From text to structured information-automatic processing of medical reports

This paper describes the analysis and processing programs for a set of natural language texts in a medical area (x-ray reports on patients with breast cancer). The programs convert the information in the text into a tabular form suitable for further automatic information processing (e.g., editing of records, question answering on the data collected, or statistical summaries of the data). To set up a tabular form appropriate for the data, we first perform a manual linguistic analysis on a sample of the texts. From this we obtain the word classes and the form of the table (called an information format) for this type of material. We then apply the series of processing programs to the sentences of the texts. Each sentence is parsed with the Linguistic String Parser English grammar in order to obtain its grammatical structure; certain standard English transformations are then applied to regularize the grammatical form of the sentence; and finally a set of "formatting transformations" map the words of the sentence into the slots of the format or table, in such a way that the sentence is reconstructible (up to paraphrase) from its representation in the table. The results of applying these programs to a corpus are described. This procedure enables us to convert a natural language corpus into a structured data base.