RESUMO
One in every 200 people worldwide cannot express orally because of cognitive, motor, neurological, or emotional problems. Assistive technologies can help people with impairments to use computers to perform their daily life activities independently and to communicate with others. This paper presents a Hidden Markov Model-based word prediction method that allows keyboard emulation software to predict words so that children with disabilities can type texts more quickly. The proposed system involved the development of a keyboard emulator, the construction and processing of a corpus, as well as a word prediction algorithm. Children with different cognitive profiles had to produce a text and type it twice: first with free typing, second using the virtual keyboard's word prediction. Results indicated the word prediction of the keyboard emulator software reduced typing efforts. However, the software initially increased the typing time when the corpus was not well adapted to users. The total amount of clicks with word prediction decreased by around 26.2%. Regarding execution time using prediction, 61% typed the text in less time. The tests performed with literate volunteers indicated a reduction in the number of clicks by up to 51.3%. This result surpasses the 15% achieved in the previous study by Free Virtual Keyboard with word prediction based on pure statistics. Moreover, all volunteers required fewer clicks to perform the task. People with impairments, especially children, could use the system and demonstrate their knowledge and abilities. The entire system is available on the Internet and users have unrestricted and free access to it.
