STTR Phase I: Authoring Tools for Virtual Role-Play

点击翻译按钮获取中文摘要

This SBIR Phase I project will develop a revolutionary new tool for creating engaging language-learning content. Educational publishers, teachers, and even students will be able to create artificially intelligent "virtual role-players" that converse in foreign languages, responding to either speech or text input. Large-scale on-line communities of educators will be able to create, customize, and share their own virtual role-play content, thus changing how people learn foreign languages. This will have a major impact on world language education and will help the US develop a more globally competent workforce. Many schools, especially in rural and underserved areas, have difficulty maintaining foreign language programs due to lack the resources to hire sufficient qualified teachers. This innovation will greatly increase the availability of high-quality interactive language learning resources that engage learners. These will be available online at little or no cost to schools. This could have a broad impact on world language enrollments as well as student retention in world language programs. In the longer term, virtual role-play can have a broad disruptive impact on training and workforce development. Employers are increasingly concerned about the skill gaps in the workforce, and communication and other 21st century skills are coming to the fore. This project will establish the technical and commercial feasibility of a new tool and platform that enables non-programmers to create sophisticated virtual role-play simulations for learning foreign languages and other communicative skills. Competing can construct linear role-play scenarios, but are not suitable for creating interactive role-play simulations in which learners can take the initiative and where there are multiple possible outcomes. To break down the problem and make authoring feasible for teachers and students, the project will focus first on a significant, commonly occurring component of such dialogs, namely context-specific natural-language question-answering. The tool will automatically train a statistical language understanding system to recognize target questions so the virtual role-player can respond to them. This data-driven approach ensures that as users create content the body of training data continues to grow, making it ever easier to create new content. The toolset will help manage the necessary training data. The approach has the potential to change world language education from a teacher-led, classroom-centric activity to a personalized, learner-centric activity. It can flip the language classroom in ways that watching videos online cannot, by helping learners develop their conversational skills outside of class.