24.1.3 Language as Joint Action: Grounding

The previous section suggested that each turn or utterance could be viewed as an action by a speaker. But dialogue is not a series of unrelated independent acts. Instead, dialogue is a collective act performed by the speaker and the hearer. One implication of joint action is that, unlike in monologue, the speaker and hearer must constantly establish common ground (Stalnaker, 1978), the set of things that are mutually believed by both speakers. The need to achieve common ground means that the hearer must ground the speaker's utterances, making it clear that the hearer has understood the speaker's meaning and intention.

As Clark (1996) points out, people need closure or grounding for non-linguistic actions as well. For example, why does a well-designed elevator button light up when it's pressed? Because this indicates to the would-be elevator traveler that she has successfully called the elevator. Clark phrases this need for closure as follows (after (Norman, 1988)):

Principle of closure. Agents performing an action require evidence, sufficient for current purposes, that they have succeeded in performing it.

Grounding is also important when the hearer needs to indicate that the speaker has not succeeded in performing an action. If the hearer has problems in understanding, she must indicate these problems to the speaker, again so that mutual understanding can eventually be achieved.

How is closure achieved? Clark and Schaefer (1989) introduce the idea that each joint linguistic act or contribution has two phases, called presentation and acceptance. In the first phase, a speaker presents the hearer with an utterance, performing a sort of speech act. In the acceptance phase, the hearer has to ground the utterance, indicating to the speaker whether understanding was achieved.

What methods can the hearer (call her B) use to ground the speaker A's utterance? Clark and Schaefer (1989) discuss five main types of methods, ordered from weakest to strongest:

1. Continued attention: B shows she is continuing to attend and therefore remains satisfied with A's presentation.

2. Relevant next contribution: B starts in on the next relevant contribution.

3. Acknowledgement: B nods or says a continuer like uh-huh, yeah, or the like, or an assessment like that's great.

4. Demonstration: B demonstrates all or part of what she has understood A to mean, for example by reformulating (paraphrasing) A's utterance, or by collaborative completion of A's utterance.

5. Display: B displays verbatim all or part of A's presentation.

Let's look for examples of these in a human-human dialogue example. We'll be returning to this example throughout the chapter; in order to design a more sophisticated machine dialogue agent, it helps to look at how a human agent performs similar tasks. Fig. 24.4 shows part of a dialogue between a human travel agent and a human client.

| $ C_{1} $: | ... I need to travel in May. |

| --- | --- |

| $ A_{1} $: | And, what day in May did you want to travel? |

| $ C_{2} $: | OK uh I need to be there for a meeting that's from the 12th to the 15th. |

| $ A_{2} $: | And you're flying into what city? |

| $ C_{3} $: | Seattle. |

| $ A_{3} $: | And what time would you like to leave Pittsburgh? |

| $ C_{4} $: | Uh hmm I don't think there's many options for non-stop. |

| $ A_{4} $: | Right. There's three non-stops today. |

| $ C_{5} $: | What are they? |

| $ A_{5} $: | The first one departs PGH at 10:00am arrives Seattle at 12:05 their time. The second flight departs PGH at 5:55pm, arrives Seattle at 8pm. And the last flight departs PGH at 8:15pm arrives Seattle at 10:28pm. |

| $ C_{6} $: | OK I'll take the 5ish flight on the night before on the 11th. |

| $ A_{6} $: | On the 11th? OK. Departing at 5:55pm arrives Seattle at 8pm, U.S. Air flight 115. |

| $ C_{7} $: | OK. |

| Figure 24.4 Part of a conversation between a travel agent (A) and client (C). | |

Utterance $ A_{1} $, in which the agent repeats in May, repeated below in boldface, shows the strongest form of grounding, in which the hearer displays their understanding by repeating verbatim part of the speakers words:

C_{1}: ...I need to travel in May.

A $ _{1} $: And, what day in May did you want to travel?

This particular fragment doesn't have an example of an acknowledgement, but there's an example in another fragment:

C: He wants to fly from Boston

A: Mm hmm

C: to Baltimore Washington International

The word mm-hmm here is a continuer, also often called a backchannel or an acknowledgement token. A continuer is a (short) optional utterance which acknowledges the content of the utterance of the other, and which doesn't require an acknowledgement by the other (Yngve, 1970; Jefferson, 1984; Schegloff, 1982; Ward and Tsukahara, 2000).

In Clark and Schaefer's third method, the speaker starts in on their relevant next contribution. We see a number of examples of this in the sample dialogue above, for example where the speaker asks a question and the hearer answers it. We mentioned these adjacency pairs above; other examples include PROPOSAL followed by ACCEPTANCE or REJECTION, APOLOGY followed by ACCEPTANCE/REJECTION, SUMMONS followed by ANSWER, and so on.

In a more subtle but very important kind of grounding act, the speaker can combine this method with the previous one. For example notice that whenever the client answers a question, the agent begins the next question with And. The And indicates to the client that the agent has successfully understood the answer to the last question:

And, what day in May did you want to travel?

And you're flying into what city?

And what time would you like to leave Pittsburgh?

As we will see in Sec. 24.5, the notions of grounding and contributions can be combined with speech acts to give a more sophisticated model of joint action in conversation; these more sophisticated models are called dialogue acts.

Grounding is just as crucial in human-machine conversation as it is in human conversation. The examples below, from Cohen et al. (2004), suggest how unnatural it sounds when a machine doesn't ground properly. The use of Okay makes (24.7) a much more natural response than (24.8) to ground a user's rejection:

(24.7) System: Did you want to review some more of your personal profile?

Caller: No.

System: Okay, what's next?

(24.8) System: Did you want to review some more of your personal profile?

System: What's next?

Indeed, this kind of lack of grounding can cause errors. Stifelman et al. (1993) and Yankelovich et al. (1995) found that humans get confused when a conversational system system doesn't give explicit acknowledgements.