Date and time: 29/01/26, 5pm CET
Speaker: Fernando Alva Manchego, School of Computer Science and Informatics, Cardiff University
Presenter: Paloma Martínez, Carlos III University
Abstract:
Automatic text simplification aims to make information more accessible without altering its original meaning. In this talk, I will focus on simplification with readability control, that is, on how to adapt texts to readers’ different levels of linguistic proficiency.
I will present the main challenges and advances in readability control, including how to represent and predict complexity levels using readability metrics and models. I will discuss approaches for incorporating this control into automatic simplification systems, ranging from models specifically trained to adjust to different readability levels, to the use of Large Language Models via prompting strategies.
I will also analyse the evaluation challenges in this context, such as balancing simplicity, meaning preservation, and fluency, as well as the limitations of traditional evaluation metrics. I will base the discussion on recent empirical results, including trends observed in the TSAR 2025 shared task and comparisons between different readability control methods.
The talk will conclude with a discussion on the potential of these technologies in real-world applications, such as education and public communication, and on the outstanding challenges in developing simplification systems that adapt effectively and responsibly to their readers.
Bio:
Fernando Alva Manchego is a Lecturer at Cardiff University, UK, where he is a member of the Natural Language Processing group. His research focuses on artificial intelligence technologies applied to education and information accessibility, with particular attention to automatic text generation (such as text simplification, machine translation, and summarisation), natural language generation evaluation, and writing assistants.
His work encompasses the creation of linguistic resources, the design of evaluation methodologies and metrics, and the implementation of machine learning models aimed at improving information clarity and readability for diverse audiences. He has contributed to projects and publications on readability-controlled text simplification, the evaluation of automatic metrics, and corpora and tools for NLP tasks.
Link to the talk: https://zoom.us/webinar/register/WN_aGqoWVQJQxycg-7O40Ua4w
