• Publications
  • Influence
Supervised Text-based Geolocation Using Language Models on an Adaptive Grid
TLDR
We define an alternative grid construction using k-d trees that more robustly adapts to data, especially with larger training sets. Expand
  • 182
  • 42
  • PDF
Wizard of Wikipedia: Knowledge-Powered Conversational agents
TLDR
In open-domain dialogue intelligent agents should exhibit the use of knowledge, however there are few convincing demonstrations of this to date. Expand
  • 149
  • 41
  • PDF
Inclusive yet Selective: Supervised Distributional Hypernymy Detection
TLDR
We test the Distributional Inclusion Hypothesis, which states that hypernyms tend to occur in a superset of contexts in which their hyponyms are found. Expand
  • 149
  • 17
  • PDF
What makes a good conversation? How controllable attributes affect human judgments
TLDR
We examine two controllable neural text generation methods, conditional training and weighted decoding, in order to control four important attributes for chit-chat dialogue: repetition, specificity, response-relatedness and question-asking. Expand
  • 68
  • 11
  • PDF
Hearst Patterns Revisited: Automatic Hypernym Detection from Large Text Corpora
TLDR
We find that simple pattern-based methods consistently outperform distributional methods on several hypernymy tasks, including detection, direction prediction, and graded entailment ranking. Expand
  • 54
  • 9
  • PDF
Recipes for building an open-domain chatbot
TLDR
We provide recipes for building opendomain chatbots that perform well in human evaluations in terms of engagingness and humanness measurements. Expand
  • 43
  • 9
  • PDF
Representing Meaning with a Combination of Logical and Distributional Models
TLDR
We adopt a hybrid approach that combines logical and distributional semantics using probabilistic logic, specifically Markov Logic Networks. Expand
  • 47
  • 8
  • PDF
Neural Text Generation with Unlikelihood Training
TLDR
We show that the likelihood objective itself is at fault, resulting in a model that assigns too much probability to sequences containing repeats and frequent words, unlike those from the human training distribution. Expand
  • 55
  • 7
  • PDF
ACUTE-EVAL: Improved Dialogue Evaluation with Optimized Questions and Multi-turn Comparisons
TLDR
We introduce ACUTE-EVAL, a method that combines the benefits, and attempts to mitigate the deficiencies, of the above two approaches by introducing a pairwise relative comparison setup for multi-turn dialogues. Expand
  • 34
  • 7
  • PDF
MGNC-CNN: A Simple Approach to Exploiting Multiple Word Embeddings for Sentence Classification
TLDR
We introduce a novel, simple, scalable CNN architecture - multi-group norm constraint CNN (MGNC-CNN) that capitalizes on multiple sets of word embeddings for sentence classification. Expand
  • 81
  • 6
  • PDF