Dialogue act classification identifies the communicative function of each utterance in a conversation. This is crucial for dialogue systems, chatbots, and conversational AI.
Dialogue Act (Speech Act): The action a speaker performs through their utterance, beyond the literal meaning of the words.
Key Insight: "Can you pass the salt?" is literally a question but functions as a request.
| Act | Description | Example |
|---|---|---|
| Statement | Asserting information | "The meeting starts at 3pm." |
| Question | Requesting information | "What time is the meeting?" |
| Request | Asking someone to do something | "Please send me the report." |
| Promise | Committing to future action | "I'll send it by tomorrow." |
| Acknowledgment | Confirming understanding | "Okay, got it." |
| Greeting | Social opening/closing | "Hi, how are you?" |
Classify each utterance with its primary dialogue act.
Should "Can you provide your order number?" be classified as a question or a request?
Sometimes the form doesn't match the function.
Should annotation capture the literal form or the intended function?
This utterance performs multiple acts. Identify all that apply:
The same words can have different functions in different contexts.
Context A: At a restaurant
Context B: In the desert, lost
How much context should annotators have access to?
What was the hardest utterance to classify? Why?
Compare your annotations with your group. Where did you disagree?
Why is dialogue act classification difficult?
Dialogue acts bridge form and function—what we say vs. what we do with words.