Table 2 Performance (F-score) of chunkers and their combination trained on subsets of different size of the GENIA GSC and on the GSC subset supplemented with an SSC, for noun-phrase and verb-phrase recognition.
From: Training text chunkers on a silver standard corpus: can silver replace gold?
| Â | Lingpipe | OpenNLP | Yamcha | Combined | ||||
|---|---|---|---|---|---|---|---|---|
GSC size | GSC | GSC+SSC | GSC | GSC+SSC | GSC | GSC+SSC | GSC | GSC+SSC |
Noun phrases | Â | Â | Â | Â | Â | Â | Â | Â |
10 | 65.8% | 80.8% | 83.0% | 87.9% | 82.7% | 85.6% | 86.8% | 90.7% |
25 | 72.2% | 81.1% | 85.7% | 88.3% | 84.3% | 86.0% | 87.9% | 90.9% |
50 | 76.8% | 81.3% | 87.5% | 88.6% | 85.4% | 86.2% | 88.9% | 91.2% |
100 | 78.2% | 81.9% | 87.9% | 88.9% | 85.6% | 86.6% | 89.3% | 91.5% |
250 | 82.4% | 82.8% | 88.3% | 89.3% | 86.7% | 87.2% | 90.6% | 92.0% |
500 | 84.5% | n.a | 89.7% | n.a | 88.1% | n.a | 92.8% | n.a |
Verb phrases | Â | Â | Â | Â | Â | Â | Â | Â |
10 | 64.1% | 86.9% | 84.3% | 93.6% | 86.2% | 92.5% | 91.3% | 94.6% |
25 | 73.8% | 87.3% | 88.8% | 94.0% | 89.7% | 92.9% | 93.0% | 94.9% |
50 | 79.2% | 87.6% | 92.1% | 94.4% | 91.7% | 93.1% | 94.4% | 95.5% |
100 | 83.6% | 87.9% | 93.6% | 94.7% | 92.3% | 93.4% | 95.4% | 95.8% |
250 | 88.3% | 88.7% | 95.0% | 95.3% | 93.8% | 93.9% | 95.8% | 96.0% |
500 | 90.3% | n.a | 95.7% | n.a | 94.1% | n.a | 96.3% | n.a |