Table A2.
Template selection accuracies. The models are different for each language, and the number of candidate templates also differs across tasks. The best result for each task is highlighted in bold.
| Accuracy | N-Grams (N = 3) | GPT-2 1 Epoch | GPT-2 2 Epochs |
|---|---|---|---|
| es_verb_time | 52.86 | 63.93 | 80.75 |
| es_verb_num | 55.56 | 77.78 | 86.87 |
| es_det | 26.55 | 49.09 | 96.00 |
| es_det_verb | 29.26 | 60.37 | 98.15 |
| es_food | 30.00 | 10.00 | 60.00 |
| fr_verb_time | 59.69 | 58.44 | 68.94 |
| fr_verb_num | 50.00 | 64.17 | 82.50 |
| fr_det_pron | 12.50 | 36.85 | 44.80 |
| fr_food | 36.79 | 39.26 | 53.33 |
| no_verb_prep | 26.47 | 76.94 | 73.86 |