Logical constraints on relevance in spontaneous conversation

Abstract

The question of relevance in human conversation still remains highly mysterious. The range of what can be considered as relevant at a given point of a conversation is always small and sometimes dramatically restricted, compared to the number of conceivable utterances which are merely related to the subject of the conversation. Whenever these constraints of relevance are violated (for experimental purposes or after a misunderstanding) we observe reactions like "Why do you say this?" or "So what?" that may even be aggressive.

Relevance is a fundamental phenomenon. We may consider it as lying in the core of the language process : in many situations it dictates what you will say next, depending on what you know. Moreover, now that computers acquire some linguistic capabilities, understanding relevance will prove to be crucial for making them look somewhat smart. Despite this, relevance remains poorly understood.

We want here to study some logical aspects of relevance. We suggest that part of the relevance phenomenon may be analyzed and predicted by means of a logical description of the content of arguments. We studied more than 30 hours of casual conversation at the logical level, and we arrived at the conclusion that speakers non consciously obey very specific constraints when dealing with a new topic. According to our model of this conversational behavior, relevance requires the new topic to be presented as problematic. Following interventions then tend to reduce what makes the topic problematic.

Keywords: conversation, relevance, logic, argumentation, Artificial Intelligence

• The natural use of natural language
• Constraints on relevance
• Constraints and spontaneous conversation
• Are spontaneous conversations logical?
• Logical context and conversational breakdowns
• Conversational logic
• What makes a topic relevant ?
• What makes a reply relevant
• Discussion and conclusion

• Contents