- train.json: training set containing 516 CAs with only single-template label per sentence.
- dev.json: dev set containing 128 CAs with only single-template label per sentence.
- test.json: test set containing 134 CAs with at least one multi-template label sentence, the other sentences have single-template labels.
- pm_id: id for the initial-argument.
- lo_id: id for the counter-argument.
- pm_speech: essay of an initial-argument.
- lo_speech: essay of a counter-argument, divided by sentences.
- ptns: template label for each sentence.
- segments: segments for different templates. A segment is considered as one or more consecutive sentences with the same template.
- key for each segment: template id.
- range: start_index, end_index of of counter-argument sentences in lo_speech.
- slots: slot-fillers for the corresponding template.
In the test set, we do not have a segment field since there are multiple ways of deciding segments due to the presence of multi-template labels. Therefore, the test set has the following fields instead.
- slots: non-expert annotators raw annotations of slot-fillers for the corresponding CA.
- workers_raw_annotations: non-expert annotators raw annotations of logic templates for corresponding CA.
- filtered_slots: slot-fillers filtered from the raw slots based on the rules 1) discarding slots that are cannot be exactly found from the CA essay; 2) discarding slots that subset of other slots.
- 1: Mitigation (Mig)
- 2: Alternative (Alt)
- 3: No evidence (No Evi)
- 4: Another true cause (ATC)
- 5: Missing mechanism #1 (MM1)
- 6: Missing mechanism #2 (MM2)
- 7: No need to address (NNA)
- 8: Negative effect due to y (Neg eff)
- 9: Positive effects of a different perspective from y #1 (Dif Per1)
- 10: Positive effects of a different perspective from y #2 (Dif Per2)
- 11: Others
- 100: Non-argumentative