Chapter 8: Dialogue Processing

Key Focus: Collaboration, Gricean Maxims, Common Ground, & Audience Design

• Traxler (2nd ed.), Chapter 8 Link to Chapter 8

Overview

This chapter examines dialogue as a collaborative joint activity where speakers and listeners coordinate to achieve shared goals. We explore the mechanisms that enable smooth conversation, including Gricean maxims, turn-taking, common ground building, and audience design. The chapter explains how we manage the complexities of real-time interaction and why successful dialogue requires mutual adaptation between conversation partners.

Learning Goals

After studying this chapter, you should be able to:

• Differentiate between dialogue and monologue in terms of collaboration and adaptation
• Apply Grice’s four maxims (Quantity, Quality, Relation, Manner) to analyze conversational implicatures
• Explain the mechanisms and timing of turn-taking in natural conversation
• Describe how common ground is built and maintained through the grounding process
• Analyze how audience design operates at lexical, syntactic, and prosodic levels
• Evaluate the evidence for egocentric bias in speaker adaptation
• Understand how listener perspective-taking enhances ambiguity resolution and memory
• Apply dialogue processing principles to explain real-world communication successes and failures

1. Introduction to Dialogue

1.1 What Makes Dialogue Unique?

Dialogue is collaborative—unlike monologue (speech), it requires adaptation and shared goals.

Foundational Example

A: Do you know where the nearest printer is?
B: The one in the break room? It was broken this morning.
A: Oh, thanks—any other options?
B: There’s one on the 3rd floor, by the copy center.

• Collaboration: B clarifies ambiguity (“the one in the break room?”) and offers an alternative—A signals ongoing need (“any other options?”).

1.2 The “Coordination Puzzle”

Core question: How do we coordinate so smoothly? (e.g., A says “Grab dinner?” → B responds with time/location, not random comments). Answer: Implicit mechanisms like Gricean maxims and common ground.

2. Gricean Maxims: The “Social Contract” of Dialogue

2.1 Cooperative Principle

Speakers/listeners follow an unspoken rule: “Make your contribution fit the conversation’s goal.” Violations (not mistakes) convey hidden meaning (implicatures).

2.2 Four Maxims (Compliance + Violations)

2.3 Implicatures: Hidden Meaning

Implicatures rely on shared cooperation. Example:

A: Should we invite Mike?
B: Mike’s busy with work.

• Literal: Mike has a heavy schedule.
• Implicature: Don’t invite Mike (B violates Quantity to avoid direct “no”).

3. Dialogue is Interactive (3 Non-Negotiable Traits)

3.1 Turn-Taking: The “Rhythm” of Dialogue

• Key Finding: Turns switch in 200–300 ms (faster than planning an utterance). Listeners predict turn ends using prosody (falling pitch, pauses).
• Example:

  A: Do you want to—
  B: Yes! I’ve been wanting to see that movie too.
  (B responds mid-A’s turn—proves prediction.)

• Breakdown Consequence: Gaps >1 second or overlapping speech cause confusion (e.g., A pauses → B asks “Meet at where?”).

3.2 Mutual Adaptation

• Speaker Adaptation: Adjust to listeners’ knowledge:
  • Doctor → Patient: “High blood pressure” vs. Doctor → Colleague: “160 mmHg systolic.”
• Listener Adaptation: Adjust interpretations to context:
  • Speaker says “The red one is mine” → Listener infers “red shirt” (not book) if looking at clothes.
• Evidence: Repeated speaker-listener pairs use shorter phrases over time (e.g., “red shirt” → “red one”) and fewer clarifications (Garrod & Anderson, 1987).

3.3 Repair Mechanisms: Fixing Breakdowns

Repair is shared—speakers and listeners fix misunderstandings:

• Listener-Initiated: A: “Pass the thingamajig.” → B: “The spatula?” → A: “Yes.”
• Speaker-Initiated: A: “Meet at the park—wait, no, the café (park’s closed).” → B: “Café works.”
• Why It Matters: Prevents small issues from escalating (e.g., passing the wrong tool).

4. Common Ground: The “Shared Knowledge Pool”

4.1 Definition

Common ground = mutually known information (community + personal):

• Community: Shared by a group (e.g., “subway = transit” for NYC residents).
• Personal: Shared by individuals (e.g., “we went to the same college” for friends).
• Example: Coworkers who attended the 10 AM meeting say “Did you hear the boss’s announcement?” (no need to explain “which meeting”).

4.2 Building Common Ground: The Grounding Process (Clark & Wilkes-Gibbs, 1986)

1. Speaker presents info: “Meet at the Main Street café.”
2. Listener acknowledges: “Got it—Main Street.”
3. Speaker confirms: “Yep, blue awning.”
4. Info joins common ground: Future references = “the café.”

Experimental Proof: The Director Task

• Setup: Director guides Matcher to select objects (grid of items).
• Trial 1 (No Common Ground): Directors use long descriptions: “Red cup, top row, 2nd left.”
• Trial 6 (Common Ground): Directors use “top-left red one” or “A.”
• Result: Descriptions 40% shorter; errors drop from 25% to 5%.

4.3 Heuristics to Verify Common Ground

Speakers use fast judgments to avoid over-explaining:

1. Community Membership: Assume group knowledge (e.g., NYC → “subway = transit”).
2. Prior Dialogue: Assume past info is shared (e.g., A mentions “dog Max” → B says “How’s Max?”).
3. Physical Co-Presence: Assume visible info is shared (e.g., pointing to rain → “Wait inside”).

• Failure: Tourist in NYC asks “Subway entrance?” → Listener (also tourist) confuses with underground walkway.

5. Audience Design: Tailoring to Listeners

5.1 Definition

Audience Design = speakers proactively adapt to listeners’ needs (mandatory for understanding—unlike robot scripts).

5.2 Three Levels of Adaptation

Experimental Evidence

• Brennan & Clark (1996): Speakers use “X” (labeled stapler) only if listeners see the label (70% of time) → avoid unshared words.
• Kraljic & Brennan (2005): Speakers use prosody to disambiguate 85% of time → listeners interpret correctly 90% of time (vs. 50% without prosody).

5.3 Limit: Egocentric Bias

Speakers over/underestimate listeners’ knowledge:

• Overestimate: Engineer says “API is down” to manager (no “API” knowledge).
• Underestimate: Teacher explains “2+2=4” to high school student.
• Why: Hard to “turn off” own knowledge (Keysar et al., 2003).
• Fix: Listener feedback (e.g., “What’s an API?”) or experience.

6. Listeners’ Perspective-Taking

Listeners actively take the speaker’s perspective—shaping comprehension in 3 ways:

6.1 Faster Ambiguity Resolution

Perspective focuses on speaker intent, not literal meaning.

• Example: Speaker hates red shirt, likes blue → says “Pass the nice shirt.” Listener takes perspective → passes blue shirt.
• Study (Keysar et al., 1998): Listeners who take perspective pass the visible small box (speaker doesn’t know hidden one) 80% of time—30% error rate without perspective.

6.2 Enhanced Memory for Speaker-Relevant Info

Listeners remember details that matter to the speaker.

• Example: Speaker says “Notebook for biology class” → Listener remembers “biology” (relevant) vs. “blue” (irrelevant).
• Study (Horton & Keysar, 1996): 30% better recall for speaker-relevant details.

6.3 Reduced Egocentric Interpretation

Perspective helps listeners set aside their own knowledge.

• Example: Listener knows café is closed → still understands speaker saying “Let’s go” (speaker doesn’t know).
• Neuro Evidence (Van Berkum et al., 2008): Smaller N400 (less confusion) when listeners take perspective—aligns with speaker intent.

Quick Review Questions

1. What is the key difference between dialogue and monologue?
2. How does violating the Maxim of Quantity create an implicature? Give an example.
3. What is common ground? Name one heuristic speakers use to verify it.
4. What is Audience Design? How do speakers adapt at the lexical level?
5. How does perspective-taking help listeners resolve ambiguity?