Computational optimization and prediction strategies for increasing communication rate in phoneme-based augmentative and alternative communication (AAC)
MetadataShow full item record
Up to 1.2% of the population is unable to meet daily communication needs using typical speech and may use augmentative and alternative communication (AAC) strategies to communicate, including manual sign language, facial gestures, and aided strategies such as selecting targets on an onscreen keyboard. However, for individuals whose impairments affect both speech and non-speech motor systems (e.g., spinal cord injury, amyotrophic lateral sclerosis, multiple sclerosis), their ability to use manual sign and access computer systems are impacted. AAC access methods in this population remain inherently slow and effortful (e.g., eye-tracking, head-tracking, mechanical switches). Thus, optimizing communication interfaces for alternate access methods may provide significant improvements in communication rates and quality of life. In this series of studies, we developed and evaluated methods for improving communication rates through optimization and prediction in communication interfaces. These interfaces enabled participants to select sounds (phonemes) instead of letters and were computationally optimized offline via a model of human movement in order for targets likely to be selected together to be in close proximity. Online prediction was implemented such that likely targets were dynamically enlarged. Computational simulations suggested that optimized phonemic interfaces could increase communication rates by up to 30.9% compared to random phonemic interfaces. Communication rates were empirically evaluated in 36 participants without motor impairment using an alternate computer access method to produce messages with phonemic interfaces over 12 sessions. Results suggested that optimization increased communication rates by 10.5–23.0% compared to a random phonemic interface. Prediction increased communication rates during training sessions, but was not a significant factor in communication rates during the final session. Empirical evaluations in individuals with motor impairment revealed that all participants strongly agreed that they would improve with practice, and four out of six participants strongly preferred the interface with prediction. Results of these studies suggest that optimized and predictive phonemic interfaces may provide increased communication rates for individuals with motor impairments affecting both oral communication and computer access. Methods for dynamically enlarging targets may also be applicable to other (non-phonemic) interfaces to increase communication rates. Further research is needed to fully translate these results into clinical practice.