Development of Question Answering Techologies on Large-Scale Text

大规模文本问答是一种通过参考存储的大规模文本来回答用户任何问题的技术。它允许用户在考虑他们的问题中所表达的兴趣和观点的情况下，获得必要的信息。本研究旨在将问答发展成为一种成熟的以用户为中心的信息获取手段，取得了以下成果：首先，研究了一系列相关问题的交互式问答技术，取代了现有的孤立问题问答技术。假设这样的技术有助于报告的写作，我们检查了那里提出了什么样的问题，以及回答这些问题需要什么程序。在测试结果的基础上，我们设计了一种新的任务来测量互动问答的能力，并构建了该任务的测试集。这项任务对最新技术的分析表明，我们仍然有很大的空间来处理这些情况，并且在处理列表类型的问题时仍然存在一些理论问题，这要求系统列举所有并且只列出正确的答案。其次，将图表的使用等可视化模式融入到当前的基于文本的QA中。在这一部分的研究中，我们证明了通过考虑用户的意图来合理使用图表是可能的，并提出了一种通过对话来引出这种意图的方法。我们将这种多模式问答向前推进，提出了一个涵盖文本摘要、信息可视化和问答等领域的新的研究课题--趋势信息的摘要与可视化。我们提出了一个研究框架，为研究构建了一个数据集，其中包括最初设计的有助于摘要和可视化的编码系统的语义注释报纸文章语料库。

Question Answering (QA) on large-scale text is technology for answering any question from users by referring to stored large-scale text. It allows users to obtain the necessary information itself with consideration of the interest and perspective expressed in their questions. Our research aims at to develop QA and to establish it as a sophisticated user-centered information access means, and achieved the following results.First, QA technologies for answering series of related questions interactively were examined, which would replace the current QA technologies for isolated questions. Assuming that such a technology helps report writing, we examined what kind of questions are made there and what procedures are needed for answering those questions. Based on the findings of the examination, we designed a novel task for measuring the abilities useful for interactive QA has been and constructed a test set of that task. Analysis of the state-of-art technologies by this task revealed that we have still have a large room to handling those situations, and some theoretical problems remain in handling list-type questions, which request systems to enumerate all and only correct answers.Second, incorporating a visual mode such as usage of charts into the current text-based QA was examined. In this part of the research, we showed that an appropriate use of charts get possible by considering users' intention, and proposed a method for eliciting such an intention thorough dialogues. We pushed this multi-modal QA forward and proposed a novel research topic, "summarization and visualization on trend information", that covers the fields of text summarization, information visualization and QA. Presenting a framework of research, we constructed a data set for the research, which includes a corpus of semantically annotated newspaper articles of an originally designed coding system useful to summarization and visualization.