Evaluation by Patterns
Problems associated with patterns:
- don�t differentiate between documents
- don�t penalize �answer stuffing�
- errors correlated with system functionality
Useful if limitations understood
- system rankings produced from lenient- assessor and pattern evaluations highly correlated
- Kendall t of .944 (24 swaps) for 250-byte
- Kendall t of .894 (28 swaps) for 50-byte