AFRIMGSM-TT: Translated Grade School Math Problems from 17 Languages

AFRIMGSM-TT is an evaluation dataset containing translations of the GSM8K grade school math word problems into English from 16 African languages and one high-resource language using the NLLB model. It includes test sets across all 17 source languages and was created by the Masakhane organization. The dataset was last updated on May 21, 2024.

Use Cases

• Benchmarking machine translation systems based on the described multilingual test sets.
• Evaluating cross-lingual transfer learning for mathematical reasoning tasks based on the translated GSM8K problems.
• Analyzing translation quality for low-resource African languages based on the described language coverage.

Strengths

• Covers 17 languages, including 16 African languages, providing a specific multilingual evaluation resource.
• Based on the established GSM8K benchmark for grade school math problems.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

Masakhane

Collection Method

Translations generated using the NLLB model from the original GSM8K dataset.

Freshness

Last updated 2024-05-21 05:20:38; freshness should be verified.

Geography

Languages from Africa and one unspecified high-resource language.