ACE-LP: Augmenting Communication using Environmental Data to drive Language Prediction

Nonspeaking disabled people who use Voice Output Communication Aids (VOCAs) speak an average of between 8 and 12 words per minute (wpm) compared to typical adult speech of 150-190 wpm 1. For some, e.g., Prof Stephen Hawking who relies on a single switch access, rates fall to 1-2 wpm 2. Several research projects have demonstrated the potential of conversational language models to increase communication rates for personal narrative. However, system using these models require hand scripted paragraphs and training users to remember the existence and location of these 3. Other systems use data-to-text sentence generators to populate a narrative ontology with conversational
topics 4,5. The ACE-LP project which started in early 2016 aims to develop a communication system for nonspeaking people which will improve on the speed of communication by automatically populating the system with appropriate conversational items within an adaptive interface that provides control over timing and delivery whilst minimising physical and cognitive load. To achieve this, we will combine probabilistic language prediction with contextual data from sources such as video, audio and positional sensors, supplemented with user profiles and past experience. The ACE-LP project is in its very early stages and with this poster we will introduce our idea to seek discussion with and feedback from a specialist audience to inform our research and design.