Insertion-based Decoding with Automatically Inferred Generation Order

arXiv: Computation and Language, pp. 661-676, 2019.

Cited by: 30|Bibtex|Views82|DOI:https://doi.org/10.1162/tacl_a_00292
EI
Other Links: academic.microsoft.com|dblp.uni-trier.de|arxiv.org

Abstract:

Conventional neural autoregressive decoding commonly assumes a fixed left-to-right generation order, which may be sub-optimal. In this work, we propose a novel decoding algorithm -- InDIGO -- which supports flexible sequence generation in arbitrary orders through insertion operations. We extend Transformer, a state-of-the-art sequence gen...More

Code:

Data:

Full Text
Your rating :
0

 

Tags
Comments