Datasets:

Owos
/

paws-x

language:
  - sw
  - ur
license: cc-by-sa-4.0
task_categories:
  - text-classification
tags:
  - paraphrase
  - multilingual
  - swahili
  - urdu
  - paws-x
dataset_info:
  - config_name: sw
    features:
      - name: id
        dtype: int64
      - name: sentence1
        dtype: string
      - name: sentence2
        dtype: string
      - name: label
        dtype: int64
    splits:
      - name: test
        num_examples: 2000
  - config_name: ur
    features:
      - name: id
        dtype: int64
      - name: sentence1
        dtype: string
      - name: sentence2
        dtype: string
      - name: label
        dtype: int64
    splits:
      - name: test
        num_examples: 2000
configs:
  - config_name: sw
    data_dir: data/sw
    default: true
  - config_name: ur
    data_dir: data/ur

PAWS-X – Swahili & Urdu

Machine-translated Swahili and Urdu subsets of the PAWS-X (Paraphrase Adversaries from Word Scrambling – Cross-lingual) benchmark. This dataset was translated using Google Machine Translate.

Dataset Description

PAWS-X is a multilingual paraphrase identification benchmark. Given two sentences, the task is to determine whether they are paraphrases of each other (label 1) or not (label 0). This repository contains Swahili (sw) and Urdu (ur) translations.

Splits

Config	Split	Language	# Examples
sw	test	Swahili	2,000
ur	test	Urdu	2,000

Data Format (CSV)

Columns: id, sentence1, sentence2, label

Usage

from datasets import load_dataset

# Load Swahili
ds_sw = load_dataset("Owos/paws-x", "sw")

# Load Urdu
ds_ur = load_dataset("Owos/paws-x", "ur")

Source

Derived from hgissbkh/paws-x.

Citation

@article{owodunni2025continually,
  title={Continually adding new languages to multilingual language models},
  author={Owodunni, Abraham Toluwase and Kumar, Sachin},
  journal={arXiv preprint arXiv:2509.11414},
  year={2025}
}