README.md

---
dataset_info:
- config_name: armis
  features:
  - name: id
    dtype: string
  - name: text
    dtype: string
  - name: lang
    dtype: string
  - name: annotations
    dtype: int64
  - name: soft_label
    list: float64
  splits:
  - name: train
    num_bytes: 170524
    num_examples: 657
  - name: validation
    num_bytes: 36479
    num_examples: 141
  - name: test
    num_bytes: 37966
    num_examples: 145
  download_size: 142736
  dataset_size: 244969
- config_name: conv_abuse
  features:
  - name: id
    dtype: string
  - name: prev_agent
    dtype: string
  - name: prev_user
    dtype: string
  - name: agent
    dtype: string
  - name: user
    dtype: string
  - name: lang
    dtype: string
  - name: annotations
    dtype: int64
  - name: soft_label
    list: float64
  splits:
  - name: train
    num_bytes: 418334
    num_examples: 2398
  - name: validation
    num_bytes: 145527
    num_examples: 812
  - name: test
    num_bytes: 150336
    num_examples: 840
  download_size: 263301
  dataset_size: 714197
- config_name: csc
  features:
  - name: id
    dtype: string
  - name: context
    dtype: string
  - name: response
    dtype: string
  - name: lang
    dtype: string
  - name: annotations
    dtype: int64
  - name: soft_label
    list: float64
  splits:
  - name: train
    num_bytes: 1954431
    num_examples: 5628
  - name: validation
    num_bytes: 247727
    num_examples: 704
  - name: test
    num_bytes: 243899
    num_examples: 704
  download_size: 694211
  dataset_size: 2446057
- config_name: hs_brexit
  features:
  - name: id
    dtype: string
  - name: text
    dtype: string
  - name: lang
    dtype: string
  - name: annotations
    dtype: int64
  - name: soft_label
    list: float64
  splits:
  - name: train
    num_bytes: 123650
    num_examples: 784
  - name: validation
    num_bytes: 26186
    num_examples: 168
  - name: test
    num_bytes: 26459
    num_examples: 168
  download_size: 119565
  dataset_size: 176295
- config_name: md_agreement
  features:
  - name: id
    dtype: string
  - name: text
    dtype: string
  - name: lang
    dtype: string
  - name: annotations
    dtype: int64
  - name: soft_label
    list: float64
  splits:
  - name: train
    num_bytes: 1182092
    num_examples: 6592
  - name: validation
    num_bytes: 201159
    num_examples: 1104
  - name: test
    num_bytes: 549739
    num_examples: 3057
  download_size: 1240448
  dataset_size: 1932990
- config_name: mp
  features:
  - name: id
    dtype: string
  - name: post
    dtype: string
  - name: reply
    dtype: string
  - name: lang
    dtype: string
  - name: annotations
    dtype: int64
  - name: soft_label
    list: float64
  splits:
  - name: train
    num_bytes: 4552552
    num_examples: 12017
  - name: validation
    num_bytes: 1127746
    num_examples: 3005
  - name: test
    num_bytes: 1353129
    num_examples: 3756
  download_size: 4458000
  dataset_size: 7033427
- config_name: varierrnli
  features:
  - name: id
    dtype: string
  - name: context
    dtype: string
  - name: statement
    dtype: string
  - name: lang
    dtype: string
  - name: annotations
    dtype: int64
  - name: entailment
    dtype: float64
  - name: neutral
    dtype: float64
  - name: contradiction
    dtype: float64
  splits:
  - name: train
    num_bytes: 90191
    num_examples: 388
  - name: validation
    num_bytes: 11490
    num_examples: 50
  - name: test
    num_bytes: 10921
    num_examples: 50
  download_size: 83332
  dataset_size: 112602
configs:
- config_name: armis
  data_files:
  - split: train
    path: armis/train-*
  - split: validation
    path: armis/validation-*
  - split: test
    path: armis/test-*
- config_name: conv_abuse
  data_files:
  - split: train
    path: conv_abuse/train-*
  - split: validation
    path: conv_abuse/validation-*
  - split: test
    path: conv_abuse/test-*
- config_name: csc
  data_files:
  - split: train
    path: csc/train-*
  - split: validation
    path: csc/validation-*
  - split: test
    path: csc/test-*
- config_name: hs_brexit
  data_files:
  - split: train
    path: hs_brexit/train-*
  - split: validation
    path: hs_brexit/validation-*
  - split: test
    path: hs_brexit/test-*
- config_name: md_agreement
  data_files:
  - split: train
    path: md_agreement/train-*
  - split: validation
    path: md_agreement/validation-*
  - split: test
    path: md_agreement/test-*
- config_name: mp
  data_files:
  - split: train
    path: mp/train-*
  - split: validation
    path: mp/validation-*
  - split: test
    path: mp/test-*
- config_name: varierrnli
  data_files:
  - split: train
    path: varierrnli/train-*
  - split: validation
    path: varierrnli/validation-*
  - split: test
    path: varierrnli/test-*
source_datasets: []
license: other
---

# lewidi

Learning with Disagreements (LeWiDi, SemEval-2023 Task 11 and its 2025 edition): soft labels from every annotator.

One config per dataset: md_agreement (offensiveness, 5 annotators), hs_brexit (hate speech, 6), armis
(Arabic misogyny and sexism, 3), conv_abuse (abuse in user turns of chatbot dialogues, 3 or more), csc
(sarcasm rated 1-6), mp (MultiPICo irony, multilingual) and varierrnli (NLI where each annotator may accept
several labels). ``soft_label`` lists the share of annotators per label in ``labels`` order; varierrnli
instead gives, per NLI label, the share of annotators who accepted it. Text fields are unpacked from the
harmonized JSON. The Paraphrase set (an 11-level scale over 500 items) is left out. The original datasets'
licenses and terms apply unchanged; see the LeWiDi repository (https://github.com/Le-Wi-Di/le-wi-di.github.io)
and each dataset's paper.

Repackaged as parquet for [tasksource](https://github.com/sileod/tasksource) by [scripts/upload_repackaged.py](https://github.com/sileod/tasksource/blob/main/scripts/upload_repackaged.py).