TL;DR
This paper introduces a novel two-level dialogue representation system designed to capture the syntactic structure and underlying goal structure of conversations. This framework aims to provide a more robust foundation for computational dialogue systems, particularly in handling communication failures.
What Problem Does This Paper Address?
Computational modeling of discourse has long grappled with effectively representing the complex, multi-layered nature of human dialogue. While some approaches have focused on syntactic descriptions or specific communicative acts, there's a need for a unified system that can model both the surface structure of conversations and the deeper, intentional goals driving them. This is crucial for building intelligent dialogue systems that can understand, generate, and recover from communication breakdowns. The authors highlight the limitations of purely syntactic models and the necessity of incorporating semantic interpretations based on participant goals.
Methods
The proposed system employs a two-level representation:
- Syntactic Level: This level describes the structure of a dialogue using a taxonomy of communicative acts and moves, building upon Burton's (1981) work. It defines hierarchical units like interactions, transactions, exchanges, moves, and acts. Moves are further categorized into types such as delineating, sketching, opening, supporting, challenging, bound-opening, and re-opening. Acts are classified based on their linguistic structure and function, with 15 distinct acts identified (e.g., marker, elicitation, informative).
- Semantic Level: This level captures the underlying goal structure of the dialogue. Two types of goals are identified:
- Substantive Goals (S-goals): These relate to the actual content or task of the dialogue.
- Communicative Goals (C-goals): These relate to the manner in which the content is conveyed (e.g., ensuring clarity, managing turn-taking). The system posits hierarchical relationships between goals and subgoals, with S-goals being more dominant at higher dialogue levels and C-goals at lower levels.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- Two-Level Representation is Key: The core insight is that a comprehensive dialogue model requires both a syntactic description of communicative acts and a semantic interpretation of participant goals.
- Goals Drive Dialogue Structure: The paper argues that dialogue constituents (like exchanges) are motivated by underlying goals, and the structure of these goals directly influences the conversational flow.
- S-Goals vs. C-Goals Bifurcation: The distinction between substantive and communicative goals provides a powerful lens for analyzing dialogue. This bifurcation becomes apparent as one moves down the hierarchy of dialogue units, with S-goals dominating at higher levels and C-goals at lower levels.
- Exchanges as Topic-Bearing Units: Exchanges are identified as the fundamental units for topic management, with their structure reflecting topic shifts and re-introductions.
- Moves as Rhetorical Elements: At the move level, the focus shifts from substantive goals to rhetorical functions, such as signaling topic changes or continuations.
Key Results
The paper presents a theoretical framework and a descriptive model rather than empirical results from a deployed system. The "results" are the conceptual advancements:
- A formal description of dialogue units (interactions, transactions, exchanges, moves, acts) and their hierarchical relationships.
- A taxonomy of 15 communicative acts with definitions based on both structure and function.
- A conceptual model for distinguishing and relating substantive (S-goals) and communicative (C-goals) structures within dialogues.
- The proposed framework is designed to be applicable to both person-person and person-machine dialogues.
Limitations and Future Work
- Inference Mechanism: The paper acknowledges that it does not define the process for inferring the goal structure from the syntactic structure of a dialogue. This is identified as a crucial area for future research.
- Grammar Specification: Further research is needed to specify a more detailed grammar for exchanges and moves.
- Goal Correspondence: Establishing a stricter correspondence between more types of goals and dialogue units is also a future direction.
- Formal Devices for Inference: A key challenge for future work is to specify the formal devices by which the semantic (goal) structure can be inferred from utterance and act elements and "raised" to higher dialogue units.
Summary updated