Glossary
Plain-English definitions of the translation, AI and language-data terms behind SourceTarget. Each entry explains one concept and links to a fuller definition.
A
AI translationAI and machine learning
Translation produced by artificial intelligence models, used as the first draft in a human-reviewed workflow.
AUSITCertification and standards
The Australian Institute of Interpreters and Translators, the profession's national association and ethics body.
ApostilleCertification and standards
An international certificate authenticating an official document for use in another Hague Convention country.
B
BLEU scoreQuality metrics
An automatic metric that rates machine translation by how much it overlaps with a human reference translation.
Back translationCore localisation
Translating content back into the source language to check that meaning was preserved.
C
CALDLanguage data and demographics
Culturally and linguistically diverse: the standard Australian term for communities that speak languages other than English.
CAT toolCore localisation
Computer-assisted translation software that supports a human translator, not a machine that translates for them.
COMET scoreQuality metrics
A neural quality metric that predicts how a human would rate a translation, rather than counting matching words.
Certified translationCertification and standards
A translation accepted as evidence by authorities, produced and stamped by a credentialled translator.
Community languageLanguage data and demographics
A language spoken by a migrant or diaspora community in a country where it is not the main language.
Confidence scoreAI and machine learning
A machine translation engine's own estimate, at generation time, of how reliable its output is.
Cultural adaptationCore localisation
Adjusting content so it fits the target community's context and beliefs, beyond literal translation.
Custom MT engineAI and machine learning
A machine translation engine adapted to a client's domain, terminology and past translations.
chrFQuality metrics
A quality metric that compares translations at the character level, fairer to languages with rich word forms.
F
First languageLanguage data and demographics
The language a person learned first and usually knows best, key to how information should reach them.
Fluency and adequacyQuality metrics
The two basic dimensions of translation quality: reads naturally, and preserves the source meaning.
H
HallucinationAI and machine learning
When an AI produces fluent, confident output that is not supported by its input, a key risk in machine translation.
Human-in-the-loopAI and machine learning
A workflow where AI produces the draft and a qualified person reviews it before it is published.
I
ISO 17100Certification and standards
The international standard for human translation services, requiring qualified translators and independent revision.
ISO 18587Certification and standards
The international standard for post-editing machine translation, defining light and full post-editing.
ISO 27001Certification and standards
The international standard for information security management, relevant when translating sensitive content.
Indigenous languagesLanguage data and demographics
The languages of First Nations peoples, many endangered, needing specialist and community-led translation.
L
LOTELanguage data and demographics
Languages Other Than English: the Australian education and policy term for community and second languages.
Language accessLanguage data and demographics
The principle and practice of making services usable by people who do not speak the dominant language.
Language pairCore localisation
The source and target languages of a translation, and the direction between them.
Language quality assuranceQuality metrics
The structured human review step that checks a translation against defined quality criteria before delivery.
Large language modelAI and machine learning
A general-purpose AI model trained on vast text, increasingly used for translation as well as other language tasks.
LocaleCore localisation
A specific combination of language and region that determines formats, conventions and wording.
LocalisationCore localisation
Adapting content for a specific locale: language, formats, references and cultural expectations.
Low-resource languageLanguage data and demographics
A language with little digital text available to train translation systems, which limits machine quality.
M
MQMQuality metrics
Multidimensional Quality Metrics: a standard framework for scoring translation quality by categorised human-marked errors.
MTPECore localisation
Machine translation post-editing: a human editor corrects machine output to publishable quality.
Machine translationAI and machine learning
Automatic translation of text by software, without a human translating each sentence.
N
NAATI certificationCertification and standards
Australia's national credential for translators, required by most authorities for official document translation.
Neural machine translationAI and machine learning
Machine translation built on neural networks, the current mainstream approach, more fluent than older methods.
Notarised translationCertification and standards
A translation where a notary public verifies the signer's identity, adding a layer some authorities require.
P
Parallel corpusLanguage data and demographics
A large collection of text paired with its translation, the raw material machine translation learns from.
Post-editingCore localisation
Human correction of machine translation output, either light (meaning only) or full (publishable).
Q
Quality estimationQuality metrics
Predicting translation quality without a human reference, so machine output can be judged on the fly.
Quality scoreQuality metrics
A composite 0 to 100 rating of a finished translation, combining several metrics weighted by content type.
R
Recognised translatorCertification and standards
A NAATI status for languages where certification testing is not available, recognising demonstrated practice.
Reference translationQuality metrics
A human translation used as the gold standard that automatic metrics compare machine output against.
S
Statistical machine translationAI and machine learning
An older machine translation approach that assembled translations from phrase probabilities, now largely superseded.
Sworn translationCertification and standards
A translation given legal force by a court-authorised translator, used in many civil-law countries.
T
TERQuality metrics
Translation edit rate: how many edits it would take to turn machine output into the reference translation.
TermbaseCore localisation
A curated database of approved terms and their translations, used to keep terminology consistent.
TranscreationCore localisation
Recreating a message for another language and culture so it lands with the same effect, not the same words.
Translation memoryCore localisation
A database of past translations reused to keep new work consistent and cheaper.