CAREER: Plan-based Simulation of Human Story Understanding

讲故事是人类经验的一个基本组成部分：我们几乎每天都在讲故事和解释故事，以分享观点，互相教导，并以更引人注目的方式进行交流。但是，尽管叙事是我们大脑倾向于接受的一种信息形式，但一个悬而未决的问题仍然存在：人们如何理解故事？ 通过为专注于预测的叙事科学奠定基础，这一努力旨在为（科学家，技术人员和更广泛的公众）提供系统地构建故事的能力，这些故事能够与观众产生共鸣。这在我们已经看到叙事的使用时是有用的，当在其设计中具有更高程度的预测控制将有利于社会时：个性化学习、康复治疗和医疗保健通信、智能分析、自动化新闻生成和人类感知人工智能（AI）的进步。在AI中，发明能够模拟和解释我们如何处理故事的系统是“故事理解”的长期重大挑战。然而，这一挑战已被广泛采用的方法，忽略了人类理解故事的认知过程。要将叙事设计从不精确的实践提升为系统和可预测的方法，需要广泛的、跨学科的、以认知为基础的努力来重新构建故事理解人工智能的基础。 该项目概述了一条使用人工智能规划生成叙事的途径，这些叙事可预测地引发心理效应的轨迹，这些心理效应在形成其基础的三个关键认知过程中随着时间的推移塑造了个人的故事理解：基于事件的心理模型更新，推理和记忆。研究团队将设计算法来预测（1）在什么条件下人们会对他们所读的内容产生推断，这是保持他们参与的关键，（2）故事结构如何帮助或阻碍更新一个人对故事的心理模型，相对于人类的推断和记忆，以及（3）理解是如何通过与一个人处理故事的经验和技能有关的结构来介导的（这是故事心理学中目前尚未回答的问题）。沿着这一过程，这项工作将为人工智能做出基础性贡献，方法是开发一个生成基于计划的故事所需时间的正式模型，并重新定义叙事计划过程，以模拟人类如何在叙事过程中反复修改他们对故事的信念。与领域专家和心理学家一起，研究团队将在两个领域试点，完善和评估人工智能软件：技能培训的互动叙述（教育的组成部分）和公共科学传播（对外宣传不可或缺的一部分）--两者都需要设计故事，以实现更多--该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准。

Storytelling is a fundamental part of the human experience: we tell stories and interpret stories almost daily to share perspectives, teach one another, and communicate between ourselves in more compelling ways. But while narratives are a form of information to which our minds are predisposed, an open question remains: how do people understand stories? By establishing the foundations for a science of narrative that is focused on prediction, this effort aims to afford (to scientists, technologists, and the broader public) the ability to systematically construct stories that resonate with audiences as one intends. This is useful where we already see the use of narrative, when having a higher degree of predictive control in its design would benefit society: the advancement of personalized learning, rehabilitation therapy and healthcare communication, intelligence analysis, automated news generation, and human-aware artificial intelligence (AI).In AI, inventing systems that can model and explain how we process stories is the long-standing grand challenge of “story understanding.” However, this challenge has been broadly approached with methods that ignore the cognitive processes through which humans understand stories. To elevate narrative design from imprecise practices into a systematic and predictable methodology requires a broad, interdisciplinary, and cognitively-grounded effort to reformulate the foundation for story understanding AI. This project outlines a pathway toward using AI planning to generate narratives that predictably elicit a trajectory of mental effects that shape an individual’s story understanding over time across three key cognitive processes that form its basis: event-based mental model updating, inferencing, and memory. The research team will architect algorithms that predict (1) under what conditions people generate inferences about what they read, key to maintaining them engaged, (2) how story structure helps or hinders updating a person’s mental model of a story relative to human inferencing and memory, and (3) how understanding is mediated by said structure in relation to a person’s experience and skill at processing stories (a presently unanswered question in story psychology). Along the way, the effort will make foundational contributions to AI by developing a formal model of time needed to generate plan-based stories, and re-defining the narrative planning process to simulate how humans iteratively revise their beliefs about a story over the course of its narration. Alongside domain experts and psychologists, the research team will pilot, refine, and evaluate the AI software in two domains: interactive narratives for skills training (integral to education) and public science communication (integral to outreach)—both require the engineering of stories for the purpose of more-effective communication.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.