Confidentiality, Ethics, Copyright, Authenticity and the Future of the Profession — Artificial Intelligence in Translation and Interpretation

Gains:

Ability to apply the principles of confidentiality, NDA, personal data (KVKK/GDPR) and choosing the right tool to translation work
Ability to adopt transparency, accountability, copyright and originality responsibility and source verification reflex in machine use
Understanding why human verification is indispensable and the transformation of the profession in the age of artificial intelligence

We close this module not with a technical skill, but with principles that protect the integrity of the profession. AI gives great power to the translator; But this power also brings new and serious responsibilities regarding confidentiality, ethics, copyright and originality. One misshared confidential document, one concealed machine use, one copyright infringement, or one fraudulent "human translation" claim can destroy not just one business, but a career and customer trust. In this unit, you will learn about privacy, professional ethics, copyright and data issues, why human verification is indispensable, and the future of the profession in the AI age.

Privacy and data: immutable principles

Most documents that come into the hands of a translator are confidential, and confidentiality is a fundamental commitment of the profession. Four rules in the age of AI:

Entering confidential/personal/unpublished text into a public tool whose data policy does not provide assurance. The text you enter is processed on the external server; It can be used and stored in training the model.
Comply with NDA (non-disclosure agreement). Many jobs come with an NDA; Giving the text to an external tool is a direct violation and gives rise to legal liability.
Choose the right tool. Corporate, contracted tools that don't just "store/train data" for confidential work; general tools for public works.
Protect personal data (KVKK/GDPR scope) privately. Texts containing name, health, financial, and identity information require additional care; Anonymization or secure environment required.

KVKK (Personal Data Protection Law — Türkiye) and GDPR (Europe's data protection regulation) determine by law how personal data will be processed; violation is not only an ethical but also a legal problem.

Caution: "Just a quick translation, no one will see it" is the most common justification for privacy violations. Once out, information cannot be taken back. If in doubt, do not export the text to an external tool; choose the safe way.

Ethics and transparency: not hiding machine use

The new axis of professional ethics in the AI age is transparency. The customer has the right to know how his job is done. Two rules of honesty:

Hiding machine use. If you did a job with MTPE and the client expected "pure human translation", it is unethical to hide it. Pricing and expectations should be shaped accordingly.
State your limits honestly. It is risky to undertake a job outside your field of expertise (for example, advanced law or medicine) thinking that you can "handle it" with AI; AI does not close your knowledge gap.

Also accountability: the translator is responsible for every text delivered. "The AI translated it that way" is not an excuse; You are responsible to the customer.

Copyright, data and originality

Three separate but related topics:

Copyright: The text to be translated may have copyright; Translation is also a derivative work. Translating and publishing a work without permission is a violation. AI "rewriting" a text does not eliminate copyright issues.
Training data and consent: AI tools are trained with publicly available and sometimes copyrighted texts; This is a controversial ethical issue in the industry. Your contribution: do not "gift" your customer's data to training a model without permission.
Originality and hallucination: Especially in human/literary texts, AI can produce made-up quotes, fake sources, non-existent idioms. Every cultural/factual claim of the text you submit must be verified and authentic.

Tip: Set yourself up with a “source verification” reflex: When the AI gives a name, date, quote, source, or “known fact,” don’t put it into the text without independently verifying it. A fabricated reference in humanities destroys the credibility of the entire work.

Human verification: the indispensable last link

The common lesson of each unit of this module: AI accelerates, humans verify and bear responsibility. Verification is not an "extra step" but an integral part of the job. In high-risk texts (legal, medical, security, brand) verification is more important than the translation itself. An AI output delivered without human verification is fast but vulnerable.

The future of the profession: transformation, not displacement

AI does not make translators unemployed; It changes the nature of work. As routine, straightforward, repetitive translation tasks shift to the machine, the value of the translator shifts to layers that the machine cannot do, such as post-editing expertise, quality judgment, cultural depth, expert domain knowledge, transcreation and verification. Translator of the future; He is a person who uses AI skillfully, but knows where not to trust, and makes a difference with his field expertise and cultural judgment. Those who reject AI and those who blindly surrender to it will be left behind; The one who tames it as a tool comes first.

three mini cases

Case 1 — Transparency built trust. A translator explained to the customer upfront that he would do the job with MTPE, which parts would be human-intensive, and the price accordingly. The client was pleased with the transparency and provided regular business; If he had hidden it, trust would have been broken at the first problem.

Case 2 — Breach of confidentiality became career risk. A freelance translator entered a confidential contract into a public tool; Once the customer discovered the NDA violation, the business relationship ended and the reference was lost. A single convenience destroyed long-term trust.

Case 3 — Validation saved authenticity. In one academic translation, AI added a "citation" that was not in the source and a made-up author's name. The translator's source verification reflex caught this; If it were submitted, it would be a violation of academic integrity and loss of reputation.

Four copyable templates

1) Privacy/tool availability decision:

Your role: data privacy advisor.I will describe what a translation job involves. Tell me (1) whether this text is confidential/personal/copyrighted, (2) whether it is appropriate to process it in a public AI tool, (3) if not, what secure alternative I should use.Content: [...]

2) Draft transparency note to client:

Write a short, professional note to a client explaining honestly how I will do this job: machine translation + human post-editing, which parts will be human-intensive, how verification will be done, price/time accordingly. Don't exaggerate, give confidence. Type of work: [...]

3) Source/fact-checking list:

Omit all factual elements that need to be verified in the following translation: names, dates, numbers, quotations, source/attributions, "truth" claims, idioms/proverbs. List each one as "confirm with independent source". Don't trust the text, only what can be verified. Translation: [...]

4) Ethics/copyright pre-check:

I will describe the work below. Flag me possible ethical and copyright issues: copyright status of the text, derivative works of the translation, personal data, protection of customer data, need for transparency. List risks and precautions.Work: [...]

Weak prompt / Strong prompt

Weak: "Translate this document." (Hidden text can go to an external tool without questioning confidentiality, copyright or tool compatibility.)

Güçlü: "First evaluate this: this document is a patient report; it contains personal health data. Should it be processed in a public tool or does a secure/institutional environment be required? What should I pay attention to in terms of KVKK? Give the translation plan after determining the appropriate environment."

The difference: strong prompt starts with privacy and legal control; Translation comes after the correct and safe environment is determined.

Ethical dimensions table

Size

Risk

precaution

Privacy

Leaked document, NDA violation

Secure tool, data policy

personal data

KVKK/GDPR violation

Anonymize/secure environment

Unauthorized derivative work

Check permission/rights

transparency

Concealed machine use

Be open to the customer

originality

Hallucination, fake source

source verification

accountability

The "AI did it" excuse

Take responsibility

Common mistakes

Entering secret text into a public tool "to make it quick". Irrevocable violation.
Hiding machine usage from the customer. When trust is broken, it is difficult to recover.
Ignoring copyright and derivative works rights. Legal risk.
Putting the source/quote made up by AI without verifying it. Loss of originality and reputation.
Putting the responsibility on AI. The owner of the delivered work is the translator.

In summary

In the age of AI, the responsibility of the translator increases as well as his power. Privacy is inviolable: confidential, personal and copyrighted text does not enter the unsecured medium. Ethics requires transparency and accountability: machine use is not hidden, responsibility is embraced. Copyright and originality are taken seriously; Sources and phrases that the AI can make up are verified. Human verification is the indispensable last link of the work. The future of the profession belongs to the translator who uses AI skillfully but knows its limits and makes a difference with his cultural and expert judgment.

Application task

Do a full “responsible translation” check for a job at hand (or a sample): Identify the right tool with the “confidentiality/tool suitability decision” template; List risks with “ethics/copyright pre-check”; draft a “transparency memo to client”; and extract a “source/fact-checking list” in a sample translation and independently verify at least three items. Add these four steps to your own standard workflow.

checklist

[ ] I evaluated whether the text was confidential/personal/copyrighted and chose the right tool.
[ ] I paid additional attention to the data within the scope of KVKK/GDPR.
[ ] I was transparent to the customer regarding machine usage and process.
[ ] I have independently verified the sources, quotes and facts produced by YZ.
[ ] I left the responsibility of the delivered work to myself, not to the AI.

Module Exam

1. Which of the following is the most accurate positioning for artificial intelligence in translation and interpretation?

A) Artificial intelligence is a blueprint, terminology and quality tool; Responsibility for meaning, tone, culture and final delivery lies with the human being ✔
B) Artificial intelligence can directly deliver legal and medical translations without human approval
C) Artificial intelligence only works in translating words, it has nothing to do with quality and terminology
D) Since artificial intelligence is always more accurate than humans, all translation decisions should be left to it.

Description: Artificial intelligence; It is a translation draft, terminology, summary and quality aid. Responsibility for meaning, tone, culture, creativity and final delivery rests with the competent translator; An unverified output could lead to a contractual error, a medical malpractice, or a trademark crisis.

2. What does it mean if machine translation output is 'fluent but inaccurate' and why is it dangerous?

A) Fluent translation is always correct, so additional checking is unnecessary
B) Even if the sentence seems natural in the target language, it may have conveyed the meaning, number or negation incorrectly; Fluency does not guarantee accuracy ✔
C) Incorrect translations are always grammatically incorrect and are easily noticed.
D) Fluency is important only in literary texts, it is unimportant in technical texts

Explanation: Even if machine translation produces a perfect, natural sentence, it may have corrupted the number, negation, or meaning in the source. Fluency is no guarantee of accuracy; Therefore, numbers, dates, proper names and negativities should be checked separately with the source without being fooled by fluency.

3. Which of the following is the main difference between NMT (neural machine translation) and LLM-based translation?

A) NMT takes instruction but LLM cannot; so LLM only translates word for word
B) They are exactly the same, there is no practical difference between them
C) LLM understands instructions such as tone, terminology, and audience and sees the broader context; NMT is fast and consistent but poor at taking instructions ✔
D) LLM is always faster than NMT and never adds

Description: NMT is fast and consistent, but is poor at taking instructions and seeing the larger context. LLM, on the other hand, understands instructions such as tone, term list and audience and sees the context more broadly, but requires control against the risk of hallucination as it can make additions.

4. What is the difference between light post-editing and full post-editing?

A) Light post-editing has better quality, full post-editing is just a quick scan
B) There is no difference between the two, only their names are different
C) Full post-editing is used only on texts for internal use, light post-editing is used on texts to be published
D) Light post-editing makes the text accurate and understandable; Full post-editing aims for a human translation quality, publication-ready result ✔

Comment: In light post-editing, the goal is to make the text accurate and understandable; No stylistic touches, suitable for internal/temporary texts. In full post-editing, the goal is for the output to be indistinguishable from human translation; Required for texts with brand visibility that will be published.

5. Why is 'over-editing' a problem in MTPE?

A) Overcorrection always makes the translation more accurate, so it is encouraged
B) Rewriting correct and appropriate sentences unnecessarily destroys the time advantage of MTPE; Only faulty areas should be corrected ✔
C) Overcorrection only increases speed, does not affect quality at all
D) Overcorrection is a problem not only in machine translation but also in scratch translation

Explanation: Overcorrection is rewriting the correct and appropriate sentence given by the machine simply out of personal preference. This destroys the entire time advantage of MTPE. Rule: if the machine output is accurate, understandable and appropriate, do not touch it; Just fix it if it's wrong.

6. What is the main benefit of specifying a 'prohibited provision' (translation that should not be used) in a termbase?

A) It is unnecessary to specify a forbidden response because the machine never makes any mistakes anyway.
B) Forbidden equivalents are only useful in literary translation, they are unnecessary in technical text
C) Explicitly prohibiting wrong alternatives alongside the right counterpart prevents inconsistency and distortion of brand language ✔
D) Specifying a forbidden equivalent slows down the translation and provides no benefit.

Explanation: Simply writing the correct response does not prevent the use of incorrect but plausible-looking alternatives. Explicitly stating prohibited equivalents (using only 'account'; 'profile', 'membership' for 'account') is the most effective way to avoid inconsistency and broken brand language.

7. What should be written to translation memory (TM) and why?

A) Every machine output should be written to TM without verification because speed is paramount
B) No translations should be written to TM, only source texts should be kept
C) TMs of different customers should be mixed together into a single large TM
D) Only verified, certified translations should be written; If raw machine output enters TM, error will be carried over to future jobs and propagated ✔

Remark: Only verified, certified translations should be submitted to the TM. Saving the raw machine output to TM without validating it will cause the error to return over and over to future jobs, appearing to be trusted as a '100% match'. Garbage in, garbage out.

8. How should placeholders (e.g. {name}, %s) in software texts be handled in localization?

A) Placeholder is preserved as is; It is not translated, deleted, formatted, moved only if necessary and checked by QA ✔
B) Placeholders, like other words, must be translated into the target language
C) Placeholders should be deleted because they impair readability
D) Placeholders are unimportant, do not need to be taken into account at all in the translation

Description: Placeholders are signs that are filled with variables at run time. Translating, deleting, or formatting them will cause the software to crash or display raw text. Placeholder is preserved as is; only the placeholder can be moved according to the syntax of the target language and a placeholder QA tour is definitely done.

9. What does the CPS (characters per second) limit mean in subtitle translation?

A) It determines how many colors will be used, not how many seconds the subtitle will remain on the screen.
B) It is only valid for dubbing, it has nothing to do with subtitles
C) It is the speed limit of the subtitles so that the audience can read them easily; so subtitles require shortening while preserving the meaning ✔
D) The higher the CPS, the better the subtitles, there is no need to set limits

Description: CPS is the speed limit required for the viewer to read subtitles comfortably (usually ~17 characters/second for adult content). That's why subtitle translation is not a word-for-word translation, but a matter of fitting and shortening while preserving the meaning; The viewer both watches and reads the image.

10. Why shouldn't translation be started before the transcription produced by automatic speech recognition (ASR) is verified?

A) ASR is always foolproof so verification is a waste of time
B) ASR only makes grammatical errors, never distorts the meaning
C) ASR output is already translated, there is no need to translate further
D) ASR is especially mistaken in proper names, numbers, accents and noises; Incorrect transcription must be verified by voice as it will be carried over into translation ✔

Explanation: ASR makes frequent errors in noise, accent, overlapping speech, proper names, numbers, and technical terms. An incorrect transcription carries over directly to the translation and subtitling. Therefore, especially proper names, brands and numbers should be verified by comparing them with sounds.

11. What does an error framework like MQM provide in assessing translation quality?

A) It allows classifying each error by category and severity (critical/major/minor) and scoring the quality in an objective and comparable manner ✔
B) Automatically corrects the translation, completely eliminating human evaluation
C) It only catches number errors, it does not deal with semantic and terminological errors
D) Does not measure quality, only calculates the length of the translation

Description: MQM (multidimensional quality metrics) assigns each error to a category (accuracy, terminology, fluency, style, format) and a weight (critical/major/minor). This way, error scores can be given instead of 'good/bad' subjectivity, translators/engines are compared objectively, and the delivery decision can be tied to a numerical threshold.

12. What is the main limit of automatic translation metrics such as BLEU?

A) It makes human evaluation unnecessary because it understands the meaning better than humans
B) Looks at word overlap, doesn't really understand the meaning; so the system measures the trend but cannot decide the delivery of a single text ✔
C) He is always flawless and can alone decide the deliverability of a text
D) It is used only in oral translation, it has nothing to do with written translation.

Explanation: BLEU scores the translation based on word overlap with a human reference, but does not actually understand the meaning. A translation that is different from the reference but correct may receive a low score, while a translation that is similar to the reference but incorrect may receive a high score; A critical negativity error has little impact on the score. Therefore, the metric system measures the trend, not decides the delivery of a single text.

13. What is the strongest and most appropriate role for artificial intelligence in conference interpreting?

A) Eliminating the need for interpreters by fully undertaking live simultaneous translation
B) Automatically transferring the audio of confidential meetings to external services
C) Not being of any use at any stage of the translation, including preparation.
D) Strong in pre-event preparation (briefing, bilingual glossary, number and proper name cards); Live simultaneous translation is left to the human ✔

Description: In interpreting, much of the work comes in preparation, and that's where AI is strong: preparing field briefings, bilingual glossaries, speaker tone, number and proper name cards. Live simultaneous translation itself is the responsibility of the human translator due to delay, error, confidentiality and context.

14. How does transcreation (creative translation) differ from other types of translation?

A) It is fidelity to the emotional impact and purpose of the source, not to its words; One can move away from the source to create the same effect in the target culture ✔
B) Always translate the source word by word, without changing it
C) It is a type of literal translation used only in law and medical texts.
D) Completely ignore the meaning and write a new random text

Description: Transcreation, especially in advertising and brand texts, is to recreate the emotional impact and purpose of the source, not the words, in the target culture; Sometimes one deliberately moves away from the source because the goal is not literal fidelity but the same effect. AI is an idea multiplier here; The final cultural decision is made by humans.

15. What is the correct approach when translating a confidential document (for example, an unsigned contract or patient report)?

A) Enter the most practical public translation tool for speed, privacy is an afterthought
B) If the document will be translated anyway, confidentiality does not matter, any tool can be used
C) Public vehicles are not allowed; Only secure and corporate tools that do not store data are used, in case of doubt the text is not given to external tools ✔
D) There is no privacy risk if confidential documents are translated by voice only

Disclosure: Confidential, personal or unpublished texts are not entered into public tools whose data policy does not provide assurance; this is a NDA/privacy breach and legal risk in terms of KVKK/GDPR. Only corporate and secure tools that do not store/train data are used; If in doubt, the text is not given to the external tool.