You May Not Need Attention

TL;DR

This paper introduces a novel recurrent neural translation model that eliminates the need for attention mechanisms and separate encoder-decoder architecture, achieving comparable or better performance especially on long sentences.

Contribution

The paper presents an eager, low-latency translation model without attention or separate encoder-decoder, demonstrating competitive results in neural machine translation.

Findings

Performs on par with attention-based models

Outperforms on long sentences

Uses constant memory during decoding

Abstract

In NMT, how far can we get without attention and without separate encoding and decoding? To answer that question, we introduce a recurrent neural translation model that does not use attention and does not have a separate encoder and decoder. Our eager translation model is low-latency, writing target tokens as soon as it reads the first source token, and uses constant memory during decoding. It performs on par with the standard attention-based model of Bahdanau et al. (2014), and better on long sentences.