Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
Question AnsweringHung-yi Lee
李宏毅
Knowledge source question
answer
Question Answering
Who is the U.S. president?
Who should pay for the date, and why?
Is Trump older than than Obama?
• word• span in source• correct choice• paragraph
Module for Source Module for QuestionAttention
Module for Answer
Knowledge source question
answer
Question Answering
Who is the U.S. president?
Who should pay for the date, and why?
Is Trump older than than Obama?
• word• span in source• correct choice• paragraph
Module for Source Module for QuestionAttention
Module for Answer
Answer is simply a word
• bAbI• MNIST of QA
• 20 types of questions
• Synthesized
Whether a system can answer questions via chaining facts, simple induction, deduction, etc.
[Weston, et al., arXiv’15]
Answer is simply a word
Module for Source Module for QuestionAttention
Module for Answer
Knowledge source question
Classification problem!
John Mary Kitchen No Maybe
Multiple Choices
Module for SourceModule for Question
Knowledge source question
(Ignoring attention here for simplicity)
Module for Choice
Choice A
Module for Answer
Yes / No
Knowledge source question Choice A Choice B
Choice C Choice D
Span in Source / Extraction-based
• SQuAD [Rajpurkar, et al., EMNLP’16], DRCD [Shao, et al., arXiv’18]
Source w1 w2 w3 w4 w5 w6 w7
StartScore
EndScore
0.1 0.1 0.7 0.1 0.0 0.0 0.0
0.0 0.0 0.0 0.2 0.6 0.2 0.0
The answer is w3 w4 w5
Module for SourceModule for Question
AttentionKnowledge source question
0.7
0.0
0.1
0.2
0.1
0.6
0.1
0.2
0.0
0.0
Module for Answer
LSTM
start end
Before BERT:
After BERT:
• MS MARCO [Bajaj, et al., NIPS’16], DuReader [He, et al., ACL workshop’18]
Free Answer Generation
S-net[Tan, et al., AAAI’18]
並不是所有問題都有答案 ……
並不是所有問題都有答案 ……
Know what you don’t know
• SQuAD 2.0 [Rajpurkar, et al., ACL’18]
Know what you don’t know
Source w1 w2 w3 w4 w5 w6 w7
StartScore
EndScore
0.1 0.1 0.0 0.1 0.0 0.0 0.0
0.0 0.0 0.0 0.2 0.60.2 0.0
Null
0.7
0.0
This is used in the original BERT paper by considering [CLS] as Null.
[Devlin, et al., NAACL’19]
Knowledge source question
answer
Module for Source Module for QuestionAttention
Module for Answer Answerable?
Yes/No[Liu, et al., arXiv’18]
questionKnowledge source answer
Answer Verifier Yes/No
[Hu, et al., AAAI’19]
Knowledge source question
answer
Question Answering
Who is the U.S. president?
Who should pay for the date, and why?
Is Trump older than than Obama?
• word• span in source• correct choice• paragraph
Module for Source Module for QuestionAttention
Module for Answer
Internet
[Chen, et al., ACL‘17]
MS MARCO and DuReader also have several documents for each question.
Not all the documents are relevant
DrQA
V-Net [Wang, et al., ACL’18]
V-Net [Wang, et al., ACL’18]
Visual
• source: picture/video
• What is in the image?
• Are there any humans?
• What sport is being played?
• Who has the ball?
• How many players are in the image?
• Who are the teams?
• Is it raining?
source: http://visualqa.org/
A vector for each region
Audio
• TOEFL Listening Comprehension Test by Machine
Question: “ What is a possible origin of Venus’ clouds? ”
Audio Story:
Choices:
(A) gases released as a result of volcanic activity
(B) chemical reactions caused by high surface temperatures
(C) bursts of radio energy from the plane's surface
(D) strong winds that blow dust into the atmosphere
(The original story is 5 min long.)
Link: https://github.com/iamyuanchung/TOEFL-QA
[Tseng, et al., INTERSPEECH’16]
Audio – SQuAD Style
• Link: https://github.com/chiahsuan156/ODSQA
SPOKEN OPEN-DOMAIN QUESTION ANSWERING DATASET
https://github.com/chiahsuan156/ODSQA
Audio – SQuAD Style
• Link: https://github.com/chiahsuan156/ODSQA
SPOKEN OPEN-DOMAIN QUESTION ANSWERING DATASET
SOD QA
OPEN-DOMAIN SPOKEN QUESTION ANSWERING DATASET
ODS QA[Lee, et al., SLT’18]
https://github.com/chiahsuan156/ODSQA
Audio - Techniques Developed
Subword UnitsAdversarial Learning
[Lee, et al., INTERSPEECH’18] [Lee, et al., ICASSP’19]
Audio - Towards End-to-end
[Chuang, et al., arXiv’19]
Movie QA
[Tapaswi, et al., CVPR’16]
QACNN
[Liu, et al., ICCV workshop’18]
Knowledge source question
answer
Question Answering
Who is the U.S. president?
Who should pay for the date, and why?
Is Trump older than than Obama?
• word• span in source• correct choice• paragraph
Module for Source Module for QuestionAttention
Module for Answer
Types of Questions
Simple Question: Match & Extract
Complex Question: Reasoning
Dialogue QA
Difficulty
Types of Questions
Simple Question: Match & Extract
Complex Question: Reasoning
Dialogue QA
Simple Questions
SQuAD
Match & Extract
Module for Source Module for Question
Knowledge source question
merge the whole question into one vector
Attention
𝛼1 𝛼2 𝛼3 𝛼4 𝛼5
𝑛=1
5
𝛼𝑛𝑥𝑛
Query-to-context Attention
Module for Answer
Classifier?
Predicting time span?
Match
Extract
𝑥𝑛
Query-to-context Attention
End-to-end Memory Network[Sukhbaatar, et al., NIPS’15]
Module for Source Module for Question
Knowledge source question
Attention
𝑛=1
5
𝛼𝑛𝑥𝑛
Query-to-context Attention (v2)
Module for Answer
𝛼11 𝛼2
1 𝛼31 𝛼4
1 𝛼51
𝛼12 𝛼2
2 𝛼32 𝛼4
2 𝛼52
𝛼13 𝛼2
3 𝛼33 𝛼4
3 𝛼53
max
multiple vectors
Query-to-context Attention (v2)
[Xu, et al., CVPR’16]
Module for Source Module for Question
Knowledge source question
Context-to-query Attention
Attention
𝛼1 𝛼2 𝛼3
𝑛=1
3
𝛼𝑛𝑥𝑛
Module for Answer
RnetR-Net
Self-attention
[Wang, et al., ACL’17]
[Huang, et al., ICLR’18]
Self-attention
Bi-directional Attention Flow
• BiDAF
Context Context
Qu
ery
Qu
ery
[Seo, et al., ICLR’17]
Dynamic Coattention Networks
[Xiong, et al., ICLR’17]
New query
QANetBERT 家族之外的最後一道榮光
[Yu, et al., ICLR’18]
Do not use RNN
BERT
你想要的 BERT 裡面通通都有了!
Knowledge source question
start endAnswerModule
Context-to-query Query-to-context
Self-matching, self-boosting
Types of Questions
Simple Question: Match & Extract
Complex Question: Reasoning
Dialogue QA
Qangaroo[Welbl, et al., TACL’18]
Hoppot QA [Yang, et al., EMNLP’18]
懷抱惡意出題
DROP Discrete Reasoning Over the text in the Paragraph[Dua, et al., NAACL’19]
Multiple-hop
Match Extract Change
Match Extract
Match Extract
Change
……
(“frog”)
(“Brian”)
(“yellow”)
MemoryNetwork
q
……
……
Match
Extract
……
……
Match
Extract
Change what to match
Change what to match
How many hops do we need?
ReasoNet[Shen, et al., KDD’17]
Graph Neural Network
[Qiu, et al., ACL’19]
Graph neural network (part 1): https://youtu.be/eybCCtNKwzAGraph neural network (part 2): https://youtu.be/M9ht8vsVEw8
Graph Neural Network
[Qiu, et al., ACL’19]
[Shao, et al., arXiv’20]
Graph is not needed …
Types of Questions
Simple Question: Match & Extract
Complex Question: Reasoning
Dialogue QA
Dialogue QA
CoQA[Reddy, et al., TACL’19]
Dialogue QA
[Choi, et al., EMNLP’18]
QuAC
Dialogue QA
D Q1
D Q2
D Q3
QA Model
QA Model
QA Model
A1
A2
A3
……
……
Dialogue QA
D Q1
D Q2
D Q3
l-th layer (l+1)-th layer
RNN
RNN
[Huang, et al., ICLR’19]
Dialogue QA
D Q1
D Q2
D Q3
l-th layer (l+1)-th layer
0.1
0.3
0.6
Attention Weight
[Qu, et al., CIKM’19]
Problem Solved?
https://rajpurkar.github.io/SQuAD-explorer/
Train on SQuAD, Test on bAbI
感謝吳侑學同學提供實驗結果
[Kaushik, et al., EMNLP’18]
[Feng, et al., EMNLP’18]
機器居然料事如神!
Human Performance
機器超越人類?
VQA [Agrawal, et al., CVPR’18]
VQA
(question type only)
(question only)
[Agrawal, et al., CVPR’18]
There is still a long way to go.
Reference
• Jason Weston, Antoine Bordes, Sumit Chopra, Alexander M. Rush, Bart van Merriënboer, Armand Joulin and Tomas Mikolov. Towards AI Complete Question Answering: A Set of Prerequisite Toy Tasks, arXiv, 2015
• Pranav Rajpurkar, Jian Zhang, Konstantin Lopyrev, Percy Liang, SQuAD: 100,000+ Questions for Machine Comprehension of Text, EMNLP, 2016
• Payal Bajaj, Daniel Campos, Nick Craswell, Li Deng, Jianfeng Gao, Xiaodong Liu, Rangan Majumder, Andrew McNamara, Bhaskar Mitra, Tri Nguyen, Mir Rosenberg, Xia Song, Alina Stoica, Saurabh Tiwary, and Tong Wang, MS MARCO: A Human Generated MAchine Reading COmprehension Dataset, NIPS, 2016
• Chih Chieh Shao and Trois Liu and Yuting Lai and Yiying Tseng and Sam Tsai, DRCD: a Chinese Machine Reading Comprehension Dataset, arXiv, 2018
• Wei He, Kai Liu, Jing Liu, Yajuan Lyu, Shiqi Zhao, Xinyan Xiao, Yuan Liu, YizhongWang, Hua Wu, Qiaoqiao She, Xuan Liu, Tian Wu, Haifeng Wang, DuReader: a Chinese Machine Reading Comprehension Dataset from Real-world Applications, ACL workshop, 2018
• Chuanqi Tan, Furu Wei, Nan Yang, Bowen Du, Weifeng Lv, Ming Zhou, S-Net: From Answer Extraction to Answer Generation for Machine Reading Comprehension, AAAI, 2018
http://arxiv.org/abs/1502.05698https://arxiv.org/search/cs?searchtype=author&query=Zhang%2C+Jhttps://arxiv.org/search/cs?searchtype=author&query=Lopyrev%2C+Khttps://arxiv.org/search/cs?searchtype=author&query=Liang%2C+Phttps://arxiv.org/search/cs?searchtype=author&query=He%2C+Whttps://arxiv.org/search/cs?searchtype=author&query=Liu%2C+Khttps://arxiv.org/search/cs?searchtype=author&query=Liu%2C+Jhttps://arxiv.org/search/cs?searchtype=author&query=Lyu%2C+Yhttps://arxiv.org/search/cs?searchtype=author&query=Zhao%2C+Shttps://arxiv.org/search/cs?searchtype=author&query=Xiao%2C+Xhttps://arxiv.org/search/cs?searchtype=author&query=Liu%2C+Yhttps://arxiv.org/search/cs?searchtype=author&query=Wang%2C+Yhttps://arxiv.org/search/cs?searchtype=author&query=Wu%2C+Hhttps://arxiv.org/search/cs?searchtype=author&query=She%2C+Qhttps://arxiv.org/search/cs?searchtype=author&query=Liu%2C+Xhttps://arxiv.org/search/cs?searchtype=author&query=Wu%2C+Thttps://arxiv.org/search/cs?searchtype=author&query=Wang%2C+Hhttps://arxiv.org/search/cs?searchtype=author&query=Tan%2C+Chttps://arxiv.org/search/cs?searchtype=author&query=Wei%2C+Fhttps://arxiv.org/search/cs?searchtype=author&query=Yang%2C+Nhttps://arxiv.org/search/cs?searchtype=author&query=Du%2C+Bhttps://arxiv.org/search/cs?searchtype=author&query=Lv%2C+Whttps://arxiv.org/search/cs?searchtype=author&query=Zhou%2C+M
Reference
• [Chen, et al., ACL‘17] Danqi Chen, Adam Fisch, Jason Weston, Antoine Bordes, Reading Wikipedia to Answer Open-Domain Questions, ACL, 2017
• Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William Cohen, Ruslan Salakhutdinov, Christopher D. Manning, HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering, EMNLP, 2018
• Pranav Rajpurkar, Robin Jia, Percy Liang, Know What You Don't Know: Unanswerable Questions for SQuAD, ACL, 2018
• Siva Reddy, Danqi Chen, Christopher D. Manning, CoQA: A Conversational Question Answering Challenge, TACL, 2019
• [Choi, et al., EMNLP’18] Eunsol Choi, He He, Mohit Iyyer, Mark Yatskar, Wen-tau Yih, Yejin Choi, Percy Liang, Luke Zettlemoyer, QuAC : Question Answering in Context, EMNLP, 2018
• [Qiu, et al., ACL’19] Lin Qiu, Yunxuan Xiao, Yanru Qu, Hao Zhou, Lei Li, WeinanZhang, Yong Yu, Dynamically Fused Graph Network for Multi-hop Reasoning, ACL, 2019
• [Shao, et al., arXiv’20] Nan Shao, Yiming Cui, Ting Liu, Shijin Wang, Guoping Hu, Is Graph Structure Necessary for Multi-hop Reasoning?, arXiv, 2020
https://arxiv.org/search/cs?searchtype=author&query=Fisch%2C+Ahttps://arxiv.org/search/cs?searchtype=author&query=Weston%2C+Jhttps://arxiv.org/search/cs?searchtype=author&query=Bordes%2C+Ahttps://www.aclweb.org/anthology/people/z/zhilin-yang/https://www.aclweb.org/anthology/people/p/peng-qi/https://www.aclweb.org/anthology/people/s/saizheng-zhang/https://www.aclweb.org/anthology/people/y/yoshua-bengio/https://www.aclweb.org/anthology/people/w/william-cohen/https://www.aclweb.org/anthology/people/r/ruslan-salakhutdinov/https://www.aclweb.org/anthology/people/c/christopher-d-manning/https://www.aclweb.org/anthology/D18-1259.pdfhttps://arxiv.org/search/cs?searchtype=author&query=Jia%2C+Rhttps://arxiv.org/search/cs?searchtype=author&query=Liang%2C+Phttps://www.aclweb.org/anthology/people/d/danqi-chen/https://www.aclweb.org/anthology/people/c/christopher-d-manning/https://www.aclweb.org/anthology/Q19-1016.pdfhttps://arxiv.org/search/cs?searchtype=author&query=He%2C+Hhttps://arxiv.org/search/cs?searchtype=author&query=Iyyer%2C+Mhttps://arxiv.org/search/cs?searchtype=author&query=Yatskar%2C+Mhttps://arxiv.org/search/cs?searchtype=author&query=Yih%2C+Whttps://arxiv.org/search/cs?searchtype=author&query=Choi%2C+Yhttps://arxiv.org/search/cs?searchtype=author&query=Liang%2C+Phttps://arxiv.org/search/cs?searchtype=author&query=Zettlemoyer%2C+Lhttps://www.aclweb.org/anthology/people/y/yunxuan-xiao/https://www.aclweb.org/anthology/people/y/yanru-qu/https://www.aclweb.org/anthology/people/h/hao-zhou/https://www.aclweb.org/anthology/people/l/lei-li/https://www.aclweb.org/anthology/people/w/weinan-zhang/https://www.aclweb.org/anthology/people/y/yong-yu/https://www.aclweb.org/anthology/P19-1617.pdfhttps://arxiv.org/search/cs?searchtype=author&query=Cui%2C+Yhttps://arxiv.org/search/cs?searchtype=author&query=Liu%2C+Thttps://arxiv.org/search/cs?searchtype=author&query=Wang%2C+Shttps://arxiv.org/search/cs?searchtype=author&query=Hu%2C+G
Reference
• Yizhong Wang, Kai Liu, Jing Liu, Wei He, Yajuan Lyu, Hua Wu, Sujian Li, Haifeng Wang, Multi-Passage Machine Reading Comprehension with Cross-Passage Answer Verification, ACL, 2018
• Bo-Hsiang Tseng, Sheng-syun Shen, Hung-Yi Lee, Lin-Shan Lee, Towards Machine Comprehension of Spoken Content: Initial TOEFL Listening Comprehension Test by Machine, INTERSPEECH, 2016
• Chia-Hsuan Lee, Szu-Lin Wu, Chi-Liang Liu, Hung-yi Lee, Spoken SQuAD: A Study of Mitigating the Impact of Speech Recognition Errors on Listening Comprehension, INTERSPEECH, 2018
• Chia-Hsuan Lee, Yun-Nung Chen, Hung-Yi Lee, "Mitigating the Impact of Speech Recognition Errors on Spoken Question Answering by Adversarial Domain Adaptation", ICASSP, 2019
• Chia-Hsuan Lee, Shang-Ming Wang, Huan-Cheng Chang, Hung-Yi Lee, ODSQA: Open-domain Spoken Question Answering Dataset, SLT, 2018
• Yung-Sung Chuang, Chi-Liang Liu, Hung-Yi Lee, Lin-shan Lee, SpeechBERT: An Audio-and-text Jointly Learned Language Model for End-to-end Spoken Question Answering, arXiv, 2019
• Makarand Tapaswi, Yukun Zhu, Rainer Stiefelhagen, Antonio Torralba, Raquel
https://www.aclweb.org/anthology/people/k/kai-liu/https://www.aclweb.org/anthology/people/j/jing-liu/https://www.aclweb.org/anthology/people/w/wei-he/https://www.aclweb.org/anthology/people/y/yajuan-lyu/https://www.aclweb.org/anthology/people/h/hua-wu/https://www.aclweb.org/anthology/people/s/sujian-li/https://www.aclweb.org/anthology/people/h/haifeng-wang/https://www.aclweb.org/anthology/P18-1178.pdf
Reference
• Tzu-Chien Liu, Yu-Hsueh Wu, Hung-Yi Lee, Query-based Attention CNN for Text Similarity Map, ICCV workshop, 2018
• Hsin-Yuan Huang, Eunsol Choi, Wen-tau Yih, FlowQA: Grasping Flow in History for Conversational Machine Comprehension, ICLR, 2019
• Xiaodong Liu, Wei Li, Yuwei Fang, Aerin Kim, Kevin Duh, Jianfeng Gao, Stochastic Answer Networks for SQuAD 2.0, arXiv, 2018
• Jacob Devlin, Ming-Wei Chang, Kenton Lee, Kristina Toutanova, BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding, NAACL, 2019
• Minghao Hu, Furu Wei, Yuxing Peng, Zhen Huang, Nan Yang, Dongsheng Li, Read + Verify: Machine Reading Comprehension with Unanswerable Questions, AAAI, 2019
• Dheeru Dua, Yizhong Wang, Pradeep Dasigi, Gabriel Stanovsky, Sameer Singh, Matt Gardner, DROP: A Reading Comprehension Benchmark Requiring Discrete Reasoning Over Paragraphs, NAACL, 2019
• Johannes Welbl, Pontus Stenetorp, Sebastian Riedel, Constructing Datasets for Multi-hop Reading Comprehension Across Documents, TACL, 2018
https://openreview.net/profile?email=eunsol%40cs.washington.eduhttps://openreview.net/profile?email=scottyih%40allenai.orghttps://arxiv.org/search/cs?searchtype=author&query=Li%2C+Whttps://arxiv.org/search/cs?searchtype=author&query=Fang%2C+Yhttps://arxiv.org/search/cs?searchtype=author&query=Kim%2C+Ahttps://arxiv.org/search/cs?searchtype=author&query=Duh%2C+Khttps://arxiv.org/search/cs?searchtype=author&query=Gao%2C+Jhttps://arxiv.org/search/cs?searchtype=author&query=Wei%2C+Fhttps://arxiv.org/search/cs?searchtype=author&query=Peng%2C+Yhttps://arxiv.org/search/cs?searchtype=author&query=Huang%2C+Zhttps://arxiv.org/search/cs?searchtype=author&query=Yang%2C+Nhttps://arxiv.org/search/cs?searchtype=author&query=Li%2C+Dhttps://arxiv.org/search/cs?searchtype=author&query=Dua%2C+Dhttps://arxiv.org/search/cs?searchtype=author&query=Wang%2C+Yhttps://arxiv.org/search/cs?searchtype=author&query=Dasigi%2C+Phttps://arxiv.org/search/cs?searchtype=author&query=Stanovsky%2C+Ghttps://arxiv.org/search/cs?searchtype=author&query=Singh%2C+Shttps://arxiv.org/search/cs?searchtype=author&query=Gardner%2C+Mhttps://arxiv.org/search/cs?searchtype=author&query=Stenetorp%2C+Phttps://arxiv.org/search/cs?searchtype=author&query=Riedel%2C+S
Reference
• Minjoon Seo, Aniruddha Kembhavi, Ali Farhadi, Hannaneh Hajishirzi, Bidirectional Attention Flow for Machine Comprehension, ICLR, 2017
• Adams Wei Yu, David Dohan, Minh-Thang Luong, Rui Zhao, Kai Chen, Mohammad Norouzi, Quoc V. Le, QANet: Combining Local Convolution with Global Self-Attention for Reading Comprehension, ICLR, 2018
• Hsin-Yuan Huang, Chenguang Zhu, Yelong Shen, Weizhu Chen, FusionNet: Fusing via Fully-aware Attention with Application to Machine Comprehension, ICLR, 2018
• Huijuan Xu, Kate Saenko. Ask, Attend and Answer: Exploring Question-Guided Spatial Attention for Visual Question Answering, CVPR, 2016
• Sainbayar Sukhbaatar, Arthur Szlam, Jason Weston, Rob Fergus, End-To-End Memory Networks, NIPS, 2015
• Caiming Xiong, Victor Zhong, Richard Socher, Dynamic Coattention Networks For Question Answering, ICLR, 2017
• Wenhui Wang, Nan Yang, Furu Wei, Baobao Chang, Ming Zhou, Gated Self-Matching Networks for Reading Comprehension and Question Answering, ACL, 2017
https://arxiv.org/search/cs?searchtype=author&query=Kembhavi%2C+Ahttps://arxiv.org/search/cs?searchtype=author&query=Farhadi%2C+Ahttps://arxiv.org/search/cs?searchtype=author&query=Hajishirzi%2C+Hhttps://arxiv.org/search/cs?searchtype=author&query=Yu%2C+A+Whttps://arxiv.org/search/cs?searchtype=author&query=Dohan%2C+Dhttps://arxiv.org/search/cs?searchtype=author&query=Luong%2C+Mhttps://arxiv.org/search/cs?searchtype=author&query=Zhao%2C+Rhttps://arxiv.org/search/cs?searchtype=author&query=Chen%2C+Khttps://arxiv.org/search/cs?searchtype=author&query=Norouzi%2C+Mhttps://arxiv.org/search/cs?searchtype=author&query=Le%2C+Q+Vhttps://openreview.net/profile?email=momohuang%40gmail.comhttps://openreview.net/profile?email=chezhu%40microsoft.comhttps://openreview.net/profile?email=yeshen%40microsoft.comhttps://openreview.net/profile?email=wzchen%40microsoft.comhttps://arxiv.org/search/cs?searchtype=author&query=Zhong%2C+Vhttps://arxiv.org/search/cs?searchtype=author&query=Socher%2C+Rhttps://www.aclweb.org/anthology/people/w/wenhui-wang/https://www.aclweb.org/anthology/people/n/nan-yang/https://www.aclweb.org/anthology/people/f/furu-wei/https://www.aclweb.org/anthology/people/b/baobao-chang/https://www.aclweb.org/anthology/people/m/ming-zhou/https://www.aclweb.org/anthology/P17-1018.pdf
Reference
• Chen Qu, Liu Yang, Minghui Qiu, Yongfeng Zhang, Cen Chen, W. Bruce Croft, Mohit Iyyer, Attentive History Selection for Conversational Question Answering, CIKM, 2019
• Aishwarya Agrawal, Dhruv Batra, Devi Parikh, Aniruddha Kembhavi, Don't Just Assume; Look and Answer: Overcoming Priors for Visual Question Answering, CVPR, 2018
• Shi Feng, Eric Wallace, Alvin Grissom II, Mohit Iyyer, Pedro Rodriguez, Jordan Boyd-Graber, Pathologies of Neural Models Make Interpretations Difficult, EMNLP, 2018
• Divyansh Kaushik, Zachary C. Lipton, How Much Reading Does Reading Comprehension Require? A Critical Investigation of Popular Benchmarks, EMNLP, 2018
• Yelong Shen, Po-Sen Huang, Jianfeng Gao, Weizhu Chen, ReasoNet: Learning to Stop Reading in Machine Comprehension, KDD, 2017
https://arxiv.org/search/cs?searchtype=author&query=Qu%2C+Chttps://arxiv.org/search/cs?searchtype=author&query=Yang%2C+Lhttps://arxiv.org/search/cs?searchtype=author&query=Qiu%2C+Mhttps://arxiv.org/search/cs?searchtype=author&query=Zhang%2C+Yhttps://arxiv.org/search/cs?searchtype=author&query=Chen%2C+Chttps://arxiv.org/search/cs?searchtype=author&query=Croft%2C+W+Bhttps://arxiv.org/search/cs?searchtype=author&query=Iyyer%2C+Mhttps://arxiv.org/search/cs?searchtype=author&query=Batra%2C+Dhttps://arxiv.org/search/cs?searchtype=author&query=Parikh%2C+Dhttps://arxiv.org/search/cs?searchtype=author&query=Kembhavi%2C+Ahttps://www.aclweb.org/anthology/people/e/eric-wallace/https://www.aclweb.org/anthology/people/a/alvin-grissom-ii/https://www.aclweb.org/anthology/people/m/mohit-iyyer/https://www.aclweb.org/anthology/people/p/pedro-rodriguez/https://www.aclweb.org/anthology/people/j/jordan-boyd-graber/https://www.aclweb.org/anthology/D18-1407.pdfhttps://arxiv.org/search/cs?searchtype=author&query=Huang%2C+Phttps://arxiv.org/search/cs?searchtype=author&query=Gao%2C+Jhttps://arxiv.org/search/cs?searchtype=author&query=Chen%2C+W