Datasets:
metadata
dataset_info:
- config_name: dan
features:
- name: orig
list: string
- name: norm
list: string
splits:
- name: train
num_bytes: 280976
num_examples: 719
- name: test
num_bytes: 64073
num_examples: 181
download_size: 164904
dataset_size: 345049
- config_name: deu
features:
- name: orig
list: string
- name: norm
list: string
splits:
- name: train
num_bytes: 272039
num_examples: 1628
- name: test
num_bytes: 92741
num_examples: 583
- name: validation
num_bytes: 89123
num_examples: 573
download_size: 217092
dataset_size: 453903
- config_name: eng
features:
- name: orig
list: string
- name: norm
list: string
splits:
- name: train
num_bytes: 631212
num_examples: 2360
- name: test
num_bytes: 526979
num_examples: 1967
- name: validation
num_bytes: 164519
num_examples: 590
download_size: 677203
dataset_size: 1322710
- config_name: hrv
features:
- name: orig
list: string
- name: norm
list: string
splits:
- name: train
num_bytes: 971981
num_examples: 4760
- name: test
num_bytes: 276506
num_examples: 1586
- name: validation
num_bytes: 342128
num_examples: 1588
download_size: 812715
dataset_size: 1590615
- config_name: ind
features:
- name: orig
list: string
- name: norm
list: string
splits:
- name: train
num_bytes: 697988
num_examples: 3016
- name: test
num_bytes: 179636
num_examples: 861
- name: validation
num_bytes: 85344
num_examples: 431
download_size: 435304
dataset_size: 962968
- config_name: ind_eng
features:
- name: orig
list: string
- name: norm
list: string
splits:
- name: train
num_bytes: 247175
num_examples: 495
- name: test
num_bytes: 76432
num_examples: 165
- name: validation
num_bytes: 84002
num_examples: 165
download_size: 210136
dataset_size: 407609
- config_name: ita
features:
- name: orig
list: string
- name: norm
list: string
splits:
- name: train
num_bytes: 228436
num_examples: 593
- name: test
num_bytes: 36353
num_examples: 100
download_size: 138740
dataset_size: 264789
- config_name: jpn
features:
- name: orig
list: string
- name: norm
list: string
splits:
- name: train
num_bytes: 1155890
num_examples: 2132
- name: test
num_bytes: 422488
num_examples: 609
- name: validation
num_bytes: 202539
num_examples: 305
download_size: 567677
dataset_size: 1780917
- config_name: kor
features:
- name: orig
list: string
- name: norm
list: string
splits:
- name: train
num_bytes: 340854
num_examples: 1701
- name: test
num_bytes: 40615
num_examples: 214
- name: validation
num_bytes: 49262
num_examples: 212
download_size: 269680
dataset_size: 430731
- config_name: nld
features:
- name: orig
list: string
- name: norm
list: string
splits:
- name: train
num_bytes: 218649
num_examples: 907
- name: test
num_bytes: 92867
num_examples: 308
- name: validation
num_bytes: 69557
num_examples: 308
download_size: 206708
dataset_size: 381073
- config_name: rus
features:
- name: orig
list: string
- name: norm
list: string
splits:
- name: train
num_bytes: 274896
num_examples: 1544
- name: test
num_bytes: 95267
num_examples: 386
download_size: 174399
dataset_size: 370163
- config_name: slv
features:
- name: orig
list: string
- name: norm
list: string
splits:
- name: train
num_bytes: 768742
num_examples: 4670
- name: test
num_bytes: 256900
num_examples: 1557
- name: validation
num_bytes: 261687
num_examples: 1557
download_size: 631982
dataset_size: 1287329
- config_name: spa
features:
- name: orig
list: string
- name: norm
list: string
splits:
- name: train
num_bytes: 131515
num_examples: 568
- name: test
num_bytes: 125214
num_examples: 531
download_size: 145378
dataset_size: 256729
- config_name: srp
features:
- name: orig
list: string
- name: norm
list: string
splits:
- name: train
num_bytes: 967757
num_examples: 4138
- name: test
num_bytes: 295265
num_examples: 1377
- name: validation
num_bytes: 305680
num_examples: 1379
download_size: 753207
dataset_size: 1568702
- config_name: tha
features:
- name: orig
list: string
- name: norm
list: string
splits:
- name: train
num_bytes: 3814903
num_examples: 1750
- name: test
num_bytes: 1095005
num_examples: 500
- name: validation
num_bytes: 544527
num_examples: 250
download_size: 1166871
dataset_size: 5454435
- config_name: tur
features:
- name: orig
list: string
- name: norm
list: string
splits:
- name: train
num_bytes: 141822
num_examples: 570
- name: test
num_bytes: 36070
num_examples: 143
download_size: 125137
dataset_size: 177892
- config_name: tur_deu
features:
- name: orig
list: string
- name: norm
list: string
splits:
- name: train
num_bytes: 255826
num_examples: 800
- name: test
num_bytes: 70352
num_examples: 229
download_size: 180596
dataset_size: 326178
- config_name: vie
features:
- name: orig
list: string
- name: norm
list: string
splits:
- name: train
num_bytes: 1793840
num_examples: 8371
- name: test
num_bytes: 232579
num_examples: 1045
- name: validation
num_bytes: 240371
num_examples: 1050
download_size: 645576
dataset_size: 2266790
configs:
- config_name: dan
data_files:
- split: train
path: dan/train-*
- split: test
path: dan/test-*
- config_name: deu
data_files:
- split: train
path: deu/train-*
- split: test
path: deu/test-*
- split: validation
path: deu/validation-*
- config_name: eng
data_files:
- split: train
path: eng/train-*
- split: test
path: eng/test-*
- split: validation
path: eng/validation-*
- config_name: hrv
data_files:
- split: train
path: hrv/train-*
- split: test
path: hrv/test-*
- split: validation
path: hrv/validation-*
- config_name: ind
data_files:
- split: train
path: ind/train-*
- split: test
path: ind/test-*
- split: validation
path: ind/validation-*
- config_name: ind_eng
data_files:
- split: train
path: ind_eng/train-*
- split: test
path: ind_eng/test-*
- split: validation
path: ind_eng/validation-*
- config_name: ita
data_files:
- split: train
path: ita/train-*
- split: test
path: ita/test-*
- config_name: jpn
data_files:
- split: train
path: jpn/train-*
- split: test
path: jpn/test-*
- split: validation
path: jpn/validation-*
- config_name: kor
data_files:
- split: train
path: kor/train-*
- split: test
path: kor/test-*
- split: validation
path: kor/validation-*
- config_name: nld
data_files:
- split: train
path: nld/train-*
- split: test
path: nld/test-*
- split: validation
path: nld/validation-*
- config_name: rus
data_files:
- split: train
path: rus/train-*
- split: test
path: rus/test-*
- config_name: slv
data_files:
- split: train
path: slv/train-*
- split: test
path: slv/test-*
- split: validation
path: slv/validation-*
- config_name: spa
data_files:
- split: train
path: spa/train-*
- split: test
path: spa/test-*
- config_name: srp
data_files:
- split: train
path: srp/train-*
- split: test
path: srp/test-*
- split: validation
path: srp/validation-*
- config_name: tha
data_files:
- split: train
path: tha/train-*
- split: test
path: tha/test-*
- split: validation
path: tha/validation-*
- config_name: tur
data_files:
- split: train
path: tur/train-*
- split: test
path: tur/test-*
- config_name: tur_deu
data_files:
- split: train
path: tur_deu/train-*
- split: test
path: tur_deu/test-*
- config_name: vie
data_files:
- split: train
path: vie/train-*
- split: test
path: vie/test-*
- split: validation
path: vie/validation-*
language:
- da
- de
- en
- hr
- id
- ja
- ko
- nl
- ru
- sl
- es
- sr
- th
- tr
- vi
Multilingual Lexical Normalization
This is the HuggingFace version of the MultiLexNorm dataset.
The original repository (https://bitbucket.org/robvanderg/multilexnorm/) contains all data pre-processed in the same format:
Most most
social social
pple people
r are
troublesome troublesome
Some of the languages include annotation for word splits and merges. When a word is split, the normalization column include a white-space character, and with a merge the normalization is only included for the first word:
if If
i i
have have
a a
head headache
ache
tomorro tomorrow
ima i'm going to
be be
pissed pissed
This Huggingface version instead can be loaded like:
from datasets import load_dataset
ds = load_dataset(
"robvanderg/MultiLexNorm",
"dan",
split="train",
)
print(ds)
Dataset({
features: ['orig', 'norm'],
num_rows: 719
})
The internal format is the same, so the norm list might contain empty elements or whitespaces (whereas the orig shouldn't)
Citation
The data in this benchmark comes from many different sources, we collected existing datasets and converged them to the same formate. This started with the original MultiLexNorm, and did the same for MultiLexNorm++. We added Russian data later.
For the original 12 language variants:
@inproceedings{multilexnorm,
title= {{MultiLexNorm}: A Shared Task on Multilingual Lexical Normalization},
author = "van der Goot, Rob and Ramponi, Alan and Zubiaga, Arkaitz and Plank, Barbara and Muller, Benjamin and San Vicente Roncal, I\~{n}aki and Ljube\v{s}i\'{c}, Nikola and \c{C}etino\u{g}lu, {\"O}zlem and Mahendra, Rahmad and \c{C}olako\u{g}lu, Talha and Baldwin, Timothy and Caselli, Tommaso and Sidorenko, Wladimir",
booktitle = "Proceedings of the 7th Workshop on Noisy User-generated Text (W-NUT 2021)",
year = "2021",
publisher = "Association for Computational Linguistics",
address = "Punta Cana, Dominican Republic"
}
For the additional five Asian languages in MultiLexNorm++:
@article{10.1145/3812651,
author = {Buaphet, Weerayut and Nguyen, Thanh-Nhi and Kondo, Risa and Kajiwara, Tomoyuki and Kim, Yumin and Lee, Jimin and Lee, Hwanhee and Lovenia, Holy and Limkonchotiwat, Peerat and Nutanong, Sarana and Van der Goot, Rob},
title = {MultiLexNorm++: A Unified Benchmark and a Generative Model for Lexical Normalization for Asian Languages},
year = {2026},
issue_date = {July 2026},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
volume = {25},
number = {7},
issn = {2375-4699},
url = {https://doi.org/10.1145/3812651},
doi = {10.1145/3812651},
journal = {ACM Trans. Asian Low-Resour. Lang. Inf. Process.},
month = jul,
articleno = {61},
numpages = {14},
}
For Russian:
@InProceedings{10.1007/978-3-032-04958-2_13,
author="Feoktistov, Grigorii
and Morozov, Dmitry",
editor="Bakaev, Maxim
and Bolgov, Radomir
and Chizhik, Anna
and Chugunov, Andrei
and Demareva, Valeriia
and Kabanov, Yury
and Pereira, Roberto
and R., Elakkiya
and Zhang, Wei",
title="TL; DR: Text Normalization for Social Media Corpus",
booktitle="Internet and Modern Society",
year="2026",
publisher="Springer Nature Switzerland",
address="Cham",
pages="168--176",
}
We would also strongly encourage you to cite the original papers if you use this data:
We use the MultiLexNorm dataset~\cite{multilexnorm,10.1145/3812651,10.1007/978-3-032-04958-2_13,plank-etal-2020-dan,sidarenka2013rule,baldwin-etal-2015-shared,alegria2013introduccion,11356/1170,barik-etal-2019-normalization,van-der-goot-etal-2020-norm,dutchNorm,11356/1123,11356/1171,colakoglu-etal-2019-normalizing,van-der-goot-etal-2021-lexical}
The remaining bib files are listed below:
@inproceedings{plank-etal-2020-dan,
title = "{D}a{N}+: {D}anish Nested Named Entities and Lexical Normalization",
author = "Plank, Barbara and
Jensen, Kristian N{\o}rgaard and
van der Goot, Rob",
booktitle = "Proceedings of the 28th International Conference on Computational Linguistics",
month = dec,
year = "2020",
address = "Barcelona, Spain (Online)",
publisher = "International Committee on Computational Linguistics",
url = "https://www.aclweb.org/anthology/2020.coling-main.583",
doi = "10.18653/v1/2020.coling-main.583",
pages = "6649--6662",
}
@inproceedings{sidarenka2013rule,
title={Rule-based normalization of {G}erman {T}witter messages},
author={Sidarenka, Uladzimir and Scheffler, Tatjana and Stede, Manfred},
booktitle={Proc. of the GSCL Workshop Verarbeitung und Annotation von Sprachdaten aus Genres internetbasierter Kommunikation},
year={2013}
}
@inproceedings{baldwin-etal-2015-shared,
title = "Shared Tasks of the 2015 Workshop on Noisy User-generated Text: {T}witter Lexical Normalization and Named Entity Recognition",
author = "Baldwin, Timothy and
de Marneffe, Marie Catherine and
Han, Bo and
Kim, Young-Bum and
Ritter, Alan and
Xu, Wei",
booktitle = "Proceedings of the Workshop on Noisy User-generated Text",
month = jul,
year = "2015",
address = "Beijing, China",
publisher = "Association for Computational Linguistics",
url = "https://www.aclweb.org/anthology/W15-4319",
doi = "10.18653/v1/W15-4319",
pages = "126--135",
}
@inproceedings{alegria2013introduccion,
title={Introducci{\'o}n a la Tarea Compartida {Tweet-Norm} 2013: Normalizaci{\'o}n L{\'e}xica de Tuits en {E}spa{\~n}ol.},
author={Alegria, Inaki and Aranberri, Nora and Fresno, V{\'\i}ctor and Gamallo, Pablo and Padr{\'o}, Lluis and San Vicente, Inaki and Turmo, Jordi and Zubiaga, Arkaitz},
booktitle={Tweet-Norm@ SEPLN},
pages={1--9},
year={2013}
}
@misc{11356/1170,
title = {Croatian {T}witter training corpus {ReLDI}-{NormTagNER}-hr 2.0},
author = {Ljube{\v s}i{\'c}, Nikola and Erjavec, Toma{\v z} and Mili{\v c}evi{\'c}, Maja and Samard{\v z}i{\'c}, Tanja},
url = {http://hdl.handle.net/11356/1170},
note = {Slovenian language resource repository {CLARIN}.{SI}},
copyright = {Creative Commons - Attribution 4.0 International ({CC} {BY} 4.0)},
year = {2017}
}
@inproceedings{barik-etal-2019-normalization,
title = "Normalization of {I}ndonesian-{E}nglish Code-Mixed {T}witter Data",
author = "Barik, Anab Maulana and
Mahendra, Rahmad and
Adriani, Mirna",
booktitle = "Proceedings of the 5th Workshop on Noisy User-generated Text (W-NUT 2019)",
month = nov,
year = "2019",
address = "Hong Kong, China",
publisher = "Association for Computational Linguistics",
url = "https://www.aclweb.org/anthology/D19-5554",
doi = "10.18653/v1/D19-5554",
pages = "417--424",
}
@inproceedings{van-der-goot-etal-2020-norm,
title = "Norm It! Lexical Normalization for {I}talian and Its Downstream Effects for Dependency Parsing",
author = "van der Goot, Rob and
Ramponi, Alan and
Caselli, Tommaso and
Cafagna, Michele and
De Mattei, Lorenzo",
booktitle = "Proceedings of the 12th Language Resources and Evaluation Conference",
month = may,
year = "2020",
address = "Marseille, France",
publisher = "European Language Resources Association",
url = "https://www.aclweb.org/anthology/2020.lrec-1.769",
pages = "6272--6278",
language = "English",
ISBN = "979-10-95546-34-4",
}
@mastersthesis{dutchNorm,
author = {Schuur, Youri},
title = {Normalization for {D}utch for improved POS tagging},
school = {University of Groningen},
year = 2020,
url = {http://robvanderg.github.io/doc/normalization_dutch.pdf}
}
@misc{11356/1123,
title = {{CMC} training corpus {Janes-Tag} 2.0},
author = {Erjavec, Toma{\v z} and Fi{\v s}er, Darja and {\v C}ibej, Jaka and Arhar Holdt, {\v S}pela and Ljube{\v s}i{\'c}, Nikola and Zupan, Katja},
url = {http://hdl.handle.net/11356/1123},
note = {Slovenian language resource repository {CLARIN}.{SI}},
copyright = {Creative Commons - Attribution-{ShareAlike} 4.0 International ({CC} {BY}-{SA} 4.0)},
year = {2017}
}
@misc{11356/1171,
title = {Serbian {T}witter training corpus {ReLDI}-{NormTagNER}-sr 2.0},
author = {Ljube{\v s}i{\'c}, Nikola and Erjavec, Toma{\v z} and Mili{\v c}evi{\'c}, Maja and Samard{\v z}i{\'c}, Tanja},
url = {http://hdl.handle.net/11356/1171},
note = {Slovenian language resource repository {CLARIN}.{SI}},
copyright = {Creative Commons - Attribution 4.0 International ({CC} {BY} 4.0)},
year = {2017}
}
@inproceedings{colakoglu-etal-2019-normalizing,
title = "Normalizing Non-canonical {T}urkish Texts Using Machine Translation Approaches",
author = {{\c{C}}olako{\u{g}}lu, Talha and
Sulubacak, Umut and
Tantu{\u{g}}, Ahmet C{\"u}neyd},
booktitle = "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics: Student Research Workshop",
month = jul,
year = "2019",
address = "Florence, Italy",
publisher = "Association for Computational Linguistics",
url = "https://www.aclweb.org/anthology/P19-2037",
doi = "10.18653/v1/P19-2037",
pages = "267--272"
}
@inproceedings{van-der-goot-etal-2021-lexical,
title = "Lexical Normalization for Code-switched Data and its Effect on {POS} Tagging",
author = "van der Goot, Rob and
\c{C}etino{\u{g}}lu, {\"O}zlem",
booktitle = "Proceedings of the 16th Conference of the {E}uropean Chapter of the Association for Computational Linguistics: Volume 1, Long Papers",
month = feb,
year = "2021",
publisher = "Association for Computational Linguistics"
}