You deliver a 5,000-word light PE job. Three hours of work, you left the robotic phrasing intact as briefed. The PM replies: “Why weren’t the idioms adapted? Why does it still sound like a machine?” You thought that was precisely the point of light PE. They assumed light PE meant something closer to full PE. Nobody wrote it down.
This is the most common failure mode in MTPE projects - not poor editing, but mismatched expectations about scope. Let’s sort that out.
What light post-editing actually means¶
Light post-editing (light PE or LPE) is one of two post-editing levels the industry recognizes. The goal is a translation that is comprehensible (the reader understands the message) and accurate (the meaning matches the source) - but not necessarily polished, natural-sounding, or stylistically consistent.
The TAUS MT Post-Editing Guidelines are explicit: the output of light PE can sound like it was produced by a machine. That’s acceptable. What’s not acceptable is content that misleads, confuses, or offends the reader.
Full post-editing targets human-quality output. The reader shouldn’t be able to tell the text started as MT. Full PE fixes terminology consistency, flow, register, idioms, sentence rhythm - everything light PE deliberately skips.
| Light PE | Full PE | |
|---|---|---|
| Goal | Comprehensible + accurate | Human-quality output |
| Sounds robotic? | Acceptable | Not acceptable |
| Fix style/tone? | No | Yes |
| Fix all grammar errors? | Only if they block meaning | Yes |
| Typical use cases | Internal docs, gisting, knowledge bases | Marketing, legal, customer-facing content |
| Words per day | 5,000-8,000 | 3,000-5,000 |
| Typical rate per word | $0.03-0.06 | $0.08-0.15 |
The core principle: in light PE, you’re doing triage, not restoration. You fix what prevents understanding and leave everything else alone.
Why the scope is so hard to pin down¶
Here’s the structural problem: there is no official standard for light post-editing scope.
ISO 18587:2017 is the only international standard that addresses MTPE. But it covers full post-editing only. Light PE is acknowledged in the standard as a concept, but ISO 18587 explicitly doesn’t specify requirements for it. So when a client asks for “ISO 18587-compliant MTPE,” they’re asking for full PE quality - not light.
As a result, every agency and every client defines light PE differently. According to Argo Translation’s analysis, the most common source of disputes is exactly this: the brief says “light PE” but the client’s actual expectations sit somewhere between light and full.
On ProZ’s MTPE forum, translators regularly argue about whether fixing awkward word order counts as light or full PE. There’s no consensus - because there’s no universal definition.
The practical consequence: you can’t rely on the label “light PE” to know your scope. You need a written brief that defines what counts as a fixable error for each specific project.
What you must fix in light PE¶
These are the categories where you have no choice - fix them regardless of how “light” the brief says to be.
Meaning errors¶
Anything that changes the factual content of the source. Added information the source doesn’t contain. Missing information the source does contain. Inverted meaning - a prohibition turned into a permission, or vice versa. These are not stylistic issues, they’re accuracy failures, and accuracy is the minimum bar for any PE level.
The dropped negation is the classic case: “the tenant shall not be liable” becomes “the tenant shall be liable” in the MT output. That’s a meaning reversal, not a style problem. Catch and fix it in any PE level.
Sentences that are incomprehensible¶
If a sentence is so badly garbled that a native-language reader cannot parse what it means, it needs to be fixed. “Comprehensible” is the minimum standard for light PE, so anything below that minimum gets corrected.
Note: a sentence can be grammatically broken and still comprehensible. “She go to shop yesterday buy milk” is awkward but the meaning is clear - in light PE, that might stay. “The expiration of limit until product by date” - that sentence means nothing, so it goes.
Missing or untranslated segments¶
If a segment wasn’t translated at all and is still in the source language, that’s a content gap, not a style issue. Fix it.
Wrong numbers, dates, and names¶
MT systems make consistent specific errors with numbers, units, date formats, and proper names. A product spec where “100 kg” became “100 lb” is a meaning error regardless of PE level. Check these specifically in every document.
Offensive or culturally unacceptable content¶
MT can produce culturally offensive output or inadvertently harmful phrasing. Fixing this isn’t about style - it’s about basic acceptability. Always within light PE scope.
What you leave alone¶
This is the part that takes discipline for experienced translators. The instinct is to fix everything you notice. In light PE, the skill is deliberate restraint.
Robotic phrasing¶
Light PE output is allowed to sound like a machine translation. Stiff structures, overly literal phrasings, lack of flow - if the meaning is clear and accurate, these are not your problem in light PE.
As Argo Translation notes: “The text may not sound as natural as a human translation, but it should still be usable for the intended purpose.” If you start smoothing every awkward phrase, you’ve crossed into full PE territory and are now working at full PE effort for light PE pay.
Stylistic inconsistency¶
If the MT uses five different phrasings for the same concept across a document, and each is comprehensible on its own, you don’t standardize them in light PE. Terminology consistency and register consistency are full PE tasks.
Minor punctuation issues¶
A missing comma, a non-standard semicolon, quotation marks in the wrong style - if these don’t affect meaning, they stay. The document survives with non-standard punctuation.
Word order that’s unusual but understandable¶
Some MT systems produce grammatically valid sentences with unusual word order. “The document, once signed, by both parties must be filed” is odd but not incomprehensible. In full PE, you’d rewrite it. In light PE, it stays.
Tone and register mismatches¶
If the MT produced formal language where informal was expected (or vice versa), that’s a full PE correction. Light PE doesn’t fix register mismatch unless the mismatch makes the text offensive or culturally unacceptable.
The ISO 18587 gap - what the standard actually says¶
ISO 18587:2017 is worth knowing even though it doesn’t directly govern light PE, because clients sometimes cite it as a quality benchmark.
The standard defines full post-editing requirements in detail: terminology accuracy, syntactic correctness, formatting adherence, semantic fidelity, style adaptation. It sets criteria that a post-editor’s output must meet.
For light PE, the standard acknowledges the concept exists but explicitly declines to standardize it. The reasoning: light PE is too context-dependent. What counts as “acceptable” for an internal FAQ is different from what’s acceptable for a training manual or an employee handbook.
The standard’s drafters noted that organizations using light PE “must negotiate their own custom standards based on competing factors of time, cost, and quality.” That means no one is going to tell you exactly where the line is. You establish it in each project briefing.
This gap has real consequences. The 2025 GTS Translation survey found that 85.99% of freelance translators believe MTPE pricing has deteriorated - and scope ambiguity is one of the key drivers. When “light PE” means something different on every project, estimating time accurately is nearly impossible, and rates calibrated for real light PE scope end up covering near-full PE work.
How to protect yourself: scope clarity before the project starts¶
Before accepting any light PE project, get answers to these questions in writing:
What is the purpose of this translation? Internal use, external-facing, machine gisting, or publication? The answer tells you how much quality actually matters for this specific use case and what “comprehensible” means in context.
What error types must be fixed? Get a list. If the client can’t provide one, send them yours (based on the “must fix” categories above) and ask them to confirm it in writing.
What’s explicitly out of scope? “Style, tone, and terminology consistency are not in scope for this project” - that sentence protects you when the client comes back asking why the output sounds robotic.
What’s the MT quality like? Ask for a sample of 30-50 segments before committing to a rate. Light PE on clean output from a modern MT engine (DeepL, GPT-4o) takes around 60-90 seconds per segment. Light PE on low-quality output from an older system can take as long as full PE - because you’re essentially retranslating garbled text.
As one experienced translator put it in a ProZ thread:
I’ve had light PE projects where the MT was so bad that every segment needed a complete rewrite. At that point you’re doing full PE work for light PE pay. I now always ask for 50 test segments before quoting.
That’s the right call. Don’t discover this halfway through a 10,000-word project.
A practical workflow for light PE sessions¶
The biggest risk when post-editing is slipping into full PE mode by instinct. Here’s a workflow that keeps you anchored to light PE scope.
Step 1: Read through the MT without editing first. Get a sense of the overall quality. This takes 10-15 minutes for a 2,000-word document and tells you whether you’re dealing with clean MT or something closer to unusable.
Step 2: Set your error filter. Before touching the text, remind yourself: you’re fixing meaning errors, missing content, nonsensical sentences, wrong numbers/names, and offensive output. That’s the list. Everything else stays.
Step 3: First pass - flag critical errors only. Don’t fix as you go. Flag segments with meaning errors using a comment or highlight. This prevents you from sliding into translation mode where you start rewriting clean segments by reflex.
Step 4: Second pass - fix only what you flagged. Open the source and fix the flagged segments. Don’t fix anything else, even if you notice something while you’re there.
Step 5: Spot-check numbers and names. Run a quick pass specifically for numeric and name accuracy. MT systems produce these errors consistently across the whole document.
Step 6: Final read-through. Read the translation only (not alongside the source) and ask: does this make sense? If yes, you’re done.
Total time for a clean 2,000-word document at light PE level: around 90 minutes to 2 hours. If you’re regularly taking longer, you’ve probably slipped into full PE mode - or the MT quality isn’t good enough for the brief.
FAQ¶
What’s the difference between light and full post-editing?¶
Light PE targets comprehensibility and accuracy only - the text must make sense and convey the source meaning, but robotic phrasing is acceptable. Full PE targets human quality - the text must read as if a human translated it from scratch, with correct style, terminology, register, and natural flow. For internal or informational content, light PE is often enough. For anything customer-facing or legally significant, it typically isn’t.
Does ISO 18587 apply to light post-editing?¶
No. ISO 18587:2017 defines requirements only for full post-editing. Light PE is mentioned in the standard but not standardized. If a client asks for “ISO 18587-compliant” work, they’re describing full PE quality expectations - not light PE.
How many words per hour is realistic for light post-editing?¶
On clean MT output from a modern engine, light PE typically runs at 800-1,200 words per hour, or 5,000-8,000 words per day. Full PE runs at 400-600 words per hour. If your actual throughput is closer to the full PE range, the MT quality is likely too poor for a light PE brief - that’s a conversation to have with the client before you continue.
What if the MT quality is so bad that light PE isn’t workable?¶
Flag it before starting work, not after. Ask for test segments, estimate your actual time, and let the client know the MT quality requires full PE scope. Either the rate goes up, the content scope narrows, or the project gets re-briefed. Don’t silently deliver full PE work at a light PE rate, and don’t submit inadequate output just to stay within an impossible brief.
Should I fix robotic-sounding sentences in light PE?¶
Only if the robotic phrasing makes the sentence genuinely incomprehensible. “Users can clicking settings icon for access menu options” - that’s unclear, fix it. “Users may access the menu options by clicking the settings icon” - that sounds like translated text but the meaning is perfectly clear, so in light PE it stays exactly as is.
How is light PE priced compared to full PE?¶
Industry rates in 2025 run from roughly $0.03-0.06 per word for light PE and $0.08-0.15 for full PE - according to current market data. Light PE is about 40-60% cheaper than full PE, reflecting the reduced scope and faster throughput. If you’re being offered light PE rates but doing full PE work due to scope ambiguity, that gap directly hits your effective hourly earnings.