An important problem in artificial intelligence is capturing, from natural language, formal representations that can be used by a reasoner to compute an answer. Many researchers have studied this problem by developing algorithms addressing specific phenomena in natural language interpretation, but few have studied (or cataloged) the types of failures associated with this problem. Knowledge of these failures can help researchers by providing a road map of open research problems and help practitioners by providing a checklist of issues to address in order to build systems that can achieve good performance on this problem. In this paper, we present a study – conducted in the context of the Halo Project – cataloging the types of failures that occur when capturing knowledge from natural language. We identified the categories of failures by examining a corpus of questions posed by naïve users to a knowledge based question answering system and empirically demonstrated the generality of our categorizations. We also describe available technologies that can address some of the failures we have identified.

Coreference resolution is a classic NLP problem and has been studied extensively by many researchers. Most existing studies, however, are generic in the sense that they are not focused on any specific text. In the past few years, opinion mining became a popular topic of research because of a wide range of applications. However, limited work has been done on coreference resolution in opinionated text. In this paper, we deal with object and attribute coreference resolution. Such coreference resolutions are important because without solving it a great deal of opinion information will be lost, and opinions may be assigned to wrong entities. We show that some important features related to opinions can be exploited to perform the task more accurately. Experimental results using blog posts demonstrate the effectiveness of the technique. 1

The knowledge acquisition bottleneck remains the fundamental obstacle to building intelligent reasoning systems. However, there is an ideal solution to this problem: creating a system that constructs knowledge bases by extracting information from texts. This approach, called Learning-by-Reading, is promising because a vast body of knowledge is already expressed in text, and – in terms of human effort – it is the least costly method of communicating information to computers. Despite this advantage, Learning-by-Reading has been infeasible due to limitations with Natural Language Understanding (NLU). In this proposal, we introduce an approach to overcome this difficulty, based on the following hypothesis: although current technologies of NLU produce imperfect semantic representations for individual texts, combining these imperfect representations drawn from multiple texts (all on the same topic) can significantly improve the overall quality of the resulting knowledge base. This approach addresses the difficulty of NLU in two ways: (1) the information drawn from one text can help reading the other texts and (2) an incorrect interpretation on one text can be repaired by the other texts with redundant content. Key in this approach is knowledge integration, the ability to coherently relate knowledge extracted from multiple texts. This paper introduces our ongoing work on a Learning-by-Reading system, Kleo, which has a sophisticated knowledge integration facility. Kleo shows that knowledge integration is effective and merits further investigation. We propose using knowledge integration to address the brittleness of key steps in natural language processing, such as parsing and semantic interpretation. First, we introduce a semantic representation language, called compact representation, which can succinctly represent the alternative interpretations of a sentence, as well as the constraints among them. Second, we present a knowledge integration algorithm that combines multiple compact representations to prune the implausible interpretations. Finally, we propose a plan for evaluating compact representations and the Kleo knowledge integration system. 1