Grounding in communication

The foundational framework for how conversational participants establish mutual understanding. Clark and Brennan introduce the concept of common ground and the grounding process: the moment-by-moment coordination through which speakers and listeners confirm that they have understood each other well enough to proceed. A prerequisite for thinking clearly about where AI conversation falls short.

Published 1991.

Summary

Common ground Mutual knowledge, mutual beliefs, and mutual assumptions shared by participants. All collective actions are built on common ground and its accumulation. Common ground must be updated moment by moment.

Grounding The collective process by which participants try to establish mutual belief that what has been said has been understood to a criterion sufficient for current purposes. Grounding is what updates common ground.

Grounding criterion The contributor and partners mutually believe that the partners have understood what the contributor meant to a criterion sufficient for current purposes. Understanding need not be perfect, only sufficient.

The contribution model (Clark & Schaefer, 1989) A contribution has two required phases:

• Presentation phase: A presents utterance u for B to consider, on the assumption that B’s evidence e will allow A to infer understanding.
• Acceptance phase: B accepts u by giving evidence e that she believes she understands.

Both phases must complete for a contribution to be complete. Contributions can be embedded hierarchically within both phases.

Four understanding states (B’s possible states)

• State 0: B did not notice A uttered anything.
• State 1: B noticed A uttered something (but not state 2).
• State 2: B correctly heard u (but not state 3).
• State 3: B understood what A meant by u.

Three forms of positive evidence of understanding

1. Acknowledgments: back-channel responses (continuers: uh huh, yeah, m; assessments: gosh, really; head nods). Signal that B is passing up the opportunity to initiate repair.
2. Initiation of the relevant next turn: answering a question, responding to a request. Constitutes positive evidence only if the response is appropriate.
3. Continued attention: unbroken or undisturbed attention. Broken attention (looking away, hanging up) signals loss of understanding.

Principle of least collaborative effort Participants try to minimize their collaborative effort, defined as the total work done by both parties from initiation of each contribution to its mutual acceptance. Replaces the speaker-only “principle of least effort” (Cooperative principle).

Three reasons the speaker-only version fails: (1) time pressure leads to improper utterances; (2) errors require repair; (3) ignorance of the interlocutor makes proper utterances impossible.

Grounding changes with purpose

For referential identity:

• Alternative descriptions: partner presents a different description of the referent to demonstrate identification.
• Indicative gestures: pointing, looking, touching.
• Referential installments: present reference as a separate installment, pause for acceptance before continuing (left-dislocation construction).
• Trial references: present uncertain reference with try marker, pause for confirmation or correction.

For verbatim content:

• Verbatim displays: partner repeats exact content back.
• Installments: divide content into chunks, receive confirmation on each.
• Spelling: spell out critical words explicitly.

Grounding changes with medium

Eight constraints media impose on grounding:

1. Copresence: shared physical environment.
2. Visibility: participants visible to each other.
3. Audibility: communication by speech.
4. Cotemporality: B receives at roughly the same time A produces.
5. Simultaneity: both parties can send and receive at once.
6. Sequentiality: turns cannot get out of order.
7. Reviewability: B can review A’s messages later.
8. Revisability: A can revise messages privately before sending.

Seven media mapped to constraints:

• Face-to-face: copresence, visibility, audibility, cotemporality, simultaneity, sequentiality.
• Telephone: audibility, cotemporality, simultaneity, sequentiality.
• Video teleconference: visibility, audibility, cotemporality, simultaneity, sequentiality.
• Terminal teleconference: cotemporality, sequentiality, reviewability.
• Answering machine: audibility, reviewability.
• Email: reviewability, revisability.
• Letters: reviewability, revisability.

Eleven grounding costs that vary by medium:

1. Formulation costs: time and effort to plan utterances.
2. Production costs: physical effort of producing output (speaking less than typing less than handwriting).
3. Reception costs: effort of listening or reading; waiting costs.
4. Understanding costs: effort to interpret; increased when contextual clues are absent.
5. Start-up costs: cost of initiating discourse (getting attention, establishing channel).
6. Delay costs: cost of pausing to plan in cotemporal media; interpreted as dropout or turn completion. Nil in non-cotemporal media (email, letters).
7. Asynchrony costs: loss of timing precision without copresence, visibility, audibility, or simultaneity.
8. Speaker change costs: cost of handing off the floor; lowest face-to-face, highest in email, letters, answering machines.
9. Display costs: cost of gestures, gaze, pointing; increases without copresence.
10. Fault costs: costs of errors and missayings; higher in non-cotemporal media where revision was possible.
11. Repair costs: cost of fixing faults; self-repair preferred over other-repair; very high in non-cotemporal media.

Related

• Is conversation still a useful metaphor?