Automatic response generation with reasoning, personalized knowledge graphs and emotional intelligence.

点击翻译按钮获取中文摘要

Conversation, an interactive verbal communication between people, is an important part of conveying information, socializing, and developing companionship. Conversation plays a vital role in human life to express emotions, exchange knowledge and ideas, and share opinions. This form of communication expresses a clear manifestation of human intelligence. The attempt to automatically generate conversations indistinguishable from human ones dates back to the early stages of Artificial Intelligence. Head-to-head dialogue with a masked machine has been the ultimate test to evaluate machine intelligence. However, this endeavour to synthetically produce fluent dialogue interactions still stands as a significant challenge. The release of popular conversational agents conducting some sort of dialogue with humans, such as Siri, Alexa, Google Assistant, and Cortana, by respectively, Apple, Yahoo, Google, and Microsoft are deceiving: They mislead the public to believe that the chatbot problem is solved. However, all these chatbots are goal-oriented grounded on rule-based hard-coded templates to treat requests such as ordering products, reserving a table at a restaurant, booking a flight ticket etc., or answering factoid questions by accessing existing knowledge bases, such as who is the premier of Ontario? or what is the weather like in Vancouver? In other words, existing conversational agents are task-oriented question-answering systems that are unable to entertain a typical day-to-day chitchat conversation. Our research program focuses on solving pragmatic and foundational problems related to generating realistic human-like conversation interactions, that we call open-ended conversations. The aim is to target practical issues like generating on-topic interactions, transitioning subjects in a long conversation, verbally expressing emotion, generating conversational humour in response to previous context, exploiting context knowledge, reasoning with facts, arguing convincingly, etc. The motivation for this research is the development of more natural human-machine interaction interfaces and providing support systems such as companionship to the elderly. The need for companionship has been identified by many people and experts in psychology and geriatrics in different countries and cultures. This need to have a companion is even more important for older people. Indeed, loneliness and social isolation can predict, for the elderly, declining health and poor quality of life. The population is aging globally and while many elderly prefer staying at home, for others who are admitted in available nursing homes, there is not enough staff or volunteers to assist with their entertainment or distraction, and boredom sets in quickly. Social interactions can reduce the mentioned effects of loneliness and social isolation. One solution is a software agent that could intelligently converse and be embedded in tablets or other domestic appliances.