0

Neural Machine Translation without Embeddings

Using one-hot byte representations in NLP models improves performance in byte-to-byte translation, equaling character-level and subword-level models, due to the effect of token dropout.

Year
2020
Venue
NAACL 2021 4
Authors
2
Hosting
Abstract onlyARXIV-DEFAULT

Cite

Notes

Only stored in your browser.

Attribution

Abstract & full text
arxiv.org/abs/2008.09396v2ARXIV-DEFAULT
TL;DR
Semantic Scholar
Attribution policy →

Abstract

Many NLP models operate over sequences of subword tokens produced by hand-crafted tokenization rules and heuristic subword induction algorithms. A simple universal alternative is to represent every computerized text as a sequence of bytes via UTF-8, obviating the need for an embedding layer since there are fewer token types (256) than dimensions. Surprisingly, replacing the ubiquitous embedding layer with one-hot representations of each byte does not hurt performance; experiments on byte-to-byte machine translation from English to 10 different languages show a consistent improvement in BLEU, rivaling character-level and even standard subword-level models. A deeper investigation reveals that the combination of embeddingless models with decoder-input dropout amounts to token dropout, which benefits byte-to-byte models in particular.

Authors

2