Abstract
This paper reports the recently made progress in building a hybrid cognitive architecture and its use in Web-based embodied conversational agents. The cognitive architecture TRIPLE integrates three engines which run in parallel and are responsible, for connectionist-like retrieval from memory, reasoning, and emotional response, respectively. TRIPLE is based on a large memory and communicates with the environment (e.g. Internet) via RDF based messages, which contain input and output information. Each engine has a specific role and is computationally optimized to cope with real time tasks. Retrieval from memory is performed by fast linear algebra calculations and reasoning is based on RDF based inference. Any specified task triggers a process of retrieval from memory of relevant knowledge. Retrieval is based on connectionist principles (spreading of activation) and assessment of relatedness to the current task. Goal completion is reached by transferring knowledge from memory, reasoning, and usage of resources external to the agent. This paper focuses on the presentation of the whole architecture and discussion of its main principles and functionality with a focus on its cognitive specificity.