Disambiguation of Period Characters in Clinical Narratives

Abstract

The period character's meaning is highly ambiguous due to the frequency of abbreviations that require to be followed by a period. We have developed a hybrid method for period character disambiguation and the identification of abbreviations, combining rules that explore regularities in the right context of the period with lexicon-based, statistical methods which scrutinize the preceding token. The texts under scrutiny are clinical discharge summaries. Both abbreviation detection and sentence delimitation showed an accuracy of about 93%. An error analysis demonstrated potential for further improvements.

Detection of sentence boundaries and abbreviations in clinical narratives

“…Due to the fact that we achieved comparable results to the cTakes sentence detection tool (applied to English clinical text) using OpenNLP, a direct comparison between the approach presented in this paper, and a retrained version of the OpenNLP sentence detection tool for our German texts would be interesting for a supervised approach in general. Additionally, an enhanced version of the preliminary approach described in [ 5 ] could be further evaluated. Furthermore, the applicability to other clinical subdomains would be of interest, as different document types (e.g.…”

“…Based on a preliminary study [ 5 ], having applied a unsupervised statistical approach together with a rule-based method for the disambiguation of the period character within clinical narratives, we are focusing in this work on a supervised method exploiting support vector machines for the two different tasks, viz . sentence delimitation and abbreviation detection.…”

“…Nevertheless, our distinction seems justifiable in the light of the particularities of German language, especially medical sub-language, for which - to the best of our knowledge - this investigation constitutes the first study on sentence delineation with the additional focus on abbreviation detection. In a preliminary study [ 5 ] we addressed the problem of sentence boundary detection together with abbreviation detection on similar texts. We had combined an unsupervised statistical approach together with a rule-based method for the disambiguation of the period character.…”