Agent Communication Language (ACL)

The fundamental units of an ACL message, based on Speech Act Theory. Each act performs an action through communication.

• Inform: Assert a proposition believed to be true (e.g., inform(sender, receiver, price=50)).
• Request: Ask another agent to perform an action.
• Propose: Submit an offer during a negotiation.
• CFP (Call for Proposals): Solicit bids, as used in the Contract Net Protocol.
• Accept/Reject: Respond to a proposal or request.

These acts define the illocutionary force—the speaker's intent—of a message.

The payload of a communicative act, representing the information exchanged. Content must be expressed in a separate Content Language that all participating agents understand.

• Syntax: Often uses a knowledge representation language like KIF (Knowledge Interchange Format) or FIPA-SL.
• Semantics: The meaning of the content expression must be unambiguous.
• Example: In inform(sender, receiver, (price stock-A 50)), the content (price stock-A 50) is a proposition in KIF stating a fact.

A wrapper containing the metadata necessary for message delivery and processing in a distributed system. It is separate from the communicative act and its content.

• Sender/Receiver: Agent identifiers.
• Message ID: Unique identifier for the message and for linking replies (via in-reply-to).
• Protocol: The Interaction Protocol governing this conversation (e.g., fipa-contract-net).
• Ontology: References the shared ontology that defines the vocabulary for the content.
• Encoding & Transport Details: Specifies how the content is serialized (e.g., string, XML) and transport mechanism (e.g., HTTP, IIOP).

Pre-defined, structured sequences of communicative acts designed to achieve a specific coordination goal. They define the legal flow of conversation.

• Request Protocol: A simple request followed by agree/refuse and then failure/inform-done.
• Contract Net Protocol: A complex protocol involving cfp, propose, accept-proposal, reject-proposal, and inform.
• Auction Protocols: Define sequences for English, Dutch, or Vickrey auctions.
• Specification: Often modeled as finite state machines or using Agent UML sequence diagrams.

Two layers that give meaning to the data inside a message.

Content Language: The formal syntax for expressing propositions, actions, or objects (e.g., KIF, RDF, FIPA-SL). It provides the grammatical structure.

Ontology: A shared conceptualization. It defines the vocabulary (terms like 'Price', 'Stock'), their types, properties, and relationships. An ontology reference in the message envelope allows agents to map terms to a common understanding, which is critical for semantic interoperability.

An Interaction Protocol defines a structured sequence of permissible message exchanges between agents to achieve a specific communicative purpose, such as a negotiation or auction. It specifies the legal ordering of communicative acts (like request and propose) and is often modeled using finite state machines or agent UML sequence diagrams. This provides the concrete "conversation rules" that implement the abstract semantics defined by an ACL like FIPA ACL.

• Example: A Contract Net Protocol is a specific interaction protocol for decentralized task allocation, defining the sequence of call-for-proposal, propose, accept-proposal, and reject-proposal messages.

Speech Act Theory is the linguistic and philosophical framework that underpins most ACLs. It models communication as the performance of actions (e.g., informing, requesting, promising), not just the transmission of information. An ACL message is a speech act with three components:

• Locutionary Act: The literal utterance/syntax.
• Illocutionary Act: The intended force (e.g., a request, an inform).
• Perlocutionary Act: The effect on the listener.

In ACLs like FIPA ACL, the performative (e.g., inform, request) explicitly encodes the illocutionary force, making agent intent machine-readable.

Social Commitments are normative constructs that create obligations between agents, forming a foundation for trust and reliable coordination. A commitment (C(debtor, creditor, antecedent, consequent)) means the debtor agent is obliged to the creditor agent to bring about the consequent if the antecedent condition holds. ACL messages often create, cancel, or discharge these commitments.

• Example: An accept-proposal message creates a commitment for the sender to perform the agreed-upon action. This moves coordination beyond simple message-passing to a model of verifiable social interaction.

An Electronic Institution is a computational framework that defines the norms, rules, and structured interaction spaces governing autonomous agents. It provides the "social infrastructure" in which ACL-based communication occurs. Components include:

• Dialogical Framework: The ontology and ACL used.
• Scene Protocols: Interaction protocols for specific activities (like a trading floor).
• Normative Rules: Rules that define permissions, obligations, and sanctions.

This ensures that agent interactions within the institution are orderly, goal-directed, and compliant with institutional rules, making open multi-agent systems manageable.

A Content Language is a formal language used to express the actual content or proposition of an ACL message. The ACL defines the envelope (sender, receiver, performative), while the Content Language defines the payload. An ACL must be paired with a content language for meaningful communication.

• Common Examples: KIF (Knowledge Interchange Format), FIPA-SL (Semantic Language), RDF, or OWL. For example, an inform message's content field might contain a KIF expression like (price stock-xyz 150).

In the context of ACLs, an Ontology provides a shared, formal specification of the concepts, relationships, and terminology within a domain. It is critical for semantic interoperability, ensuring that agents interpret the content of messages consistently. The ontology is referenced within the ACL message, allowing receivers to unambiguously decode terms.

• Role: Defines the vocabulary for the Content Language.
• Example: In a logistics domain, an ontology would precisely define concepts like Container, hasLocation, isLoadedOn, so an inform message about a container's status is understood correctly by all agents.