An ascription-based approach to speech acts

Mark Lee, Yorick Wilks

Abstract

The two principal areas of natural language processing research in pragmatics are belief modelling and speech act processing. Belief modelling is the development of techniques to represent the mental attitudes of a dialogue participant. The latter approach, speech act processing, based on speech act theory, involves viewing dialogue in planning terms. Utterances in a dialogue are modelled as steps in a plan where understanding an utterance involves deriving the complete plan a speaker is attempting to achieve. However, previous speech act based approaches have been limited by a reliance upon relatively simplistic belief modelling techniques and their relationship to planning and plan recognition. In particular, such techniques assume precomputed nested belief structures. In this paper, we will present an approach to speech act processing based on novel belief modelling techniques where nested beliefs are propagated on demand.

Introduction

Traditional dialogue systems often struggle with the "infinite regression" of mutual belief—the computationally expensive loop of "I think that you think that I think..." required for true mutual understanding. This paper challenges this conventional model, arguing that such deep nesting is both computationally inefficient and psychologically unrealistic for natural human conversation, which rarely exceeds two levels of nesting.

Instead of precomputing complex nested belief stacks, the authors propose an "ascription-based" approach. This method treats utterances as actions within a plan but simplifies the underlying belief modeling by generating nested beliefs only when necessary. This shift allows for more robust handling of misconceptions and "badly formed" plans without the overhead of maintaining an objective, external "mutual knowledge" state.

Methodology: The ViewGen System

The proposed framework is built upon ViewGen, a system designed to represent mental attitudes—beliefs, goals, and intentions—within nested "intensional environments." The core innovation lies in the mechanism of Ascription, which replaces precomputed mutual belief with on-demand propagation.

Default Ascription

The system operates on a principle of cognitive economy: it assumes other agents share its common-sense beliefs about the world unless explicit evidence suggests otherwise. Beliefs are "pushed" from the system's outer environment into the inner model of another agent as needed.

Figure 2: ViewGen default ascription mechanism showing belief propagation.

Stereotypical Ascription

Beyond general world knowledge, the system can attribute clusters of beliefs based on an agent's role or category. For instance, if an agent is identified as a "Doctor," the system automatically ascribes a set of medical beliefs (e.g., knowledge about pneumonia and its treatments) to that agent's environment.

Figure 3: Stereotypical ascription of medical knowledge based on agent roles.

Speech Act Taxonomy and Planning

The authors define a taxonomy of 20 speech act types (such as questions, requests, and informs) based on the mental preconditions required for a speaker to perform them. These acts are integrated into a non-linear Partial Order Causal Link (POCL) planner. By reasoning within ascribed belief environments, the system can effectively simulate another agent's planning process to better understand or predict dialogue moves.

Key Insights: Solipsism and Minimalism

A central theme of this work is "solipsism as a strategy." In this context, solipsism refers to the system basing its actions entirely on its own finite representations of the speaker or hearer. This avoids the need for an objective "mutual knowledge" state, making the system more scalable.

Furthermore, the approach emphasizes minimalism in speech act definitions. Acts assume only the bare minimum conditions for success. Complex interpretations, such as irony or deception, are not hard-coded but are instead derived through the ascription process when the context demands it.

Results and Future Directions

While the paper focuses on architectural and theoretical contributions, several key advantages emerge:

  • Computational Efficiency: By avoiding infinite regression and precomputation, the model is better suited for real-time dialogue processing.
  • Robustness to Misconceptions: Explicitly differentiating between speaker and hearer belief sets allows the system to detect and repair misunderstandings that "mutual belief" models might overlook.
  • Extensibility: The hierarchical taxonomy of speech acts, where specific acts (like "Correction") inherit properties from more general ones (like "Inform"), provides an elegant framework for future expansion.

Future work will focus on integrating a robust dialogue parser to process raw text and empirically validating the belief sets against real-world data, specifically within medical domains using platforms like LaSie.

Summary updated

Abstract

The two principal areas of natural language processing research in pragmatics are belief modelling and speech act processing. Belief modelling is the development of techniques to represent the mental attitudes of a dialogue participant. The latter approach, speech act processing, based on speech act theory, involves viewing dialogue in planning terms. Utterances in a dialogue are modelled as steps in a plan where understanding an utterance involves deriving the complete plan a speaker is attempting to achieve. However, previous speech act based approaches have been limited by a reliance upon relatively simplistic belief modelling techniques and their relationship to planning and plan recognition. In particular, such techniques assume precomputed nested belief structures. In this paper, we will present an approach to speech act processing based on novel belief modelling techniques where nested beliefs are propagated on demand.
BibTeX
@misc{lee1999ascription,
  author = {Mark Lee and Yorick Wilks},
  title = {{An ascription-based approach to speech acts}},
  year = {1999},
  eprint = {cs/9904009},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/cs/9904009},
}
Figure