Gains:
- Being able to distinguish where artificial intelligence saves time in the translation workflow (draft, term, summary) and where decisions such as meaning, tone, culture and responsibility are left to the human, depending on the level of risk.
- Ability to apply a discipline that verifies each translation output by comparing it with the source, confirming terminology and consistency, and passing it through the meaning-tone-culture filter.
- Understand why confidentiality, personal data protection and unverified output must be taken into account in translation from the very beginning.
You are in the morning at a translation desk. In your inbox you have an urgent translation of a legal contract, 800 lines of interface text of a software, a preparation file for a two-day conference and a subtitle revision of a TV series. The customer asks "why is this taking so long when there is artificial intelligence?" he asks, and you know that a mistranslated contractual clause can cost compensation, and a made-up term can cost brand reputation. Translation and interpreting (the common name of the professions of translator who translates written texts and interpreter who instantly transmits oral speech) is a field that is multilingual in nature, context-dependent, cultural and highly responsible. Artificial intelligence (AI - software that can extract patterns from historical data, produce, translate and summarize text) gives you serious speed in this volume and time pressure. But the very beginning of this module is clear: AI is an assistant, translation draft generator and terminology/quality assistant; You are the competent translator who takes responsibility for the meaning, tone, culture and final delivery.
In this first unit we will focus on discipline, not the tool. You will learn where AI saves real time in the translation workflow, where it is dangerous, how to verify each output, how to protect the client's confidential document, and how to stay on the ethical-copyright line. Without laying this foundation, subsequent units remain in the air — because an unverified output in translation is not just a "bad sentence"; It could mean a false obligation in a contract, a dangerous dosing error in a drug package insert, or a brand being ridiculed.
Where does AI come in handy in the translation workflow?
Let's divide translation work into two large clusters. The first cluster: voluminous, repetitive, pattern-extractable, and language-intensive tasks. First draft translation of a long technical manual, consistent transfer of repetitive sentences, extraction of a list of terms, simplification of a text for readability, comparison of two texts in two languages, first draft of subtitle time codes. In these jobs, AI reduces hours to minutes and does not get tired.
The second cluster: meaning, tone, culture, creativity, and responsible decisions. What an idiom means in the target culture, the emotional impact of an advertising slogan, the exact equivalent of a legal term between two legal systems, the voice of a literary text, the final approval of the text to be delivered to a client. These decisions require expertise, cultural knowledge and human judgment. Here AI proliferates options, generates drafts — but the final say and signature for delivery is yours.
Let's clarify the distinction in one sentence: AI is strong on "what is the first draft of this text and its possible counterparts" questions; The decision is yours when it comes to questions such as "Does this really carry the right meaning, the right tone and the right cultural impact? Can I deliver this to my client?"
Tip: Before outsourcing a job to an AI, ask: “What do I lose if this output is wrong?” If the answer is "a few minutes of refactoring", delegate comfortably. If the answer is "an incorrect contract clause, incorrect medical information, or a brand crisis," let the AI produce the draft and you make the decision and verification.
Basic concepts: MT, NMT, LLM, MTPE
Let's define the field's four key terms from the beginning. MT (Machine Translation) is the general name of software that automatically translates a text from one language to another. NMT (Neural Machine Translation) is the modern generation of MT using artificial neural networks and is the basis of tools such as Google Translate and DeepL; It is fluent and fast on a sentence basis. LLM (Large Language Model) are general-purpose, instruction-driven, flexible models in context and tone, such as ChatGPT and Claude; They can understand instructions such as "translate this text with formal tone and this list of terms." MTPE (Machine Translation Post-Editing) is a form of work in which a human translator corrects the raw output of machine translation and is the most common working model in the industry today. We will cover these terms in depth in subsequent units; For now, keep in mind the "AI produces the raw draft, the human makes it trustworthy" chain in mind.
Privacy: inviolable principle
The texts that come into the hands of the translator are often confidential: an unsigned contract, a patient report, a company's financial statement, a court file, a product that has not yet been announced. Rule of thumb here: No confidential, personal or unpublished client documents are entered into a public AI tool whose confidentiality is not guaranteed. Text you paste into a public translation service is processed on that service's servers; You often cannot control who sees it, where it is stored, and whether it is used in training the model. Many translation contracts include an NDA (Non-Disclosure Agreement) and giving the confidential text to an external tool is a violation of this agreement and even causes legal liability.
Professional AI use therefore walks in two worlds: generic tools can be used for public domain texts; For confidential/contracted texts, only secure systems are used that ensure that the data does not leak out and that institutional/contracted, "we do not use the data in education" assurance. Confusing the two is the most dangerous mistake of this profession.
Attention: "The AI translated it that way" is not a justification. If there is a mistranslation, a made-up term, or a leaked document, the responsibility lies not with the AI, but with the translator who delivered that output without verifying it or entered the secret text into the tool.
Verification discipline: three steps
AI produces fluent and confident translation; That doesn't mean it's true. AI occasionally produces hallucinations — that is, adds a sentence that doesn't exist in the source, changes a number, turns a negative into a positive (replacing "is" instead of "is"), or makes up a term that doesn't exist. In a contract or medical text this is disastrous. Apply a three-step reflex to each output:
- Compare with source. Read the equivalent of each sentence in the source. In particular, check numbers, dates, negative suffixes, proper names and currencies exactly.
- Confirm terminology and consistency. Does it match the organization/client's list of terms and previous translations? Is the same term translated the same throughout the text?
- Filter it through meaning, tone and culture. Is the sentence natural in the target language? Is the tone right? Is it culturally appropriate? Your expert judgment is the final filter.
three mini cases
Case 1 — MTPE saved time. One translator translated a 24,000-word user manual by post-editing NMT output rather than from scratch. The work, which was estimated to take 8 days from scratch, was reduced to 3.5 days thanks to the large number of repetitions. But every security warning sentence has been revised one by one.
Case 2 — Validation caught a semantic error. An expert had AI translate a contract clause. While the source said "the seller shall NOT be liable", the output said "the seller shall NOT be liable"; negativity had decreased. The comparison with the source step prevented a dangerous mistake that would have reversed liability.
Case 3 — Return from privacy breach. A new translator pasted a patient's medical report onto a public translation site for speed. His senior colleague realized: it was personal health data and it was out. The incident was reported as a privacy breach; the job was redone only on the corporate secure vehicle.
Four copyable templates
1) Job suitability assessment:
Your role: senior translation project manager. I will describe the job below. Tell me (1) whether this is draft/term work that can be delegated to the AI, or critical work that requires meaning, tone, and accountability, (2) the potential cost of incorrect output, (3) the verification I need to do before and after delegation. Job: [insert job here]
2) Request for translation faithful to the source:
Translate the following text from [source language] to [target language].Rule: do not add or remove any information. Preserve number, date, proper name and negations verbatim. Mark the term you are not sure about in [square brackets]. Do not make up a sentence that is not in the source. Text: [write text here]
3) Privacy pre-check:
Evaluate the following text BEFORE processing: could it contain personal data, confidential business information or unpublished content? If so, which parts? Warn me if it should not be processed in a public means.Text: [insert text here]
4) Tone and culture filter:
Evaluate the following translation in terms of naturalness, tone and cultural appropriateness in the target language. Mark the sentences that sound like a translation, carry a shadow of the source language (smell like a translation), and suggest a more natural alternative. Changing the meaning.Translation: [insert text here]
Weak prompt / Strong prompt
Weak: "Translate this to English." (No tone, no audience, no terms, no context; the machine produces a default translation, with a twist.)
Strong: "Translate this marketing email from Turkish to English. Audience: US corporate customers. Tone: professional but warm. Use these terms: 'subscription' → 'subscription', 'trial' → 'trial'. Keep numbers and dates. Produce natural English that doesn't smell like translation."
Difference: strong prompt gives direction, audience, tone, term and constraint; the output directly approximates the usable draft.
Roles and risk level table
business type
Role of AI
Risk
human verification
Draft technical guide
Produces raw translation
medium
Term + number check
Contract/law
Draft + term proposal
very high
Sentence by sentence, expert approval
medical text
draft
very high
Dose/negativity one to one
Marketing / slogan
idea multiplier
high
Creative, cultural approval
Summary of internal correspondence
Summary/draft
low
Quick review
Common mistakes
- Delivering raw machine output without verification. The most common and most dangerous mistake; "It looks smooth" doesn't mean it's true.
- Pasting confidential document into public tool. NDA breach and legal risk.
- Asking for a translation without giving context. Without mass, tone and term, the output will always be default and superficial.
- Not checking number, date and negativity. This is where MT's most insidious errors hide.
- Sacrificing quality just because "AI is fast". Speed is not a substitute for verification.
In summary
AI is a powerful assistant that greatly accelerates voluminous, repetitive and language-intensive tasks in translation and interpreting; but the responsibility for meaning, tone, culture and delivery lies with the competent translator. Internalize three things from the start: verifying each output against the source, never giving secret text to an unsecured tool, and positioning the AI as a draft generator, not a decision maker. The next units will be based on this discipline.
Application task
Select a text you have (or a sample) and follow these three steps: (1) Determine its risk level with the business suitability template. (2) Do a privacy pre-check and decide which tool you will be working with. (3) Get a draft with a faithful translation template, then manually perform three-step verification (source-term-tone) and note any errors you find.
checklist
- [ ] I determined the risk level of the job (low/medium/high).
- [ ] I checked if the text is hidden and selected the correct tool.
- [ ] I gave the translation mass, tone and terminological context.
- [ ] I compared the number, date, proper name and negations with the source.
- [ ] I left the responsibility for the final delivery to myself, not the AI.