Automated error detection and correction in Augmentative and Alternative Communication (AAC) Systems Using NLP

This research examines the application of natural language processing (NLP) methods in the field of error detection and correction attached to the weakly constrained domains of Augmentative and Alternative Communication (AAC) systems designed for people with several communicative difficulties. To train and evaluate the effectiveness of the NLP model, we present a simulated dataset containing typical input AAC errors, such as symbol errors, typographical errors, and context arguments that do not suit the input; this way, we do not use real user data. The methodology consists of using simulated queries to train an NLP model that has been designed to work in real time and corrects speech recognition errors. Key results include the provision of accuracy evidence of the model in detecting and correcting the input errors for which the model was built to perform based on the data provided by the simulated test. Evaluation measurements such as precision, recall and F1-score also assure on the model performance which explains the ability of the new model to enhance the usability and accessibility of AAC systems. Limitations are mostly related to difficulties in extrapolating the data to the real world, however results demonstrate great advances in AAC error correction technology. Further studies aim at improving these models based on the real-life feedback received.