Context Aware Language Models
Code for building language models that adapt to different contexts. This code was originally written to support the experiments in the paper Improving Context Aware Language Models. It has since been modified to support experiments for the paper Low-Rank RNN Adaptation for Context-Aware Language Modeling (to appear in TACL). Read the paper for a complete description of the model.
The main idea is that metadata or other context information can be used to adapt or control a language model. The adaptation takes place by allowing a context embedding to transform the weights of the recurrent layer of the model. We call this model the FactorCell model.
I will work on documenting the code more. Send me a message if you want some help getting started.