Learning to Compute Word Embeddings On the Fly
This addresses the challenge of handling rare words in NLP tasks, which is incremental as it builds on existing embedding methods.
The paper tackles the problem of poor representations for rare words in natural language by introducing a method to predict embeddings on the fly from small auxiliary data, improving results in reading comprehension, textual entailment, and language modeling tasks.
Words in natural language follow a Zipfian distribution whereby some words are frequent but most are rare. Learning representations for words in the "long tail" of this distribution requires enormous amounts of data. Representations of rare words trained directly on end tasks are usually poor, requiring us to pre-train embeddings on external data, or treat all rare words as out-of-vocabulary words with a unique representation. We provide a method for predicting embeddings of rare words on the fly from small amounts of auxiliary data with a network trained end-to-end for the downstream task. We show that this improves results against baselines where embeddings are trained on the end task for reading comprehension, recognizing textual entailment and language modeling.