Abstract

This paper describes XNMT, the eXtensible Neural Machine Translation toolkit.XNMT distin- guishes itself from other open-source NMT toolkits by its focus onmodular code design, with the purpose of enabling fast iteration in researchand replicable, reliable results. In this paper we describe the design of XNMTand its experiment configuration system, and demonstrate its utility on thetasks of machine translation, speech recognition, and multi-tasked machinetranslation/parsing. XNMT is available open-source athttps://github.com/neulab/xnmt

 

Quick Read (beta)