The complexity of official and digital texts has become a barrier for millions of people. Easy Read, based on UNE Standard 153101:2018 EX, aims to remove these difficulties. Artificial intelligence (AI) is emerging as a tool to streamline the adaptation of content into more understandable versions.
The study, conducted by 31 working groups from the UPM, analyzed how chatbots detect comprehension obstacles and adapt texts. Results indicate that, on average, they correctly identified 68% of cognitive accessibility issues. Accuracy varied by guideline evaluated, excelling in detecting gerunds and adverbs ending in "-mente" (80%), while passive voice was identified 70% of the time.
However, chatbots showed weaknesses in visual design, detecting only 38% of issues like the use of "etc." or ellipsis, and a similar percentage for the semicolon. This is because AI processes text as abstract sequences, without perceiving the document's spatial layout. For instance, starting each sentence on a new line achieved only a 43% success rate.
Regarding text adaptation, chatbots removed key context such as figures and acronyms instead of explaining them, generating artificial texts. Although "mathematically" easy, they are not always comprehensible to a person.
To overcome these limitations, teams proposed a three-phase workflow: preprocessing to fragment the document, interpretation of complex concepts by a model like Gemini, and recombination and style adjustment by another model like ChatGPT, followed by mandatory human review.
Javier Manobanda from the UPM concludes that general chatbots do not replace cognitive accessibility professionals but act as assistants to expedite preliminary analysis and the initial adaptation of large content. Researcher Mari Carmen Suárez de Figueroa, head of the study, emphasizes that human validation remains essential, as "making the complex simple continues to be one of the greatest communication challenges."
The need for specific tools like FACILE, an AI-based tool developed at UPM by the Ontology Engineering Group (OEG), is highlighted.
