Repetitive context patterns in the data sets. In CoNLL, a large percentage of organizations occur in sports scores. In MUC-6, a large percentage of PER entities are preceded by an honorific.
| . | Honorifics % PER . | Sports Scores % ORG . |
|---|---|---|
| CoNLL train | 2.74 | 25.89 |
| CoNLL testa | 2.19 | 22.13 |
| CoNLL testb | 2.59 | 32.37 |
| Wikipedia | 4.65 | 0.00 |
| MUC-6 train | 27.46 | 0.00 |
| MUC-6 test | 35.08 | 0.00 |
| . | Honorifics % PER . | Sports Scores % ORG . |
|---|---|---|
| CoNLL train | 2.74 | 25.89 |
| CoNLL testa | 2.19 | 22.13 |
| CoNLL testb | 2.59 | 32.37 |
| Wikipedia | 4.65 | 0.00 |
| MUC-6 train | 27.46 | 0.00 |
| MUC-6 test | 35.08 | 0.00 |