SmithE.M. x 6
bookRéférences 6
Multiple-attribute text style transfer
… Firstly, while previous approaches operate at the sentence level by assuming that every...
2019-09-22 20:07:01
… We demonstrate the technique using a set of finegrained sentiment labels such as happy,...
Hi, my name is Martha: Using names to measure and mitigate bias in generative dialogue models
… finetuned on several dialogue datasets that were designed to impart the model with a wi...
" I'm sorry to hear that": finding bias in language models with a holistic descriptor dataset
… By inspection, utterances marked as offensive tend to be those that respond to negative...
Controlling style in generated dialogue
… We do not make use of the sentiment labels from ED in this work. They might be somewhat...
“I'm sorry to hear that”: Finding New Biases in Language Models with a Holistic Descriptor Dataset
… This effect is not due solely to template length, as seen when comparing longer, emotio...