EAGER: Collaborative Research: Interactive Dialog Agents for Social Language Development and Listening Comprehension

EAGER：协作研究：用于社交语言发展和听力理解的交互式对话代理

基本信息

批准号：
1748056
负责人：
Marilyn Walker
金额：
$ 20.32万
依托单位：
University of California-Santa Cruz
依托单位国家：
美国
项目类别：
Standard Grant
财政年份：
2017
资助国家：
美国
起止时间：
2017-10-01 至 2021-12-31
项目状态：
已结题

来源：
https://www.nsf.gov/awardsearch/showAward?AWD_ID=1748056&HistoricalAwards=false
关键词：
EAGER Collaborative Research Interactive Dialog

项目摘要

Practicing oral language skills supports children's later ability to process written text. Low socioeconomic status children receive lower levels of quality home language input, which negatively affects language development. Many children already suffer from a significant deficit in reading comprehension by the time they begin formal education, a deficit they may never overcome. This EArly Grant for Exploratory Research (EAGER) project will investigate improving children's home language environment with pedagogical learning agents, animated story characters and a child-like narrator who carry on a dialog with a child while telling a story. Story characters will talk directly to the child, tell the story from different perspectives, ask the child questions, phrase those questions in different ways using different vocabulary, and thus engage the child in understanding the story and improving their social language and listening comprehension skills. Pedagogical learning agents are a cheap and easily deployable technology that could readily be provided to children in any socioeconomic group.This project will develop technology for expressive interactive storytelling agents that model human conversational storytelling with a high level of social engagement. This will require methods to effectively manage a spoken dialog with a child, using speech recognition and text-to-speech output. The project will scaffold from a deep computational representation of narrative to generate different dialogs around the same story. Novel algorithms for natural language generation will support asking different types of questions at different story points and generation of story events with first person and direct speech to allow the characters to talk. The humanoid child narrator will model high immediacy nonverbal behaviors shown to increase engagement and positive affect. The project's prototype will be informed by regular evaluation by children at the UC Davis Reading and Academic Development Center, as the project assesses the promise of this exploratory technology to improve social language and comprehension skills.

练习口语技能可以支持孩子以后处理书面文本的能力。社会经济地位较低的儿童获得的家庭语言输入质量较低，这会对语言发展产生负面影响。许多孩子在开始正规教育时就已经患有阅读理解方面的严重缺陷，而他们可能永远无法克服这种缺陷。这项早期探索性研究资助 (EAGER) 项目将研究通过教学学习代理、动画故事人物和在讲故事时与孩子进行对话的儿童般的叙述者来改善儿童的家庭语言环境。故事人物将直接与孩子交谈，从不同的角度讲述故事，向孩子提出问题，使用不同的词汇以不同的方式表达这些问题，从而让孩子理解故事并提高他们的社交语言和听力理解能力。教学学习代理是一种廉价且易于部署的技术，可以很容易地提供给任何社会经济群体的儿童。该项目将开发用于表达性互动讲故事代理的技术，该代理以高水平的社会参与来模拟人类对话讲故事。这将需要使用语音识别和文本到语音输出来有效管理与孩子的口头对话的方法。该项目将从叙事的深度计算表示开始，围绕同一故事生成不同的对话。用于自然语言生成的新颖算法将支持在不同的故事点提出不同类型的问题，并使用第一人称和直接语音生成故事事件，以允许角色说话。人形儿童叙述者将模仿高度直接的非语言行为，以增加参与度和积极影响。该项目的原型将通过加州大学戴维斯分校阅读和学术发展中心的儿童定期评估来确定，因为该项目评估了这种探索性技术在提高社交语言和理解技能方面的前景。

项目成果

期刊论文数量（8）

专著数量（0）

科研奖励数量（0）

会议论文数量（0）

专利数量（0）

Curate and Generate: A Corpus and Method for Joint Control of Semantics and Style in Neural NLG

DOI：
10.18653/v1/p19-1596
发表时间：
2019-06
期刊：
ArXiv
影响因子：
0
作者：
Shereen Oraby;Vrindavan Harrison;Abteen Ebrahimi;M. Walker
通讯作者：
Shereen Oraby;Vrindavan Harrison;Abteen Ebrahimi;M. Walker

Neural MultiVoice Models for Expressing Novel Personalities in Dialog

在对话中表达新颖个性的神经多语音模型

DOI：
10.21437/interspeech.2018-2174
发表时间：
2018
期刊：
Proceedings of Interspeech 2018
影响因子：
0
作者：
Oraby, Shereen;Reed, Lena;T.S., Sharath;Tandon, Shubhangi;Walker, Marilyn
通讯作者：
Walker, Marilyn

Can Neural Generators for Dialogue Learn Sentence Planning and Discourse Structuring?

对话神经生成器可以学习句子规划和话语结构吗？

DOI：
发表时间：
2018
期刊：
Proceedings of the 11th International Conference on Natural Language Generation
影响因子：
0
作者：
Reed, Lena;Oraby, Shereen;Walker, Marilyn
通讯作者：
Walker, Marilyn

Neural Generation of Diverse Questions using Answer Focus, Contextual and Linguistic Features