• Publications
  • Influence
Inducing Domain-Specific Sentiment Lexicons from Unlabeled Corpora
TLDR
We combine domain-specific word embeddings with a label propagation framework to induce accurate domain- specific sentiment lexicons using small sets of seed words. Expand
  • 195
  • 38
  • PDF
What Does BERT Look At? An Analysis of BERT's Attention
TLDR
We propose methods for analyzing the attention mechanisms of pre-trained models and apply them to BERT, showing that certain attention heads correspond well to linguistic notions of syntax and coreference. Expand
  • 275
  • 32
  • PDF
Deep Reinforcement Learning for Mention-Ranking Coreference Models
TLDR
We propose using reinforcement learning to directly optimize a neural mention-ranking model for coreference evaluation metrics, obviating the need for hyperparameters that must be carefully selected. Expand
  • 176
  • 20
  • PDF
Semi-Supervised Sequence Modeling with Cross-View Training
TLDR
We propose Cross-View Training (CVT), a semi-supervised learning algorithm that improves the representations of a Bi-LSTM sentence encoder using a mix of labeled and unlabeled data. Expand
  • 142
  • 19
  • PDF
Improving Coreference Resolution by Learning Entity-Level Distributed Representations
TLDR
We present a neural network based coreference system that produces high-dimensional vector representations for pairs of coreference clusters. Expand
  • 193
  • 18
  • PDF
Entity-Centric Coreference Resolution with Model Stacking
TLDR
We show that the scores produced by such models can be aggregated to define powerful entity-level features between clusters of mentions. Expand
  • 139
  • 14
  • PDF
Large-scale Analysis of Counseling Conversations: An Application of Natural Language Processing to Mental Health
TLDR
We present a large-scale, quantitative study on the discourse of text-message-based counseling conversations and discover actionable conversation strategies that are associated with better conversation outcomes. Expand
  • 96
  • 11
  • PDF
BAM! Born-Again Multi-Task Networks for Natural Language Understanding
TLDR
We propose a way of applying knowledge distillation where single-task models teach a multi-task model by training the student to imitate the teacher’s outputs. Expand
  • 65
  • 10
  • PDF
RevMiner: an extractive interface for navigating reviews on a smartphone
TLDR
We introduce RevMiner - a novel smartphone interface that utilizes Natural Language Processing techniques to analyze and navigate reviews. Expand
  • 43
  • 7
  • PDF
ELECTRA: Pre-Training Text Encoders As Discriminators Rather Than Generators
Masked language modeling (MLM) pre-training methods such as BERT corrupt the input by replacing some tokens with [MASK] and then train a model to reconstruct the original tokens. While effective whenExpand