A Practical Partial Parser for Biomedical Literature Summarization

We present a partial parser called TeLePaPa (TextLens Partial Parser) to identify subjects and predicate verbs (SPVs) in a sentence of abstracts of MEDLINE citations. The performance of TeLePaPa is the precision of 96.7% and 97.1% for the SPV detection, respectively, and the recall of 91.3% and 94.9%, respectively. We found that there was a similarity in the distribution of the pairs of SPV over different research topics in the domain. In addition, we found that the power law holds for the relationship of the number of citations uncovered by SPV pairs and its rank. That is, only a half of the pairs covered about 90% of all the citations. This fact enables us to efficiently scan the huge amount of biomedical literature.