An automatic natural language sentence generation from images

While traditional Augmentative and Alternative Communication (AAC) device allows users with communication disorder to communicate via images, users are required to pick the images in the right order. Therefore, in this study, an algorithm that independent from the sequence of input images has been proposed. To develop such an algorithm, Statistical Natural Language Processing together with Word Bigram and Class Bigram was utilized. Using this algorithm, users can generate simple sentences in Thai language by inputting images in any order which allowing them to arbitrarily communicate their messages. The algorithm incorporates Nanci Bell's 12 questions protocol to categorize selected images. Then, the system generates sentences based on Thai grammar. The responses from the system were compared with target sentences to estimate the correctness of each input sentence. The results showed the correctness at 91.11% with 3 input images, 81.67% with 4 input images, 63.33% with 5 input images, and 41.67% with 6 input images.