• Publications
  • Influence
ERASER: A Benchmark to Evaluate Rationalized NLP Models
TLDR
This work proposes the Evaluating Rationales And Simple English Reasoning (ERASER) a benchmark to advance research on interpretable models in NLP, and proposes several metrics that aim to capture how well the rationales provided by models align with human rationales, and also how faithful these rationales are. Expand
Attention is not Explanation
TLDR
This work performs extensive experiments across a variety of NLP tasks to assess the degree to which attention weights provide meaningful “explanations” for predictions, and finds that they largely do not. Expand
Learning to Faithfully Rationalize by Construction
TLDR
Variations of this simple framework yield predictive performance superior to ‘end-to-end’ approaches, while being more general and easier to train. Expand
Structured Disentangled Representations
TLDR
Experiments on a variety of datasets demonstrate that the proposed two-level hierarchical objective can not only disentangle discrete variables, but that doing so also improves disentanglement of other variables and, importantly, generalization even to unseen combinations of factors. Expand
SciREX: A Challenge Dataset for Document-Level Information Extraction
TLDR
SciREX is introduced, a document level IE dataset that encompasses multiple IE tasks, including salient entity identification and document level N-ary relation identification from scientific articles, and a neural model is developed as a strong baseline that extends previous state-of-the-art IE models to document-level IE. Expand
Hierarchical Disentangled Representations
TLDR
This work synthesizes a generalization of the evidence lower bound that explicitly represents the trade-offs between sparsity of the latent code, bijectivity of representations, and coverage of the support of the empirical data distribution. Expand
An Analysis of Attention over Clinical Notes for Predictive Tasks
TLDR
This work performs experiments to explore whether inclusion of attention mechanisms is critical for neural encoder modules that operate over notes fields in order to yield competitive performance, but unfortunately, while these boost predictive performance, it is decidedly less clear whether they provide meaningful support for predictions. Expand
Question Answering over Knowledge Base using Factual Memory Networks
TLDR
Factual Memory Network is introduced, which learns to answer questions by extracting and reasoning over relevant facts from a Knowledge Base, and improves the run-time efficiency of the model using various computational heuristics. Expand
Cross Lingual Sentiment Analysis using Modified BRAE
TLDR
It is shown that the Recursive Autoencoder architecture is used to develop a Cross Lingual Sentiment Analysis tool using sentence aligned corpora between a pair of resource rich (English) and resource poor (Hindi) language and significantly outperforms state of the art systems for Sentiment analysis. Expand
An Empirical Comparison of Instance Attribution Methods for NLP
TLDR
It is found that simple retrieval methods yield training instances that differ from those identified via gradient-based methods (such as IFs), but that nonetheless exhibit desirable characteristics similar to more complex attribution methods. Expand
...
1
2
...