Free tools Windows power users keep installed
One-click scans. No signup required.
AI translation terminology is easier to use when teams separate three questions: how a translation is generated, which terms it should use, and how people judge the result. Neural machine translation (NMT) and large language models (LLMs) are different approaches; a glossary or term base helps standardize approved wording; and fluency alone does not show that a translation preserves the source meaning. This glossary gives support leaders, agents, operations teams, and localization specialists a shared vocabulary for choosing and reviewing AI-assisted translation.
How to use this glossary
The definitions below are practical working explanations, not claims that every term has one universally accepted definition. Standards and vendor guidance have narrower scopes: ISO 12616-1:2021 addresses translation-oriented terminology collections, while ISO 5060:2024 addresses evaluation of human translation, post-edited machine translation, and unedited machine translation. Microsoft Learn discusses NMT and LLM translation as approaches, including relevant trade-offs. These sources inform the distinctions here; they do not establish one universal taxonomy or prescribe a complete support-team workflow.
As an Amazon Associate I earn from qualifying purchases.
For everyday decisions, first identify the translation approach, then the terminology resources and locale requirements, and finally the review method. Keep the source text (the original) and target text (the translated result) distinct when discussing omissions, additions, or meaning changes.
What does NMT mean?
Neural machine translation (NMT)
NMT means neural machine translation: machine translation produced using neural-network methods. Microsoft describes NMT as an approach used by current translation applications, including Microsoft Translator, and says NMT systems are designed specifically for translation. NMT is one kind of machine translation, not a synonym for every automated translation system.
#1 Best Overall
Machine translation (MT)
Machine translation means translation produced by a computer system. It is the broad category: systems may use NMT or an LLM-based workflow, among other approaches. Use “machine translation” when the distinction between methods does not matter; name the method when discussing terminology controls, review, or system behavior.
Large language model (LLM)
A large language model is a general-purpose language model that can be prompted to translate as well as perform other language tasks. Microsoft contrasts LLMs with NMT systems designed specifically for translation. Its guidance identifies potential trade-offs involving speed, cost, terminology, and fabrication risk; these are considerations, not a universal benchmark for all models, languages, or implementations.
Machine translation versus LLM-assisted translation
The terms are not opposites: LLM-assisted translation is a form of machine-assisted translation. “NMT” names a translation approach, while “LLM” names a broader model type that can be applied to translation. In practice, a support team should assess the particular system and workflow rather than assume that one label guarantees better quality.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstall| Question | NMT | LLM-assisted translation |
|---|---|---|
| What is it designed to do? | Microsoft describes NMT as optimized specifically for translation. | LLMs are general language models that can be used for translation and other language tasks. |
| How might terminology be applied? | Microsoft says existing glossaries and term bases can be easier to integrate. | Integration may be harder, depending on the implementation. |
| What should reviewers watch for? | Check meaning, omissions, and approved terminology. | Check meaning and terminology, and also whether the output adds content absent from the source. |
| What affects suitability? | Language pair, domain, customization, and terminology controls. | Language pair, task flexibility, cost, latency, and human review. |
This is a qualitative comparison based on Microsoft Learn guidance, not a product ranking or a numerical performance comparison. Results depend on language, domain, model, and workflow.
How do we keep translated support terms consistent?
Glossary and term base
A glossary or term base is a maintained collection of terminology and associated information used to encourage consistent language. For support teams, entries can identify an approved target-language term and clarify its intended meaning or use. The key is that the resource is maintained and applied in the translation workflow; a list that is outdated or disconnected from the work cannot reliably guide every result.
ISO 12616-1:2021 covers fundamentals and recommendations for producing sound bilingual or multilingual terminology collections. Its scope includes translation-oriented terminography—the work of setting goals, collecting, researching, documenting, using, and maintaining terminology data. It is a terminology-work standard, not a guarantee that a particular translation system will apply every entry correctly.
Terminology management and terminography
Terminology management is the ongoing work of establishing terminology goals, gathering and researching terms, documenting decisions, applying the data, and keeping it current. Terminography refers to the systematic work of compiling and managing terminology information. In a support operation, these activities can help prevent different agents or translated workflows from using inconsistent labels for the same product feature, account action, or policy concept.
Useful terminology-entry details
A terminology collection is more useful when it captures enough context to guide a real translation decision, not merely a source word and its proposed equivalent. Teams can document the approved form, the relevant language or locale, the meaning in the product or support context, and any distinctions that affect usage. This is a practical implementation suggestion; ISO 12616-1:2021 supports sound terminology collection practices but does not prescribe a particular support-system data schema.
Review and quality terms
Machine translation post-editing (MTPE)
Machine translation post-editing (MTPE) is human revision of machine-translated text. The human reviewer compares the target text with the source and corrects problems according to the intended use and required review depth. ISO 5060:2024 explicitly includes evaluation of post-edited machine translation output.
Post-editor
A post-editor is the person who reviews and corrects machine translation. The level of language expertise and review required should reflect the content’s consequences and the organization’s policy. ISO 5060:2024 addresses evaluator qualifications and competence, but it does not define a staffing model for support teams.
Translation quality evaluation
Translation quality evaluation means assessing output against defined criteria or error categories. ISO 5060:2024 describes an analytic approach that uses error types and penalty points to produce an error score and quality rating. It covers evaluation of human translation, post-edited machine translation, and unedited machine translation. A score is meaningful only in the context of the evaluation method and intended use; it should not be treated as a universal measure across unrelated systems or workflows.
Sampling
Sampling means reviewing selected translation outputs rather than every item. ISO 5060:2024 includes guidance relevant to evaluation and sampling. For a support team, sampled review can help identify recurring terminology or meaning problems over time, but sample results should not be confused with proof that every unreviewed translation is correct.
Fidelity risks and locale terms
Hallucination or fabrication in translation
A hallucination or fabrication in this context is text produced by an AI system that does not appear in the source. Microsoft warns that LLMs can introduce words or phrases absent from the source and that plausible-sounding output may still mislead. Reviewers should therefore check not only whether the target text reads naturally, but also whether it preserves the source’s meaning without unsupported additions.
Source text and target text
The source text is the original content submitted for translation. The target text is the resulting translation. Comparing them directly helps reviewers spot altered meaning, omitted conditions, or added instructions—issues that a fluent target text may conceal.
Localization
Localization is adapting content for a target locale, including language variety and context, rather than only converting words from one language to another. Microsoft discusses localization alongside translation and notes that distinctions such as Portugal Portuguese and Brazilian Portuguese can present challenges for LLMs, while NMT may be optimized for variants. Treat that comparison as Microsoft’s guidance, not a timeless rule: behavior varies by model, language pair, domain, and workflow.
A practical workflow for translated support content
The following is an adaptable operating practice, not an end-to-end workflow mandated by either ISO standard. ISO 12616-1:2021 supports disciplined terminology collection and maintenance; ISO 5060:2024 supports structured evaluation of translation outputs.
Best Value
- Define the use case and target locale. Decide what kind of support content is being translated and which language variety or locale the target text must serve. Distinguish routine conversational content from text with legal, safety, billing, identity, or account-access consequences.
- Maintain approved terminology with context. Record preferred terms and the context needed to use them consistently across relevant languages and variants. Assign responsibility for researching, documenting, using, and maintaining those entries.
- Select the translation approach for the task. Compare the language pair, domain, terminology controls, customization needs, latency, cost, and human-review plan. Microsoft’s guidance suggests NMT may make existing glossary integration easier, while LLMs offer broader language-task flexibility but can introduce unsupported content. The relevant trade-off depends on the implementation.
- Review target text against the source. Check whether the target preserves the source meaning, retains required terminology, and avoids omissions or additions. For high-consequence content, route work to qualified language review under the organization’s policy.
- Evaluate and sample over time. Use defined error categories or criteria, and review selected outputs periodically for recurring issues. Record findings in a way that can inform terminology maintenance, workflow changes, or reviewer guidance.
Is there one authoritative AI translation glossary for support teams?
The sources cited here establish useful adjacent foundations rather than one authoritative glossary dedicated to AI translation in customer support. ISO 12616-1:2021 addresses translation-oriented terminology collections; ISO 5060:2024 addresses translation evaluation; NIST and ITU publish glossaries relevant to trustworthy AI and machine-learning terminology. ITU-T Y Supplement 97 (2025), for example, compiles machine-learning definitions from ITU-T and other standards. These resources serve different purposes, so teams should define working terms for their own translation process instead of assuming every organization uses one settled vocabulary.
Frequently Asked Questions
Does fluent translated text mean the translation is accurate?
No. Natural-sounding wording does not prove fidelity to the source. Compare the target text with the source for meaning, omissions, terminology, and unsupported additions.
Is a term base the same thing as a translation memory?
They serve different purposes. A term base organizes approved terminology and related information; a translation memory stores previously translated text segments for reuse. The material here supports the terminology-resource distinction but does not specify how a particular translation platform implements either resource.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesShould support teams use NMT or an LLM?
There is no universal winner established by the cited guidance. The choice depends on the language pair, subject matter, terminology integration, task flexibility, latency and cost considerations, and the available review process.
What does ISO 5060:2024 cover?
It gives guidance on evaluating human translation output, post-edited machine translation output, and unedited machine translation output, including an analytic approach based on error types and penalty points.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
