← 学习库 Speech and Language Processing 本册目录

24.5.1 Dialogue Acts

As we implied above, the speech acts as originally defined by Austin don't model key features of conversation such as grounding, contributions, adjacency pairs and so on. In order to capture these conversational phenomena, we use an extension of speech acts called dialogue acts (Bunt, 1994) (or dialogue moves or conversational moves (Power, 1979; Carletta et al., 1997b). A dialogue act extends speech acts with internal structure related specifically to these other conversational functions (Allen and Core, 1997; Bunt, 2000).

A wide variety of dialogue act tagsets have been proposed. Fig. 24.17 shows a very domain-specific tagset for the Verbmobil two-party scheduling domain, in which speakers were asked to plan a meeting at some future date. Notice that it has many very domain-specific tags, such as SUGGEST, used for when someone proposes a particular date to meet, and ACCEPT and REJECT, used to accept or reject a proposal for a date. Thus it has elements both from the presentation and acceptance phases of the Clark contributions discussed on page 7.

TagExample
THANKThanks
GREETHello Dan
INTRODUCEIt's me again
BYEAllright bye
REQUEST-COMMENTHow does that look?
SUGGESTfrom thirteenth through seventeenth June
REJECTNo Friday I'm booked all day
ACCEPTSaturday sounds fine,
REQUEST-SUGGESTWhat is a good day of the week for you?
INITI wanted to make an appointment with you
GIVE_REASONBecause I have meetings all afternoon
FEEDBACKOkay
DELIBERATELet me check my calendar here
CONFIRMOkay, that would be wonderful
CLARIFYOkay, do you mean Tuesday the 23rd?
DIGRESS[we could meet for lunch] and eat lots of ice cream
MOTIVATEWe should go to visit our subsidiary in Munich
GARBAGEOops, I-
Figure 24.17 The 18 high-level dialogue acts used in Verbmobil-1, abstracted over a total of 43 more specific dialogue acts. Examples are from Jekat et al. (1995).

There are a number of more general and domain-independent dialogue act tagsets. In the DAMSL (Dialogue Act Markup in Several Layers) architecture inspired by the

原书第 966 页
Chapter 24. Dialogue and Conversational Agents
Act typeSample Acts
turn-taking grounding core speech acts argumentationtake-turn, keep-turn, release-turn, assign-turn acknowledge, repair, continue inform, wh-question, accept, request, offer elaborate, summarize, question-answer, clarify
Figure 24.18Conversation act types, from Traum and Hinkelman (1992).

work of Clark and Schaefer (1989) and Allwood et al. (1992), Allwood (1995) each utterance is tagged for two functions, forward looking functions like speech act functions, and the backward looking function, like grounding and answering, which ‘look back’ to the interlocutor’s previous utterance (Allen and Core, 1997; Walker et al., 1996; Carletta et al., 1997a; Core et al., 1999).

Traum and Hinkelman (1992) proposed that the core speech acts and grounding acts that constitute dialogue acts could fit into an even richer hierarchy of conversation acts. Fig. 24.18 shows the four levels of act types they propose, with the two middle levels corresponding to DAMSL dialogue acts (grounding and core speech acts). The two new levels include turn-taking acts and argumentation relation, a conversational version of the coherence relations of Ch. 21.

The acts form a hierarchy, in that performance of an act at a higher level (for example a core speech act) entails performance of a lower level act (taking a turn). We will see the use of conversational acts in generation later on in this section, and will return to the question of coherence and dialogue structure in Sec. 24.7.

← 24.4.2 Dialogue System Evaluation24.5.2 Interpreting Dialogue Acts →