9314963 Boggess This is a standard award in the NSF/ARPA Human Language Technology for cooperative research involving Mississippi State University and the Chemical Abstracts Service of the American Chemical Society for the purpose of investigating means of partially automating the indexing process of scientific text. The system assists the rapid determination of the key topics and contents of scientific articles in a specified domain and in the production of key phrases describing the content. The domain, which is the literature used in the Chemical Abstracts Service, is analyzed making use of the uniform standards set for abstracting, the well-understood terms and concepts applied to manual indexing, and the hierarchy of concepts already established within that domain. Software used at Chemical Abstracts Service and the KUDZU system - a knowledge-based natural language processing system - are modified, extended and applied to the task of indexing articles in the field of physical chemistry, with the goal of better than 80% success rate. The systems also highlights all relevant material in an article being indexed in a manner which aids a human document analyst to verify the correctness of the automatically produced index phrases and to enhance the rapid correction of any erroneous or omitted index phrases.
