Paper Stash
The papers I've read and notes on them
This is a place for me to document some of the papers I’ve read and remind myself what they were about for future reference. For some papers, I’ve included my quick notes on them or slides I’ve made for them.
PowerTransformer: Unsupervised Controllable Revision for Biased Language Correction
Uses a joint training objective on a GPT based system to revise biased sentences. These objectives are mask & reconstruct and paraphrase. They aim to revise bias in the agency and power of characters.
RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models
Creates a dataset of prompts for language models in order to analyze whether or not the generation models will be toxic. Uses dataset to analyze the effectiveness of detoxifying methods.
End-to-End Slot Alignment and Recognition for Cross-Lingual NLU
Introduces dataset and model for cross-lingual transfer learning. Model allows for slot-filling without external word alignment tools.