Datasets:
metadata
license: cc0-1.0
language:
- en
- es
- it
- fr
- pt
- de
- nl
- sv
- id
- eo
- sw
- vi
- tr
- pl
- cy
- tl
- ga
- mt
- ro
- fi
- hu
- eu
- is
- sq
- mg
- ha
- et
- lv
- lt
- ka
- hy
- yo
- ig
- am
- km
- lo
- ar
- fa
- ur
- hi
- bn
- ru
- uk
- el
- he
- sr
- zh
- ja
- ko
- th
- ms
- ta
- te
- kn
- ml
- si
- ne
- mr
- gu
- pa
- or
tags:
- translation
- machine-translation
- multilingual
- parallel-corpus
- synthetic
- controlled-generation
- low-resource
- low-resource-languages
- parquet
- cc0
task_categories:
- translation
size_categories:
- 100K<n<1M
About
A synthetic multilingual translation dataset generated with GPT-5.5 XHigh. The goal is to teach AI models accurate translation from English to target languages — ranging from well-documented languages like Spanish and French, to lower-resource languages like Swahili and Yoruba.
Size
273,980 examples across 61 languages.
Quality
Each example uses complex sentence structures including subordinate clauses and conditionals, making the data richer and more generalizable than simple phrase datasets. All translations were verified using multi-agent pipelines to ensure quality across all language pairs. Generation Synthetically generated and verified using GPT-5.5 XHigh sub-agents.