Machine Translation: Which Type Works Best? (2018)

28 Jun 2018
by Donald A. DePalma, Dr. Arle Lommel

Just a few years ago it seemed that the decade-old battle between rule-based and statistical machine translation had ended, with statistical approaches dominating and rule-based systems occupying a niche role. Hybrid technology that combines the two had become common. Then along came neural and upended the market.

In this report we analyze the state of the various approaches currently in production or on the horizon. It discusses solutions in chronological order of development – rule-based, statistical, hybrid, and neural MT. It then discusses the role of adaptive MT, characterizes deployment choices, and compares the types and lists major providers for each one. Rather than take this journey through the MT epochs, you can go straight to the approach that interests you. For most people today, it will be neural MT.

This report is an update of “Machine Translation: Which Type Works Best?” published on July 28, 2016. CSA Research periodically re-examines and updates research to account for changing market and technological factors. We revised this report to reflect major developments in the machine translation market since 2016.

 

Related Research

 

Page Count: 12