r-three-moto/multilingual-addition

Multilingual Addition

The MoTo paper uses this dataset as a minimal case study of how tokenization affects arithmetic across languages.

Part of the MoTo release: paper · code · models.

Synthetic dataset of addition problems of the form a+b=answer, where a and b are written-form representations of integers in 25 languages, plus a 26th split using raw digit strings.

Task format

Each sample contains:

field	type	description
a_str	str	written-form (or digit) representation of a
a_digit	int	integer value of a
b_str	str	written-form (or digit) representation of b
b_digit	int	integer value of b
answer	str	written-form (or digit) of a + b
answer_digit	int	integer value of a + b
text	str	"{a_str}+{b_str}={answer}" (completion target)
question	str	"{a_str}+{b_str}=" (prompt)
lang	str	language tag, e.g. eng_Latn, or digit

Numbers range from 0 to 999 for both a and b (answers up to 1998).

Languages

lang	train	val
eng_Latn	900,000	100,000
dan_Latn	900,000	100,000
swe_Latn	900,000	100,000
vie_Latn	900,000	100,000
hun_Latn	900,000	100,000
fas_Arab	900,000	100,000
tur_Latn	900,000	100,000
ces_Latn	900,000	100,000
arb_Arab	900,000	100,000
ell_Grek	900,000	100,000
ind_Latn	900,000	100,000
nld_Latn	900,000	100,000
pol_Latn	900,000	100,000
por_Latn	900,000	100,000
ita_Latn	900,000	100,000
jpn_Jpan	900,000	100,000
fra_Latn	900,000	100,000
spa_Latn	900,000	100,000
deu_Latn	900,000	100,000
cmn_Hani	900,000	100,000
rus_Cyrl	900,000	100,000
nob_Latn	900,000	100,000
fin_Latn	900,000	100,000
ben_Beng	900,000	100,000
kor_Hang	900,000	100,000
digit	900,000	100,000

Generation

python create_multilingual_addition_data.py \
    --hf_repo_id flexitok/multilingual-addition \
    --publish_to_hf \
    --a_min 0 --a_max 999 \
    --seed 42 --train_ratio 0.9

License

Released under MIT.

Citation

If you use this data, please cite MoTo and the tools it was built with.

@inproceedings{altintas2026moto,
  title={{MoTo}: Mixture of Tokenizers Towards Fair Multilingual Language Modeling},
  author={G{\"u}l Sena Alt{\i}nta{\c{s}} and Colin Raffel},
  booktitle={The Fortieth Annual Conference on Neural Information Processing Systems},
  year={2026},
  url={https://openreview.net/forum?id=avjeUhPuC6}
}

@electronic{num2words,
  author = {num2words},
  date-added = {2026-05-07 03:51:44 -0400},
  title = {num2words library - Convert numbers to words in multiple languages},
  url = {https://pypi.org/project/num2words/}}