Relevant answers to polar questions

Area of Science:

• Cognitive Science
• Artificial Intelligence
• Linguistics

Background:

• Understanding relevance in communication is complex.
• Overinformative answers are common but poorly understood.
• Existing models struggle to capture nuanced relevance.

Purpose of the Study:

• To introduce a probabilistic cognitive model (PRIOR-PQ) for overinformative answers.
• To formalize the role of Theory of Mind (ToM) in relevance.
• To improve AI's ability to generate and interpret relevant answers.

Main Methods:

• Developed the PRIOR-PQ probabilistic cognitive model.
• Grounded pragmatics in inferences about questioner goals.
• Evaluated the model against human behavior in three case studies.
• Compared model performance with state-of-the-art large language models.

Main Results:

• PRIOR-PQ accurately predicts nuanced relevance in overinformative answers.
• The model outperforms existing computational models, including LLMs.
• Inferences about the questioner's goal are key to relevance.
• A novel chain-of-thought prompting method improves LLM performance.

Conclusions:

• Theory of Mind (ToM) plays a crucial mechanistic role in communication pragmatics.
• Goal-sensitive reasoning enhances AI dialogue systems.
• Integrating cognitive models with machine learning is vital for socially grounded AI.