dataset

RO-STS: Romanian Semantic Textual Similarity Dataset

This folder contains the two types of files released:

1. Textual similarity

In the text-similarity folder there are 3 tsv files:

• RO-STS.train.tsv containing the train portion of the dataset
• RO-STS.dev.tsv containing the development portion of the dataset
• RO-STS.test.tsv containing the test portion of the dataset

The train/dev/test splits are identical to the original English STS corpus splits.

All files have the same tab-separated format:

Similarity score [TAB] Sentence 1 [TAB] Sentence 2 

where Similarity score is a floating point number from 0 to 5, with 5 being the maximum similarity.

For example:

1.5	Un bărbat cântă la harpă.	Un bărbat cântă la claviatură.
1.8	O femeie taie cepe.	O femeie taie tofu.
3.5	Un bărbat merge pe o bicicletă electrică.	Un bărbat merge pe bicicletă.
2.2	Un bărbat cântă la tobe.	Un bărbat cântă la chitară.
2.2	Un bărbat cântă la chitară.	O doamnă cântă la chitară.

2. Parallel corpus (RO-EN)

The parallel corpus is found in the ro-en folder, and contains 6 files:

• RO-STS.train.ro and RO-STS.train.en representing the train portion of the dataset
• RO-STS.dev.ro and RO-STS.dev.en representing the development portion of the dataset
• RO-STS.test.ro and RO-STS.test.en representing the test portion of the dataset

The files are raw text, one-sentence per line, each file pair containing corresponding ro - en sentences.