In an increasingly information-driven world, ensuring equitable access to this information is a right, particularly for individuals with cognitive impairments, intellectual disabilities, older adults, those with low literacy levels, and non-native language speakers. The HULAT research group has addressed this need by developing a suite of tools and resources aimed at content simplification, thereby facilitating understanding and access to critical information in areas such as e-government, health, and education. This article introduces two main tools: EASIER, focused on lexical simplification, and Access2meet, an accessible videoconferencing solution, alongside the EASIER corpus, which has been enriched through annotation and validation by experts and individuals with intellectual disabilities. These technologies incorporate Natural Language Processing (NLP) and Artificial Intelligence (AI) approaches and specific datasets in easy reading, highlighting the importance of direct participation by disabled individuals in the development of accessible technologies.
tool for Intellectual Disability for each identified complex word, including synonyms, definitions, and pictograms tailored to the word’s context. These invaluable resources are consolidated into a The Easier system 23 provides a specialized tool designed Glossary facility specifically aimed at supporting people to enhance the understanding and readability of text con- with intellectual disabilities. tent for individuals with intellectual disabilities. This The foundation of the Easier system is a robust maapproach combines methods from the disciplines of Arti- chine learning model that has been extensively trained ifcial Intelligence (AI) and Human-Computer Interaction using a wide array of texts annotated by experts in Spanto generate accessible interfaces with simplified texts. ish Easy Reading and Plain Language and validation with Specifically, the system employs Natural Language Pro- individuals with disabilities [1] as well as the use of the cessing (NLP) techniques and resources from the domains EASIER corpus that is described in section 3. of Easy Reading and Plain Language to assess and classify To ensure the generation and selection of contextually words in Spanish texts as complex or straightforward. appropriate synonyms, a comprehensive suite of tools,
The process begins when a user inputs Spanish text including dictionaries, semantic similarity techniques, into a textbox. The Easier system identifies complex and paraphrasing resources, was developed. Additionwords within the text, highlighting them for easy recogni- ally, for the critical task of providing accurate definitions tion (see Figure 1). Upon hovering over these highlighted for complex words within their specific contexts, a disamwords, the system displays a tooltip that suggests a sim- biguation process using the BERT (Bidirectional Encoder pler replacement or synonym, efectively aiding the user Representations from Transformers) word embedding in improving their reading and comprehension skills. To model was employed [2]. further assist, the system provides a variety of resources The Easier system draws its definitions from authoritative sources such as the "Diccionario Fácil"4 (Easy Dictionary) for Easy Reading definitions and the Royal Spanish
3https://github.com/LURMORENO/easier
Academy (RAE) dictionaries 5. For the visual represen- independence, with an average of 15 sentences per docutation of concepts, pictograms supplied by ARASAAC 6 ment, rendering the EASIER corpus a highly adaptable are utilized, further enhancing the system’s accessibility. resource. It includes 260 documents with 8,155 complex
In designing the Easier user interface, Web Content words and 5,130 words associated with context-aware Accessibility Guidelines (WCAG) and the guidelines es- synonyms, all annotated in accordance with easy-to-read tablished by the “Cognitive and Learning Disabilities Ac- standards. cessibility Task Force (W3C-COGA)” were followed [3]. To facilitate the annotation process, sixteen guidelines Additionally, specific Design Patterns were employed to were established to diferentiate between complex and optimize the system’s design for cognitive accessibility, simple words, thereby enhancing the corpus’s utility and ensuring that it meets the highest standards of accessibil- supporting subsequent annotations. The criteria for idenity and usability [4]. The web user interface of the Easier tifying complex words encompass uncommon vocabusystem boasts a responsive design, making it accessible lary, syllable complexity, technical jargon, and abstract on a wide range of devices, from desktop computers to terms. Utilizing their extensive experience in converting mobile devices. Moreover, to extend the system’s utility, conventional texts into formats that are easier to read, the extensions for Chrome and Mozilla browsers have been annotators manually adapted the texts through a methoddeveloped. ology that actively involved individuals with intellectual
The EASIER system was evaluated by individuals with disabilities throughout the process. intellectual disabilities and older people. Fifty partici- The validity of the corpus was verified through an pants were recruited in partnership with an organization inter-annotator agreement test, resulting in a Fleiss serving people with intellectual disabilities. The evalua- Kappa coeficient of 0.641, which signifies a moderate tion indicated that the chosen texts were more challeng- level of agreement among annotators. Furthermore, a ing for those with intellectual disabilities than for older qualitative evaluation with 45 users, including those with adults, suggesting the proposal was more beneficial for intellectual disabilities and older adults, demonstrated the intellectually disabled group [5]. moderate efectiveness in complex word identification and considerable approval for the suggested synonyms.
The proliferation of accessible information presents challenges for individuals with ageing-related or intellectual disabilities, as well as for non-native speakers who encounter dificulties in reading comprehension. To address these challenges, text simplification emerges as a crucial tool for improving information accessibility. The application of Natural Language Processing (NLP) techniques, often involving machine learning, requires resources such as corpora for the development and validation of lexical simplification methods. Motivated by this need, the EASIER Corpus 7 has been developed as a resource aimed at fostering lexical simplification strategies for Spanish texts without being confined to any specific domain [6]. This corpus, resulting from the collaboration of three expert Spanish-speaking annotators from Plena Inclusión Madrid 8 and Grupo Amas Fácil 9, comprises 260 diverse news articles from the "60 y más" magazine 10. These articles span a broad array of topics and maintain domain 5https://www.rae.es/ 6https://arasaac.org/ 7https://github.com/LURMORENO/EASIER_CORPUS 8https://plenainclusionmadrid.org/ 9https://amasfacil.org/ 10https://imserso.es/el-imserso/documentacion/publicaciones/ publicaciones-periodicas/60-mas
Since the COVID-19 pandemic began, video conferencing tools have become deeply embedded in our daily routines, emphasizing their role as vital conduits for social inclusion. Research by Transparency Market Research forecasts that the value of the video conferencing market is set almost to double its 2019 figure, reaching an estimated $11.56 billion by 2027. However, concerns have arisen regarding the accessibility of these tools for people with disabilities. Studies highlight a significant gap in accessibility for these individuals [7] [8] [9]. These technologies involve interaction with people through user interfaces, and this implies designing accessible user interfaces that ensure access and management of systems and their contents for all people regardless of their visual, auditory, cognitive, or motor abilities. Videoconferencing systems should be designed following accessibility standards and User Experience (UX) requirements to ensure access to people with disabilities. Intuitive user interfaces, text simplification, and summarization services, and subtitling are essential accessibility requirements that must be met.
ACCESS2MEET is a project focused on the application of accessibility techniques in designing interactive video conferencing systems for all users, including those with disabilities. This interdisciplinary research draws on Natural Language Processing (NLP) and Human-Computer These requirements and their design patterns have reInteraction (HCI), particularly addressing the complexity sulted in the development of the Access2meet accessible of medical texts for patients. To design the user interface videoconferencing tool, tailored for medical consultaof a videoconferencing system, first, an exhaustive review tions within the Spanish healthcare domain, which is of accessibility requirements was carried out following showcased in this demonstration (see Figure 2 and Figure standards, and based on these requirements, design pat- 3). terns were defined [ 10]. From the author’s prior research, In this demonstration, we will focus on describing the platform was designed to comply with standards such aspects of the real-time subtitling service that can be actias WCAG, UAAG, and EN 301549 [11], with essential re- vated during a medical appointment involving the doctor, quirements including: nurse, patient, family member, and, if necessary, a sign language interpreter, as well as the text simplification • An accessible interface for magnifiers for individ- services to facilitate the readability and understanding uals with low vision, screen readers for the blind, of health information. and device-independent input access for people Access2meet is an innovative platform architected with physical-motor disabilities. for accessibility that integrates technology with user• Subtitling services for those with hearing impair- centered design to deliver an accessible videoconferencments, with the option for sign language inter- ing experience. Its backend is powered by Express.js and preter participation. enriched by WebRTC for real-time video interactions, • An interface that is intuitive and cognitively ac- along with socket.io for instantaneous data exchange. cessible for people with cognitive challenges. Python APIs facilitated by the Django Rest Framework Also, it was required that the content and infor- are employed for AI features. The platform distinguishes mation exchanged during meetings be accessible. itself by ofering integrated automatic subtitling, utilizing To address this: AI-driven speech-to-text conversion, and a text simplifi– The text content must be simplified into cation service that processes complex medical documents plain language. through AI and NLP techniques. The frontend, crafted – Give the possibility of providing sum- with React, ensures barrier-free user engagement thanks maries as a meeting outcome. to the accessible components developed in the project. It – Provide the possibility of simplifying the features real-time subtitle displays during video calls and chat messages to plain language. user interfaces that facilitate interaction with simplified texts. Users have access to a straightforward toggle option for subtitles and can efortlessly upload documents for simplification. Smooth data flow is paramount, with Socket.io ensuring prompt backend-to-frontend communication for subtitle and text updates. Regarding the subtitling and simplification services: • Simplification module : This module is in charge of translating a medical note elaborated by the doctor into a clearer and simpler one, following plain language and easy-to-read guidelines. It is based on current transformer architectures, particularly decoder-based models. There are not any Spanish generative models, and consequently, we have adopted a multilingual Llama2 open-source, [12] from Huggingface (Llama-2-7b-hf). Llama2 has shown considerable potential when performing dificult processing tasks requiring specialized knowledge in a variety of domains, including generating text from commands and instructions. Apart from fine-tuning with a collection of clinical cases and its easy-to-read simplifications, several prompts have been tested to translate text input to simplified content. These approaches have been described in [13]. • Subtitling module: The integration of real-time transcription and captioning functionalities addresses the needs of people with hearing impairments, ensuring access to spoken content during videoconference. By providing synchronized text-based representations of verbal communication, this feature enhances comprehension and retention for patients. The system has been developed using Node.js and boasts a comprehensive API comprising REST and Web Sockets endpoints.
The Web Sockets API serves a critical role in providing live transcriptions for the incoming audio of sessions in real-time, leveraging the streaming recognition functionality of Google Cloud Speech API. Consequently, each audio stream has a dedicated streaming recognizer and a corresponding Web Socket to facilitate seamless data transmission and reception. Before transmitting the transcriptions, the output from the Speech API undergoes further processing through an in-house trained Transformers model, thereby automating the punctuation of the transcribed text. This step is of paramount importance, given that the Speech API’s punctuation functionality yields limited results for the Spanish language.
To include punctuation in transcriptions, we experimented [14] with various encoder-based pretrained models using Spanish-only models, particularly RoBERTa-bne-base [15] that has been trained with corpora from the Spanish National Library at the Barcelona Supercomputer Center.
As the next step, we are developing an advanced tool built on the integrated services of the EASIER system aimed at helping professionals who specialize in adapting texts to Easy-to-Read and Plain Language standards. This tool will provide a comprehensive suite of lexical resources and support functionalities designed to streamline the text adaptation process.
Moreover, we are testing the Access2meet tool with users, including those with cognitive impairments and other disabilities, healthcare professionals, and relatives, to assess its real-world eficacy and inform further development. Simultaneously, we’re enhancing the platform to create easy summaries from subtitle transcriptions and to simplify chat interactions with plain language and emojis, thereby improving digital communication accessibility for all users. video conferencing platforms. universal access in the information society (in press) (2024). [8] M. Hersh, B. Leporini, M. Buzzi, et al., Accessibility evaluation of video conferencing tools to support disabled people in distance teaching, meetings and other activities, in: Future Perspectives of AT, eAccessibility and eInclusion, ICCHP, 2020, pp. 133– 139. [9] J. Boland, S. Banks, R. Krabbe, S. Lawrence, T. Murray, T. Henning, M. Vandenberg, A covid-19-era rapid review: using zoom and skype for qualitative group research., Public Health Research & Practice 32 (2022). [10] L. Moreno, P. Martínez, R. Alarcon, Requirements and design patterns for an accessible video conferencing tool, in: Proceedings of the XXII International Conference on Human Computer Interaction, 2022, pp. 1–9. [11] CEN, CENELEC, ETSI, Accessibility requirements for ict products and services, 2021. URL: https: //www.etsi.org/deliver/etsi_en/301500_301599/ 301549/03.02.01_60/en_301549v030201p.pdf . [12] H. Touvron, L. Martin, K. Stone, P. Albert, A. Almahairi, Y. Babaei, N. Bashlykov, S. Batra, P. Bhargava, S. Bhosale, et al., Llama 2: Open foundation and finetuned chat models, arXiv preprint arXiv:2307.09288 (2023). [13] P. Martínez, A. Ramos, L. Moreno, Exploring large language models to generate easy-to-read content. [1] R. Alarcon, L. Moreno, P. Martínez, Lexical simpli- artificial intelligence and digital accessibility (subifcation system to improve web accessibility, IEEE mitted), Frontiers in Computer Science (2024).
Access 9 (2021) 58755–58767. [14] M. Perez-Enriquez, J. Masiello-Ruiz, J. Lopez[2] R. Alarcon, L. Moreno, P. Martínez, Word-sense Cuadrado, P. M. I. Gonzalez-Carrasco, B. Ruizdisambiguation system for text readability, in: Pro- Mezcua, Automatic punctuation model for spanish ceedings of the 9th International Conference on live transcriptions, LREC-COLING 2024 (2024). Software Development and Technologies for En- [15] A. Gutiérrez-Fandiño, J. Armengol-Estapé, hancing Accessibility and Fighting Info-exclusion, M. Pàmies, J. Llop-Palao, J. Silveira-Ocampo, C. P. 2020, pp. 147–152. Carrino, A. Gonzalez-Agirre, C. Armentano-Oller, [3] W3C, W3c web accessibility initiative (wai), 2019. C. Rodriguez-Penagos, M. Villegas, Maria: Spanish
URL: https://www.w3.org/WAI/. language models, arXiv preprint arXiv:2107.07253 [4] L. Moreno, H. Petrie, P. Martínez, R. Alarcon, De- (2021).
signing user interfaces for content simplification aimed at people with cognitive impairments, Universal Access in the Information Society (2023) 1– 19. [5] R. Alarcon, L. Moreno, P. Martínez, J. A. Macías,
Easier system. evaluating a spanish lexical simplification proposal with people with cognitive impairments, International Journal of Human–Computer
Interaction 40 (2024) 1195–1209. [6] R. Alarcon, L. Moreno, P. Martínez, Easier corpus:
A lexical simplification resource for people with cognitive impairments, Plos one 18 (2023) e0283622. [7] L. Moreno, A. Diaz-Redondo, P. Martínez, H. Ochoa,
Assessing standards-driven accessibility in top