Individual results

View docs

View in-depth performance of a single language model on a single test suite.

Region-by-region surprisal
Sample item for Negative Polarity Licensing (ever; with subject relative clause)
Item
Condition
Licensor np compl rc_verb rc_dp rc_subj has npi continuation
Item Condition Licensor np compl rc_verb rc_dp rc_subj has npi continuation
1 neg_pos No author that liked the senators has ever been popular
1 neg_neg No author that liked no senators has ever been popular
1 pos_pos The author that liked the senators has ever been popular
1 pos_neg The author that liked no senators has ever been popular
Prediction performance for RNNG on Negative Polarity Licensing (ever; with subject relative clause)
Accuracy
Formula
Description
AccuracyPredictionDescription
100.00% (587,neg_pos/8,npi) < (585,pos_pos/8,npi) No description provided.
100.00% (586,neg_neg/8,npi) < (588,pos_neg/8,npi) No description provided.
7.89% (587,neg_pos/8,npi) < (588,pos_neg/8,npi) No description provided.