metadata

license: apache-2.0
tags:
  - natural-language-understanding
language_creators:
  - expert-generated
  - machine-generated
multilinguality:
  - multilingual
pretty_name: Polyglot or Not? Fact-Completion Benchmark
size_categories:
  - 100K<n<1M
task_categories:
  - text-generation
  - fill-mask
  - text2text-generation
dataset_info:
  features:
    - name: dataset_id
      dtype: string
    - name: stem
      dtype: string
    - name: 'true'
      dtype: string
    - name: 'false'
      dtype: string
    - name: relation
      dtype: string
    - name: subject
      dtype: string
    - name: object
      dtype: string
  splits:
    - name: English
      num_bytes: 3474255
      num_examples: 26254
    - name: Spanish
      num_bytes: 3175733
      num_examples: 18786
    - name: French
      num_bytes: 3395566
      num_examples: 18395
    - name: Russian
      num_bytes: 659526
      num_examples: 3289
    - name: Portuguese
      num_bytes: 4158146
      num_examples: 22974
    - name: German
      num_bytes: 2611160
      num_examples: 16287
    - name: Italian
      num_bytes: 3709786
      num_examples: 20448
    - name: Ukrainian
      num_bytes: 1868358
      num_examples: 7918
    - name: Polish
      num_bytes: 1683647
      num_examples: 9484
    - name: Romanian
      num_bytes: 2846002
      num_examples: 17568
    - name: Czech
      num_bytes: 1631582
      num_examples: 9427
    - name: Bulgarian
      num_bytes: 4597410
      num_examples: 20577
    - name: Swedish
      num_bytes: 3226502
      num_examples: 21576
    - name: Serbian
      num_bytes: 1327674
      num_examples: 5426
    - name: Hungarian
      num_bytes: 865409
      num_examples: 4650
    - name: Croatian
      num_bytes: 1195097
      num_examples: 7358
    - name: Danish
      num_bytes: 3580458
      num_examples: 23365
    - name: Slovenian
      num_bytes: 1299653
      num_examples: 7873
    - name: Dutch
      num_bytes: 3732795
      num_examples: 22590
    - name: Catalan
      num_bytes: 3319466
      num_examples: 18898
  download_size: 27093258
  dataset_size: 52358225
language:
  - en
  - fr
  - es
  - de
  - uk
  - bg
  - ca
  - da
  - hr
  - hu
  - it
  - nl
  - pl
  - pt
  - ro
  - ru
  - sl
  - sr
  - sv
  - cs

Dataset Card

Homepage: https://bit.ly/ischool-berkeley-capstone
Repository: https://github.com/daniel-furman/Capstone
Point of Contact: daniel_furman@berkeley.edu

Dataset Summary

This is the dataset for Polyglot or Not?: Measuring Multilingual Encyclopedic Knowledge Retrieval from Foundation Language Models.

Test Description

Given a factual association such as The capital of France is Paris, we determine whether a model adequately "knows" this information with the following test:

Step 1: prompt the model to predict the likelihood of the token Paris following The Capital of France is
Step 2: prompt the model to predict the average likelihood of a set of false, counterfactual tokens following the same stem.

If the value from 1 is greater than the value from 2 we conclude that model adequately recalls that fact. Formally, this is an application of the Contrastive Knowledge Assessment proposed in [[1][bib]].

For every foundation model of interest (like LLaMA), we perform this assessment on a set of facts translated into 20 languages. All told, we score foundation models on 303k fact-completions (results).

We also score monolingual models (like GPT-2) on English-only fact-completion (results).

Languages

The dataset covers 20 languages, which use either the Latin or Cyrillic scripts: bg, ca, cs, da, de, en, es, fr, hr, hu, it, nl, pl, pt, ro, ru, sl, sr, sv, uk.

Data Splits

The dataset splits correspond to the 20 languages above.

Source Data

We sourced the English cut of the dataset from [1] and [2] and used the Google Translate API to produce the other 19 language cuts.

Licensing Information

The dataset is licensed under the Apache 2.0 license and may be used with the corresponding affordances without limit.

Citation Information

@misc{polyglot_or_not,
  author = {Daniel Furman and Tim Schott and Shreshta Bhat},
  title = {Polyglot or Not?: Measuring Multilingual Encyclopedic Knowledge Retrieval from Foundation Language Models},
  year = {2023}
  publisher = {GitHub},
  howpublished = {\url{https://github.com/daniel-furman/Capstone}},
}

Bibliography

[1] Dong, Qingxiu, Damai Dai, Yifan Song, Jingjing Xu, Zhifang Sui, and Lei Li. "Calibrating Factual Knowledge in Pretrained Language Models". In Findings of the Association for Computational Linguistics: EMNLP 2022. [arXiv:2210.03329][cka] (2022).

@misc{dong2022calibrating,
      doi = {10.48550/arXiv.2210.03329},
      title={Calibrating Factual Knowledge in Pretrained Language Models}, 
      author={Qingxiu Dong and Damai Dai and Yifan Song and Jingjing Xu and Zhifang Sui and Lei Li},
      year={2022},
      eprint={2210.03329},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

[2] Meng, Kevin, Arnab Sen Sharma, Alex Andonian, Yonatan Belinkov, and David Bau. "Mass Editing Memory in a Transformer." arXiv preprint [arXiv:2210.07229][memit] (2022).

@misc{meng2022massediting,
      doi = {10.48550/arXiv.2210.07229},
      title={Mass-Editing Memory in a Transformer}, 
      author={Kevin Meng and Arnab Sen Sharma and Alex Andonian and Yonatan Belinkov and David Bau},
      year={2022},
      eprint={2210.07229},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}