizamos cookies para mejorar su experiencia y las funcionalidades de esta web. Ver política de cookies