Dataset card for Neo-GATE

Homepage: https://mt.fbk.eu/neo-gate/

Dataset summary

Neo-GATE is a bilingual corpus designed to benchmark the ability of machine translation (MT) systems to translate from English into Italian using gender-inclusive neomorphemes. It is built upon GATE (Rarrick et al., 2023), a benchmark for the evaluation of gender rewriters and gender bias in MT.

Neo-GATE includes 841 test entries (Neo-GATE.tsv) and 100 dev entries (Neo-GATE-dev.tsv).

Each entry is composed of an English source sentence, three Italian references which only differ for the presence of either masculine/feminine/nonbinary words, and the annotation of the target words that are relevant for the evaluation of gender-inclusive MT.

The source sentences are gender-ambiguous, i.e. they provide no information about the gender of human referents. In this setting, words referring to human entities in the target language should be rendered with neomorphemes, special characters or symbols that replace masculine and feminine inflectional morphemes.

Neo-GATE allows for the evaluation of any neomorpheme paradigm in Italian. For more details see the Adaptation section below.

Data Fields

Neo-GATE.tsv includes the following columns:

• #: Neo-GATE unique identifier.
• GATE-ID: A unique identifier of the entry in GATE, composed of a prefix indicating the subset of origin within GATE (e.g., IT_2_variants) followed by a serial number indicating the position of the entry within that subset (i.e., 001, 002, etc.).
• SOURCE: The English source sentence.
• REF-M: The Italian reference where all gender-marked terms are masculine.
• REF-F: The Italian reference where all gender-marked terms are feminine.
• REF-TAGGED: The Italian reference where all gender-marked terms are tagged with Neo-GATE's annotation.
• ANNOTATION: The word level annotation.

Dataset creation

Please refer to the original paper for full details on dataset creation.

Curation rationale

Neo-GATE was designed to allow for the evalution of gender-inclusive MT and to be adaptable to any neomorpheme paradigm in Italian. To this aim, the original Italian references found in GATE were edited so as to have placeholder tags in place of gendered morphemes and function words (articles, possessive adjectives, etc.) referred to human entities. The tags were designed to cover all parts of the grammar which express grammatical gender, and to be replaced with corresponding forms in the desired neomorpheme paradigm.

Adaptation

To adapt Neo-GATE to the desired neomorpheme paradigm, a .json file mapping Neo-GATE's tagset to the desired forms is required. See schwa.json or asterisk.json for an example. For more information on the tagset, see Table 8 in the original paper.

To create the adapted references and annotations, use the neo-gate_format.py script with the following syntax:

python neo-gate_adapt.py --tagset JSON_FILE_PATH --out OUTPUT_FILE_NAME

This command will create two files: OUTPUT_FILE_NAME.ref, containing the adapted references, and OUTPUT_FILE_NAME.ann, containing the adapted annotations.

For instance, to generate the references and the annotations adapted to the Schwa paradigm provided in the example file schwa.json, the following command can be used:

python neo-gate_adapt.py --tagset schwa.json --out neogate_schwa

This will create the two files neogate_schwa.ref and neogate_schwa.ann.

By default, the script will adapt references and annotations found in Neo-GATE.tsv. If the Neo-GATE.tsv file is located in a different directory, or if you wish to use a different file (e.g., the dev set split file Neo-GATE-dev.tsv), you can specify the path to the file with the optional argument --neogate.

Evaluation

The evaluation code will soon be available at fbk-NEUTR-evAL.

Dataset Curators

• Andrea Piergentili (FBK): apiergentili@fbk.eu
• Beatrice Savoldi (FBK): bsavoldi@fbk.eu
• Luisa Bentivogli (FBK): bentivo@fbk.eu

Licensing Information

The Neo-GATE corpus is released under a Creative Commons Attribution 4.0 International license (CC BY 4.0). See the LICENSE file for details.

Citation

If you use Neo-GATE in your work, please cite the following paper:

    @misc{piergentili2024enhancing,
    title={Enhancing Gender-Inclusive Machine Translation with Neomorphemes and Large Language Models}, 
    author={Andrea Piergentili and Beatrice Savoldi and Matteo Negri and Luisa Bentivogli},
    year={2024},
    eprint={2405.08477},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}

Contributions

Thanks to @apiergentili for adding this dataset.