최근 지능형 서비스 플랫폼으로 빠르게 발전하고 있는 아마존의 에코 서비스를 비롯해 애플의 시리, 마이크로소프트 코타나 등 글로벌 거대 IT 기업들의 다양한 가상비서(Virtual Assistant) 서비스들은 새로운 플랫폼 전쟁을 예고하고 있다. 언어처리 및 인공지능 기술의 발전은 사용자의 문제를 해결하고 상황 적합한 정보제공을 능동적으로 수행하는 가상비서뿐 아니라 마이크로소프트 테이처럼 스스로 학습하고 성장 가능한 챗봇과 이를 활용한 새로운 사업모델의 확산을 가속시키고 있다. 본 튜토리얼에서는 다양한 질의응답 시스템과 가상비서, 챗봇들의 개발 현황과 서비스 및 기술 특징들을 소개, 분석하고 관련한 발전 방향을 제시함으로 지능형 서비스 구현 목표를 가진 국내의 인공지능 연구자와 개발자들에게 도움을 주고자 한다.
Lecture-2: "Semantic interpretation of natural language questions for Korean quiz show challenge" (Pum-Mo Ryu)
Natural language questions of TV quiz shows embed multiple constraints in explicit or implicit manners. The question interpretation module for the TV quiz shows consists of three parts. The first question focus interpretation part identifies question focus, semantic answer type and lexical answer type for a given natural language question. The second semantic constraints interpretation part identifies predefined semantic relations between the question focus and other constraints. The semantic relations includes temporal & spatial constraint, ordinal constraint, language constraint, definition & apposition constraints. The last lexical frame interpretation part generates predicate-arguments structures from given questions. The structure consists of a predicate and multiple arguments including a question focus. The interpretation serves as both the input of SPARQL generation module for KBQA and the input of IRQA.
Lecture-3: "QA over linked data" (Christina Unger)
Knowledge bases and linked data have become important information sources for search. However, accessing the billions of RDF triples already available requires proficiency in the query language SPARQL as well as familiarity with the datasets available and their underlying schemas. As a result, there is a growing amount of research on interaction paradigms that allow end users to access linked data in natural language, hiding the complexity of Semantic Web standards behind an easy-to-use interface. Especially question answering systems play a major role, facing the key task of interpreting a user's information need with respect to the data that is queried. This talk will give an overview of the main challenges involved in question answering over linked data, as well as current developments that make it a rapidly evolving area of research and how they relate to OKBQA.
The second day program of the tutorial aims at helping audience become more familiar and better acquainted with the OKBQA framework and its core modules, so that they can in practice participate in cloning and making changes to the individual modules as well as to the entire framework throughout the hackathon later on. This day's tutorial will not only consist of technical introduction but also a series of hands-on exercises focusing on each and every module of the OKBQA framework. In the end of the tutorial, the audience are expected to come up with a better understanding and more ideas of how to implement a new QA system and how to improve existing systems and modules. This tutorial is indeed an essential course to take for maximizing the productivity of your participation in the hackathon.