Using salience rankings of entities and tasks to aid computer interpretation of natural language input

摘要

A computing device receives a message that is part of a session. The message includes natural language content derived from input by a user. The computing device analyzes the natural language content to determine an intent corresponding to a given task the user desires the computing device to perform. The analyzing is based on a ranked list of tasks that indicates the current relevance of tasks in the session. A definition for the given task that includes an entity slot is retrieved from a data store and a specific entity is assigned to the slot based on a ranked list of entities that indicates current relevance of the entities in the session. The computing device invokes an agent to perform the given task based on the specific entity assigned to the entity slot.