إنتقل إلى المحتوى الرئيسي

Automatic translation

Automatic translation creates missing language variants after source content exists. Requests return quickly while background workers populate additional locales, and dashboard and guest surfaces rely on progress indicators instead of blocking until everything completes.

When translation runs​

Two patterns coexist:

  1. Placeholder completion — When new authored copy lacks an English variant yet, Scoutello may enqueue lightweight tasks that synthesize a starting English row before broader localization proceeds.

  2. Scoped translation — Teams trigger translation for a specific object (landing page, tour, tour step, event, PDF, and similar). Each scope fans out to the appropriate processor so domain-specific fields translate consistently.

Queue processing​

Jobs execute through shared workers that:

  • Identify source rows eligible for translation.
  • Honor flags that skip machine translation entirely.
  • Avoid duplicate target locales when they already exist.
  • Batch work with configured concurrency limits.
  • Record progress as batches finish.

The durable outcome is updated translation rows (and optional audio), not the transient progress meter itself.

Translation providers​

String batches are sent to managed translation models with conservative decoding settings and automatic retries when providers fail transiently. Translated rows keep metadata that shows whether a human or automation produced the text.

Specific vendor defaults can change between releases; quote capabilities—not vendor names—when speaking with customers.

Audio and narration​

Tour-heavy locales often pair translated copy with generated narration. Audio pipelines may lag behind text; separate progress messaging avoids implying silence equals failure.

Progress tracking​

Long jobs expose polling-friendly progress so dashboards and guest waiting screens can show percentages or staged statuses.

Progress records are temporary; only finished translations belong in publishing workflows.

Skipping translation​

Mark content that must remain untouched:

  • Brand names and proper nouns that should not drift across locales.
  • Codes, identifiers, or deliberately neutral tokens.
  • Legal clauses awaiting counsel review.
  • Mixed-language snippets already curated manually.

Operational notes​

  • Jobs should be safe to retry: existing locales must not duplicate endlessly.
  • Clients should poll progress endpoints rather than waiting on a single long HTTP request.
  • Manual edits after automation become the canonical text for that locale.

Related concepts: How translations work, Tours multilingual audio, Languages and localization.