Grand Challenge: Producing Meaningful Texts

We will develop an automatic system capable of understanding scientific and humanistic texts, identifying novel ideas and key concepts, and then producing specialized summaries understandable for different audiences. The system will collect information from research publications, identify major topics, and produce readable summaries that can be targeted towards, for instance, other researchers, the general public, school children, or people with reading impairments. This task requires contributions from diverse fields, e.g. Computational Linguistics, Artificial Intelligence, Cognitive Science, and Science of Science. Subtasks include the automatic linguistic analysis of texts/recordings, production of text from abstract representations of meaning, identification of key topics in documents and of texts with a high potential for being cited, etc. While major progress has been made in some subtasks, major limitations remain: automatic generation systems often produce incoherent/unreadable text, while readable text can be reliably produced only in very limited areas such as fantasy football reporting. As such, the approaches we develop will catapult computational language understanding into a new era. We will produce a system that generates press releases summarizing university research that are judged (1) understandable and (2) indistinguishable from human-authored press releases by at least 50% of adult readers. The system can be adapted to rephrase medical texts, patents, intelligence information from a wide range of sources, etc. IU is ideally suited to address this challenge because we have a critical mass of researchers from different fields working on those subareas, e.g., Computer Science, Computational Linguistics, Cognitive Science, and Information and Library Science. Grand Challenge: Producing Meaningful Texts 1. The Grand Challenge We will develop an automatic system capable of understanding scientific and humanistic texts, identifying novel ideas and key concepts, and then producing specialized summaries understandable for different audiences. The system will collect information from research publications, identify major topics, and produce readable summarizes that can be targeted towards, for instance, other researchers, the general public, primary and secondary school children, or people with reading impairments. This task requires contributions from diverse fields, e.g. Computational Linguistics, Artificial Intelligence, Cognitive Science, and Science of Science, as well as from domain experts from the sciences and the humanities. Subtasks include the automatic linguistic analysis of texts and recordings, production of text from abstract representations of meaning, identification of key topics in documents and of texts with a high potential for being cited, etc. While major progress has been made in some subtasks, major limitation remain: automatic generation systems often produce incoherent/unreadable text, while readable text can be reliably produced only in very limited areas such as fantasy football reporting. As such, the approaches we develop will catapult computational text understanding into a new era.