• Publications
  • Influence
Repurposing Entailment for Multi-Hop Question Answering Tasks
TLDR
Multee is introduced, a general architecture that can effectively use entailment models for multi-hop QA tasks and outperforms QA models trained only on the target QA datasets and the OpenAI transformer models when using an entailment function pre-trained on NLI datasets.
Author Masking through Translation
TLDR
This notebook paper documents the approach adopted by the team for Author Masking Task in PAN 2016, which uses a simple translation based approach for obfuscating the identity of author of the text.
DeFormer: Decomposing Pre-trained Transformers for Faster Question Answering
TLDR
DeFormer, a decomposed transformer, is introduced, which substitutes the full self-attention with question-wide and passage-wideself-attentions in the lower layers, which allows for question-independent processing of the input text representations.
Is Multihop QA in DiRe Condition? Measuring and Reducing Disconnected Reasoning
TLDR
This work formalizes such undesirable behavior as disconnected reasoning across subsets of supporting facts as well as introducing an automatic transformation of existing datasets that reduces the amount of disconnected reasoning.
♫ MuSiQue: Multihop Questions via Single-hop Question Composition
TLDR
A bottom–up approach is introduced that systematically selects composable pairs of single-hop questions that are connected, that is, where one reasoning step critically relies on information from another, to create MuSiQue-Ans, a new multihop QA dataset with 25K 2–4 hop questions.
Controlling Information Aggregation for Complex Question Answering
TLDR
Empirical evaluation on an elementary science exam benchmark shows that the proposed methods enables effective aggregation even over larger graphs and demonstrates the complementary value of information aggregation for answering complex questions.
What Ingredients Make for an Effective Crowdsourcing Protocol for Difficult NLU Data Collection Tasks?
TLDR
It is found that asking workers to write explanations for their examples is an ineffective stand-alone strategy for boosting NLU example difficulty and that training crowdworkers, and then using an iterative process of collecting data, sending feedback, and qualifying workers based on expert judgments is an effective means of collecting challenging data.
IrEne: Interpretable Energy Prediction for Transformers
TLDR
IrEne is an interpretable and extensible energy prediction system that accurately predicts the inference energy consumption of a wide range of Transformer-based NLP models and can be used to conduct energy bottleneck analysis and to easily evaluate the energy impact of different architectural choices.
Measuring and Reducing Non-Multifact Reasoning in Multi-hop Question Answering
TLDR
An automated sufficiency-based dataset transformation that considers all possible partitions of supporting facts, capturing disconnected reasoning is introduced, formalizing this form of disconnected reasoning and proposing contrastive support sufficiency as a better test of multifact reasoning.
...
1
2
...