Dataset Preview
Duplicate
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed because of a cast error
Error code:   DatasetGenerationCastError
Exception:    DatasetGenerationCastError
Message:      An error occurred while generating the dataset

All the data files must have the same columns, but at some point there are 7 new columns ({'channel_name', 'message ID', 'dehumanisation', 'language', 'country', 'date', 'message'}) and 9 missing columns ({'Extreme negative evaluation', 'Deindividuated entitative perception', 'Explicit comparison to an animal, machine or non-human substance', 'Examples', 'Denial of prosocial values and communality', 'Moral disgust or indifference', 'Denial of agency', 'Derogative word', 'Type'}).

This happened while the csv dataset builder was generating data using

hf://datasets/rehumanizer/telegram_dehumanisation22/full_labelled_dataset.csv (at revision 516a1d98bf53c3c53c5a3e4373859d9d5b97ecb9), ['hf://datasets/rehumanizer/telegram_dehumanisation22@516a1d98bf53c3c53c5a3e4373859d9d5b97ecb9/dehumanising keywords.csv', 'hf://datasets/rehumanizer/telegram_dehumanisation22@516a1d98bf53c3c53c5a3e4373859d9d5b97ecb9/full_labelled_dataset.csv', 'hf://datasets/rehumanizer/telegram_dehumanisation22@516a1d98bf53c3c53c5a3e4373859d9d5b97ecb9/manually labelled posts.csv']

Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1856, in _prepare_split_single
                  writer.write_table(table)
                  ~~~~~~~~~~~~~~~~~~^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 811, in write_table
                  self._write_table(pa_table, writer_batch_size=writer_batch_size)
                  ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 822, in _write_table
                  pa_table = table_cast(pa_table, self._schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2381, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2309, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              message ID: int64
              message: string
              dehumanisation: double
              country: string
              language: string
              channel_name: string
              date: string
              -- schema metadata --
              pandas: '{"index_columns": [{"kind": "range", "name": null, "start": 0, "' + 1103
              to
              {'Derogative word': Value('string'), 'Explicit comparison to an animal, machine or non-human substance': Value('float64'), 'Moral disgust or indifference': Value('float64'), 'Extreme negative evaluation': Value('float64'), 'Denial of prosocial values and communality': Value('float64'), 'Deindividuated entitative perception': Value('float64'), 'Denial of agency': Value('float64'), 'Type': Value('string'), 'Examples': Value('string')}
              because column names don't match
              
              During handling of the above exception, another exception occurred:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1379, in compute_config_parquet_and_info_response
                  parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
                                                                        ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      builder, max_dataset_size_bytes=max_dataset_size_bytes
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 956, in stream_convert_to_parquet
                  builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
                  ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1702, in _prepare_split
                  for job_id, done, content in self._prepare_split_single(
                                               ~~~~~~~~~~~~~~~~~~~~~~~~~~^
                      gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  ):
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1858, in _prepare_split_single
                  raise DatasetGenerationCastError.from_cast_error(
                  ...<4 lines>...
                  )
              datasets.exceptions.DatasetGenerationCastError: An error occurred while generating the dataset
              
              All the data files must have the same columns, but at some point there are 7 new columns ({'channel_name', 'message ID', 'dehumanisation', 'language', 'country', 'date', 'message'}) and 9 missing columns ({'Extreme negative evaluation', 'Deindividuated entitative perception', 'Explicit comparison to an animal, machine or non-human substance', 'Examples', 'Denial of prosocial values and communality', 'Moral disgust or indifference', 'Denial of agency', 'Derogative word', 'Type'}).
              
              This happened while the csv dataset builder was generating data using
              
              hf://datasets/rehumanizer/telegram_dehumanisation22/full_labelled_dataset.csv (at revision 516a1d98bf53c3c53c5a3e4373859d9d5b97ecb9), ['hf://datasets/rehumanizer/telegram_dehumanisation22@516a1d98bf53c3c53c5a3e4373859d9d5b97ecb9/dehumanising keywords.csv', 'hf://datasets/rehumanizer/telegram_dehumanisation22@516a1d98bf53c3c53c5a3e4373859d9d5b97ecb9/full_labelled_dataset.csv', 'hf://datasets/rehumanizer/telegram_dehumanisation22@516a1d98bf53c3c53c5a3e4373859d9d5b97ecb9/manually labelled posts.csv']
              
              Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Derogative word
string
Explicit comparison to an animal, machine or non-human substance
null
Moral disgust or indifference
float64
Extreme negative evaluation
float64
Denial of prosocial values and communality
float64
Deindividuated entitative perception
float64
Denial of agency
float64
Type
string
Examples
null
бандерорвцы
null
1
null
1
null
null
dehumanising ukrainians
null
бендерорвцы
null
1
null
1
null
null
dehumanising ukrainians
null
боевик
null
null
null
1
null
0.5
dehumanising ukrainians
null
враг, противник
null
null
1
null
null
null
dehumanising ukrainians
null
враг, противник (with respect to a whole group)
null
null
1
null
1
null
dehumanising ukrainians
null
всушник
null
1
null
null
null
null
dehumanising ukrainians
null
вфу
null
1
null
null
null
1
dehumanising ukrainians
null
вышивaнка
null
1
null
null
null
null
dehumanising ukrainians
null
вышивaта
null
1
null
null
null
null
dehumanising ukrainians
null
живая сила (противника)
null
null
null
null
1
null
dehumanising ukrainians
null
западэнцы
null
1
null
null
null
null
dehumanising ukrainians
null
вой
null
0.5
null
null
null
null
dehumanising ukrainians
null
зверства
null
null
null
1
null
null
dehumanising ukrainians
null
зе
null
1
null
null
null
null
dehumanising ukrainians
null
зсушник
null
1
null
null
null
null
dehumanising ukrainians
null
майданутый
null
1
null
null
null
1
dehumanising ukrainians
null
майдаун
null
1
null
null
null
1
dehumanising ukrainians
null
малороссия
null
1
null
null
null
0.5
dehumanising ukrainians
null
марионеточный режим
null
0.5
0.5
null
null
0.5
dehumanising ukrainians
null
нацбат
null
null
null
null
null
1
dehumanising ukrainians
null
нацик
null
1
1
1
null
null
dehumanising ukrainians
null
националист
null
null
null
null
null
0.5
dehumanising ukrainians
null
нацист
null
null
1
1
null
null
dehumanising ukrainians
null
нацэлементы
null
null
null
null
1
0.5
dehumanising ukrainians
null
ненька
null
1
null
null
null
null
dehumanising ukrainians
null
новороссия
null
1
null
null
null
1
dehumanising ukrainians
null
правосек
null
1
null
null
null
null
dehumanising ukrainians
null
рагуля / рогуля
null
1
null
null
null
1
dehumanising ukrainians
null
угил
null
1
1
null
null
null
dehumanising ukrainians
null
укра
null
1
null
null
null
null
dehumanising ukrainians
null
укро…
null
1
null
null
null
null
dehumanising ukrainians
null
укроп
null
1
null
null
null
1
dehumanising ukrainians
null
укрфашист / укронацист
null
1
1
1
1
null
dehumanising ukrainians
null
укры
null
1
null
null
1
null
dehumanising ukrainians
null
хохлы
null
1
null
null
null
null
dehumanising ukrainians
null
ЦИПсО
null
null
null
1
null
null
dehumanising ukrainians
null
киевская клика
null
null
0.5
null
null
0.5
dehumanising ukrainians
null
киевский/украинский режим
null
null
0.5
null
null
0.5
dehumanising ukrainians
null
укрорейх
null
1
1
1
1
null
dehumanising ukrainians
null
зверства
null
0.5
null
null
1
null
dehumanising russians
null
Z-нацист
null
null
1
1
null
null
dehumanising russians
null
Zапутинист / Zапутiнист
null
1
1
null
null
null
dehumanising russians
null
zасранец
null
1
1
null
null
null
dehumanising russians
null
zло
null
null
1
1
null
null
dehumanising russians
null
абизяна
null
1
null
1
null
1
dehumanising russians
null
вата
null
1
null
1
1
1
dehumanising russians
null
ватан
null
1
null
null
null
null
dehumanising russians
null
ватка
null
1
null
1
1
1
dehumanising russians
null
ватник
null
1
null
null
null
null
dehumanising russians
null
ввх
null
1
null
null
null
null
dehumanising russians
null
великороссийский / великоросийськi
null
1
null
null
null
null
dehumanising russians
null
великорусский / великоруськi
null
1
null
null
null
null
dehumanising russians
null
(подняли) вой
null
0.5
null
null
null
null
dehumanising russians
null
враг/ворог, противник
null
null
1
null
null
null
dehumanising russians
null
враг/ворог, противник (with respect to a whole group)
null
null
1
null
1
null
dehumanising russians
null
живая сила (противника)
null
null
null
null
1
null
dehumanising russians
null
кацап
null
1
null
null
null
null
dehumanising russians
null
макака
null
1
null
1
null
1
dehumanising russians
null
мерZота
null
1
1
null
null
null
dehumanising russians
null
мордор
null
1
1
1
1
1
dehumanising russians
null
москаль
null
1
null
null
null
null
dehumanising russians
null
москвота
null
1
null
null
1
1
dehumanising russians
null
московит
null
1
null
null
null
null
dehumanising russians
null
московія
null
1
null
null
null
null
dehumanising russians
null
орда
null
null
1
null
1
null
dehumanising russians
null
орк
null
1
1
1
null
1
dehumanising russians
null
путлер
null
1
1
null
null
null
dehumanising russians
null
раша, рашка
null
1
null
null
null
null
dehumanising russians
null
рашист, рашизм
null
1
1
1
1
null
dehumanising russians
null
русня
null
1
null
null
1
null
dehumanising russians
null
свинособаки / 🐷🐶
null
1
null
1
null
1
dehumanising russians
null
свинья
null
1
null
1
null
1
dehumanising russians
null
скоторосійський
null
1
1
1
1
1
dehumanising russians
null
скоторусы, свинораша
null
1
1
1
1
1
dehumanising russians
null
фашиzм
null
null
1
1
null
null
dehumanising russians
null
фашист
null
null
1
1
null
null
dehumanising russians
null
ху*ло
null
0.5
0.5
null
null
null
dehumanising russians
null
хуйло
null
0.5
0.5
null
null
null
dehumanising russians
null
эрефия/ерефiя
null
1
null
null
null
null
dehumanising russians
null
эрэфия
null
1
null
null
null
null
dehumanising russians
null
гасить
null
1
0
0
1
1
killing euphemism
null
дать прикурить
null
0
0
0
0
1
killing euphemism
null
ебашить
null
1
0
0
0
1
killing euphemism
null
зачистить
null
0
0
0
1
1
killing euphemism
null
закончить
null
1
0
0
1
1
killing euphemism
null
заминусовать / задвухсотить
null
1
0
0
1
1
killing euphemism
null
ликвидировать
null
1
0
0
0
1
killing euphemism
null
мінуснуть
null
1
0
0
1
1
killing euphemism
null
накрыть
null
0
0
0
0
1
killing euphemism
null
обезвредить/знешкодити
null
1
0
0
0
1
killing euphemism
null
обнулить
null
1
0
0
1
1
killing euphemism
null
отработать
null
1
0
0
1
1
killing euphemism
null
отрправить к Бандере
null
1
0
1
0
1
killing euphemism
null
перемолоть
null
1
0
0
1
1
killing euphemism
null
поджарить
null
1
0
0
0
1
killing euphemism
null
покрошить
null
1
0
0
1
1
killing euphemism
null
потери личного состава
null
0
0
0
1
0
killing euphemism
null
прижмурить
null
1
0
0
0
1
killing euphemism
null
приземлить
null
1
0
0
1
1
killing euphemism
null
приуныть
null
1
0
0
0
1
killing euphemism
null
End of preview.

Dataset Description

Overview

The Russian- and Ukrainian-Language Telegram Dehumanisation Dataset (2022) is a multilingual research corpus designed to support computational and social-scientific research on dehumanising language in wartime online communication. It comprises posts published by leading Russian and Ukrainian news and political Telegram channels during 2022, following Russia's full-scale invasion of Ukraine.

The dataset addresses a gap in existing resources for studying online dehumanisation, particularly the limited availability of annotated data in Russian and Ukrainian. It provides a theory-informed foundation for examining how dehumanising language is expressed in political and media communication, how it targets outgroups, and how its prevalence varies across different types of communication channels.

The dataset release comprises three complementary resources:

  1. Large-scale annotated corpus: 238,190 Telegram posts collected from leading Russian and Ukrainian news and political Telegram channels during 2022. Posts are classified using LLM-assisted annotation into three categories: non-dehumanising, marginally dehumanising, and dehumanising.

  2. Manually annotated evaluation set: A collection of 1,000 posts manually annotated for six theoretically informed manifestations of dehumanisation, the targeted outgroup, and overall dehumanisation. This resource supports the evaluation of automated classification approaches and more fine-grained analyses of dehumanising language.

  3. Dictionary of dehumanising and derogatory keywords: A lexicon comprising 108 indicative dehumanising and derogatory keywords frequently used in posts targeting the two national groups. Each keyword is labelled according to the six theoretically informed dimensions of dehumanisation and the target outgroup (Russians or Ukrainians). The dictionary provides an additional resource for lexicon-based identification and analysis of dehumanising expressions, as well as for comparing the vocabulary used to target the two national groups.

Content Warning

This dataset contains offensive, hateful, and dehumanising language targeting national and other outgroups. Some posts contain explicit derogatory expressions, slurs, animalistic metaphors, and other forms of hostile language. The material is retained for research purposes and should be approached with appropriate care.

Dataset Composition

The main corpus contains 238,190 posts collected from leading Russian and Ukrainian news and political Telegram channels during 2022, covering both languages. The posts are classified into three levels:

  • Non-dehumanising (0): Posts not classified as dehumanising.
  • Marginally dehumanising (0.5): Posts classified within the intermediate category of dehumanisation.
  • Dehumanising (1): Posts classified as dehumanising.

The three-category classification is intended to capture variation in the presence and intensity of dehumanising language, rather than treating dehumanisation as a strictly binary phenomenon.

The accompanying manually annotated evaluation set contains 1,000 posts and provides more detailed annotations covering six manifestations of dehumanisation, the targeted outgroup, and overall dehumanisation.

Annotation

The large-scale corpus was constructed using LLM-assisted annotation, applying a theoretically informed classification framework to the collected Telegram posts.

The manually annotated evaluation set provides an additional resource for assessing automated classification performance and studying more specific manifestations and targets of dehumanisation.

The two resources serve complementary purposes: the large-scale corpus enables quantitative analyses across a substantial collection of wartime communication, while the manually annotated set enables more detailed evaluation and investigation of the phenomenon.

Intended Uses

The dataset is intended for academic and non-commercial research, particularly in:

  • Computational detection and classification of dehumanising and hateful language.
  • Computational linguistics and multilingual NLP.
  • Research on political communication and wartime discourse.
  • Comparative studies of Russian- and Ukrainian-language communication.
  • Research on intergroup relations, outgroup derogation, and dehumanisation.
  • Analysis of the prevalence and dynamics of dehumanising language across different types of media and political communication channels.
  • Evaluation of LLM-assisted and other automated approaches to hate-speech and dehumanisation detection.

The dataset may also support research into the relationship between dehumanising discourse and broader patterns of wartime communication.

Citation

If you use this dataset in your research, publications, or other academic outputs, please cite the original paper introducing the dataset:

Chernenko, E., & Vedres, B. (2026). A Multilingual Dataset for Dehumanisation Detection in Russian and Ukrainian Wartime Communication. In Proceedings of the 10th Workshop on Online Abuse and Harms (WOAH) (pp. 346–360). Association for Computational Linguistics.

https://doi.org/10.18653/2026.woah-1.30

BibTeX:

@inproceedings{chernenko-vedres-2026-multilingual,
    title = "A Multilingual Dataset for Dehumanisation Detection in {R}ussian and {U}krainian Wartime Communication",
    author = "Chernenko, Elizaveta and Vedres, Balazs",
    editor = "Calabrese, Agostina and Davidson, Thomas and de Kock, Christine and Khurana, Urja and Marchiori Manerba, Marta and Piot, Paloma and Talat, Zeerak",
    booktitle = "Proceedings of the 10th Workshop on Online Abuse and Harms ({WOAH})",
    month = oct,
    year = "2026",
    address = "Budapest, Hungary",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2026.woah-1.30/",
    doi = "10.18653/2026.woah-1.30",
    pages = "346--360",
    ISBN = "979-8-89176-487-3"
}

Scope and Limitations

The dataset covers selected Russian and Ukrainian news and political Telegram channels and is restricted to communication published during 2022. It should therefore not be considered representative of all Telegram users, the entirety of Russian- or Ukrainian-language online communication, or public opinion in either country.

The large-scale corpus relies on LLM-assisted annotation. As with other automated annotation approaches, classifications may contain errors, including false positives and false negatives. The manually annotated evaluation set provides a complementary resource for assessing classification performance, but its size and scope should also be considered when interpreting evaluation results.

The dataset focuses specifically on dehumanising language as operationalised in the accompanying annotation framework. It does not capture every form of hate speech, hostility, or negative outgroup representation.

Responsible Use

The dataset is released to facilitate research into harmful language and intergroup hostility, not to promote, reproduce, or endorse the expressions it contains.

Users should take care when reproducing examples, presenting quotations, or building models using the data. In particular, the presence of a dehumanising expression in a post should be interpreted in its linguistic and communicative context rather than automatically treated as evidence of an author's endorsement.

Researchers should also consider the limitations of automated classification and avoid using dataset labels as definitive assessments of individuals or groups.

Downloads last month
54