The main module of a chatbot is the dialogue manager. This module controls the flow of human-machine interaction. It receives the user's request as input and decides what the system's response should be. It will memorize the dialogue context in some form, for example, through pairs of key values, to manage the conversation on several steps between the user and the system.

For the dialogue manager to be able to choose the right answer for the request made by the user, it is necessary to understand the user's intention. In the most advanced chatbots, which are able to understand human language, the user's expressions will be translated into a semantic representation consisting of user intent and entity. This operation will be carried out by a module of natural language comprehension. This module must be previously trained to understand a series of user intent previously identified by the developer. This module is based on the natural language understanding (NLU) component.

In the case of voice input, the system must also be equipped with a voice recognition module that can translate the input into text before passing it to the natural language comprehension module. At the end of the operation, the system response (output) must first be treated with a speech synthesizer module that converts the textual response of the system into speech.

When user input is understood, the dialogue manager takes actions. To perform the action or generate the response, the dialogue manager retrieves the required information from the data sources. After this, a response message is generated by the response generation component and sent back to the user. The following diagram is a chatbot architecture scheme:

To keep track of the context, the dialogue manager keeps the conversation status to know if the request is related to the previous conversation or introduces a new topic into the conversation.