Unsupervised Frame Induction: Event Type Hierarchies and Complex Event Types (FInd)

FIND项目旨在使用概率框架建模和机器学习的技术，无监督地归纳由动词描述的事件的框架(即事件类型)。这里使用的术语“无监督的”是指关于语义框架的信息，即系统只接受句法信息作为输入(例如词性标签和依存性)。除了研究一般的无监督框架诱导技术外，FIND还特别关注框架之间的层次关系和由动词交替产生的递归结构框架。此外，从英语框架诱导的方法开始，FIND将把这些扩展到多语言环境，重点是某些类型的动词多词表达的框架诱导。

The project FInd aims at the unsupervised induction of frames (i.e., event types) for events described by verbs, using techniques from probabilistic frame modeling and from machine learning. The term ``unsupervised" is used here with respect to information about semantic frames, i.e., the system takes only syntactic information as input (for instance part-of-speech tags and dependencies). Beyond investigating techniques of unsupervised frame induction in general, FInd focuses in particular on hierarchical relations between frames and on recursively structured frames that arise from verb alternations. Furthermore, starting from approaches to frame induction for English, FInd will extend these to a multilingual setting with a focus on frame induction for certain types of verbal multi-word expressions.