TL;DR
This paper proposes a formal model for simulating cooperative dialogs between two partners solving a problem. It introduces the concepts of a "communicative cycle" and "communicative strategy" to describe how participants guide each other's mental states (P-parameters) to achieve their goals, aiming to bridge the gap between human-human and human-computer communication.
What Problem Does This Paper Address?
The paper tackles the challenge of creating computational models that can effectively simulate human-like dialog. While previous work has focused on generating dialog structures or grammars, this research aims to go deeper by modeling the underlying process of communication. Specifically, it addresses how participants in a dialog, acting as cooperative partners, influence each other's decision-making and understanding to collaboratively solve a problem. The ultimate goal is to make human-computer interaction more natural and intuitive.
Methods
The core of the proposed model lies in the concepts of the Communicative Cycle (CC) and Communicative Strategy (CS).
- Communicative Cycle (CC): This is defined as a dialog arising between two partners while solving a problem. It's characterized by alternating turn-takings, where each turn is composed of communicative acts. A minimal CC involves a couple or triplet of turns (e.g., question-answer, question-answer-acceptance).
- Communicative Strategy (CS): This is a general scheme guiding a dialog participant towards achieving their goal. A CS is formalized by specifying:
- The relevant P-parameters (psychic parameters) of the partner that are crucial for the goal. These include assessments, knowledge, interests, and requirements.
- Whether the goal requires increasing or lowering the value of each relevant P-parameter.
- An immediate goal (Gi) which is the specific P-parameter to be changed.
- How to change the value of the immediate goal P-parameter, either directly through communicative acts or indirectly by influencing other P-parameters.
The model suggests that a participant (A) aims to change the partner's (B's) P-parameters in a way that leads B to a desired decision. This involves A constructing a sequence of communicative acts that subtly influences B's knowledge, assumptions, assessments, and wishes. The process is iterative, with each participant updating their model of the other based on the received information.
Datasets
Not discussed in the paper. The paper describes a theoretical model and mentions its implementation in a pilot human-computer dialog system (TARLUS) but does not detail specific datasets used for training or evaluation.
Key Insights and Ideas
- Dialog as Goal-Oriented Influence: The paper frames dialog not just as an exchange of information, but as a strategic process where one participant actively tries to influence the other's mental state (P-parameters) to achieve a specific outcome.
- P-Parameters as the Core of Influence: The concept of "P-parameters" (psychic parameters) is central. These represent the internal states of a participant (knowledge, beliefs, desires, assessments) that drive their decisions. The effectiveness of a communicative strategy hinges on its ability to manipulate these parameters.
- Indirect Influence is Key: Achieving complex dialog goals often requires indirect manipulation of P-parameters. For instance, to make someone want to do something, you might first influence their assessment of the positive consequences of that action, rather than directly commanding them.
- Dialog Models Need to Capture Process, Not Just Structure: The authors argue that a true dialog model must describe the process of communication and the underlying cognitive mechanisms, not just the surface-level structure of the dialog.
Key Results
The paper presents a formal model for simulating cooperative dialogs. It outlines the components of a communicative strategy and the steps involved in a communicative cycle. The model has been used to underpin the pilot human-computer dialog system TARLUS, which can perform morphological, syntactic, and semantic analysis/generation for Estonian texts and recognize descriptions of criminal events.
Limitations and Future Work
- Complexity of P-Parameters: The paper acknowledges that a full explication of all possible P-parameters requires extensive investigation, particularly psychological research.
- Qualitative Theory Needed: The development of communicative strategies presupposes a qualitative theory of P-parameters and how they influence decision-making.
- Future Work: The authors suggest that further work is needed to fully develop the theory of P-parameters and their influence on decisions. The TARLUS system, while functional, represents an initial realization of the proposed model.
Summary updated