← Garden Patch Home · Citations
Clark and Brennan argue that communication is not transmission but a joint activity. Participants must continuously coordinate their understanding — a process they call “grounding.” Every contribution to a conversation proceeds in two phases: the speaker presents an utterance, and the addressee provides evidence of understanding (continued attention, acknowledgment, paraphrase, or relevant next turn). Until acceptance occurs, the contribution remains incomplete.
The paper’s central framework identifies eight constraints that communication media impose on the grounding process: copresence, visibility, audibility, contemporality, simultaneity, sequentiality, reviewability, and revisability. Face-to-face conversation provides copresence, visibility, audibility, contemporality, simultaneity, and sequentiality — but not reviewability or revisability. Written correspondence provides reviewability, revisability, and sequentiality — but lacks the rest. Each combination of constraints determines which repair and evidence techniques are available.
The “principle of least collaborative effort” governs strategy selection: participants choose techniques that minimize total effort across both parties, not just the speaker’s effort. This explains why people adapt their communication style to the medium rather than using one universal strategy.
Two-phase contributions: Presentation alone does not constitute a successful contribution. The acceptance phase — where the addressee signals understanding — completes it. This distinguishes grounding from information-theoretic models where transmission equals communication.
Eight grounding constraints: These are properties of the communication medium, not the participants. Copresence enables pointing and shared reference. Visibility enables gesture and facial expression. Audibility enables tone and prosody. Contemporality means messages arrive as produced. Simultaneity allows concurrent send and receive. Sequentiality preserves turn order. Reviewability allows re-reading. Revisability allows editing before sending.
Least collaborative effort: The principle operates across both participants. A speaker may invest more effort in precise wording (higher speaker cost) to reduce addressee confusion (lower addressee cost), if the net effort is lower. Medium constraints shift this calculus — revisability lets speakers optimize their message before sending, while contemporality enables rapid back-and-forth repair.
Grounding criterion: Understanding need not be perfect — it must be sufficient for current purposes. A casual conversation tolerates more ambiguity than a surgical instruction. The required level of grounding shifts with the stakes.
“Grounding takes one shape in face-to-face conversation but another in personal letters.”
The grounding framework provides the theoretical foundation for understanding why different communication channels produce different quality of mutual understanding. For agent protocol design and Deep Context Architecture, it reframes error handling as a grounding problem: agents need evidence of uptake, not just message delivery. The eight constraints map directly to properties of agent communication channels, predicting which channels will support which kinds of repair and confirmation.
relates_to::[[Errors as Negotiation Opportunities]] The two-phase contribution model means errors are not failures but incomplete grounding sequences. An error description is a presentation; the recipient’s corrective action is the acceptance. Natural-language error descriptions support grounding; error codes do not.
relates_to::[[Communication Theory]]↑ Clark and Brennan’s model extends speech act theory into collaborative territory. Communication is joint action, not speaker action plus listener reception.