1 paper · 2 filters
Nicholas Dingwall, Christopher Potts
We present a simple extension of the GloVe representation learning model that begins with general-purpose representations and updates them based on data from a specialized domain.…