Reading comprehension activities are an authentic task including a rich, language-based context, which makes them an interesting reallife challenge for research into automatic content analysis. For textual entailment research, content assessment of reading comprehension exercises provides an interesting opportunity for extrinsic, real-purpose evaluation, which also supports the integration of context and task information into the analysis. In this paper, we discuss the first results for content assessment of reading comprehension activities for German and present results which are competitive with the current state of the art for English. Diving deeper into the results, we provide an analysis in terms of the different question types and the ways in which the information asked for is encoded in the text. We then turn to analyzing the role of the question and argue that the surface-based account of information that is given in the question should be replaced with a more sophisticated, linguistically informed analysis of the information structuring of the answer in the context of the question that it is a response to. 1

Martha Palmer, whose guidance and support, and the personal time she has invested throughout my time as a graduate student, are much appreciated. Dan Gildea has been instrumental in helping me develop and focus my dissertation research topic. I would also like to thank Mitch Marcus, Fernando Pereira, and Ben Taskar, for accepting my invitation to participate in my thesis dissertation as members of the thesis committee. Finally, I would like to thank my wife, my parents, and my two brothers for their unwavering love, affection and support. ii

Seas of web pages in the Internet contain free texts in natural language that are only read by human beings. To be understandable for machines, these pages should be annotated with semantic markups. Manually annotating large amounts of pages is an arduous work. This has made automatic semantic annotation an urgent challenge. In this paper, we propose a machine-learning based automatic annotation approach. This approach can be trained for different domains and requires nearly no manual rules. The annotation is on the sentence level and is in RDF format. We adopt a dependency grammar -- Link Grammar [2] -- for this purpose. ALPHA system, a prototype of this approach has been developed with IBM China Research Lab. We expect many improvements are possible for this approach and our work may be selectively adopted or enhanced. 1

Seas of web pages in the Internet contain free texts in natural language that are only read by human beings. To be understandable for machines, these pages should be annotated with semantic markups. Manually annotating large amounts of pages is an arduous work. This has made automatic semantic annotation an urgent challenge. In this paper, we propose a machine-learning based automatic annotation approach. This approach can be trained for different domains and requires nearly no manual rules. The annotation is on the sentence level and is in RDF format. We adopt a dependency grammar -- Link Grammar [2] -- for this purpose. ALPHA system, a prototype of this approach has been developed with IBM China Research Lab. We expect many improvements are possible for this approach and our work may be selectively adopted or enhanced.

This paper considers weighted kernel functions for support vector machine learning with string data. More precisely,

family bias

We introduce a new kernel for Support Vector Machine learning in a natural language setting. As a case study to incorporate domain knowledge into a kernel, we consider the problem of resolving Prepositional Phrase attachment ambiguities. The new kernel is derived from a distance function that proved to be succesful in memory-based learning. We start with the Simple Overlap Metric from which we derive a Simple Overlap Kernel and extend it with Information Gain Weighting. Finally, we combine it with a polynomial kernel to increase the dimensionality of the feature space. The closure properties of kernels guarantee that the result is again a kernel. This kernel achieves high classification accuracy and is efficient in both time and space usage. We compare our results with those obtained by memory-based and other learning methods. They make clear that the proposed kernel achieves a higher classification accuracy. 1

This paper presents the outcomes of research into using lingual parts of music in an automatic mood classification system. Using a collection of lyrics and corresponding user-tagged moods, we build classifiers that classify lyrics of songs into moods. By comparing the performance of different mood frameworks (or dimensions), we examine to what extent the linguistic part of music reveals adequate information for assigning a mood category and which aspects of mood can be classified best. Our results show that word oriented metrics provide a valuable source of information for automatic mood classification of music, based on lyrics only. Metrics such as term frequencies and tf*idf values are used to measure relevance of words to the different mood classes. These metrics are incorporated in a machine learning classifier setup. Different partitions of the mood plane are investigated and we show that there is no large difference in mood prediction based on the mood division. Predictions on the valence, tension and combinations of aspects lead to similar performance. 1.