blimp

  • Description:

BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars.

Split Examples
'train' 1,000
  • Features:
FeaturesDict({
    'UID': Text(shape=(), dtype=tf.string),
    'field': Text(shape=(), dtype=tf.string),
    'lexically_identical': tf.bool,
    'linguistics_term': Text(shape=(), dtype=tf.string),
    'one_prefix_method': tf.bool,
    'pair_id': tf.int32,
    'sentence_bad': Text(shape=(), dtype=tf.string),
    'sentence_good': Text(shape=(), dtype=tf.string),
    'simple_LM_method': tf.bool,
    'two_prefix_method': tf.bool,
})
@article{warstadt2019blimp,
  title={BLiMP: A Benchmark of Linguistic Minimal Pairs for English},
  author={Warstadt, Alex and Parrish, Alicia and Liu, Haokun and Mohananey, Anhad and Peng, Wei, and Wang, Sheng-Fu and Bowman, Samuel R},
  journal={arXiv preprint arXiv:1912.00582},
  year={2019}
}

blimp/adjunct_island (default config)

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm adjunct_island.

  • Download size: 350.86 KiB

  • Dataset size: 366.47 KiB

blimp/anaphor_gender_agreement

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm anaphor_gender_agreement.

  • Download size: 426.51 KiB

  • Dataset size: 332.31 KiB

blimp/anaphor_number_agreement

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm anaphor_number_agreement.

  • Download size: 440.29 KiB

  • Dataset size: 341.07 KiB

blimp/animate_subject_passive

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm animate_subject_passive.

  • Download size: 451.46 KiB

  • Dataset size: 345.50 KiB

blimp/animate_subject_trans

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm animate_subject_trans.

  • Download size: 422.95 KiB

  • Dataset size: 329.27 KiB

blimp/causative

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm causative.

  • Download size: 309.73 KiB

  • Dataset size: 324.36 KiB

blimp/complex_NP_island

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm complex_NP_island.

  • Download size: 383.17 KiB

  • Dataset size: 398.78 KiB

blimp/coordinate_structure_constraint_complex_left_branch

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm coordinate_structure_constraint_complex_left_branch.

  • Download size: 558.30 KiB

  • Dataset size: 410.43 KiB

blimp/coordinate_structure_constraint_object_extraction

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm coordinate_structure_constraint_object_extraction.

  • Download size: 357.47 KiB

  • Dataset size: 372.10 KiB

blimp/determiner_noun_agreement_1

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm determiner_noun_agreement_1.

  • Download size: 457.66 KiB

  • Dataset size: 356.93 KiB

blimp/determiner_noun_agreement_2

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm determiner_noun_agreement_2.

  • Download size: 477.40 KiB

  • Dataset size: 357.01 KiB

blimp/determiner_noun_agreement_irregular_1

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm determiner_noun_agreement_irregular_1.

  • Download size: 463.80 KiB

  • Dataset size: 365.08 KiB

blimp/determiner_noun_agreement_irregular_2

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm determiner_noun_agreement_irregular_2.

  • Download size: 479.35 KiB

  • Dataset size: 361.76 KiB

blimp/determiner_noun_agreement_with_adj_2

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm determiner_noun_agreement_with_adj_2.

  • Download size: 514.46 KiB

  • Dataset size: 379.92 KiB

blimp/determiner_noun_agreement_with_adj_irregular_1

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm determiner_noun_agreement_with_adj_irregular_1.

  • Download size: 487.95 KiB

  • Dataset size: 384.67 KiB

blimp/determiner_noun_agreement_with_adj_irregular_2

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm determiner_noun_agreement_with_adj_irregular_2.

  • Download size: 516.14 KiB

  • Dataset size: 384.54 KiB

blimp/determiner_noun_agreement_with_adjective_1

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm determiner_noun_agreement_with_adjective_1.

  • Download size: 492.85 KiB

  • Dataset size: 385.25 KiB

blimp/distractor_agreement_relational_noun

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm distractor_agreement_relational_noun.

  • Download size: 513.33 KiB

  • Dataset size: 391.45 KiB

blimp/distractor_agreement_relative_clause

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm distractor_agreement_relative_clause.

  • Download size: 551.53 KiB

  • Dataset size: 416.14 KiB

blimp/drop_argument

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm drop_argument.

  • Download size: 297.07 KiB

  • Dataset size: 311.70 KiB

blimp/ellipsis_n_bar_1

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm ellipsis_n_bar_1.

  • Download size: 402.32 KiB

  • Dataset size: 417.06 KiB

blimp/ellipsis_n_bar_2

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm ellipsis_n_bar_2.

  • Download size: 417.53 KiB

  • Dataset size: 432.47 KiB

blimp/existential_there_object_raising

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm existential_there_object_raising.

  • Download size: 537.77 KiB

  • Dataset size: 423.00 KiB

blimp/existential_there_quantifiers_1

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm existential_there_quantifiers_1.

  • Download size: 348.95 KiB

  • Dataset size: 363.58 KiB

blimp/existential_there_quantifiers_2

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm existential_there_quantifiers_2.

  • Download size: 350.80 KiB

  • Dataset size: 365.43 KiB

blimp/existential_there_subject_raising

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm existential_there_subject_raising.

  • Download size: 385.21 KiB

  • Dataset size: 399.84 KiB

blimp/expletive_it_object_raising

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm expletive_it_object_raising.

  • Download size: 573.88 KiB

  • Dataset size: 437.58 KiB

blimp/inchoative

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm inchoative.

  • Download size: 291.71 KiB

  • Dataset size: 306.34 KiB

blimp/intransitive

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm intransitive.

  • Download size: 298.33 KiB

  • Dataset size: 312.96 KiB

blimp/irregular_past_participle_adjectives

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm irregular_past_participle_adjectives.

  • Download size: 434.10 KiB

  • Dataset size: 345.74 KiB

blimp/irregular_past_participle_verbs

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm irregular_past_participle_verbs.

  • Download size: 410.27 KiB

  • Dataset size: 327.21 KiB

blimp/irregular_plural_subject_verb_agreement_1

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm irregular_plural_subject_verb_agreement_1.

  • Download size: 449.91 KiB

  • Dataset size: 366.17 KiB

blimp/irregular_plural_subject_verb_agreement_2

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm irregular_plural_subject_verb_agreement_2.

  • Download size: 442.75 KiB

  • Dataset size: 354.70 KiB

blimp/left_branch_island_echo_question

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm left_branch_island_echo_question.

  • Download size: 471.31 KiB

  • Dataset size: 348.84 KiB

blimp/left_branch_island_simple_question

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm left_branch_island_simple_question.

  • Download size: 335.40 KiB

  • Dataset size: 351.01 KiB

blimp/matrix_question_npi_licensor_present

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm matrix_question_npi_licensor_present.

  • Download size: 447.08 KiB

  • Dataset size: 354.14 KiB

blimp/npi_present_1

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm npi_present_1.

  • Download size: 427.75 KiB

  • Dataset size: 339.68 KiB

blimp/npi_present_2

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm npi_present_2.

  • Download size: 412.24 KiB

  • Dataset size: 329.11 KiB

blimp/only_npi_licensor_present

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm only_npi_licensor_present.

  • Download size: 448.41 KiB

  • Dataset size: 349.50 KiB

blimp/only_npi_scope

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm only_npi_scope.

  • Download size: 570.04 KiB

  • Dataset size: 408.52 KiB

blimp/passive_1

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm passive_1.

  • Download size: 332.30 KiB

  • Dataset size: 346.93 KiB

blimp/passive_2

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm passive_2.

  • Download size: 301.12 KiB

  • Dataset size: 315.75 KiB

blimp/principle_A_c_command

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm principle_A_c_command.

  • Download size: 515.32 KiB

  • Dataset size: 388.54 KiB

blimp/principle_A_case_1

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm principle_A_case_1.

  • Download size: 466.05 KiB

  • Dataset size: 370.87 KiB

blimp/principle_A_case_2

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm principle_A_case_2.

  • Download size: 481.42 KiB

  • Dataset size: 370.88 KiB

blimp/principle_A_domain_1

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm principle_A_domain_1.

  • Download size: 488.15 KiB

  • Dataset size: 371.62 KiB

blimp/principle_A_domain_2

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm principle_A_domain_2.

  • Download size: 481.63 KiB

  • Dataset size: 365.92 KiB

blimp/principle_A_domain_3

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm principle_A_domain_3.

  • Download size: 501.84 KiB

  • Dataset size: 359.74 KiB

blimp/principle_A_reconstruction

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm principle_A_reconstruction.

  • Download size: 337.40 KiB

  • Dataset size: 353.00 KiB

blimp/regular_plural_subject_verb_agreement_1

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm regular_plural_subject_verb_agreement_1.

  • Download size: 441.26 KiB

  • Dataset size: 359.56 KiB

blimp/regular_plural_subject_verb_agreement_2

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm regular_plural_subject_verb_agreement_2.

  • Download size: 445.78 KiB

  • Dataset size: 354.47 KiB

blimp/sentential_negation_npi_licensor_present

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm sentential_negation_npi_licensor_present.

  • Download size: 479.49 KiB

  • Dataset size: 372.30 KiB

blimp/sentential_negation_npi_scope

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm sentential_negation_npi_scope.

  • Download size: 600.52 KiB

  • Dataset size: 431.18 KiB

blimp/sentential_subject_island

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm sentential_subject_island.

  • Download size: 357.25 KiB

  • Dataset size: 372.86 KiB

blimp/superlative_quantifiers_1

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm superlative_quantifiers_1.

  • Download size: 372.25 KiB

  • Dataset size: 360.02 KiB

blimp/superlative_quantifiers_2

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm superlative_quantifiers_2.

  • Download size: 504.02 KiB

  • Dataset size: 360.07 KiB

blimp/tough_vs_raising_1

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm tough_vs_raising_1.

  • Download size: 334.99 KiB

  • Dataset size: 349.62 KiB

blimp/tough_vs_raising_2

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm tough_vs_raising_2.

  • Download size: 355.54 KiB

  • Dataset size: 370.17 KiB

blimp/transitive

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm transitive.

  • Download size: 449.50 KiB

  • Dataset size: 334.45 KiB

blimp/wh_island

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm wh_island.

  • Download size: 438.12 KiB

  • Dataset size: 343.47 KiB

blimp/wh_questions_object_gap

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm wh_questions_object_gap.

  • Download size: 378.35 KiB

  • Dataset size: 392.99 KiB

blimp/wh_questions_subject_gap

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm wh_questions_subject_gap.

  • Download size: 380.84 KiB

  • Dataset size: 395.47 KiB

blimp/wh_questions_subject_gap_long_distance

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm wh_questions_subject_gap_long_distance.

  • Download size: 451.82 KiB

  • Dataset size: 466.53 KiB

blimp/wh_vs_that_no_gap

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm wh_vs_that_no_gap.

  • Download size: 374.28 KiB

  • Dataset size: 388.91 KiB

blimp/wh_vs_that_no_gap_long_distance

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm wh_vs_that_no_gap_long_distance.

  • Download size: 431.08 KiB

  • Dataset size: 445.74 KiB

blimp/wh_vs_that_with_gap

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm wh_vs_that_with_gap.

  • Download size: 359.16 KiB

  • Dataset size: 373.79 KiB

blimp/wh_vs_that_with_gap_long_distance

  • Config description: BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. This configuration includes the paradigm wh_vs_that_with_gap_long_distance.

  • Download size: 416.00 KiB

  • Dataset size: 430.63 KiB