Methods and Applications for Relation Detection Potential and Limitations of Automatic Learning in IE

The detection of relation instances is a central functionality for the extraction of structured information from unstructured textual data and for gradually turning texts into semi-structured information. Experience from many years of shared-task efforts in the MUC and ACE frameworks has led to promising initial results but also to frustrating barriers. But the systematic collective efforts have also yielded valuable insights into the complexity of the task and the limitations of existing approaches. An entire research area has emerged from the numerous efforts to increase the sophistication of the various approaches to relation extraction and from the obtained empirical results. In the meantime, it has become clear that there cannot be a single best method for relation extraction since there are many types of relations differing in complexity and in their reflection in the vocabulary of the language. The scale of complexity ranges from simple binary relations of frequent entity types all the way to complex embeddings of relations of various arity. Types of opinions and complex events are just special types of relations. For some relations the language provides prepositions, verbs or other lexemes that allow a concise and compact encoding. Others have to be described by a combination of different words and constructions. Relation extraction tasks do not only differ in the complexity and the linguistic inventory associated with the relevant relations. They also differ with respect to the size and nature of the available data for training and application. With respect to the applied mathematical methods, we find discrete (or symbolic) and non-discrete approaches. The latter are usually statistical methods. We also witness a growing tendency to combine different methods. With respect to the acquisition of the classifiers or detection grammars, the existing approaches fall in three large categories: i. detection by classifiers/grammars acquired through intellectual huma- n labor ii. detection by classifiers/grammars acquired through supervised learning iii. detection by classifiers/grammars acquired through unsupervised or minimally supervised learning In the talk we will provide examples for the classes of approaches and summarize their respective advantages and disadvantages. We will argue that different relation detection tasks require different methods or even different combinations of methods.