MINDS - Project Goals


Text summarization is aimed at providing quick access to a large document or a large number of documents. The need for automatic summarization is especially strong if the source text is in a language different from the one(s) in which readers are most fluent. The state of the art in machine translation (MT) does not yet provide texts of excellent quality in a target language. Rather than dealing with a poorly constructed translation of the same length as the source text, having available a shorter summary generated automatically would be preferable in many situations. What is even more desirable is to have an integrated system in which one can choose between a short summary, the full source document(s), a full translation in a different language, or information extracted from the document(s) in the form of lists or templates, depending on the purpose and the level of detail at which one needs to comprehend a document.

As core research and technology, we plan to develop a flexible, multi-engine, TIPSTER-compliant system for multi-lingual summarization that integrates a core summarization engine and three innovative engines that extend the core methods to multilingual and multiple-document summarization. The core engine combines keyword and statistical methods with text structure, morphological and syntactic analysis techniques. The three new engines will be based on machine translation techniques based on glossaries and bilingual lexicons, information extraction methods, and ontological mapping. Both language generation and other compilation techniques will be used to produce both textual and non-textual summaries. Key areas of innovation in summarization being proposed include: