What Is Machine Translation Post-Editing?

Also known as: MTPE

Definition

Machine Translation Post-Editing (MTPE), also called PEMT, is a hybrid translation workflow where raw machine translation output is reviewed and corrected by professional human linguists to achieve publication-quality results. There are two industry-standard levels: light post-editing produces text that is accurate and understandable, typically for internal or low-visibility content, while full post-editing produces output indistinguishable from human translation, suitable for client-facing and published materials. MTPE is significantly faster and less expensive than full human translation from scratch — typically 30-50% cheaper with 40-70% faster turnaround — making it ideal for large-volume technical documentation, knowledge bases, support content, and product catalogs where speed and cost matter alongside quality. The trade-off is that MTPE is less appropriate for creative, marketing, or brand-critical content where the nuance, cultural adaptation, and creative choices of full human translation remain superior.

Quick Facts

Also known as
Post-Editing Machine Translation (PEMT), AI translation post-editing
Levels
Light post-editing and full post-editing (ISO 18587)
Typical cost
30-50% cheaper than full human translation
Best for
Technical docs, knowledge bases, support content, product info

Light vs full post-editing

The distinction between light and full post-editing is defined by ISO 18587 and determines the quality level of the final deliverable. Light post-editing focuses on making machine translation output accurate, understandable, and free from critical errors, without rewriting for style, fluency, or native-sounding phrasing. Sentences may read somewhat machine-like but contain correct information. Light post-editing is appropriate for internal documents, knowledge base articles for search and reference, product catalog entries where users skim rather than read deeply, and high-volume content where cost and speed matter more than polish. Full post-editing produces output indistinguishable from human translation, with full stylistic adaptation, cultural localization, terminology consistency, and native-sounding phrasing. Full post-editing is appropriate for customer-facing marketing content, user-facing product documentation, training materials, and any content representing the brand to external audiences. The cost differential between light and full post-editing is typically 20-30%, with full post-editing approaching but rarely exceeding traditional human translation prices.

When to use MTPE vs human translation

MTPE fits best when source content is technical, repetitive, and well-structured; when large volumes are involved (typically 50,000+ words per project or per month); when timeline pressure requires faster turnaround than traditional human translation; when the target audience will skim or reference the content rather than read it deeply; and when the content has limited brand or creative impact. Full human translation remains the better choice when source content is creative, persuasive, or emotional such as marketing copy, advertising, and brand content; when cultural nuance, idiomatic phrasing, or creative adaptation matters; when legal, medical, or regulatory content carries liability risk; when translating into languages with limited machine translation quality (some minority languages have MT quality too low for efficient post-editing); and when volumes are small enough that the overhead of setting up MT workflows outweighs benefits. Good localization vendors recommend the right mix for each project and language pair rather than defaulting to one approach.

MTPE quality and risk management

High-quality MTPE depends on three factors: the underlying machine translation engine, the post-editor's skill, and the quality assurance process. Modern neural MT engines (DeepL, Google, Microsoft, and custom-trained engines) produce substantially better raw output than statistical MT of the past, making post-editing more productive and the final quality higher. However, MT quality varies significantly by language pair — European language pairs like English-Spanish or English-German produce very high raw quality, while English to Japanese, Korean, or minority languages requires substantially more post-editing effort. Post-editors need specific training beyond traditional translation skills to recognize and efficiently correct MT-specific errors like over-literal translations, gender disagreement in languages with grammatical gender, inconsistent terminology across long documents, and subtle meaning distortions that sound plausible but contradict the source. Leading MTPE vendors use two-tier review (post-editor plus senior linguist), CAT tool QA checks for terminology and formatting, and periodic back-translation sampling for quality verification, producing consistent results measurable against defined quality standards.

Step-by-Step

How machine translation post-editing works

  1. 1

    Content analysis

    Assess source content to determine if MTPE is appropriate (volume, type, target audience).

  2. 2

    MT engine selection

    Choose the best MT engine for the language pair (DeepL, Google, Microsoft, custom).

  3. 3

    Raw MT generation

    Process source content through the selected MT engine to produce raw output.

  4. 4

    Post-editing

    Human linguist edits the MT output to the agreed level (light or full) per ISO 18587.

  5. 5

    QA check

    Automated checks for terminology consistency, tags, numbers, and formatting.

  6. 6

    Review (for full PE)

    Senior linguist reviews post-edited content for final quality before delivery.

  7. 7

    Delivery

    Final files delivered with any required project reports and TM updates.

Key Terminology

Related terms

Machine Translation (MT)
Automated translation by AI/neural engines like Google Translate, DeepL, or custom-trained models
Human Translation (HT)
Translation performed entirely by professional human translators without MT
CAT tool
Computer-assisted translation software like Trados, memoQ, or Phrase that aids translators
Translation Memory (TM)
Database of previously translated segments reused to ensure consistency
Termbase
Approved terminology database ensuring consistent use of technical terms
ISO 18587
International standard defining MTPE quality levels and process requirements

Frequently Asked Questions

Machine Translation Post-Editing FAQs

MTPE pricing typically runs 30-50% below full human translation rates, with light post-editing at the lower end and full post-editing at the upper end. Indicative pricing ranges include light MTPE at US$0.04-0.08 per source word for major European language pairs, full MTPE at US$0.07-0.12 per source word for the same pairs, versus human translation at US$0.10-0.18 per source word. Asian and RTL language pairs command 20-40% premiums across all tiers due to higher complexity. Low-resource languages where MT quality is poor may be priced at or near human translation rates because post-editing effort approaches translating from scratch. Volume discounts apply to large projects with typical breakpoints at 50,000, 100,000, and 500,000 words. Ongoing programs with monthly minimums receive additional discounts. The cost savings from MTPE are real but not automatic — they require appropriate content selection, good MT engines, and skilled post-editors. Vendors charging significantly less than these ranges typically produce quality below ISO 18587 standards or use untrained editors.

Try Us Risk-Free — Send 1 File

Send us one file. We'll DTP it for free — so you can judge our quality before committing. No strings attached.

99.5% on-time delivery  ·  125,000+ projects  ·  Avg. 2hr response