Belebele-FTP

A first-token-prediction (FTP) reframing of facebook/belebele. Each example is a single text sequence ending in Answer: so a model can predict the answer as one token (A/B/C/D).

Format

Example (eng_Latn):

P: <passage>
Q: <question>
A. <choice 1>
B. <choice 2>
C. <choice 3>
D. <choice 4>
Answer: 

Schema: question_id: int, text: str, answer_label: str (one of A/B/C/D).

Reframing

ANSWER_LABELS = ["A", "B", "C", "D"]
RESPONSE_TEMPLATE = "Answer: "

def format_ftp_example(example, idx):
    passage = f"P: {example['flores_passage']}"
    question = f"Q: {example['question'].strip()}"
    choices_str = "\n".join(
        f"{ANSWER_LABELS[i]}. {example[f'mc_answer{i + 1}']}"
        for i in range(4)
    )
    answer_label = ANSWER_LABELS[int(example["correct_answer_num"]) - 1]
    return {
        "question_id": idx,
        "text": f"{passage}\n{question}\n{choices_str}\n{RESPONSE_TEMPLATE}",
        "answer_label": answer_label,
    }

Splits

test only (matches upstream — Belebele has no train/validation).

Citation

@inproceedings{bandarkar-etal-2024-belebele,
    title = "The Belebele Benchmark: a Parallel Reading Comprehension Dataset in 122 Language Variants",
    author = "Bandarkar, Lucas  and
      Liang, Davis  and
      Muller, Benjamin  and
      Artetxe, Mikel  and
      Shukla, Satya Narayan  and
      Husa, Donald  and
      Goyal, Naman  and
      Krishnan, Abhinandan  and
      Zettlemoyer, Luke  and
      Khabsa, Madian",
    booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = aug,
    year = "2024",
    address = "Bangkok, Thailand and virtual meeting",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.acl-long.44",
    pages = "749--775",
}

License & credits

All credit to the original authors of Belebele (Bandarkar et al., Meta AI). This release inherits the upstream license (CC-BY-SA-4.0) and adds no new annotations — only a reformatting of the existing data.