Table 3.

Comparison of model evaluations under corpus-based methods (i.e. % match score) and ratings-based methods (i.e. % judged as acceptable)

	Corpus-based	ratings
	(strict; lenient)
RC00	90%; 93%	90%
CDP++	78%; 80%	82%
Sequitur	76%; 81%	90%

To allow the fairest comparison, match scores from the corpus-based method in Table 1 were recomputed on the basis of the 799 nonwords analysed in 3: Sequitur’s performance improved as a consequence of eliminating the false negatives introduced by the 9→$ substitution