Datasets:

NeTSlab
/

BLiMP-IT

language:
  - it
license: cc-by-sa-4.0
task_categories:
  - text-classification
  - text-ranking
  - question-answering
pretty_name: BLiMP-IT
size_categories:
  - 10K<n<100K
tags:
  - linguistics
  - syntax
  - morphology
  - acceptability
  - minimal-pairs
  - language-model-evaluation
  - italian
configs:
  - config_name: blimp_it_agreement_and_inflection_dp
    data_files:
      - split: test
        path: data/test/blimp_it_original_agreement_and_inflection_dp/test.jsonl
  - config_name: blimp_it_non_local_dependencies_adjunctisland
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_adjunctisland/test.jsonl
  - config_name: blimp_it_non_local_dependencies_atb_affirmative2
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_atb_affirmative2/test.jsonl
  - config_name: blimp_it_non_local_dependencies_atb_affirmative_mi
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_atb_affirmative_mi/test.jsonl
  - config_name: blimp_it_non_local_dependencies_atb_affirmative_ti
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_atb_affirmative_ti/test.jsonl
  - config_name: blimp_it_non_local_dependencies_atb_interrogative_gap_nogap_clitic
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_atb_interrogative_gap_nogap_clitic/test.jsonl
  - config_name: blimp_it_non_local_dependencies_atb_interrogative_gap_nogap_np_aux
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_atb_interrogative_gap_nogap_np_aux/test.jsonl
  - config_name: blimp_it_non_local_dependencies_atb_interrogative_nogap_gap_clitic
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_atb_interrogative_nogap_gap_clitic/test.jsonl
  - config_name: blimp_it_non_local_dependencies_atb_interrogative_nogap_gap_np_aux
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_atb_interrogative_nogap_gap_np_aux/test.jsonl
  - config_name: blimp_it_non_local_dependencies_atb_interrogative_nogap_nogap_clitic
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_atb_interrogative_nogap_nogap_clitic/test.jsonl
  - config_name: blimp_it_non_local_dependencies_atb_interrogative_nogap_nogap_np_aux
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_atb_interrogative_nogap_nogap_np_aux/test.jsonl
  - config_name: blimp_it_non_local_dependencies_atb_rc_object_gap_nogap
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_atb_rc_object_gap_nogap/test.jsonl
  - config_name: blimp_it_non_local_dependencies_atb_rc_object_nogap_gap
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_atb_rc_object_nogap_gap/test.jsonl
  - config_name: blimp_it_non_local_dependencies_atb_rc_object_nogap_nogap
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_atb_rc_object_nogap_nogap/test.jsonl
  - config_name: blimp_it_non_local_dependencies_complexnpisland
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_complexnpisland/test.jsonl
  - config_name: >-
      blimp_it_non_local_dependencies_coordinatestructureconstraint_complex_object_extraction
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_coordinatestructureconstraint_complex_object_extraction/test.jsonl
  - config_name: blimp_it_non_local_dependencies_left_branch_island_echo_question
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_left_branch_island_echo_question/test.jsonl
  - config_name: blimp_it_non_local_dependencies_parasiticgap_adjunctisland1
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_parasiticgap_adjunctisland1/test.jsonl
  - config_name: blimp_it_non_local_dependencies_parasiticgap_adjunctisland2
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_parasiticgap_adjunctisland2/test.jsonl
  - config_name: blimp_it_non_local_dependencies_rc_object
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_rc_object/test.jsonl
  - config_name: blimp_it_non_local_dependencies_rc_object_nogap
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_rc_object_nogap/test.jsonl
  - config_name: blimp_it_non_local_dependencies_rc_subject
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_rc_subject/test.jsonl
  - config_name: blimp_it_non_local_dependencies_rc_subject_nogap
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_rc_subject_nogap/test.jsonl
  - config_name: blimp_it_non_local_dependencies_sententialsubjectisland
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_sententialsubjectisland/test.jsonl
  - config_name: blimp_it_non_local_dependencies_wh_extraction_embedd2_clitic_inanimate
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_wh_extraction_embedd2_clitic_inanimate/test.jsonl
  - config_name: blimp_it_non_local_dependencies_wh_extraction_embedd2_np_inanimate
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_wh_extraction_embedd2_np_inanimate/test.jsonl
  - config_name: blimp_it_non_local_dependencies_wh_extraction_embedd2_np_inanimate_dem
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_wh_extraction_embedd2_np_inanimate_dem/test.jsonl
  - config_name: blimp_it_non_local_dependencies_wh_extraction_embedd_clitic_inanimate
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_wh_extraction_embedd_clitic_inanimate/test.jsonl
  - config_name: blimp_it_non_local_dependencies_wh_extraction_embedd_np_inanimate
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_wh_extraction_embedd_np_inanimate/test.jsonl
  - config_name: blimp_it_non_local_dependencies_wh_extraction_embedd_np_inanimate_dem
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_wh_extraction_embedd_np_inanimate_dem/test.jsonl
  - config_name: blimp_it_non_local_dependencies_wh_island_embedd
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_wh_island_embedd/test.jsonl
  - config_name: blimp_it_non_local_dependencies_wh_island_root
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_wh_island_root/test.jsonl
  - config_name: blimp_it_non_local_dependencies_wh_island_root_affirmative_dove
    data_files:
      - split: test
        path: >-
          data/test/blimp_it_original_non_local_dependencies_wh_island_root_affirmative_dove/test.jsonl
  - config_name: conversa_a_agreement_dp
    data_files:
      - split: test
        path: data/test/conversa_a_agreement_dp/test.jsonl
  - config_name: conversa_a_agreement_past_particple_pre_v_cl
    data_files:
      - split: test
        path: data/test/conversa_a_agreement_past_particple_pre_v_cl/test.jsonl
  - config_name: conversa_a_agreement_past_particple_unacc
    data_files:
      - split: test
        path: data/test/conversa_a_agreement_past_particple_unacc/test.jsonl
  - config_name: conversa_a_agreement_psych_verbs_piacere
    data_files:
      - split: test
        path: data/test/conversa_a_agreement_psych_verbs_piacere/test.jsonl
  - config_name: conversa_a_agreement_psych_verbs_preoccupare
    data_files:
      - split: test
        path: data/test/conversa_a_agreement_psych_verbs_preoccupare/test.jsonl
  - config_name: conversa_a_agreement_subject_nominal_predicate
    data_files:
      - split: test
        path: data/test/conversa_a_agreement_subject_nominal_predicate/test.jsonl
  - config_name: conversa_a_agreement_subject_nominal_predicate_attraction
    data_files:
      - split: test
        path: >-
          data/test/conversa_a_agreement_subject_nominal_predicate_attraction/test.jsonl
  - config_name: conversa_a_agreement_subject_verb_attraction
    data_files:
      - split: test
        path: data/test/conversa_a_agreement_subject_verb_attraction/test.jsonl
  - config_name: conversa_a_agreement_subject_verb_cumulative
    data_files:
      - split: test
        path: data/test/conversa_a_agreement_subject_verb_cumulative/test.jsonl
  - config_name: conversa_a_agreement_subject_verb_trans
    data_files:
      - split: test
        path: data/test/conversa_a_agreement_subject_verb_trans/test.jsonl
  - config_name: conversa_a_agreement_subject_verb_unacc
    data_files:
      - split: test
        path: data/test/conversa_a_agreement_subject_verb_unacc/test.jsonl
  - config_name: conversa_a_agreement_subject_verb_unerg
    data_files:
      - split: test
        path: data/test/conversa_a_agreement_subject_verb_unerg/test.jsonl
  - config_name: conversa_a_agreement_subject_verb_unerg_trans
    data_files:
      - split: test
        path: data/test/conversa_a_agreement_subject_verb_unerg_trans/test.jsonl
  - config_name: conversa_a_agreement_verb_subject_unacc
    data_files:
      - split: test
        path: data/test/conversa_a_agreement_verb_subject_unacc/test.jsonl
  - config_name: conversa_b_argument_structure_b_auxiliary_selection_ditranstitives
    data_files:
      - split: test
        path: >-
          data/test/conversa_b_argument_structure_b_auxiliary_selection_ditranstitives/test.jsonl
  - config_name: >-
      conversa_b_argument_structure_b_auxiliary_selection_passive_active_diathesis
    data_files:
      - split: test
        path: >-
          data/test/conversa_b_argument_structure_b_auxiliary_selection_passive_active_diathesis/test.jsonl
  - config_name: conversa_b_argument_structure_b_auxiliary_selection_transtitives
    data_files:
      - split: test
        path: >-
          data/test/conversa_b_argument_structure_b_auxiliary_selection_transtitives/test.jsonl
  - config_name: conversa_b_argument_structure_b_auxiliary_selection_unaccusatives
    data_files:
      - split: test
        path: >-
          data/test/conversa_b_argument_structure_b_auxiliary_selection_unaccusatives/test.jsonl
  - config_name: conversa_b_argument_structure_b_auxiliary_selection_unergatives
    data_files:
      - split: test
        path: >-
          data/test/conversa_b_argument_structure_b_auxiliary_selection_unergatives/test.jsonl
  - config_name: conversa_b_argument_structure_b_theta_roles
    data_files:
      - split: test
        path: data/test/conversa_b_argument_structure_b_theta_roles/test.jsonl
  - config_name: conversa_b_argument_structure_theta_roles
    data_files:
      - split: test
        path: data/test/conversa_b_argument_structure_theta_roles/test.jsonl
  - config_name: conversa_c_pronouns_1_2_person_answers
    data_files:
      - split: test
        path: data/test/conversa_c_pronouns_1_2_person_answers/test.jsonl
  - config_name: conversa_c_pronouns_1_2_person_elicited_declaratives
    data_files:
      - split: test
        path: >-
          data/test/conversa_c_pronouns_1_2_person_elicited_declaratives/test.jsonl
  - config_name: conversa_c_pronouns_clitics_answering_with_acc_cl
    data_files:
      - split: test
        path: data/test/conversa_c_pronouns_clitics_answering_with_acc_cl/test.jsonl
  - config_name: conversa_c_pronouns_clitics_coordinating_with_acc_cl
    data_files:
      - split: test
        path: >-
          data/test/conversa_c_pronouns_clitics_coordinating_with_acc_cl/test.jsonl
  - config_name: conversa_c_pronouns_clitics_coordinating_with_dat_cl
    data_files:
      - split: test
        path: >-
          data/test/conversa_c_pronouns_clitics_coordinating_with_dat_cl/test.jsonl
  - config_name: conversa_c_pronouns_reflexives_other
    data_files:
      - split: test
        path: data/test/conversa_c_pronouns_reflexives_other/test.jsonl
  - config_name: conversa_c_pronouns_reflexives_psych_verbs
    data_files:
      - split: test
        path: data/test/conversa_c_pronouns_reflexives_psych_verbs/test.jsonl
  - config_name: conversa_c_pronouns_reflexives_unacc
    data_files:
      - split: test
        path: data/test/conversa_c_pronouns_reflexives_unacc/test.jsonl
  - config_name: conversa_d_questions_d_answering_questions_with_rc_obj
    data_files:
      - split: test
        path: >-
          data/test/conversa_d_questions_d_answering_questions_with_rc_obj/test.jsonl
  - config_name: conversa_d_questions_d_answering_questions_with_rc_subj
    data_files:
      - split: test
        path: >-
          data/test/conversa_d_questions_d_answering_questions_with_rc_subj/test.jsonl
  - config_name: conversa_d_questions_d_answering_wh_adjunct_questions
    data_files:
      - split: test
        path: >-
          data/test/conversa_d_questions_d_answering_wh_adjunct_questions/test.jsonl
  - config_name: conversa_d_questions_d_answering_wh_argument_questions
    data_files:
      - split: test
        path: >-
          data/test/conversa_d_questions_d_answering_wh_argument_questions/test.jsonl
  - config_name: >-
      conversa_d_questions_d_answering_wh_argument_questions_number_disambiguation
    data_files:
      - split: test
        path: >-
          data/test/conversa_d_questions_d_answering_wh_argument_questions_number_disambiguation/test.jsonl
  - config_name: conversa_d_questions_d_answering_why_questions
    data_files:
      - split: test
        path: data/test/conversa_d_questions_d_answering_why_questions/test.jsonl
  - config_name: conversa_d_questions_d_answering_yes_no_questions
    data_files:
      - split: test
        path: data/test/conversa_d_questions_d_answering_yes_no_questions/test.jsonl
  - config_name: conversa_d_questions_d_question_formation_subject_position
    data_files:
      - split: test
        path: >-
          data/test/conversa_d_questions_d_question_formation_subject_position/test.jsonl

BLiMP-IT

Dataset Summary

BLiMP-IT is a linguistically motivated benchmark for evaluating Italian language models through minimal pairs. Each example consists of a grammatical sentence paired with a minimally different ungrammatical counterpart that isolates a single morphosyntactic contrast. The benchmark is designed to evaluate whether language models assign higher probability to the grammatical sentence than to the ungrammatical one.

The benchmark is inspired by the English BLiMP while extending and adapting it to Italian morphosyntax. It combines phenomena from the original BLiMP with additional contrasts derived from Italian linguistic literature and the COnVERSA psycholinguistic assessment battery.

Rather than measuring downstream task performance, BLiMP-IT measures the acquisition of abstract morphosyntactic generalizations.

Supported Tasks

The dataset is intended for:

Language model evaluation through sentence probability comparison
Minimal-pair acceptability evaluation
Psycholinguistically motivated syntactic evaluation
Cross-model comparison
Cross-linguistic comparison with BLiMP and related benchmarks

Typical evaluation consists of assigning higher probability (or lower perplexity) to the grammatical member of each minimal pair.

Dataset Structure

The repository contains one YAML configuration per linguistic phenomenon.

Macro-phenomena

The benchmark covers four major linguistic domains.

Macro-phenomenon	Number of phenomena
Agreement & Inflection	31
Verbal Class & Argument Structure	10
Pronouns	25
Non-local Dependencies	64

Overall the benchmark contains 130 phenomena.

Some phenomena originate from the original BLiMP benchmark, while others are adapted from the COnVERSA battery or specifically developed for Italian.

Phenomena

Agreement & Inflection

Original BLiMP

DP agreement
- blimp_it_original_agreement_and_inflection_dp

COnVERSA

Determiner–noun agreement
- conversa_a_agreement_dp
- conversa_a_agreement_in_dp
Subject–verb agreement
- conversa_a_agreement_subject_verb_trans
- conversa_a_agreement_subject_verb_unacc
- conversa_a_agreement_subject_verb_unerg
- conversa_a_agreement_subject_verb_unerg_trans
- conversa_a_agreement_subject_verb_attraction
- conversa_a_agreement_subject_verb_cumulative
- conversa_a_agreement_verb_subject_unacc
Predicate agreement
- conversa_a_agreement_subject_nominal_predicate
- conversa_a_agreement_subject_nominal_predicate_attraction
Past participle agreement
- conversa_a_agreement_past_particple_pre_v_cl
- conversa_a_agreement_past_particple_unacc
- conversa_a_past_particple_pre_v_cl
- conversa_a_past_particple_unacc
Psych verbs
- conversa_a_agreement_psych_verbs_piacere
- conversa_a_agreement_psych_verbs_preoccupare
- conversa_a_psych_verbs_piacere
- conversa_a_psych_verbs_preoccupare

Verbal Class & Argument Structure

COnVERSA

Theta-role assignment
- conversa_b_theta_roles
- conversa_b_argument_structure_theta_roles
- conversa_b_argument_structure_b_theta_roles
Auxiliary selection
- conversa_b_auxiliary_selection_transitives
- conversa_b_auxiliary_selection_unaccusatives
- conversa_b_auxiliary_selection_unergatives
- conversa_b_auxiliary_selection_ditranstitives
- conversa_b_auxiliary_selection_passive_active_diathesis
Argument structure variants
- conversa_b_argument_structure_b_auxiliary_selection_transitives
- conversa_b_argument_structure_b_auxiliary_selection_unaccusatives
- conversa_b_argument_structure_b_auxiliary_selection_unergatives
- conversa_b_argument_structure_b_auxiliary_selection_ditranstitives
- conversa_b_argument_structure_b_auxiliary_selection_passive_active_diathesis

Pronouns

COnVERSA

Personal pronouns
- conversa_c_pronouns_1_2_person_answers
- conversa_c_pronouns_1_2_person_elicited_declaratives
Clitics
- conversa_c_pronouns_clitics_answering_with_acc_cl
- conversa_c_pronouns_clitics_coordinating_with_acc_cl
- conversa_c_pronouns_clitics_coordinating_with_dat_cl
Reflexives
- conversa_c_pronouns_reflexives_other
- conversa_c_pronouns_reflexives_psych_verbs
- conversa_c_pronouns_reflexives_unacc

Non-local Dependencies

COnVERSA

Yes/no questions
- conversa_d_answering_yes_no_questions
Why questions
- conversa_d_answering_why_questions
WH-questions
- conversa_d_answering_wh_adjunct_questions
- conversa_d_answering_wh_argument_questions
- conversa_d_answering_wh_argument_questions_number_disambiguation
- conversa_d_answering_questions_with_rc_subj
- conversa_d_answering_questions_with_rc_obj
- conversa_d_question_formation_subject_position

Original BLiMP

Relative clauses
- subject relatives
- object relatives
- gap vs no-gap conditions
WH extraction
- root extraction
- embedded extraction
- two-level embedded extraction
- clitic fillers
- NP fillers
- demonstrative fillers
Across-the-board movement
- affirmative
- interrogative
- relative clause
- gap/no-gap manipulations
Islands
- adjunct island
- complex NP island
- sentential subject island
- WH island
- left branch island
- coordinate structure constraint
Parasitic gaps
- adjunct island variants

Data Fields

Each dataset configuration consists of sentence pairs containing:

sentence_good
sentence_bad
grammaticality label
linguistic phenomenon identifier
minimal-pair metadata

Data Creation

Source

BLiMP-IT combines:

manually designed minimal pairs
automatic lexical generation using grammar templates
manually annotated Italian lexicon
human validation

The lexicon was derived from a 2.4M-token corpus of Italian child-directed and naturalistic language, annotated with Universal POS tags, morphological features, and animacy information. Automatic generation produces lexical variants while preserving the targeted morphosyntactic contrast.

Dataset Characteristics

The benchmark follows two design principles:

Minimality of contrasts

Each grammatical and ungrammatical sentence differs only in the manipulation responsible for (un)grammaticality.
Linguistically meaningful errors

Ungrammatical sentences contain theoretically motivated morphosyntactic violations rather than random perturbations or word permutations.

Intended Uses

BLiMP-IT is intended for

evaluating Italian language models
probing syntactic generalization
comparing language models
comparing humans and language models
psycholinguistic research
computational linguistics research

Limitations

The benchmark focuses exclusively on morphosyntax rather than semantics.
Some generated sentences may be semantically unusual while remaining syntactically well formed.
Human acceptability judgments currently exist only for subsets overlapping with COnVERSA.
Automatic lexical generation is still expanding coverage across all phenomena.

Citation

Bressan et al. (2026) Language models assessment through linguistically motivated contrasts: A benchmark for Italian (BLiMP-IT). Proceedings of GLOW 47. DOI: 10.11576/glow-1242

@inproceedings{bressan2026blimpit,
  title={Language models assessment through linguistically motivated contrasts: A benchmark for Italian (BLiMP-IT)},
  author={Bressan, Veronica and Barbini, Matilde and Fusco, Achille and Neri, Sofia and Piccini Bianchessi, Maria Letizia and Rossi, Sarah and Chesi, Cristiano},
  booktitle={Proceedings of GLOW 47},
  year={2026}
}

License

CC BY-SA 4.0