Immersive Translate

Renders original and translated text in a vertical split-pane layout on web pages, using DOM parsing to identify main content areas and paragraph boundaries. Detects semantic paragraph units rather than line breaks, preserving context for accurate translation. The extension intercepts page rendering, injects translated content alongside originals, and applies CSS-based layout adjustments to maintain readability without disrupting page structure.

Abstracts translation requests across 20+ backend services (DeepL, OpenAI, Google Translate, Microsoft, Tencent, Claude, Gemini, etc.) through a unified API interface. Routes requests to user-selected primary service, with automatic fallback to secondary services if rate limits or API errors occur. Manages API key configuration, request queuing, and response caching to minimize redundant API calls across the same page content.

Implements privacy-first translation architecture where translation requests are encrypted before transmission to backend services, and translated content is not retained on extension servers or used for model training. Supports optional local-only translation mode (if using local models) to avoid sending content to cloud services. Provides transparency reports on data handling and compliance with GDPR, CCPA, and other privacy regulations.

Tracks translation quality metrics (user satisfaction, correction frequency, service performance) and adapts translation service selection based on historical performance. Provides confidence scores for translations (if supported by service) and allows users to flag low-quality translations, which feed back into service selection algorithm. Maintains per-service quality metrics (accuracy, latency, language pair coverage) to optimize future routing decisions.

Supports batch translation of multiple documents or content blocks with automatic scheduling to respect API rate limits and quota constraints. Queues translation requests, distributes them across available translation services, and manages concurrent requests to avoid hitting rate limits. Provides progress tracking, retry logic for failed requests, and estimated completion time. Useful for translating large document collections or bulk content without manual intervention.

Analyzes webpage DOM structure using heuristics (text density, semantic HTML tags, visual layout) to identify main content areas and exclude navigation, advertisements, sidebars, and metadata from translation. Implements machine learning-based content detection (if available) to improve accuracy on complex layouts, with user override capability to manually mark content areas for translation or exclusion.

Processes PDF, ePub, DOCX, and Markdown files by extracting text content while preserving original formatting, fonts, and page layout. For scanned PDFs without embedded text, applies OCR (Optical Character Recognition) to extract text from images before translation. Exports translated documents in original format with side-by-side bilingual layout or translation-only mode, maintaining column structure, headers, footers, and page breaks.

Extracts subtitle tracks from video platforms (YouTube, Netflix, etc.) by intercepting WebVTT or SRT subtitle APIs, translates subtitle text while preserving timing codes and speaker labels, and re-injects translated subtitles into the video player. Supports both hardcoded subtitles (burned-in text) via OCR and soft subtitles (extracted tracks). Maintains synchronization between original and translated subtitles with optional dual-subtitle display.

Detects and intercepts text input fields (textarea, input[type=text], contenteditable divs) on web pages, providing real-time translation suggestions as users type. Maintains context awareness by analyzing surrounding page content and form labels to infer intent (e.g., translating a search query vs. a message). Supports both inline translation display (below input field) and side-by-side comparison, with one-click insertion of translated text into the field.

Detects images containing text on web pages, applies OCR to extract text, translates extracted text, and overlays or replaces translated text on the original image while preserving visual layout and styling. Supports both static images and dynamically loaded images (lazy-loaded, infinite scroll). Maintains image aspect ratio and text positioning, with optional original text preservation via tooltip or side-by-side display.

Maintains a curated database of website-specific translation rules and optimizations for major platforms (Google, Twitter, Reddit, YouTube, Bloomberg, Wall Street Journal, etc.). Applies custom CSS selectors, content detection heuristics, and layout adjustments tailored to each platform's DOM structure, improving translation accuracy and layout preservation. Rules include main content area detection, navigation element exclusion, and platform-specific formatting (e.g., tweet threading, comment nesting).

Implements a hover interaction model where moving the mouse over a paragraph triggers translation display below or adjacent to the original text. Uses event delegation to detect hover on paragraph elements, fetches translation for the hovered paragraph, and renders translated text in a styled overlay or inline element. Supports customizable display styles (background color, font size, opacity) and positioning (below, beside, tooltip). Caches translations to avoid redundant API calls on repeated hovers.

Syncs user preferences, translation history, and custom settings across multiple devices and browsers (Chrome, Firefox, Safari, mobile apps) via cloud backend. Stores user configuration (preferred translation service, display style, website-specific settings) in encrypted cloud storage, enabling seamless experience when switching devices. Maintains translation history for quick re-access and learning purposes, with optional export/import functionality.

Validates requested language pairs against each translation service's supported language matrix before routing requests. Detects source and target language from user input or page metadata, checks coverage against selected translation service, and either routes to service or falls back to alternative service with broader language support. Provides user feedback on language pair availability and suggests alternative services if primary service doesn't support the pair.