A dialogue system, also called a conversational agent, is software designed to hold a conversation with a person in natural language, whether typed or spoken. The category stretches from the 1960s rule-based chatbots like ELIZA through today’s voice assistants and large language model chat interfaces. A widely used survey by Chen, Liu, Yin, and Tang organizes the field around a basic split that has shaped both research and products.
On one side are task-oriented dialogue systems, built to help a user achieve a concrete goal such as booking a flight, reserving a restaurant, or controlling a device. These are traditionally built as a pipeline: understand what the user wants, keep track of the conversation so far, decide on the next action, and generate a response. On the other side are non-task-oriented or open-domain systems, sometimes called chit-chat bots, whose aim is to keep up an engaging conversation rather than complete a specific errand. The two were long treated as different engineering problems with different techniques, the goal-driven side leaning on structured language understanding and the chit-chat side on generative or retrieval models.
Voice assistants such as Siri, Alexa, and Google Assistant are essentially task-oriented dialogue systems wrapped around speech recognition and synthesis, while companion and social chatbots sit on the open-domain side. Modern large language model assistants increasingly try to do both at once with a single model.
For a general reader, the term names the whole family of “things you can talk to,” and the task-oriented versus open-domain distinction is the simplest way to understand why a bot that books your table and a bot that just keeps you company were, for decades, built so differently.