- Keywords: NLP, word embeddings, offensive language, reliability, fairness
- TL;DR: Static embeddings result in reliable downstream consistency of word embeddings
- Abstract: While pre-trained word embeddings have been shown to improve the performance of downstream tasks, many questions remain regarding their reliability: Do the same pre-trained word embeddings result in the best performance with slight changes to the training data? Do the same pre-trained embeddings perform well with multiple neural network architectures? What is the relation between downstream fairness of different architectures and pre-trained embeddings? In this paper, we introduce two new metrics to understand the downstream reliability of word embeddings. We find that downstream reliability of word embeddings depends on multiple factors, including, the handling of out-of-vocabulary words and whether the embeddings are fine-tuned.