Publications
For the most up-to-date list of publications, please visit my Google Scholar profile page
2023
The Whole Truth and Nothing But the Truth: Faithful and Controllable Dialogue Response Generation with Dataflow Transduction and Constrained Decoding. Findings of ACL 2023
Hao Fang*, Anusha Balakrishnan*, Harsh Jhamtani*, John Bufe, Jean Crawford, Jayant Krishnamurthy, Adam Pauls, Jason Eisner, Jacob Andreas, Dan Klein
Ontologically Faithful Generation of Non-Player Character Dialogues. DialDoc workshop (non-archival) at ACL 2023
Nathaniel Weir, Ryan Thomas, Randolph D'Amore, Kellie Hill, Benjamin Van Durme, Harsh Jhamtani
[Paper] [Bibtex]
2022
Target-Guided Dialogue Response Generation Using Commonsense and Data Augmentation. NAACL 2022 (Findings)
Prakhar Gupta, Harsh Jhamtani, Jeffrey Bigham
[Paper] [Bibtex]
2021
Truth-Conditional Captions for Time Series Data. EMNLP 2021
Harsh Jhamtani, Taylor Berg-Kirkpatrick
[Slides][EMNLP-talk] [EMNLP-poster]
Investigating Robustness of Dialog Models to Popular Figurative Language Constructs. EMNLP 2021. (Short Paper)
Harsh Jhamtani *, Varun Gangal *, Eduard Hovy, Taylor Berg-Kirkpatrick
[Paper] [Bibtex] [Code and Data]
[Slides] [EMNLP-Talk] [EMNLP-poster]
News coverage: [UCSD][The-Register][News8-plus][Newswise]
Improving Automated Evaluation of Open Domain Dialog via Diverse Reference Augmentation. Findings of ACL 2021. (Short Paper)
Varun Gangal *, Harsh Jhamtani *, Eduard Hovy, Taylor Berg-Kirkpatrick
[Paper] [Bibtex] [Code & Data]
[ACL-talk] [ACL-poster] [ACL-slides]
2020
Learning to Explain: Datasets and Models for Identifying Valid Reasoning Chains in Multihop Question-Answering. EMNLP 2020 . Harsh Jhamtani, Peter Clark
[Paper] [Bibtex] [Code] [Presentation] [Data] [Leaderboard][EMNLP-talk]
TLDR: 1) Prior work focuses mostly on QA results for multihop, or work on explanations for questions which can easily decompose 2) We propose new datasets for *reasoning chain* explanations for multihop question answering. 3) A simple lightweight procedure to delexicalize overlapping entities (leading to chains such as ”X is a Y” AND ”Y has Z” IMPLIES ”X has Z”) maintains performance on explanation prediction (even after losing typing info) , is more robust to certain perturbations, and the top scoring chains demonstrate interesting generally applicable reasoning patterns.
Narrative Text Generation with a Latent Discrete Plan. Findings of EMNLP 2020
Harsh Jhamtani, Taylor Berg-Kirkpatrick
[Code] [Presentation]
TLDR: 1) A discrete plan such as a sequence of keywords can lead to better coherence in generated text -- prior work approaches this with first tagging the text with 'plans' using off-the-shelf extractors and employing supervised learning. We consider the plan as latent, and train the models using variational learning. 2) An inference network constrained to use tokens from corresponding sentence only leads to much better models. A constrained non-monotonic decoder, designed to always use corresponding anchor word from sampled plan, provides much better control via plan, but at cost of some coherence.
Like hiking? You probably enjoy nature: Persona-grounded Dialog with Commonsense Expansions. EMNLP 2020
Bodhisattwa Majumder, Harsh Jhamtani, Taylor Berg-Kirkpatrick and Julian McAuley
[Code] [Presentation]
TLDR: 1) Contemporary Persona-grounded dialog models often struggle to learn simple implications of given persona 2) We propose to use common sense knowledge base such as Comet to expand the given persona. We use a discrete latent variable model which performs fine-grained selection of persona attribute rather than encoding the entire persona. 3) Leads to better perplexity, favorable human evaluation ratings, and provides controllable & interpretable generation
Domain Adaptation via Context Prediction for Engineering Diagram Search. ECIR (Short paper) 2020
Harsh Jhamtani, Taylor Berg-Kirkpatrick
2019
Learning Rhyming Constraints using Structured Adversaries. EMNLP-IJCNLP (Short paper) 2019
Harsh Jhamtani, Sanket Vaibhav Mehta, Jaime Carbonell, Taylor Berg-Kirkpatrick
[Poster]
[Slides from seminar talk at UCSD (November 2019)]
A Sociolinguistic Study of Online Echo Chambers on Twitter. NLP-CSS Workshop. NAACL 2019
Nikita Duseja*, Harsh Jhamtani*
[Poster]
2018
Learning to Describe Differences Between Pairs of Similar Images. EMNLP 2018
Harsh Jhamtani, Taylor Berg-Kirkpatrick.
[Paper]
[Bibtex]
[Poster]
[Slides from Speaking Requirement Talk at LTI, CMU (July 2018; Last updated: September 2018) ]
Learning to Generate Move-by-Move Commentary for Chess Games from Large-Scale Social Forum Data. ACL 2018
Harsh Jhamtani*, Varun Gangal*, Eduard Hovy, Graham Neubig and Taylor Berg-Kirkpatrick.
[poster]
SPINE - Sparse Interpretable Neural Embeddings. AAAI 2018.
Harsh Jhamtani *, Anant Subramanian*, Danish Pruthi*, Taylor Berg-Kirkpatrick, Eduard Hovy
[Paper] [Bibtex]
[Slides from talk at Adobe Research Bangalore] (January 2018)
2017
Charmanteau: Character Embedding Models For Portmanteau Creation. EMNLP 2017 (Short Paper).
Varun Gangal*, Harsh Jhamtani *, Graham Neubig, Eduard Hovy, Eric Nyberg.
[Code and Data] [Demo]
[Slides] [EMNLP Talk]
Portmanteau
Shakespearizing Modern Language Using Copy-Enriched Sequence to Sequence Models. Workshop on Stylistic Variation at EMNLP 2017
Harsh Jhamtani *, Varun Gangal*, Eduard Hovy, Eric Nyberg.
[Slides]
Leveraging Site Search Logs to Identify Missing Content on Enterprise Webpages. (Short paper) ECIR 2017.
Harsh Jhamtani, Rishiraj Saha Roy, Niyati Chhaya, Eric Nyberg
[Poster]
Generating Appealing Brand Name. CICLING 2017.
Gaurush Hiranandani., Pranav Manerikar, Harsh Jhamtani
Moumita Sinha, Harsh Jhamtani, Sanket Mehta, Balaji Vasan Srinivasan. Modeling End-of-Online-Session From Streaming Data. IKDD 2017. [Bibtex]
2016
Modani N., Srinivasan BV., Jhamtani H. Generating Multiple Diverse Summaries. WISE 2016. [Paper] [Bibtex]
Jhamtani, H., Varma, S., Gundapuneni M., Dutta S. A Supervised Approach for Text Illustration. ACM MM 2016. [Paper] [Bibtex]
2015
Jhamtani, H., Chhaya N., Karwa, S., Varshney, D., Kedia, D., Gupta, V.: Identifying Suggestions for Improvement of Product Features from Online Product Reviews. SocInfo-2015. [Paper] [Bibtex]
2014
Jhamtani, H., Bhogi, S., & Raychoudhury V. Word-level Language Identification in Bi-lingual Code-switched Texts. PACLIC-2014. [Paper] [Presentation] [Bibtex]
Gupta, V., Varshney, D., Jhamtani, H., Kedia, D., & Karwa, S. Identifying Purchase Intent from Social Posts. ICWSM-2014. [Paper] [Bibtex]