Hedge cues were detected using a super- vised Conditional Random Field (CRF) classifier exploiting features from the RASP parser. The CRF’s predictions were filtered using known cues and unseen in- stances were removed, increasing preci- sionwhileretainingrecall. Rulesforscope detection, based on the grammatical re- lations of the sentence and the part-of- speech tag of the cue, were manually- developed. However, another supervised CRFclassifierwasusedtorefinethesepre- dictions. As a final step, scopes were con- structed from the classifier output using a small set of post-processing rules. Devel- opmentofthesystemrevealedanumberof issues with the annotation scheme adopted by the organisers.

Click Here to View the Paper In XML Format