As NMT Matures, Machine Translation Post-Editing Goes Mainstream

Since the rise of neural machine translation (NMT) in 2016, MT usability has improved year by year. By 2020, "machine translation + human post-editing" (MTPE) had moved from a controversial novelty to the default production workflow for large volumes of commercial and technical documentation. Service design and linguist training around ISO 18587 — Translation services, Post-editing of machine translation output, Requirements — visibly accelerated during the year.

Published in 2017, ISO 18587 specifies post-editing processes, linguist competence requirements and delivery quality standards. As NMT output improved, more and more LSPs began quoting tiered services — "light post-editing" (ensuring comprehensibility) versus "full post-editing" (matching human translation quality) — and clients gradually accepted the idea of choosing a quality level by use case.

For translators, mainstream MTPE changed the shape of work: the share of pure from-scratch translation declined, while post-editing and MT quality assessment grew. Pricing models adjusted accordingly — hourly rates or discounted per-source-word rates increasingly replaced traditional full-price per-word billing.

It bears emphasizing that MTPE does not suit every scenario: legal contracts, marketing copy and literary translation — texts where accuracy and style are paramount — remain the domain of senior human translators. Seeing the boundaries of machine translation clearly is precisely where a professional translation company adds value.