Research on Advanced Natural Language Processing and Text Mining

该项目的目标是将统计建模与基于结构的符号处理相结合的方法应用于更具挑战性的任务，如深层语义处理、基于知识的信息提取和上下文处理，这种方法在句子分析中已被证明是成功的。我们在以下方面取得了显著的成果：（1）基于语言学上合理的形式主义的高效和鲁棒的深度解析，（2）生物学领域的大规模语义注释语料库（GENIA语料库），（3）信息抽取程序（命名为实体识别器和事件识别器），其联合收割机了（1）中的深度解析和结构机器学习算法，（4）以数据为中心的并行处理工作流软件（2）中的GENIA语料库被公认为生物文本挖掘研究的金标准语料库，已被世界上许多团体使用。它被两次国际共享任务竞赛（BioNLP 09和BioNLP 11）用作训练和测试语料。在（3）中开发的提取程序成功地展示了这些国际共享任务竞赛中的最先进性能。基于（1）和（4）的系统表明，本项目开发的技术对于处理真实的世界文本是实用的。我们成功地处理了整个MEDLINE（2000万篇摘要，超过20亿个句子），并在不到一周的时间内对它们进行了语义索引。MEDLINE的处理结果已通过智能文档检索系统（MEDIE）公开提供

The objective of the project was to apply the methodology of combining statistical modeling with structure-based symbolic processing, which had proven successful in sentence parsing, to more challenging tasks such as deep semantic processing, knowledge-based information extraction and contextual processing. We have achieved significant results in (1) efficient and robust deep parsing based on a linguistically sound formalism, (2) a large scale semantically annotated corpus for the biology domain (GENIA corpus), (3) information extraction programs (named entity recognizers and event recognizers) for the biology domain which combine the deep parsing in (1) and structural machine learning algorithms, and (4) Workflow software for data-centered parallel processing.The GENIA corpus in (2) has been recognized as the gold standard corpus for research of text mining for biology and has been used by many groups in the world. It was adopted as the training and test corpus for international shared task competition twice (BioNLP 09 and BioNLP 11). The extraction programs developed in (3) successfully showed the state of the art performance in these international shared task competitions. The system based on (1) and (4) showed that the technology developed by this project was practical for processing the real world text. We successfully processed the whole of MEDLINE (20 million abstracts, more than 2 billion sentences) and indexed them semantically in less than a week. The processing results of MEDLINE has been made publicly available through an intelligent document retrieval system (MEDIE)