@inproceedings{albilali-etal-2021-bert,
title = "What does {BERT} Learn from {A}rabic Machine Reading Comprehension Datasets?",
author = "Albilali, Eman and
Altwairesh, Nora and
Hosny, Manar",
editor = "Habash, Nizar and
Bouamor, Houda and
Hajj, Hazem and
Magdy, Walid and
Zaghouani, Wajdi and
Bougares, Fethi and
Tomeh, Nadi and
Abu Farha, Ibrahim and
Touileb, Samia",
booktitle = "Proceedings of the Sixth Arabic Natural Language Processing Workshop",
month = apr,
year = "2021",
address = "Kyiv, Ukraine (Virtual)",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2021.wanlp-1.4",
pages = "32--41",
abstract = "In machine reading comprehension tasks, a model must extract an answer from the available context given a question and a passage. Recently, transformer-based pre-trained language models have achieved state-of-the-art performance in several natural language processing tasks. However, it is unclear whether such performance reflects true language understanding. In this paper, we propose adversarial examples to probe an Arabic pre-trained language model (AraBERT), leading to a significant performance drop over four Arabic machine reading comprehension datasets. We present a layer-wise analysis for the transformer{'}s hidden states to offer insights into how AraBERT reasons to derive an answer. The experiments indicate that AraBERT relies on superficial cues and keyword matching rather than text understanding. Furthermore, hidden state visualization demonstrates that prediction errors can be recognized from vector representations in earlier layers.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="albilali-etal-2021-bert">
<titleInfo>
<title>What does BERT Learn from Arabic Machine Reading Comprehension Datasets?</title>
</titleInfo>
<name type="personal">
<namePart type="given">Eman</namePart>
<namePart type="family">Albilali</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nora</namePart>
<namePart type="family">Altwairesh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Manar</namePart>
<namePart type="family">Hosny</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2021-04</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Sixth Arabic Natural Language Processing Workshop</title>
</titleInfo>
<name type="personal">
<namePart type="given">Nizar</namePart>
<namePart type="family">Habash</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Houda</namePart>
<namePart type="family">Bouamor</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hazem</namePart>
<namePart type="family">Hajj</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Walid</namePart>
<namePart type="family">Magdy</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wajdi</namePart>
<namePart type="family">Zaghouani</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fethi</namePart>
<namePart type="family">Bougares</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nadi</namePart>
<namePart type="family">Tomeh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ibrahim</namePart>
<namePart type="family">Abu Farha</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Samia</namePart>
<namePart type="family">Touileb</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Kyiv, Ukraine (Virtual)</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>In machine reading comprehension tasks, a model must extract an answer from the available context given a question and a passage. Recently, transformer-based pre-trained language models have achieved state-of-the-art performance in several natural language processing tasks. However, it is unclear whether such performance reflects true language understanding. In this paper, we propose adversarial examples to probe an Arabic pre-trained language model (AraBERT), leading to a significant performance drop over four Arabic machine reading comprehension datasets. We present a layer-wise analysis for the transformer’s hidden states to offer insights into how AraBERT reasons to derive an answer. The experiments indicate that AraBERT relies on superficial cues and keyword matching rather than text understanding. Furthermore, hidden state visualization demonstrates that prediction errors can be recognized from vector representations in earlier layers.</abstract>
<identifier type="citekey">albilali-etal-2021-bert</identifier>
<location>
<url>https://aclanthology.org/2021.wanlp-1.4</url>
</location>
<part>
<date>2021-04</date>
<extent unit="page">
<start>32</start>
<end>41</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T What does BERT Learn from Arabic Machine Reading Comprehension Datasets?
%A Albilali, Eman
%A Altwairesh, Nora
%A Hosny, Manar
%Y Habash, Nizar
%Y Bouamor, Houda
%Y Hajj, Hazem
%Y Magdy, Walid
%Y Zaghouani, Wajdi
%Y Bougares, Fethi
%Y Tomeh, Nadi
%Y Abu Farha, Ibrahim
%Y Touileb, Samia
%S Proceedings of the Sixth Arabic Natural Language Processing Workshop
%D 2021
%8 April
%I Association for Computational Linguistics
%C Kyiv, Ukraine (Virtual)
%F albilali-etal-2021-bert
%X In machine reading comprehension tasks, a model must extract an answer from the available context given a question and a passage. Recently, transformer-based pre-trained language models have achieved state-of-the-art performance in several natural language processing tasks. However, it is unclear whether such performance reflects true language understanding. In this paper, we propose adversarial examples to probe an Arabic pre-trained language model (AraBERT), leading to a significant performance drop over four Arabic machine reading comprehension datasets. We present a layer-wise analysis for the transformer’s hidden states to offer insights into how AraBERT reasons to derive an answer. The experiments indicate that AraBERT relies on superficial cues and keyword matching rather than text understanding. Furthermore, hidden state visualization demonstrates that prediction errors can be recognized from vector representations in earlier layers.
%U https://aclanthology.org/2021.wanlp-1.4
%P 32-41
Markdown (Informal)
[What does BERT Learn from Arabic Machine Reading Comprehension Datasets?](https://aclanthology.org/2021.wanlp-1.4) (Albilali et al., WANLP 2021)
ACL