Darsh J Shah



I am a 5th year Phd candidate of Electrical Engineering and Computer Science at MIT.
I work on Natural Language Processing and am very fortunate to be advised by Prof. Regina Barzilay.

I graduated from IIT Bombay in 2016, with a BTech in Computer Science and Engineering.



Research

I am interested in modeling the generation of contrastive text.

Contradictions are pervasive and I study text generation problems which deal with contrastive data. This has been encapsulated in my work on - (i) Generating related work sections for Scientific Papers, where we need to deal with different relations between the past work and a new research idea; (ii) Generating consensual multi-document summaries which are cognisant of the degree of agreement/disagreement amongst all inputs and (iii) Generating fact-guided modifications in text, where an old compendium is modified using a factual claim which contradicts past information.

I am also interested in scaling machine learning models so that they generalize to many tasks and are applicable for all people.
Which has inspired my work in transfer learning, domain adaptation and de-biasing.



Publications

Generating Related Work for Scholarly Papers  
Darsh J Shah and Regina Barzilay. Preprint 2021
[ Preprint ]
 
Nutri-bullets Hybrid: Consensual Multi-document Summarization  
Darsh J Shah, Lili Yu, Tao Lei, Regina Barzilay. NAACL 2021
[ Paper ]
 
Nutri-bullets: Summarizing Health Studies by Composing Segments  
Darsh J Shah, Lili Yu, Tao Lei, Regina Barzilay. AAAI 2021
[ Paper ]
 
Are We Safe Yet? The Limitations of Distributional Features For Fake News Detection  
Tal Schuster, Roei Schuster, Darsh J Shah, Regina Barzilay. Computational Linguistics 2020
[ Paper ]
 
Automatic Fact-guided Sentence Modification  
Darsh J Shah*, Tal Schuster*, Regina Barzilay. AAAI 2020
[ Paper ] [media: MIT News, dailymail, msn, news18]
 
Capturing Greater Context for Question Generation  
Luu Anh Tuan*, Darsh J Shah*, Regina Barzilay. AAAI 2020
[ Paper ]
 
Towards debiasing fact verification models  
Tal Schuster*, Darsh J Shah*, S. Yeo, D. Filizzola, E. Santus, Regina Barzilay. EMNLP 2019
[ Paper ] [Forbes]
 
Robust Zero-Shot Cross-Domain Slot Filling with Example Values  
Darsh J Shah*, Raghav Gupta*, Amir A Fayazi, Dilek Hakkani-Tur. ACL 2019
[ Paper ]
 
Multi-Source Domain Adaptation with Mixture of Experts  
Jiang Guo, Darsh J Shah, Regina Barzilay. EMNLP 2018
[ Paper ]
 
Adversarial Domain Adaptation for Duplicate Question Detection  
Darsh J Shah, Tao Lei, A. Moschitti, S. Romeo, P. Nakov. EMNLP 2018
[ Paper ]

* Equal Contribution

Contact


darsh@csail.mit.edu