Communicative Intent

The illocutionary force is the speaker's intended action performed in saying something, such as promising, requesting, or declaring. It is the core pragmatic function of an utterance.

• Examples: 'I promise to be there' (commissive), 'Close the door' (directive), 'I name this ship...' (declarative).
• In AI, recognizing illocutionary force allows an agent to classify user prompts (e.g., a request vs. a command) and respond with appropriate action, not just literal text.

The perlocutionary effect is the actual consequence or impact the utterance has on the listener, such as persuading, surprising, or alarming them. It is the outcome achieved by saying something.

• Key Distinction: While illocutionary force is the intended action, the perlocutionary effect is the often-unpredictable result. A warning (illocution) may fail to deter someone (perlocution).
• For cooperative AI agents, this involves modeling how a generated response might influence a human user's subsequent beliefs or actions.

This dichotomy separates sentence meaning (the literal, semantic content derived from words and syntax) from speaker meaning (the intended message inferred using context and shared knowledge).

• Example: The sentence 'It's cold in here' literally describes temperature. The speaker meaning might be a request to close a window.
• AI systems must bridge this gap using pragmatic inference, relying on context, user history, and common ground to decode intent.

Common ground is the shared knowledge, beliefs, and assumptions between conversational participants that forms the basis for interpreting communicative intent.

• Components: Includes mutual perceptual environment, cultural knowledge, and the conversational history itself.
• AI agents must actively build and maintain a model of common ground with users to generate relevant, context-aware responses and avoid stating the obvious or making unwarranted assumptions.

Philosopher H.P. Grice proposed that communication operates on a Cooperative Principle: participants expect each other to make contributions that are truthful, informative, relevant, and clear. Intent is inferred when these maxims appear to be flouted.

• Example: A verbose, unclear answer (flouting the Maxim of Manner) may imply the speaker is being evasive.
• This framework provides a logical basis for AI to generate non-literal inferences and to craft its own utterances to be effectively interpreted by humans.

Intent recognition is the computational process of mapping a user's natural language input to a structured representation of their goal. It is a critical upstream task for dialog systems and agentic workflows.

• Methods: Ranges from simple pattern matching and keyword classification to sophisticated sequence models that use context and user embeddings.
• Output: Typically a discrete intent label (e.g., book_flight) and associated slots or parameters (e.g., destination: Paris), which form the basis for task execution.

The cognitive process of deriving a speaker's intended meaning from an utterance by integrating context, shared knowledge (common ground), and conversational principles that go beyond the literal semantic content. It is the computational mechanism that resolves communicative intent.

• Example: The utterance "Can you pass the salt?" is literally a question about ability, but pragmatically it is a request for action.
• Relies on assumptions of cooperative communication, often formalized by Gricean Maxims.

A set of conversational principles proposed by philosopher H.P. Grice that describe the implicit rules governing cooperative communication. Listeners assume speakers adhere to these maxims, enabling pragmatic inference.

• Maxim of Quality: Be truthful. Do not say what you believe to be false.
• Maxim of Quantity: Be as informative as required, but not more.
• Maxim of Relation: Be relevant.
• Maxim of Manner: Be clear, avoid obscurity and ambiguity.

Violations or floutings of these maxims (e.g., sarcasm) signal a deeper communicative intent that the listener must infer.

The computational process of inferring the goals or purposes behind an agent's observed actions or communications. While communicative intent focuses on linguistic acts, intent recognition generalizes to any observable behavior.

• Key Methods: Often uses plan recognition, inverse planning, or sequence classification models.
• Application: Essential for conversational AI (chatbots), proactive assistant systems, and human-robot interaction, where the system must discern a user's goal from partial or ambiguous input.

The set of knowledge, beliefs, and assumptions that are mutually known to be shared by participants in a conversation. It is the foundational context required for accurately inferring communicative intent.

• Establishment: Built through joint attention, previous discourse, and shared cultural or community membership.
• Function: Allows speakers to make efficient references (e.g., using "it" or "the project") and enables listeners to correctly resolve ellipsis and ambiguity.
• Computational Representation: A critical component in dialogue state tracking for AI agents.

The broader cognitive capacity to attribute mental states—such as beliefs, desires, intentions, and knowledge—to oneself and others. Communicative intent is a specific type of intention that Theory of Mind mechanisms are used to infer.

• Orders of Reasoning:
  • First-Order: "I believe Alice intends to inform me."
  • Second-Order: "I believe Alice intends for me to believe she is joking."
• False Belief Understanding: The ability to recognize that others can hold beliefs different from reality, a key test for ToM that is also crucial for understanding deception or irony in communication.

The task of inferring an agent's high-level plans and goals from a sequence of observed low-level actions. In communication, utterances are seen as actions within a larger plan to achieve a goal, making plan recognition a key methodology for discerning intent.

• Inverse Planning: A Bayesian approach that reasons backwards from observed actions, assuming the agent is rational and planning optimally.
• Application: Used in AI to make assistants proactive (e.g., inferring a user is planning a trip from a sequence of searches for flights and hotels), and in security for threat detection.