Translators face a specific version of the AI quality problem: AI language models generate fluent, plausible text in the target language, which means AI-generated translation errors can look correct to anyone who isn't deeply familiar with the source text's domain, the target language's professional register, or the specific client's terminology preferences. The errors that AI makes in translation — subtle terminological inaccuracies, register violations, calques from the source language structure, culturally inappropriate expressions — are precisely the errors that require professional translation expertise to catch.
The professional question is not whether to use AI in translation work but which parts of the translation workflow benefit from AI assistance and which parts require the professional expertise that AI currently cannot replicate. The answer is specific and consequential.
Where AI Provides Genuine Leverage for Translators
Terminology Orientation and Initial Research
When encountering unfamiliar specialized terminology, AI can provide useful orientation: what a technical term generally means, how it's discussed in the field, what related terms exist. This orientation helps the translator understand the source text concept before conducting authoritative terminology research.
High-reliability orientation uses:
- Understanding what a specialized concept means before researching its target language equivalent
- Getting an initial sense of the terminology landscape in an unfamiliar domain
- Identifying the vocabulary cluster around a concept to guide authoritative research
- Understanding the context in which a term appears in the source language professional community
The critical discipline: AI orientation is a starting point, not a translation decision. The target language term used in a professional translation comes from authoritative sources — professional glossaries, regulatory documents, professional association terminology databases — not from AI. AI suggests where to look; authoritative sources provide the term.
Background Research on Source Text Topics
When translating about topics outside the translator's primary subject areas, AI can provide useful background understanding of the subject matter: how a process works, what an event was, what a concept means in general terms.
High-reliability background research uses:
- Understanding the general context of a document's subject matter
- Clarifying how a described process or system works
- Getting background on historical events, institutional structures, or technical processes described in the source text
- Understanding how concepts in the source text relate to each other
The verification requirement: background research from AI should be treated as orientation, not as the ground truth for accurate translation. For subject matter where precision matters — legal, medical, technical, financial — background understanding from AI should be verified against authoritative domain sources before being used to inform translation decisions.
Target Language Editing and Proofreading Support
AI can be useful for reviewing target language text for obvious fluency issues, awkward constructions, and grammatical errors — particularly for translators who are highly fluent but not native speakers of the target language.
High-reliability editing uses:
- Flagging grammatically unusual constructions for review
- Identifying obviously awkward phrasing that needs attention
- Checking for consistency in article and preposition use in languages where these are complex
- Initial proofreading for mechanical errors before professional review
The discipline: AI editing catches some fluency issues but misses the register violations, professional community collocations, and specialized terminology errors that require professional expertise to identify. AI editing does not replace professional translation review; it supports it by catching mechanical issues that free professional review to focus on substantive quality.
Administrative and Business Writing
For the administrative and business side of a translation practice — client communications, project proposals, invoices, professional development materials — AI can draft prose from provided information.
High-reliability administrative uses:
- Drafting client email responses from bullet points
- Generating project proposal structures from provided scope information
- Drafting professional development summaries
- Creating template documents for standard client communications
Where AI Introduces Risk for Translators
Terminological Accuracy in Specialized Domains
AI-generated target language text for specialized domains will contain terminology that sounds plausible but may be incorrect, outdated, non-standard, or inconsistent with the translator's client terminology.
The problem: AI's training data for specialized terminology reflects the distribution of usage in the training corpus — which includes informal, incorrect, and outdated usage alongside authoritative usage, with no clear signals about which is which. An AI that generates "apprentissage machine" instead of the preferred "apprentissage automatique" for a French client who has specified their preference, or that uses outdated regulatory terminology that was replaced three years ago, is producing text that looks correct but is professionally wrong.
Rule: all terminology in professional translation comes from authoritative sources or verified client preferences, not from AI generation. AI-suggested terminology requires verification against authoritative domain glossaries and client terminology preferences before use.
Register and Professional Community Collocations
AI-generated text in professional registers often sounds slightly unnatural — technically correct but not the way professionals in the target language actually write in that context. The collocations that native professionals use, the syntactic patterns that are characteristic of the register, the specific phrasings that mark a text as having been produced by a professional in the field rather than translated — these are precisely the areas where AI most commonly falls short.
The problem: a contract that sounds like a translation rather than like a contract drafted in the target language; a medical report that uses technically correct but slightly unusual collocations; a financial document where the phrasing is comprehensible but not natural to professionals in the target market. These register failures are visible to professional readers and undermine the quality of the translation for its intended audience.
Rule: target language professional register quality requires the translator's professional judgment and knowledge of how the target language's professional community actually writes. AI can produce fluent text; it cannot reliably produce text that sounds like it was written by an insider to the professional community.
Culturally Specific Expressions and Adapted Content
When source text contains culturally specific references, idioms, or expressions that require cultural adaptation rather than direct translation, AI adaptation is unreliable.
The problem: cultural adaptation in translation requires genuine understanding of how both cultures work — what resonates with the target audience, what carries unintended connotations, what references require explanation or substitution for the target cultural context. AI approaches this with pattern matching from training data, which produces plausible-looking cultural adaptations that may miss the actual cultural nuance.
Subject Matter Precision
For specialized content where terminological and conceptual precision matters — legal, medical, regulatory, technical — AI-generated content cannot be trusted for precision. The AI that generates a technically fluent translation of a pharmaceutical regulatory submission may use the wrong regulatory terminology, apply concepts inappropriately, or translate a legally significant term in a way that changes its meaning.
Rule: translation of specialized content requires professional expertise in the subject area, not just fluency in both languages. AI can assist with orientation and drafting; it cannot substitute for subject matter expertise.
The Translator's AI Quality Check
For any AI-assisted output in professional translation work:
Terminology check: has every specialized term been verified against an authoritative source or client terminology record? (Not assumed to be correct because AI used it.)
Register check: does the text sound like it was written by a professional in the target language, in this specific document type? (Not just grammatically correct, but professionally idiomatic.)
Calque check: are there constructions that follow the source language structure rather than target language idiom? (AI translation is particularly prone to calques.)
Accuracy check: does the target language text convey the same meaning as the source text — including the specific legal, technical, medical, or financial meaning, not just the general sense?
These four checks are the professional quality discipline that makes AI-assisted translation professionally reliable.
A Worked Example: AI Used Well in Technical Translation
Situation: a translator is working on a software documentation project in an area where they haven't worked before (cloud infrastructure security).
Good use of AI:
Concept orientation: use AI to get a background understanding of cloud security concepts before beginning research — what "zero trust architecture" means, what "defense in depth" refers to, how "IAM" (Identity and Access Management) works. This orientation makes the subsequent terminology research more efficient.
Document structure: use AI to draft the email to the client confirming project scope, timeline, and format requirements.
Where AI doesn't substitute for professional work:
Terminology: all cloud security terminology in the translation comes from authoritative sources — official documentation, NIST standards, professional industry bodies — and is cross-checked against any client terminology preferences.
Technical accuracy: every instruction, specification, or technical claim in the translated documentation is verified to accurately convey the source text's technical meaning, not just its surface meaning.
Register: the final text is reviewed for whether it sounds like professional technical documentation in the target language, not just a translation of documentation.
Key Takeaways
-
AI orientation in specialized domains requires authoritative verification before use in translation. AI knows what terms exist and roughly what they mean; it doesn't know which is authoritative, current, or consistent with client preferences.
-
Register is the quality dimension where AI most consistently falls short. Professionally idiomatic collocations, characteristic syntactic patterns, and the markers that identify a text as produced by an insider to the professional community require the translator's professional judgment.
-
AI translation is particularly prone to calques. Structures that follow the source language syntax but are unnatural in the target language are a specific AI failure mode that professional translation review needs to catch.
-
Cultural adaptation requires genuine cultural understanding. AI pattern-matching from training data doesn't reliably produce accurate cultural adaptation; professional translators with genuine bicultural expertise are required.
-
The four-question AI quality check applies to all AI-assisted translation output. Terminology verified, register appropriate, calques caught, meaning accurate — these are the professional quality disciplines that make AI assistance reliable in professional translation work.
Conclusion
AI tools provide genuine leverage for translators in terminology orientation (starting point for authoritative research, not the research result), subject matter background research (orientation for unfamiliar domains, not precision for specialized content), target language editing support (fluency and mechanical checks that support professional review), and administrative writing. They introduce professional risk in terminological accuracy for specialized domains (AI terminology requires authoritative verification), professional register and collocations (AI produces plausible but often not professionally idiomatic text), cultural adaptation (pattern matching doesn't reliably capture bicultural nuance), and subject matter precision (AI fluency without subject matter expertise creates professional-quality-looking translation errors). The four-question quality check — terminology verified, register appropriate, calques caught, meaning accurate — is the professional discipline that makes AI assistance a quality-positive addition to professional translation work rather than a quality risk.
To go deeper, check out Clip Articles for Later Reading.