xqasem 0.1.2

Cross-lingual QASem parser for predicate-centered question-answer structures.

XQASem

XQASem is a Python package for extracting predicate-centered question-answer structures from text. It currently includes presets for French, Russian, and Hebrew models hosted on HuggingFace.

This package accompanies the paper Effective QA-Driven Annotation of Predicate-Argument Relations Across Languages.

Installation

pip install xqasem

Install the spaCy pipelines you plan to use:

python -m spacy download fr_core_news_md
python -m spacy download ru_core_news_sm

Hebrew uses spacy-stanza; on first use, Stanza may need to download its Hebrew resources.

Environment

This project was tested with the following setup:

• Python 3.10.20
• torch 2.6.0 (CUDA 12.4)
• transformers 4.57.1
• spaCy 3.7.5

Requirements

• Python 3.10+
• transformers >= 4.50
• spaCy >= 3.7
• torch >= 2.0

Note: GPU is recommended for efficient inference.

Installation from source

git clone https://github.com/JohnnieDavidov/xqasem.git
cd xqasem
pip install -e .

Basic Usage

Example

from xqasem import XQasemParser

parser = XQasemParser.from_language("fr")

sentences = [
    "Les experts ont souligné que le nouvel algorithme accélère considérablement le traitement des requêtes complexes."
]

df = parser(sentences)

print(df)

Example output:

sentence	predicate	predicate_type	question	answer
Les experts ont souligné que le nouvel algorithme accélère considérablement le traitement des requêtes complexes.	souligné	verb	qui a souligné quelque chose?	Les experts
Les experts ont souligné que le nouvel algorithme accélère considérablement le traitement des requêtes complexes.	accélère	verb	qu'est-ce qui accélère quelque chose?	le nouvel algorithme
Les experts ont souligné que le nouvel algorithme accélère considérablement le traitement des requêtes complexes.	accélère	verb	qu'est-ce que quelque chose accélère?	le traitement des requêtes complexes

The built-in language presets are:

parser_fr = XQasemParser.from_language("fr")
parser_ru = XQasemParser.from_language("ru")
parser_he = XQasemParser.from_language("he")

You can also load an explicit Hugging Face model:

parser = XQasemParser.from_pretrained(
    "YonatanDavidov/qasem-fr-claire-lora",
    spacy_lang="fr_core_news_md",
    is_adapter=True,
)

Command Line

After installation, you can run:

xqasem-parse --lang fr --output outputs/fr.csv "Les experts ont souligné que le nouvel algorithme accélère le traitement."

You can also run the script directly:

python scripts/run_parser.py --lang fr --output outputs/fr.csv

Model Presets

Language	Model	spaCy pipeline
French	YonatanDavidov/qasem-fr-claire-lora	fr_core_news_md
Russian	YonatanDavidov/qasem-ru-sambalingo-lora	ru_core_news_sm
Hebrew	YonatanDavidov/qasem-he-dictalm2-lora	he via spacy-stanza
Hebrew full model	YonatanDavidov/qasem-he-dictalm2-full	he via spacy-stanza

Citation

If you use this code or the released models, please cite:

@inproceedings{davidov-etal-2026-effective,
    title = "Effective {QA}-Driven Annotation of Predicate{--}Argument Relations Across Languages",
    author = "Davidov, Jonathan  and
      Slobodkin, Aviv  and
      Klein, Shmuel Tomi  and
      Tsarfaty, Reut  and
      Dagan, Ido  and
      Klein, Ayal",
    editor = "Demberg, Vera  and
      Inui, Kentaro  and
      Marquez, Llu{\\'i}s",
    booktitle = "Proceedings of the 19th Conference of the {E}uropean Chapter of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
    month = mar,
    year = "2026",
    address = "Rabat, Morocco",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2026.eacl-long.112/",
    doi = "10.18653/v1/2026.eacl-long.112",
    pages = "2484--2502",
    ISBN = "979-8-89176-380-7"
}

Acknowledgments

• Ayal Klein
• Paul Roit

This implementation builds on ideas and code structure from Paul Roit's qasem_parser, which provides the original English QA-Sem parsing framework that inspired this multilingual extension.