S18 - SKILL 2022 - Studierendenkonferenz Informatik

Auflistung nach:

1 - 1 von 1

Textdokument
Comparing Link Grammars and Dependency Grammars for parsing German histological reports
(SKILL 2022, 2022) Dörenberg, Julian
The availability of structured data is becoming an increasingly critical factor in medical research. Still, pathologists in Germany document their findings in running text instead of in a structured form. In order to obtain structured data from these report texts, hey have to be converted to a more useful form. Link Grammars (LGs) and Dependency Grammars (DGs) both can be used to parse the texts. Hence, LGs and DGs can be used for information extraction on histological reports. This paper aims to compare LGs and DGs, to show why DGs are superior and to evaluate the performance of a DG parser on a corpus of 200 histological reports randomly selected from breast biopsy reports. The DG parser achieved an Unlabelled Attachment Score of 96, a Labelled Accuracy of 95 and a Labelled Attachment Score of 93. Further evaluation shows that the occurrence of medical words which have not been part of the training data does not affect the parsers performance.