Alternative and Augmentative Communication (AAC) systems have long benefitted individuals who struggle with spoken and written language. In my current project, we’re taking AAC further, utilizing cutting-edge advancements in conversational AI to assist AAC users further.
Our approach incorporates two types of AI models – the “uninformed” and the “informed.” The uninformed AI is responsible for generating general open-domain conversations. We experimented with DialoGPT and BlenderBot, two well-regarded conversational AI models for this task. We implemented the 400M BlenderBot model, which impressively generates three responses per user input, offering a dynamic casual experience. After thorough testing, we found that BlenderBot outperformed, providing more engaging and contextually accurate conversations.
Understanding How The Chatbot Generates A Response
The uninformed AI serves to simulate ordinary conversation, answering questions and engaging in discussions that don’t require specific knowledge about the user. This way, AAC users can practice casual, everyday communication, improving their skills and confidence.
Stay tuned to learn more about the second part of our AI framework – the informed AI – in the next post.