Unveiling Information Through Narrative In Conversational Information Seeking

July 04, 2024

We, as humans, have the ability to create and communicate narratives. We use narratives to make sense of the world around us, share knowledge, and solve complex problems. This got us wondering: How should Conversational Information-Seeking agents present information:

(1) They should only reflect facts, (2) Cater to human cognitive preferences, (3) Strike a balance between them.

**Figure 1:** Contrasting approaches to information delivery: Factual vs. Narrative. Supposing that a conversational agent aims to play the role of a politician to address a healthcare policy.

This question led to our recent research project, which we are very excited about. In particular, our goal is to explore how narrative can help align conversational agents such as GPT-4 or Gemini with the human brain’s cognitive process to make them more understandable.

The Inspiration: Human Brain and Narrative

First, let’s consider why narratives are so powerful for human cognition. Narrative as a noun refers to a story or a description of a series of events. When we encounter a story, our brain lights up in fascinating ways. Multiple areas become engaged simultaneously – from Broca’s area to Sensory Cortices that help us process language and imagine sights, sounds, and even smells. This holistic activation allows us to become deeply absorbed in the narrative, a phenomenon known as “narrative transportation”.

**Figure 2:** Brain receiving facts (left) vs. Brain receiving stories (right).

Throughout this research, we aim to investigate how integrating narrative techniques with conversational agents can be systematically accomplished. We propose several steps that help answer this question.

Applicability of Narrative in Conversational Information-Seeking Tasks:

Inspired by findings from studies about how the brain processes narratives, which identify active brain regions involved in narrative comprehension compared to factual processing.

Narrative techniques can enhance conversational information-seeking across various domains, particularly where users appreciate examples, human experiences, and relatable contexts alongside facts.

Exploratory Information-Seeking

Sense-making and Knowledge Integration

Personal Experiences

Problem-solving and Decision-making

Persuasion

Narrative Generation for Conversational Information-seeking Systems:

By considering the conversational information-seeking tasks and requirements, we suggest different components for an optimal form of a novel conversational narrative information-seeking system in Table 1 that facilitate effective interaction between humans and conversational agents during information-seeking conversations.

**Table 1:** Desired requirements for conversational information-seeking systems and respective narrative-driven solutions.

Figure 3 shows an example of a conversational narrative information-seeking system in an exploratory search scenario, incorporating narrative techniques such as Analogical Reasoning, which identifies similarities between the target concept (information being conveyed) and a more familiar concept (analogy), highlighted in blue and black colors, respectively, and Progressive Disclosure which reveals information gradually throughout the narrative, rather than presenting it all at once, presented by line breaks.

**Figure 3:** An example of a conversational narrative information-seeking system.

Looking Ahead: Future Work

We proposed a novel framework for a conversational narrative information-seeking system by leveraging narrative techniques. We examined the applicability of the conversational narrative information-seeking system across various domains.

As a last point, we see our task in developing this idea further by designing evaluation techniques to strike a balance between accuracy, engagement, and understandability.

Vahid Sadiri Javadi contributed to this article.

Previous post
NAACL2024 was great! Next post
CAISA group welcomes Dr. Cass Zhixue Zhao