SBIR Phase I: A System for Automating End-to-End Creation of Natural Language Interfaces

小型企业创新研究(SBIR)第一阶段项目的更广泛影响/商业潜力是通过对话界面更高效地访问组织和企业的信息、商品和服务。据估计，到2017年底，美国安装了3300万台声控设备，消费者开始看到这些设备上商业服务的增长带来的好处。用自然语言与计算机交互比以往任何时候都更有效率，企业正迫切寻求将他们的服务纳入这种新的模式。然而，尽管机器学习的进步提高了建立对话界面的速度和准确性，但今天的解决方案仍然导致开发过程中的主要瓶颈。解决这些瓶颈将大大节省企业的时间和资源，并增加最终用户的便利性。这一小型企业创新研究(SBIR)第一阶段项目推动了机器学习和自然语言处理技术的发展，这些技术将能够快速创建对话界面，并提高个人与计算机交互的效率。尽管对话界面工具最近取得了重大进展，但当前在构建健壮的对话界面方面仍然存在瓶颈，因为开发过程的一部分从高度自动化转向手动编程。相反，建议的方法将实现端到端的自动化，使用创新的LSTM顺序到顺序机器学习模型来学习和模拟网页上的人类行为，以及用于开放领域问题回答的新系统。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project is to enable more efficient access to information, goods, and services from organizations and businesses through conversational interfaces. With an estimated 33 million voice-activated devices installed in the U.S. by the end of 2017, consumers are beginning to see the benefits of the rise in business services on these devices. Interacting with computers in natural language is more efficient than ever before, and business are urgently seeking to include their services on this new modality. Yet, while advances in machine learning have increased the speed and accuracy with which conversational interfaces are built, today's solutions still result in major bottlenecks in the development process. Addressing these bottlenecks would result in significant savings of time and resources for businesses, in addition to increased convenience for end users.This Small Business Innovation Research (SBIR) Phase I project advances the development of machine learning and natural language processing technologies that will enable rapid creation of conversational interfaces and increase the efficiency with which individuals interact with computers. Despite major recent advances in conversational interface tooling, current bottlenecks in building robust conversational interfaces still exist, as portions of the development process go from highly-automated to manually-programmed. Instead, the proposed approach will achieve end-to-end automation using an innovative LSTM sequence-to-sequence machine learning model to learn and simulate human behavior on a web page, as well as a new system for open-domain question answering.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.