U.S.-Switzerland Cooperative Research in Computational Linguistics

该奖项支持布兰代斯大学的James Pustejovsky教授和一名研究生与瑞士日内瓦大学语义与认知研究所（ISSCO）的Susan Armstrong等人在计算语言学研究方面的合作。他们研究的目标是解释单词组合的方式，利用不同的视角来研究这个问题，即词典学、语言学和统计学。问题是在计算语义理论和通过“基于使用”的标准直接处理建模语言使用的实际系统之间建立联系。Brandeis小组在词汇语义和自然语言生成方面带来了丰富的理论经验，而Swiss小组则提供了用于分析文本的计算平台和多语言环境，以及机器翻译系统的经验。美国和瑞士研究团队之间的合作将引入跨语言视角，利用瑞士参与者提供的多语言语料库分析所得的信息。具体地说，瑞士小组参与了一个项目，该项目旨在自动生成由瑞士气象研究所制作的多语言天气预报。这项合作研究为语义选择理论的实证基础做出了贡献。从根本上说，它关注的是自然语言中的词义是如何被形式上表征的，以便解释不同句子中单词的微妙使用，以及在新语境中单词的“创造性”使用。基于规则的技术和数据驱动技术的集成是自然语言处理研究的一个重要领域。在实践层面上，本研究将有助于多语言生成器的开发和机器翻译系统的改进。

This award supports Professor James Pustejovsky of Brandeis University and a graduate student to collaborate in computational linguistics research with Susan Armstrong and others at the Institute of Semantic and Cognitive Studies (ISSCO) of the University of Geneva, Switzerland. The goal of their research is to account for the way in which words combine, taking advantage of the different perspectives from which this issue has been approached, namely, lexicographic, linguistic and statistical. The problem is to establish a connection between computational semantic theories and practical systems dealing directly with modeling language use through `usage-based` criteria. The Brandeis group brings substantial theoretical experience on lexical semantics and natural language generation, while the Swiss group contributes the computational platform and the multilingual environment for analyzing text, as well as experience in machine translation systems. Collaboration between the U.S. and Swiss research teams will introduce a cross- linguistic perspective by taking advantage of information derived from the analyses of multilingual corpora provided by the Swiss participants. Specifically, the Swiss group is involved in a project to produce automatic multilingual generation of the weather forecasts produced by the Swiss Institute of Meteorology. This collaborative research contributes to the empirical grounding of a theory of semantic selection. Fundamentally, it looks at how word meaning in natural language might be characterized formally, in order to account for both the subtle use of words in different sentences, and the `creative` use of words in novel contexts. The integration of rule-based technique and data driven technique is a major area of interest in natural language processing research. On a practical level, this research will contribute to the development of a multilingual generator and the enhancement of machine translation systems.