Speech Act Theory

J.L. Austin's core taxonomy distinguishes three simultaneous dimensions of a speech act:

• Locutionary Act: The literal utterance of words with a specific meaning and reference (e.g., saying "The door is open").
• Illocutionary Act: The intended action performed in saying something (e.g., informing, warning, or promising). This is the core communicative force.
• Perlocutionary Act: The consequential effect achieved by saying something on the listener's thoughts or actions (e.g., persuading someone to close the door, startling them). In multi-agent systems, protocols focus on standardizing the illocutionary force to ensure predictable interpretation.

The illocutionary force is the speaker's intent (e.g., to request, commit, or declare). For an act to be successful or "felicitous," certain felicity conditions must be met:

• Preparatory Conditions: Contextual prerequisites (e.g., to promise, you must be capable of performing the promised action).
• Sincerity Condition: The speaker must possess the requisite psychological state (e.g., intention to fulfill a promise).
• Essential Condition: The utterance counts as undertaking an obligation (e.g., a promise creates an expectation of future action). Agent communication languages (ACLs) encode these conditions in message semantics to validate communicative acts.

John Searle's classification categorizes illocutionary acts by their direction of fit and psychological state:

• Directives: Acts where the speaker attempts to get the hearer to do something. Direction of fit is world-to-word (the world changes to match the words). Examples: requests, commands, questions. Associated psychological state is want or desire.
• Commissives: Acts where the speaker commits to a future course of action. Direction of fit is also world-to-word. Examples: promises, pledges, oaths. Associated psychological state is intention. These are fundamental for agent protocols involving task allocation (directives) and forming commitments (commissives).

Searle's other core categories complete the model of agent communication:

• Assertives: Acts that commit the speaker to the truth of a proposition. Direction of fit is word-to-world (words match the world). Examples: informing, stating, concluding. Psychological state is belief.
• Expressives: Acts that express the speaker's psychological state about a state of affairs. No direction of fit. Examples: thanking, apologizing, congratulating.
• Declarations: Acts that immediately change institutional reality by their successful performance. Direction of fit is both ways (words change the world). Examples: "You're fired," "I pronounce you married." These require a specific institutional role or authority, highly relevant for agents operating within electronic institutions.

A performative utterance is one where "saying is doing"—the speech act itself constitutes the action (e.g., "I promise..."). This concept is directly implemented in Agent Communication Languages (ACLs) like FIPA ACL. An ACL message explicitly encodes:

• Performative: The illocutionary force (e.g., inform, request, cfp (call-for-proposal), propose).
• Content: The propositional content of the message.
• Sender/Receiver: The participating agents.
• Protocol & Conversation ID: The interaction context. This formalization allows software agents to unambiguously interpret messages as actions, not just strings, enabling reliable coordination.

Individual speech acts are composed into structured sequences called conversation policies or interaction protocols. These define the permissible flow of communicative acts between agents to achieve a larger goal. Common patterns include:

• Request-Response: A request followed by an inform (result) or refuse.
• Contract Net: A cfp (call-for-proposal), followed by propose messages from bidders, leading to accept-proposal/reject-proposal and inform-done/failure.
• Auctions & Negotiations: Sequences of propose, counter-propose, accept, and reject. Protocols are often specified using finite state machines or UML sequence diagrams, providing a predictable framework for complex, stateful agent dialogues grounded in speech act theory.

A formal language with defined syntax, semantics, and pragmatics that enables autonomous agents to exchange knowledge and requests. It operationalizes Speech Acts into computable messages.

• Syntax: Defines the message structure (e.g., sender, receiver, performative, content).
• Semantics: Provides the precise meaning of message elements.
• Pragmatics: Governs the intended effect and context of the message exchange.
• Standard Example: The FIPA ACL (Foundation for Intelligent Physical Agents) is a widely referenced standard specifying performatives like inform, request, cfp (call-for-proposal), and propose.

A predefined, structured sequence of permissible message exchanges between agents to achieve a specific communicative goal. It choreographs Speech Acts into a predictable conversation.

• Formal Specification: Often defined using finite state machines, Petri nets, or sequence diagrams.
• Common Protocols: Contract Net Protocol (for task allocation), various auction protocols (English, Dutch, Vickrey), and negotiation protocols.
• Role: Ensures that a series of communicative acts (e.g., a call for proposals followed by bids and an award) follows a mutually understood pattern, preventing miscoordination.

Normative constructs that create obligations between agents, defining that a debtor agent is committed to a creditor agent to bring about a certain condition. They provide a formal mechanism for trust and accountability.

• Foundation: Grounded in the Speech Act of promising.
• Lifecycle: Commitments have states (e.g., active, fulfilled, violated, terminated).
• Function: Enable agents to reason about the future behavior of others, forming the basis for cooperative planning. If Agent A commits to delivering a result to Agent B, B can plan its subsequent actions accordingly.

Computational frameworks that define the norms, rules, and structured interaction spaces governing autonomous agents, ensuring orderly societal interactions. They provide the 'rules of the game' for Speech Act-based communication.

• Components: Include roles agents can play, scenes (virtual rooms for specific interactions), and normative rules that specify penalties or rewards.
• Analogy: Functions like a digital marketplace or courtroom, where protocols dictate who can speak, when, and what they can say.
• Purpose: Reduces complexity and enforces desirable global properties (e.g., fairness, non-repudiation) in open multi-agent systems.

A prominent software model for intelligent agents based on practical reasoning, where an agent's behavior is driven by its Beliefs (world model), Desires (goals), and Intentions (committed plans). Speech Acts are the primary means for BDI agents to interact.

• Reasoning Loop: Agents perceive (update beliefs), deliberate on desires, plan (form intentions), and act.
• Communication Link: An inform speech act updates another agent's beliefs. A request speech act creates a new desire or intention in the recipient.
• Framework Example: The JACK or JADE platforms implement BDI principles with built-in support for agent communication.

The set of communicative acts defined by the FIPA standard, providing a concrete vocabulary for Agent Communication Languages. Each performative corresponds to a type of Speech Act with formal semantics.

• Assertives: inform, confirm, disconfirm (commit the speaker to the truth of a proposition).
• Directives: request, cfp (call-for-proposal), subscribe (attempt to get the hearer to perform an action).
• Commissives: propose, accept-proposal, reject-proposal (commit the speaker to a future course of action).
• Declaratives: proxy (empower another agent to act).
• Example: (request :sender A :receiver B :content "open valve V1") is a directive Speech Act.