SEPLN 2023: a model for machine-generated text detection

Last September, Advacheck team participated in the SEPLN 2023 Conference. SEPLN, the Spanish Society for Natural Language Processing, hosted its XXXIX International Conference in Jaén, Spain. As a pioneering company specializing in detecting machine-generated text, Advacheck took part in the conference and came up with a model and tools for machine-generated text detection in documents from multiple domains.

Within the SEPLN 2023 congress, Advacheck actively engaged in the IberLEF 2023. IberLEF, a shared evaluation campaign of Natural Language Processing systems in Spanish and other Iberian languages, featured various tasks aimed at advancing research in text processing, understanding, and generation. Among these tasks was the AuTexTification competition, focusing on the detection and attribution of machine-generated text in multiple domains.

Advacheck’s participation in the AuTexTification competition yielded significant insights and contributions to the field. Our team presented a novel approach, outlined in a comprehensive research paper. This paper details a model for machine-generated text detection, leveraging various BERT-like encoder models, including XLM-RoBERTa, mDeBERTa, and MiniLM-V2. Through extensive preprocessing and data expansion techniques, we achieved remarkable results, enhancing the quality of text detection.
The competition results showcased Advacheck’s commitment to innovation and excellence. Our approach enabled us to achieve approximately 66% accuracy on the English binary dataset, demonstrating the effectiveness of our methodology in addressing the challenges posed by machine-generated text.

SEPLN’s commitment to fostering research and collaboration in Natural Language Processing resonates with Advacheck’s mission to combat inappropriate borrowing in academic and scholarly papers and participate in the development of ethical standards in the use of machine-generated content. By participating in events like IberLEF 2023, we contribute to the advancement of NLP technologies and reinforce our position as a leader in text detection solutions.

Experience Advacheck with a 14-day FREE trial!

  • Personal consultation on how the system works.
  • Access to a demo account tailored to your needs.
  • The top-notch plagiarism detection experience is guaranteed.