Contact emails [email protected]
Explainer https://github.com/webmachinelearning/translation-api Specification https://webmachinelearning.github.io/translation-api/#language-detector-api Summary Exposes the LanguageDetector JavaScript Web API on Chrome for Android (Clank). This allows web applications to detect text languages locally on mobile devices using fast, privacy-preserving, and offline-capable on-device machine learning models. Blink component Blink > AI > Translator & Language Detector <https://issues.chromium.org/issues?q=customfield1222907:%22Blink%20%3E%20AI%20%3E%20Translator%20%26%20Language%20Detector%22> Web Feature ID languagedetector <https://webstatus.dev/features/languagedetector> Motivation Exposing the Language Detector API on Chrome for Android allows web applications to perform fast, privacy-preserving, and offline-capable language identification directly on-device. Risks Interoperability and Compatibility Low. This brings Chrome for Android to parity with Chrome Desktop (shipped in M138) using the exact same WebIDL surface and Blink implementation. Microsoft Edge also supports the API <https://learn.microsoft.com/en-us/microsoft-edge/web-platform/languagedetector-api> with an independently-developed model and inference stack. Gecko: Negative (https://github.com/mozilla/standards-positions/issues/1015) WebKit: No signal (https://github.com/WebKit/standards-positions/issues/339) Web developers: Positive Usage & Momentum: The Language Detector API is one of the most widely used Built-in AI APIs, often used prior to translation, summarization, or content routing. - Since shipping on desktop in M138, LanguageDetector_Detect is called on ~0.23% of Chrome page loads <https://chromestatus.com/metrics/feature/timeline/popularity/5403>—comparable to established APIs like navigator.canShare (~0.23%) and approaching the Badging API (~0.31%). - Received positive feedback in the WebML CG <https://github.com/webmachinelearning/translation-api/issues/36> and WICG <https://github.com/WICG/proposals/issues/147>. Production Adoption: Telemetry and open-source codebases show active usage across several areas: - E-Commerce & Enterprise: Used by large storefronts and merchant plugins to detect the language of user reviews, search queries, and seller content. It is also used in customer support and CRM tools (e.g., hospitality, real estate, and delivery dispatch) to help agents triage and reply to multilingual inquiries in real time. - Browser Extensions: Adopted by popular translation extensions to handle live chats, social feeds, and sites built on frameworks incompatible with Chrome's built-in page translation. Running detection on-device lets extensions identify languages per paragraph or comment without network latency or API costs, serving as a web-standard alternative to chrome.i18n.detectLanguage. - Open Source: Over 140 public references on GitHub (including SAP UI5, PWAFire, and Phanpy), typically used as a fast, zero-cost check before triggering heavier tasks. Other signals: Ergonomics Inference is typically quite fast, and existing browser feature usage of the same model and inference stack does not pose performance concerns on mobile. Real-world telemetry shows developer inputs are heavily skewed toward short strings (median ~12 characters), making inference lightweight for mobile CPU and battery budgets. This API is frequently used in concert with the Translator API, and work is in progress to bring the Translator API to Android as well (https://chromestatus.com/feature/5106126157905920). Activation Developers can use LanguageDetector.availability() for feature detection and fall back to cloud services or WebAssembly/WebGPU polyfills when unavailable. Security No new security or privacy risks beyond Desktop (M138). Chrome for Android reuses the existing TFLite model already loaded in renderer processes for built-in page translation. Inference is stateless across calls and runs entirely locally inside the sandboxed renderer process over a read-only model file, so partial site isolation on Android does not introduce cross-origin state leakage. Mobile fit - Our telemetry shows that real-world developer queries are heavily concentrated on short strings (p50 median: ~12 characters, p75: ~28–37 characters). - These input sizes are compatible with typical mobile web patterns: search queries, virtual keyboard inputs, live stream chat, and short messaging threads. - Built-in AI support for Android comes up regularly in our discussions with partners and the community, especially with regards to the Language Detector and Translator APIs. - While desktop users frequently rely on browser extensions for cross-lingual assistance, extension support is missing on many mobile browsers; exposing this API on Android is therefore critical in that it allows mobile web apps to deliver on-device linguistic accessibility at no extra cost. WebView application risks Does this intent deprecate or change behavior of existing APIs, such that it has potentially high risk for Android WebView-based applications? No. blink::features::kLanguageDetectionAPI is explicitly disabled in Android WebView (aw_field_trials.cc) because WebView does not bundle or fetch the Optimization Guide language detection model. Debuggability Standard DevTools JavaScript debugging suffices: developers can call and inspect LanguageDetector.availability(), create(), detect(), downloadprogress events, and DOMExceptions directly in the Console and Sources panel, matching the Desktop launch. Will this feature be supported on all six Blink platforms (Windows, Mac, Linux, ChromeOS, Android, and Android WebView)? No — Supported on Windows, Mac, Linux, ChromeOS (shipped in M138) and Chrome for Android (shipping in M157). Not supported on Android WebView. Is this feature fully tested by web-platform-tests <https://chromium.googlesource.com/chromium/src/+/main/docs/testing/web_platform_tests.md> ? Yes — Covered by existing WPTs under external/wpt/ai/language_detection/ as well as cross-platform LanguageDetectionModelServiceBrowserTest. Finch feature name LanguageDetectionAPI Rollout plan Will ship enabled for all users Requires code in //chrome? Yes, //chrome/browser/language_detection/ wires ChromeContentBrowserClient::BindLanguageDetectionDriver and OptimizationGuideKeyedService to deliver the language detection model to the renderer. Tracking bug https://crbug.com/549896512 Launch bug https://launch.corp.google.com/launch/4508966 Measurement WebFeature::kLanguageDetector_Create, WebFeature::kLanguageDetector_Availability, WebFeature::kLanguageDetector_Detect, AI.LanguageDetector.AvailabilityV2, AI.Session.LanguageDetector.PromptResponseStatus, AI.Session.LanguageDetector.ResponseCompleteTime, AI.Session.LanguageDetector.PromptRequestSize Availability expectation Available in Chrome on Desktop (M138+) and Chrome for Android (M157+). Adoption plan Update Chrome developer documentation ( https://developer.chrome.com/docs/ai/language-detection) and MDN browser compatibility data to reflect Chrome for Android support in M157. Non-OSS dependencies This feature depends on a language detection model provided to the open-source implementation via //components/optimization_guide. Estimated milestones Platform Milestone Shipping on Android 157 Anticipated spec changes None Link to entry on the Chrome Platform Status https://chromestatus.com/feature/5087964794322944 Links to previous Intent discussions Intent to Prototype (Android): https://groups.google.com/a/chromium.org/g/blink-dev/c/saJHYqHHPSY Intent to Ship (Desktop M138): https://groups.google.com/a/chromium.org/g/blink-dev/c/sWcHBe9wpbo -- You received this message because you are subscribed to the Google Groups "blink-dev" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CANA9qLJXBxFgrd_Mw%3Dy1EyjcBzafkNpkU5d%3DuhHQ44vOXeaieA%40mail.gmail.com.
