# IndicGenBench Flores-IN: FLORES Machine Translation Sentences Human-Translated into 29 Indic Languages (Google Research)

## Facts

- Provider: Google Research (Singh et al.)
- License: CC-BY-SA-4.0 (https://github.com/google-research-datasets/indic-gen-bench)
- Size: 118 files
- Format: JSON (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/google/IndicGenBench_flores_in
- Access: public, readable without an account
- Category: Natural Language Processing
- Source: https://arxiv.org/abs/2404.16816
- Page: https://scigantic.com/archive/1b8b13a6-69d9-522c-aafb-66f3b1c29d21
- Keywords: Natural Language Processing, benchmark, JSON, Hugging Face

## About

Flores-IN is IndicGenBench's machine translation task: sentences from the FLORES benchmark paired with human translations, evaluated in both directions (English to Indic, enxx, and Indic to English, xxen). IndicGenBench is a multilingual, multi-way parallel benchmark from Google Research for measuring language generation by large language models in 29 Indic languages spanning 13 writing scripts and 4 language families, including many low-resource languages such as Assamese, Bhojpuri, Awadhi, Maithili, Santali and Konkani alongside Hindi, Bengali, Tamil, Telugu, Marathi, Gujarati, Kannada and Malayalam. Google extended existing English benchmarks by collecting human translations of their English examples into each target language. Each record has a source sentence, a target sentence, the translation direction and the language code. The repository holds 116 JSON files, one dev and one test file per language and direction, named like flores_as_en_test.json (about 70 MB); the dev and test files correspond to the FLORES dev and devtest splits. Load with the Hugging Face datasets library or plain JSON readers. Distributed under CC BY-SA 4.0, following the FLORES source license; cite Singh et al., IndicGenBench: A Multilingual Benchmark to Evaluate Generation Capabilities of LLMs on Indic Languages, ACL 2024.

## Load the data

```python
from huggingface_hub import hf_hub_download
path = hf_hub_download(repo_id="google/IndicGenBench_flores_in", filename="flores_as_en_dev.json", repo_type="dataset")
print(path)
```

## Data structure

File types in sampled files: JSON

Example file: `flores_as_en_dev.json`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/1b8b13a6-69d9-522c-aafb-66f3b1c29d21
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/1b8b13a6-69d9-522c-aafb-66f3b1c29d21
- Over MCP: `get_schema_card` with id `1b8b13a6-69d9-522c-aafb-66f3b1c29d21` at https://api.scigantic.com/mcp

## Related topics

- [JSON datasets](https://scigantic.com/topics/json)
- [Benchmark datasets](https://scigantic.com/topics/benchmark)
- [Natural Language Processing datasets](https://scigantic.com/topics/natural-language-processing)
