The AAC4ALL project will led to the development of LifeCompanion, an open software platform allowing the creation of specific modules for alternative and augmentative communication. The first version of LifeCompanion is already used in clinical practice.
The platform is accessible to different types of disabilities (motor, visual, cognitive, etc.) and is already used in many health structures (over 500 unique installations so far).
The platform will allow for :
The specifications and technical documentation of the platform are available on the LifeCompanion repository .
AAC systems incorporate ordinarily a word prediction module in order to speed up text input and thus communication. The benefit of word prediction depends largely on the system’s ability to adapt to the dialogue context and the user’s language. With this aim, we develop prediction engines adapted to children’s language, and more globally to the evolution of our language skills. These age-appropriate prediction engines will be developed and evaluated during the year 2023.
Software distribution – An adaptative word prediction engine is already integrated in the LifeCompanion platform. It will be improved in the first phase (2023) of the project.
Last results – Creation of a new French corpus for child language, enabling the traininf of specific child prediction model. With such an adapted model, the desired words are often proposed in a better position in the prediction list. This work was presented at the next HCI International conference (July 2023). We are currently working on the adaptation of the prediction to the current theme of the conversation. We are also investigating the integration of neural large language models in word prediction techniques.
Publications – Scientific presentation of our dedicated child prediction model : Ben Khelil C., Rayar F., Antoine J-Y., Hoiry M., Raynal M., Halftermeyer A. (2023) What you need is what you get: adapting word prediction of Augmentative and Alternative Communication aids to youth language. HCII International. Free version (pre-print) : https://hal.science/hal-04190258
Just like on a cell phone SMS editor, the prediction module of an AAC system displays, after each new letter input, the list of the most probable words being typed. The prediction is based on the preceding words but also on the letters already entered in the current word. Prediction may be seriously affected by the occurrence of spelling or grammatical errors. This problem quickly becomes insurmountable in case of DYS disorders (dyslexia in particular) or when the user has limited language skills. We will develop a spell correction module that adapts to the person’s language skills and/or disorders in order to improve text entry. This correction will work on the fly while typing. However, it must also be able to work in verification mode at the end of the word input (spell checking). We will particularly focus on the writings of people suffering from language disorders (DYS or other). Software distribution – A first spell correction model is already integrated in the LifeCompanion platform. Now speech therapists are investigating the ways to adapt correction rules to the needs of any specific user.
Publications – No publication yet ! Your current works are inspired by our critical analysis of the state of the art and some experiments we conducted. This work is presented here : Antoine Jean-Yves et al. (2018) Ma copie adore le vélo : analyse des besoins réels en correction orthographique sur un corpus de dictées d’enfants. Actes TALN’2019, Rennes, France. [HAL-02375246]
Design and implementation of experimental protocols in psycholinguistics and neurolinguistics (methodologies: EEG, Oculometry) on the processing of pictograms in patients with impaired communication (language disorders, cognitive disorders, motor disorders, etc)Publication of results: articles and scientific communications.Publication of protocols and data sets (stimuli, programs, data) following FAIR principles.
Translated with www.DeepL.com/Translator (free version)