[Paper Review] An investigation into the correlation of cue phrases, unfilled pauses and the structuring of spoken discourse

Janet E. Cahn|ArXiv.org|Nov 22, 1995

Speech and dialogue systems9 references11 citations

TL;DR

This study investigates how cue phrases and unfilled pauses correlate with discourse structure in spoken directions-giving, using Grosz and Sidner's discourse theory as a framework. Results show that cue phrases align with discourse structuring tasks, and pause duration correlates with specific discourse modifications, suggesting prosodic cues help organize spoken discourse.

ABSTRACT

Expectations about the correlation of cue phrases, the duration of unfilled pauses and the structuring of spoken discourse are framed in light of Grosz and Sidner's theory of discourse and are tested for a directions-giving dialogue. The results suggest that cue phrase and discourse structuring tasks may align, and show a correlation for pause length and some of the modifications that speakers can make to discourse structure.

Motivation & Objective

• To explore the relationship between cue phrases, unfilled pauses, and discourse structure in spoken language.
• To test whether cue phrases and pause duration align with discourse structuring tasks as predicted by Grosz and Sidner's theory.
• To analyze how speakers modify discourse structure and whether such modifications correlate with pause length or cue phrase usage.
• To evaluate the role of prosodic features like pauses in signaling discourse boundaries and transitions.
• To contribute empirical evidence on prosodic cues in spoken discourse organization for natural language processing.

Proposed method

• The study analyzes a corpus of directions-giving dialogues from the IRCS workshop on prosody in natural speech.
• It applies Grosz and Sidner's theory of discourse to identify discourse structure components such as topic shifts and information packaging.
• Cue phrases are manually annotated for their discourse functions (e.g., addition, contrast, conclusion).
• Unfilled pauses (e.g., 'uh', 'um') are measured for duration and location relative to discourse boundaries.
• Statistical correlation is computed between pause duration and discourse structure modifications.
• The analysis compares cue phrase usage and pause patterns across different discourse structure types.

Experimental results

Research questions

• RQ1 Do cue phrases align with specific discourse structuring tasks as predicted by Grosz and Sidner's theory?
• RQ2 Is there a significant correlation between the duration of unfilled pauses and the type of discourse structure being constructed?
• RQ3 Do speakers use cue phrases and pauses in a coordinated way to signal discourse boundaries and transitions?
• RQ4 How do modifications to discourse structure (e.g., topic shifts) relate to the presence and length of unfilled pauses?
• RQ5 Can prosodic cues like pauses reliably signal discourse structure in spoken dialogue?

Key findings

• Cue phrases were found to align with discourse structuring tasks, supporting the theoretical prediction that they signal discourse organization.
• A significant correlation was observed between pause duration and specific discourse modifications, particularly around topic shifts and information packaging changes.
• Longer unfilled pauses were more likely to occur before or around discourse boundary markers, indicating a prosodic cue to structural reorganization.
• The use of cue phrases and pauses was not random; their distribution correlated with discourse structure, suggesting functional roles in spoken planning.
• The findings support the hypothesis that prosodic features such as pauses and cue phrases serve as cues for discourse planning and structural signaling in spoken language.