Scope, delivery and quality.
Language technology improves performance when controlled by clear data governance, terminology ownership, quality sampling, and responsible human review. Translation memory supports consistent reuse; termbases protect approved terminology; AI can assist specific tasks only when risk appropriate. The current profile mentions translation memory and human-in-the-loop tools, while MTPE, annotation, and Arabic LLM evaluation appear principally as expansion opportunities.
Use of an AI tool does not establish conformance with ISO 18587, which specifically addresses full human post-editing of machine-translation output. Confidential source material should not be transmitted to unapproved providers. Claims such as zero data retention require contractual evidence, not a marketing assumption.
Service components & deliverables
Operate client-specific translation memories
Output linkage: Client-owned translation memory and glossary. Quality check: Data classification plus processing agreements.
Govern terminology databases and change requests
Output linkage: Documented AI use and disclosure statement. Quality check: Mandatory technical human sign-off.
Deliver MT post-editing in authorized scope
Output linkage: MTPE quality evaluation and correction log. Quality check: Translation memory quality checks.
Apply human-in-the-loop scientific review
Output linkage: Model evaluation rubric and error taxonomy. Quality check: Termbase ownership and controlled approvals.
Create and annotate Arabic training datasets
Output linkage: Vendor and privacy risk review. Quality check: Independent blinded evaluation dataset.
Evaluate Arabic and domain-specific language models
Output linkage: Human sign-off and release audit trail. Quality check: Source provenance and auditability record.
Automate numerical and terminology QA checks
Output linkage: Client-owned translation memory and glossary. Quality check: Data classification plus processing agreements.
Assess AI vendor privacy and security risk
Output linkage: Documented AI use and disclosure statement. Quality check: Mandatory technical human sign-off.
Typical inputs and outputs
Client AI usage policy and permissions
Client-owned translation memory and glossary
Existing TMX, TBX, and termbase assets
Documented AI use and disclosure statement
Information security and privacy classification
MTPE quality evaluation and correction log
Defined quality and performance rubrics
Model evaluation rubric and error taxonomy
Retention, deletion, and subcontractor terms
Vendor and privacy risk review
Approved model evaluation data where relevant
Human sign-off and release audit trail
Illustrative delivery workflow
Classify content sensitivity and permitted tasks
Input: Client AI usage policy and permissions. Check: Data classification plus processing agreements.
Approve tools, vendors, and conditions
Input: Existing TMX, TBX, and termbase assets. Check: Mandatory technical human sign-off.
Minimize or anonymize data where required
Input: Information security and privacy classification. Check: Translation memory quality checks.
Apply limited automation within approved boundaries
Input: Defined quality and performance rubrics. Check: Termbase ownership and controlled approvals.
Verify results by qualified human reviewers
Input: Retention, deletion, and subcontractor terms. Check: Independent blinded evaluation dataset.
Archive or delete assets according to contract
Input: Approved model evaluation data where relevant. Check: Source provenance and auditability record.
Quality and risk controls
Confidential client data exposed to a third party
Data classification plus processing agreements
Hallucinated clinical or legal statements
Mandatory technical human sign-off
Erroneous phrases propagated through TM
Translation memory quality checks
Terminology applied in the wrong context
Termbase ownership and controlled approvals
Biased Arabic model evaluation outcomes
Independent blinded evaluation dataset
Poor provenance for generated explanations
Source provenance and auditability record

