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 (any; with object relative clause)
Item
Condition
Licensor np compl rc_dp rc_subj rc_verb matrix_v npi continuation
Item Condition Licensor np compl rc_dp rc_subj rc_verb matrix_v npi continuation
1 neg_pos No author that the senators liked has had any success
1 neg_neg No author that no senators liked has had any success
1 pos_pos The author that the senators liked has had any success
1 pos_neg The author that no senators liked has had any success
Prediction performance for GPT-2 on Negative Polarity Licensing (any; with object relative clause)
Accuracy
Formula
Description
AccuracyPredictionDescription
100.00% (579,neg_pos/8,npi) < (577,pos_pos/8,npi) No description provided.
97.37% (578,neg_neg/8,npi) < (580,pos_neg/8,npi) No description provided.
97.37% (579,neg_pos/8,npi) < (580,pos_neg/8,npi) No description provided.