Journal
Notes from a working Turkmen linguist.
On translation, localization, freelancing, and the AI shift.
A 100% Match Is Only 100% in the Database
The segments you're told not to touch are the ones that come back to bite you. A hard look at why leveraged matches deserve your eyes, and who pays when they don't.
Read →The Discount Grid Is Where Your Rate Actually Gets Set
Your per-word rate means nothing until you've seen the Trados match grid underneath it. That's the real negotiation, and most of us lose it before we've said a word.
Read →There's Nothing to Post-Edit in a Transcreation
Post-editing works because there's a source to measure against. Transcreation throws that away — and that's exactly why the machines choke on it, and why the pricing model everyone loves falls apart.
Read →AI Can Draft the Contract. It Can't Sign for It.
In legal translation, the signature at the bottom isn't a formality — it's the whole product. Here's why the accountability gap keeps a human in the loop even when the draft is 80% there.
Read →When Your PM Leaves, You're a Stranger to the Agency Again
Your relationship with an agency isn't with the agency. It's with one person who has your name in their head — and when they quit, so does your history.
Read →COMET Doesn't Speak Turkmen, and My QA Panel Knows It
Automatic quality estimation is quietly reshaping how agencies triage translation — but the scores go blind on low-resource languages. Here's what actually catches errors in Turkmen work.
Read →Reading Speed Is the Real Editor
Everyone argues about word choice in subtitles. The number that actually decides what survives is characters per second — and most people never look at it.
Read →Leave the Barrel Alone: Units in Oil & Gas Translation
Converting numbers in a drilling report feels helpful. It's the fastest way to get someone hurt. When to touch a unit and when to keep your hands off it.
Read →The "Can You Sharpen Your Rate" Email Is a Test
Agencies ask freelancers to drop their rates 2.5 times more often than direct clients do. That email isn't an insult — it's a test, and there's a right way to answer it.
Read →ISO 5060 Puts a Number on Something I Do by Feel
A new standard wants to grade translation with error-based metrics. That's fine — until the score starts standing in for the reading. Here's where MQM-style scoring helps my work and where it lies about it.
Read →Spotted for English, Delivered in Turkmen
Most subtitle jobs arrive pre-timed for a language that isn't the one you're writing in. Here's why the template is quietly deciding your translation before you type a word.
Read →Whose Name Goes at the Bottom of the Page
AI can draft a legal or oil & gas translation. It can't sign one. As the EU AI Act phases in and liability stays unresolved, the human signature is turning into the actual deliverable.
Read →My Terminology Checker Thinks Grammar Is a Bug
CAT-tool QA was built for languages where a word stays the same word. Turkmen glues six things onto a stem, and the checker panics. Here's what actually breaks and how I stop trusting the red flags.
Read →They Ask Where You Live to Decide What You're Worth
Agencies dress up the "country of residence" question as compliance, but half the time it's rate profiling. Here's why your postcode is the worst possible way to price a translation — and how I refuse to hand it over as a bargaining chip.
Read →The Demand Didn't Drop. It Changed Address.
Freelance requests are down while enterprise in-house demand holds steady or climbs. The work didn't vanish — it moved into the building, and that changes who you should be positioning for.
Read →Whoever Answers First Gets the Job
Agencies don't pick vendors by who's best. They pick by who replies before lunch. Here's what that actually rewards, and how I stay reachable without living inside my inbox.
Read →Let the Score Decide What I Never See
Quality-estimation routing now sends only the 'risky' segments to a human. Fine — until the score doing the triage was never trained on Turkmen, and the confident-wrong lines sail straight to publish.
Read →Track the Hour, Not the Word
Your per-word rate tells you almost nothing about whether a client is worth keeping. The number that actually matters is the one you have to work to find — what you earn per hour once the file, the queries, and the payment chase are done.
Read →What the Viewer Never Gets to Read
Subtitling Turkmen isn't translation with a character limit stapled on. It's deciding, line by line, what meaning the viewer will never receive — and that call is the actual job.
Read →The Sentence You're Not Allowed to Translate
Some phrases in an oil & gas safety document aren't yours to render freely — they have official, legally fixed wordings. The real skill is knowing which sentences those are, and what to do when your language never got one.
Read →Which QA Checks I Keep On, and Which I Kill
The automated QA profile in your CAT tool was tuned for German and English. Run it on Turkmen and half the flags are noise — here's what I actually keep switched on, and why live preview does more real checking than any rule set.
Read →Pick a Niche, They Say — I Already Am One
The industry preaches narrow specialization to lift your rates. For a language like Turkmen, that advice runs backwards, and the numbers behind it were never collected with me in mind.
Read →The Joke Doesn't Survive the Character Limit
When you're subtitling comedy or wordplay into Turkmen at 42 characters a line, transcreation and constraint fight each other. Here's how I decide which one loses.
Read →The Rig Speaks Russian, the Contract Wants Turkmen
On the ground, Turkmenistan's oil and gas fields run on Russian and English. So what do you do when the document has to be Turkmen and the settled term simply doesn't exist yet?
Read →Tell Me What the Machine Already Touched
Agencies increasingly hand freelancers files that have already been through machine translation without saying so — and dress the job up as a 'review.' Here's why disclosure isn't a courtesy, it's a pricing input.
Read →Translating Row 47 Blind
CAT tools chopped every document into numbered strings and we called it progress. The industry's new 'context-centric' pitch admits the segment was the problem all along — but it oversells the cure.
Read →Going Direct Is a Plan for a Market I Don't Have
The whole industry is telling freelancers to fire their agencies and win direct clients at double the rate. For a rare-language translator, that advice quietly assumes a market that doesn't exist.
Read →The Minimum Fee Is the Most Honest Number on Your Rate Sheet
A 40-word job and a 4,000-word job cost roughly the same to set up. Here's why a minimum fee isn't greed — it's the only line item that tells the truth about what small work actually costs.
Read →Whose Voice Are You Cloning? AI Dubbing Meets a Language That Never Got Dubbed
The panels promise voice-cloned localization at scale for every market. That pitch quietly assumes an audience already used to hearing content in its own language — and for Turkmen, that audience was never built.
Read →The Comma That Shuts Down a Rig
In oil & gas translation, the most dangerous thing on the page isn't a word — it's a number with a decimal separator you didn't think about. A working note on why measurement is the real specialism.
Read →Your Real Client Is a Person, and She's About to Quit
For a freelance vendor, an agency isn't an institution — it's the two or three project managers who actually know your name. When they leave, so does everything they knew about you.
Read →The QA Report That Nobody Reads Past the Score
CAT-tool QA output has turned into a number everyone treats as a verdict. Here's why the summary line lies, and what a working reviewer actually looks at instead.
Read →Net 60 Is a Rate Cut You Already Agreed To
Payment terms, wire fees, and currency spread quietly eat a chunk of your rate — and most freelancers never put them in the math. Here's how I price the wait, from Dubai.
Read →Back-Translation Is Where Transcreation Goes to Die
Clients ask me to transcreate a slogan, then approve it by reading a literal back-translation of my Turkmen. That's like judging a joke by its autopsy.
Read →The Tier They Never Assign You To
Agencies now sort content into risk tiers before it reaches a translator. In oil & gas and legal work, the label on the folder decides how much your judgment is worth — and the labels are usually wrong.
Read →The Onboarding Tax: What Agencies Ask Before They've Sent a Single Word
Every new agency wants forty minutes of forms, an NDA, a portal login, and a sample test — before there's any work. Here's what's fair, what's theater, and where I've learned to push back.
Read →Garbage In, Locked Forever: The Translation Memory Nobody Audits
Everyone runs QA on the file. Almost nobody runs it on the translation memory feeding the file — and for a single-vendor language, that memory is just a record of my own past decisions, good and bad.
Read →The Rate Advice Everyone Gives Is Written for Somebody Else
Escape the bottom, go direct, charge for strategy, chase retainers — sound advice if you translate Spanish marketing. For a single-vendor pair like Turkmen, most of it doesn't apply. Here's what actually moves my rate.
Read →Fluent and Wrong: The New Failure Mode of Machine Turkmen
Old machine translation failed loudly. The new stuff fails quietly — polished, confident, and wrong in ways that slip past everyone but a native reader. That shift changes what agencies should actually be paying me for.
Read →The Machine That Grades the Machine: On Automated Quality Scores for Turkmen
CAT platforms now ship LLM-based quality scores that rate a translation from 0 to 100. For a low-resource language like Turkmen, that number measures something, but it isn't what the dashboard implies.
Read →The Alphabet Is Not a Font Setting: Script Conversion in Central Asian Localization
When a Turkic MT engine's headline feature is dual-script Uzbek support, it's telling you something: in this region, choosing the right alphabet is half the localization job. Here's why script conversion is a separate skill, and why Turkmen makes it harder than anyone budgets for.
Read →The Number You Don't Put on Your Profile
Publishing a fixed per-word rate feels transparent and professional. For a rare-language specialist, it can also quietly cap what you're allowed to earn.
Read →The Post-Editing Trap: When Cleaning Up Machine Turkmen Costs More Than Starting Over
For high-resource languages, post-editing machine translation is a genuine productivity gain. For Turkmen, it often quietly becomes a tax on the translator and a false economy for the agency. Here's how to tell the difference.
Read →Forty-Two Characters and No Room to Breathe: Subtitling Turkmen Against the Clock
Subtitling is not translation with a word limit bolted on — it's a compression problem. And for an agglutinative language like Turkmen, the math is brutal in ways most project managers never see on the spec sheet.
Read →The Conversation Only Goes One Way: Why Multilingual Budgets Broadcast but Never Listen
Global companies spend heavily to push their message into dozens of languages, then leave the replies untranslated. Treating communication as one-directional is the quiet failure mode of multilingual business.
Read →Resolve It or Preserve It: The Opposite Instincts of Technical and Legal Translation
Technical translation rewards the instinct to clarify; legal translation punishes it. Oil and gas projects hand you both kinds of text in the same folder — and the discipline is knowing which reflex to switch off.
Read →Stop Counting Words: Why the Per-Word Model Is the Wrong Way to Price the Next Decade
The word count was always a proxy for effort. Machine translation has broken that proxy — and the agencies that survive will price by risk and consequence instead of volume.
Read →You Might Be Their Only Turkmen Translator: On Being a Single-Vendor Language
For many agencies, the Turkmen pair starts and ends with one freelancer. That scarcity changes the relationship — and most vendors underuse the leverage it gives them.
Read →Leave It Alone: The Hardest Skill in Bilingual Revision
Most revision damage doesn't come from missing errors — it comes from fixing things that were never broken. A note on the discipline of restraint, with examples from Turkmen.
Read →When the QA Checker Cries Wolf: Automated Checks Versus Turkmen Morphology
Rule-based QA modules in CAT tools were built for languages that inflect lightly. Run them against agglutinative Turkmen and you drown in false positives — here's how I separate the real errors from the noise.
Read →The Terminology Vacuum: Localizing Software Into a Language Without Settled Words
Turkmen has no consolidated IT lexicon, so every UI string forces a choice between Russian habit, Turkish borrowing, English carry-over, or a neologism. Here's how I make those calls — and why a project glossary is the real deliverable.
Read →Don't Lump Turkmen In with Central Asia: A Linguist's Note on the Resource Gap
Project managers often bundle Turkmen, Kazakh, and Uzbek under one 'Central Asia' line item. That assumption quietly breaks budgets, timelines, and quality — here's why the three diverge sharply, and what to do about it.
Read →