Avatar

Ben Bogin

Post-doc

AI2/UW

Hi! I’m Ben, a post-doc at AI2 and the University of Washington. I have pursued my PhD at Tel Aviv University, under the guidance of Prof. Jonathan Berant, where I studied compositional generalization in semantic parsing and visual question answering.

These days I’m working on improving our understanding of the connection between pre-training data and generalization abilities of LLMs, and on building systems that can automatically and interactively generate code for software engineering. I’m always open to new collaborations, feel free to reach out!

Publications

Leveraging Code to Improve In-context Learning for Semantic Parsing (2023). PDF
Answering Questions by Meta-Reasoning over Multiple Chains of Thought (EMNLP, 2023). PDF
Diverse Demonstrations Improve In-context Compositional Generalization (ACL, 2023). PDF
Unobserved Local Structures Make Compositional Generalization Hard (EMNLP, 2022). PDF
COVR: A Test-Bed for Visually Grounded Compositional Generalization with Real Images (EMNLP, 2021). PDF Code Project
Text-to-SQL in the Wild: A Naturally-Occurring Dataset Based on Stack Exchange Data (Workshop on Natural Language Processing for Programming @ ACL, 2021). PDF Code
An autonomous debating system (Nature, 2021). PDF
Latent Compositional Representations Improve Systematic Generalization in Grounded Question Answering (TACL, 2020). PDF Code
Obtaining Faithful Interpretations from Compositional Neural Networks (ACL, 2020). PDF Code
Evaluating nlp models via contrast sets (Findings of EMNLP, 2020). PDF Code
MedICaT: A Dataset of Medical Images, Captions, and Textual References (Findings of ACL, 2020). PDF
Global Reasoning over Database Structures for Text-to-SQL Parsing (EMNLP, 2019). PDF Code
Representing Schema Structure with Graph Neural Networks for Text-to-SQL Parsing (ACL, 2019). PDF Code
Grammar-based Neural Text-to-SQL Generation (2019). PDF
Emergence of Communication in an Interactive World with Consistent Speakers (Emergent Communication Workshop at NIPS, 2018). PDF Code
Towards an argumentative content search engine using weak supervision (COLING, 2018). PDF
Language Generation with Recurrent Generative Adversarial Networks without Pre-training (Subword and Character level models in NLP Workshop, EMNLP, 2017). PDF Code