Lost in Multilinguality: Dissecting Cross-lingual Factual Inconsistency in Transformer Language Models
Paper
• 2504.04264 • Published
• 2
Dataset Preview
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed
Error code: DatasetGenerationError
Exception: ArrowInvalid
Message: JSON parse error: Column() changed from object to string in row 0
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 174, in _generate_tables
df = pandas_read_json(f)
^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 38, in pandas_read_json
return pd.read_json(path_or_buf, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/pandas/io/json/_json.py", line 815, in read_json
return json_reader.read()
^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/pandas/io/json/_json.py", line 1014, in read
obj = self._get_object_parser(self.data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/pandas/io/json/_json.py", line 1040, in _get_object_parser
obj = FrameParser(json, **kwargs).parse()
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/pandas/io/json/_json.py", line 1176, in parse
self._parse()
File "/usr/local/lib/python3.12/site-packages/pandas/io/json/_json.py", line 1391, in _parse
self.obj = DataFrame(
^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/pandas/core/frame.py", line 778, in __init__
mgr = dict_to_mgr(data, index, columns, dtype=dtype, copy=copy, typ=manager)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/pandas/core/internals/construction.py", line 503, in dict_to_mgr
return arrays_to_mgr(arrays, columns, index, dtype=dtype, typ=typ, consolidate=copy)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/pandas/core/internals/construction.py", line 114, in arrays_to_mgr
index = _extract_index(arrays)
^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/pandas/core/internals/construction.py", line 677, in _extract_index
raise ValueError("All arrays must be of the same length")
ValueError: All arrays must be of the same length
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 1815, in _prepare_split_single
for _, table in generator:
^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 177, in _generate_tables
raise e
File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 151, in _generate_tables
pa_table = paj.read_json(
^^^^^^^^^^^^^^
File "pyarrow/_json.pyx", line 342, in pyarrow._json.read_json
File "pyarrow/error.pxi", line 155, in pyarrow.lib.pyarrow_internal_check_status
File "pyarrow/error.pxi", line 92, in pyarrow.lib.check_status
pyarrow.lib.ArrowInvalid: JSON parse error: Column() changed from object to string in row 0
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1455, in compute_config_parquet_and_info_response
parquet_operations = convert_to_parquet(builder)
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1054, in convert_to_parquet
builder.download_and_prepare(
File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 894, in download_and_prepare
self._download_and_prepare(
File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 970, in _download_and_prepare
self._prepare_split(split_generator, **prepare_split_kwargs)
File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 1702, in _prepare_split
for job_id, done, content in self._prepare_split_single(
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 1858, in _prepare_split_single
raise DatasetGenerationError("An error occurred while generating the dataset") from e
datasets.exceptions.DatasetGenerationError: An error occurred while generating the datasetNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
name string | prompt_templates list | samples list | relation_id string |
|---|---|---|---|
applies_to_jurisdiction | [
"في أي بلد يكون لدى <subject> مصطلحًا قانونيًا؟ الجواب هو: <mask>.",
"في أي اختصاص تكون لدى <subject> مصطلح قانوني؟ الجواب هو: <mask>.",
"أين يتم تطبيق المصطلح القانوني <subject>؟ الجواب هو: <mask>.",
"أي كيان قانوني يستخدم <subject>؟ الجواب هو: <mask>.",
"في أي منطقة يعتبر <subject> مصطلحًا قانونيًا معترفً... | [
{
"subject": "رئيس الجمهورية الفرنسية",
"object": "فرنسا",
"object_candidates": "الصومال, سويسرا, رومانيا, سنغافورة, النرويج, كوبا, ليتوانيا, تايوان, موناكو, فرنسا",
"index": 5960,
"subject_en": "President of the French Republic",
"object_en": "France"
},
{
"subject": "رئيس وزراء كند... | P1001 |
capital | [
"أين يقع عاصمة <subject>? الجواب هو: <mask>",
"ما هي عاصمة <subject>? الجواب هو: <mask>",
"ما هي المدينة التي تعتبر عاصمة <subject>? الجواب هو: <mask>",
"ذكر اسم عاصمة <subject>. الجواب هو: <mask>",
"أين تحتل <subject> مكانة عاصمتها؟ الجواب هو: <mask>"
] | [
{
"subject": "أذربيجان",
"object": "باكو",
"object_candidates": "غوادالاخارا, كولمبو, برغن, سالونيك, كلكتا, مانيلا, لينكون, رالي، كارولاينا الشمالية, بابل, باكو",
"index": 6152,
"subject_en": "Azerbaijan",
"object_en": "Baku"
},
{
"subject": "نبراسكا",
"object": "لينكون",
"ob... | P36 |
capital_of | [
"ما هي الدولة أو الولاية التي <subject> هي عاصمتها؟ الجواب هو: <mask>",
"أين <subject> عاصمة؟ الجواب هو: <mask>",
"أي دولة أو ولاية لديها <subject> كعاصمة لها؟ الجواب هو: <mask>",
"أي مكان لديه <subject> كعاصمة؟ الجواب هو: <mask>",
"أين تكون <subject> خدمة كعاصمة؟ الجواب هو: <mask>"
] | [
{
"subject": "إدمونتون",
"object": "ألبرتا",
"object_candidates": "تشيكوسلوفاكيا, لاوس, كجرات, بافاريا, سراوق, بنغلاديش, تكساس, قطر, بروفنس, ألبرتا",
"index": 3750,
"subject_en": "Edmonton",
"object_en": "Alberta"
},
{
"subject": "أنقرة",
"object": "تركيا",
"object_candidates... | P1376 |
continent | ["أي قارة يقع <subject> فيها؟ الجواب: <mask>","ما هي قارة <subject>؟ ا(...TRUNCATED) | [{"subject":"تركيا","object":"آسيا","object_candidates":"أوروبا, الأمريكتان(...TRUNCATED) | P30 |
country_of_citizenship | ["أي دولة هي <subject> مواطنة منها? الجواب هو: <mask>","ما هو جنسي(...TRUNCATED) | [{"subject":"بوريس سباسكي","object":"فرنسا","object_candidates":"جمهورية أي(...TRUNCATED) | P27 |
developer | ["أي شركة هي مطور <subject>؟ الإجابة: <mask>","من قام بتطوير <subject(...TRUNCATED) | [{"subject":"ماك أوس","object":"أبل","object_candidates":"بوينغ, آي بي إم,(...TRUNCATED) | P178 |
field_of_work | ["في أي مجال يعمل <subject>? الإجابة هي: <mask>","في أي مجال من ال(...TRUNCATED) | [{"subject":"آلان تورنغ","object":"منطق","object_candidates":"علم وظائف الأ(...TRUNCATED) | P101 |
headquarters_location | ["أين يوجد مقر <subject>؟ الإجابة هي: <mask>","في أي مدينة يقع مق(...TRUNCATED) | [{"subject":"باريس سان جيرمان","object":"باريس","object_candidates":"بكين, ك(...TRUNCATED) | P159 |
instrument | ["ما الآلة الموسيقية التي يعزفها <subject>? الجواب هو: <mask>","أ(...TRUNCATED) | [{"subject":"نت كينغ كول","object":"بيانو","object_candidates":"كمان, قيثارة(...TRUNCATED) | P1303 |
language_of_work_or_name | ["ما هو اللغة الأصلية لـ <subject>؟ الجواب: <mask>.","بأي لغة كتب (...TRUNCATED) | [{"subject":"أحدب نوتردام","object":"الفرنسية","object_candidates":"الهولن(...TRUNCATED) | P407 |
End of preview.
YAML Metadata Warning: empty or missing yaml metadata in repo card
Check out the documentation for more information.
KLAR-CLC
This repository provides the dataset for:
💡 Introduction
We apply mechanistic interpretability methods to analyze cross-lingual inconsistencies in multilingual masked language models (MLMs). Our findings reveal that MLMs represent knowledge in a shared, language-independent space through most layers, transitioning to language-specific representations only in the final layers. Errors often occur during this transition, leading to incorrect predictions in the target language despite correct answers in others. These insights offer a lightweight and effective strategy for improving factual consistency across languages.
This repository hosts KLAR, the dataset we created for multilingual knowledge probing, covering 17 languages. The accompanying code is available here.
📙 Citation
If you found our work useful for your research, please cite it as follows:
@inproceedings{wang-etal-2025-lost-multilinguality,
title = "Lost in Multilinguality: Dissecting Cross-lingual Factual Inconsistency in Transformer Language Models",
author = {Wang, Mingyang and Adel, Heike and Lange, Lukas and Liu, Yihong and Nie, Ercong and Strötgen, Jannik and Schütze, Hinrich},
booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
month = jul,
year = "2025",
address = "Vienna, Austria",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.acl-long.253/",
}
- Downloads last month
- 35