Search

The insertion of linguistic material into document sentences to create new sentences is a common activity in document abstracting. We investigate a transformation-based learning method to simulate this type of operation relevant for text summarization. Our work is framed on a theory of transformation-based abstracting where an initial text summary is transformed into an abstract by the application of a number of rules learnt from a corpus of examples. Our results are as good as recent work on classification-based predicate insertion. [ABSTRACT FROM AUTHOR]

Business intelligence requires the collecting and merging of information from many different sources, both structured and unstructured, in order to analyse for example financial risk, operational risk factors, follow trends and perform credit risk management. While traditional data mining tools make use of numerical data and cannot easily be applied to knowledge extracted from free text, traditional information extraction is either not adapted for the financial domain, or does not address the issue of information integration: the merging of information from different kinds of sources. We describe here the development of a system for content mining using domain ontologies, which enables the extraction of relevant information to be fed into models for analysis of financial and operational risk and other business intelligence applications such as company intelligence, by means of the XBRL standard. The results so far are of extremely high quality, due to the implementation of primarily high-precision rules. [ABSTRACT FROM AUTHOR]

In this paper we examine the interplay between the requirements of information seekers to access information in large digital text collections and the techniques developed by natural language processing researchers to support this access. In particular we examine how language processing technologies such as question answering, single and multidocument summarisation, and ontology-guided similar event searching can assist journalists in gathering information from news archives for the purpose of writing background to a breaking news event - the Cub Reporter scenario. Our thesis is that investigating real-world tasks with complex information access requirements, such as the Cub Reporter scenario, stimulates researchers to look beyond existing search engine solutions and drives the development and evaluation of novel language processing techniques; at the same time novel developments in language processing capabilities allow both conceptual insights into how to characterise information seeking behaviour and empirical insights based on observation of information seeking behaviour using new technologies [ABSTRACT FROM AUTHOR]

Neural networks have proved to be versatile and robust for particle separation in many experiments related to particle astrophysics. We apply these techniques to separate gamma rays from hadrons for the MAGIC Čerenkov Telescope. Two types of neural network architectures have been used for the classification task: one is the MultiLayer Perceptron (MLP) based on supervised learning, and the other is the Self-Organising Tree Algorithm (SOTA), which is based on unsupervised learning. We propose a new architecture by combining these two neural networks types to yield better and faster classification results for our classification problem. [ABSTRACT FROM AUTHOR]

With its diameter of 17m, the MAGIC telescope is the largest Cherenkov detector for gamma ray astrophysics. It is sensitive to photons above an energy of 30 GeV. MAGIC started operations in October 2003 and is currently taking data. This report summarizes its main characteristics, its first results and its potential for physics. [ABSTRACT FROM AUTHOR]

Question Answering (QA) aims at providing users with short text units that answer specific, well-formed natural language questions. A two stage architecture is widely adopted for this task consisting of a document retrieval step followed by an answer extraction step. In such an approach two main problems need to be addressed to reduce the search space: better selecting answer bearing passages in the document retrieval step and better pinpointing answers in the answer extraction step. We investigate the effect of word-based and linguistic-based features for the identification of answer-bearing sentences and answer candidates in a QA system and show that both play a significant role. [ABSTRACT FROM AUTHOR]

We describe experiments on content selection for producing biographical summaries from multiple documents. The method relies on a set of patterns to identify descriptive phrases, an available co-reference resolution algorithm, and a greedy, corpus-based sentence deletion procedure for document compression. We show that in an automatic evaluation of content using ROUGE, the proposed method obtains very good performance. [ABSTRACT FROM AUTHOR]