TL;DR
This paper introduces "attitude emergence," a novel mechanism for inferring a speaker's underlying attitudes and speech intent in persuasive discourse, particularly advertisements. By constructing and assimilating "attitude models" through bridging inferences and mutual knowledge, the system can better understand the persuasive intent behind seemingly disjointed statements.
What Problem Does This Paper Address?
Interpreting persuasive discourse, especially advertisements, is challenging because speakers often don't explicitly state their intentions. Instead, they present a series of statements that, when taken together, are meant to subtly influence the listener's attitudes and beliefs. Traditional approaches, like "belief percolation," focus on how beliefs are passed down, but they struggle to capture how seemingly unrelated statements can collectively build a persuasive argument. The authors highlight the need for a mechanism that can infer the speaker's underlying attitudes (beliefs, desires, preferences) and, consequently, their speech intent, by bridging these statements through shared knowledge and inferential reasoning.
Methods
The core of the paper is the "attitude emergence" mechanism, which operates in three steps:
- Construct Initial Attitude Model (A-model): An initial model is built for each statement, representing the agent and their attitude towards the information.
- Assimilate Statements and Recognize Semantic Association: Successive statements are integrated into a coherent A-model. Crucially, bridging inferences based on mutual knowledge are used to connect these statements and identify semantic associations between them.
- Percolate Upwards for Attitude Change: The recognized semantic associations are propagated upwards through the A-model's context. This process calibrates and enhances the interpretation of attitudes in higher contexts, ultimately leading to the recognition of the proper speech intent.
This process involves constructing A-models as recursive structures and utilizing simulative reasoning within these contexts. The authors also incorporate common sense and logical reasoning to handle more complex scenarios.
Datasets
The paper mentions the implementation of a system called BUYER, which is being used to understand food advertisements from Reader's Digest. A corpus of 120 advertisements was collected for constructing and verifying the system. The paper presents an example of BUYER processing a Folgers coffee advertisement.
Key Insights and Ideas
- Bridging Inferences are Key: The persuasive power of advertisements lies not just in individual claims but in how seemingly disparate pieces of information are connected through inferential reasoning.
- Mutual Knowledge is Crucial: Understanding advertisements relies heavily on shared semantic and world knowledge between the speaker and the hearer.
- Attitude Models as Dynamic Structures: A-models are not static representations but evolve as new information is assimilated, leading to a richer understanding of the speaker's intent.
- Beyond Simple Percolation: Attitude emergence goes beyond simply "pushing down" beliefs; it emphasizes how assimilation and semantic associations change attitudes in higher contexts, which is vital for persuasive discourse.
- Ambiguity Resolution: The mechanism can handle ambiguous statements by exploring multiple potential attitude models and using contextual information and world knowledge to determine the most likely interpretation.
Key Results
The paper demonstrates the effectiveness of attitude emergence through a detailed analysis of a Folgers coffee advertisement. The system, BUYER, is shown to infer the speech intents behind statements like "Is your decaffeinated as dark as ours?" (interpreted as an intention to inform, not just a question) and "So dark and rich, shouldn't you switch?" (interpreted as a motivation to switch). The paper highlights how the system uses world knowledge and logical reasoning to resolve ambiguities and derive persuasive intents. While quantitative results are not the primary focus, the qualitative analysis showcases the system's ability to interpret complex persuasive strategies.
Limitations and Future Work
The authors acknowledge that the current formulation of attitude emergence primarily focuses on the "force" aspect of a statement. Future work aims to:
- Systematize Attitude Dynamics: Develop a fuller formulation that includes both "force" and "counter-force" to reflect the dynamics of enforcement and resistance towards expressed attitudes.
- Explore Force Dynamics: Investigate the relevance of Talmy's "force dynamics" framework to better explain the psychological reality of persuasive force and pressure.
- Further Generalization: Continue to test and refine the BUYER system with a larger corpus of advertisements to ensure the generality of the attitude emergence mechanism.
- Deeper Reasoning: While the paper incorporates common sense and logical reasoning, there's always room for deeper and more sophisticated inferential capabilities.
Summary updated