CAMeL-Lab/camelparser-barectb-BARECTB

CamelParser-BARECTB: BARECTB

A supervised Arabic dependency parser using the CATiB representation, trained on BARECTB for BARECTB: A Dependency Treebank and Parsing Benchmark for Arabic Children’s Texts (ArabicNLP 2026).

This repository contains the original paper-era barectb.model checkpoint. It is the same checkpoint distributed in the paper repository's models/ directory. The later expanded 160K experiments use separate checkpoints.

Model

• Encoder: CAMeLBERT-MSA, as recorded in the checkpoint's training configuration.
• Dependency parser: BERT-encoded biaffine parser implemented with SuPar and used through CAMeL Parser.
• Training condition: BARECTB.
• Annotation: CATiB dependency relations.
• Checkpoint format: native SuPar .model; load with CAMeL Parser/SuPar.

The original barectb.model.yaml accompanies the checkpoint. Its historical training paths are preserved as provenance; inference uses the downloaded checkpoint.

Files

File	Description
barectb.model	Original trained parser checkpoint.
barectb.model.yaml	Original training configuration.
SHA256SUMS	Checkpoint SHA-256 checksum.
CITATION.bib	Paper citation.

Inference with CAMeL Parser

Follow the installation instructions in the CAMeL Parser main branch, including its Python and CAMeL Tools data requirements. From that checkout, download this checkpoint into its models/ directory:

from huggingface_hub import hf_hub_download

hf_hub_download(
    repo_id="CAMeL-Lab/camelparser-barectb-BARECTB",
    filename="barectb.model",
    local_dir="models",
)

Then parse raw Arabic text:

python text_to_conll_cli.py -f text -m barectb.model -s "القطة نائمة."

Use -i input.txt for a file of sentences. The optional MSA-S31 morphology database requires the separate installation and access steps documented upstream.

Scope

This checkpoint supports the paper's supervised Arabic dependency parsing experiments. Use the matching original BARECTB splits when reproducing that benchmark. Training corpora and morphology resources are separate from this model repository.

Citation

Elmadani, Khalid N., Mohamed Adel, Muhammed AbuOdeh, Reem Faraj, Jamila El-Gizuli, Hanada Taha and Nizar Habash. BARECTB: A Dependency Treebank and Parsing Benchmark for Arabic Children’s Texts. In Proceedings of ArabicNLP 2026, Budapest, Hungary, 2026.

@inproceedings{elmadani-etal-2026-barectb,
  title = {{BARECTB}: A Dependency Treebank and Parsing Benchmark for Arabic Children's Texts},
  author = {Elmadani, Khalid N. and Adel, Mohamed and AbuOdeh, Muhammed and Faraj, Reem and El-Gizuli, Jamila and Taha, Hanada and Habash, Nizar},
  booktitle = {Proceedings of ArabicNLP 2026},
  address = {Budapest, Hungary},
  year = {2026}
}