Azure AI Translator

Azure AI Translator is a cloud-based, RESTful API service that provides real-time text translation between over 100 languages and dialects. It is part of the Azure Cognitive Services family, specifically under the Language services category. The service uses neural machine translation (NMT) technology, which offers high-quality, fluent translations compared to older statistical machine translation methods. NMT models consider the full context of a sentence to produce more accurate and natural-sounding translations.

Organizations operate globally and need to communicate in multiple languages. Building in-house translation systems is complex, expensive, and requires vast amounts of bilingual data. Azure AI Translator provides a pre-built, scalable solution that can be integrated into applications, websites, and workflows with minimal code. It enables scenarios like:

When you send a translation request, the following steps occur at a high level: 1. Request Reception: The API endpoint receives the text, source language (optional), target language(s), and optional parameters like profanity filtering or custom glossary. 2. Language Detection (if source not specified): The service uses a language detection model to identify the source language. The model is trained on large corpora and can detect over 100 languages. 3. Pre-processing: The text is cleaned and segmented into sentences or phrases. For HTML input, the service preserves tags and translates only text content. 4. Neural Translation: The segmented text is passed through a neural network model specific to the language pair. The model uses an encoder-decoder architecture with attention mechanisms to produce the translation. 5. Post-processing: The translated text is reassembled, and optional features like profanity masking or custom glossary terms are applied. 6. Response: The API returns a JSON object containing the translation(s), detected language (if requested), and alignment information.

- API Endpoint: https://api.cognitive.microsofttranslator.com/translate?api-version=3.0 - Authentication: Requires an Azure Cognitive Services resource key or a Translator resource key. Passed via Ocp-Apim-Subscription-Key header. - Region: For regional processing, use the regional endpoint (e.g., https://api.cognitive.microsofttranslator.com/translate?api-version=3.0&region=eastus). If using a global endpoint, region can be set via Ocp-Apim-Subscription-Region header. - Supported Languages: Over 100 languages. Full list available at GET /languages?api-version=3.0. - Text Length Limit: Each request can contain up to 50,000 characters across all text elements. Each individual text element (string) can be up to 10,000 characters. - Rate Limits: Varies by pricing tier. Free tier (F0): 2 million characters per month, 20 requests per minute. Standard tier (S1): up to 1 billion characters per month, 100 requests per minute. - Optional Parameters: - from: Source language (e.g., 'en'). If omitted, language detection is performed. - to: Target language(s). Can specify multiple, e.g., to=fr&to=de. - textType: 'plain' (default) or 'html'. - profanityAction: 'NoAction' (default), 'Marked', or 'Deleted'. - profanityMarker: 'Asterisk' (default) or 'Tag'. - includeAlignment: Boolean, returns alignment information mapping source to target words. - includeSentenceLength: Boolean, returns sentence boundaries. - Custom Glossary: You can provide a glossary as part of the request to enforce specific translations for terms. The glossary is an array of objects with text (source) and translation (target) fields.

To use Azure AI Translator, you must first create a Cognitive Services resource or a Translator resource in the Azure portal. The resource will provide an API key and endpoint. Below is a sample cURL command to perform a translation:

curl -X POST "https://api.cognitive.microsofttranslator.com/translate?api-version=3.0&from=en&to=fr" \
-H "Ocp-Apim-Subscription-Key: YOUR_KEY" \
-H "Content-Type: application/json" \
-d "[{ 'Text': 'Hello, how are you?' }]"

Response:

[
  {
    "translations": [
      {
        "text": "Bonjour, comment allez-vous ?",
        "to": "fr"
      }
    ]
  }
]

To detect language:

curl -X POST "https://api.cognitive.microsofttranslator.com/detect?api-version=3.0" \
-H "Ocp-Apim-Subscription-Key: YOUR_KEY" \
-H "Content-Type: application/json" \
-d "[{ 'Text': 'Bonjour' }]"

Response:

[
  {
    "language": "fr",
    "score": 1.0,
    "isTranslationSupported": true,
    "isTransliterationSupported": false
  }
]

For transliteration (converting text between scripts):

curl -X POST "https://api.cognitive.microsofttranslator.com/transliterate?api-version=3.0&language=ja&fromScript=Jpan&toScript=Latn" \
-H "Ocp-Apim-Subscription-Key: YOUR_KEY" \
-H "Content-Type: application/json" \
-d "[{ 'Text': 'こんにちは' }]"

Custom Translator is a feature that enables users to build translation models tailored to specific terminology and style. It requires a minimum of 10,000 parallel sentences (source-target pairs) in a .tmx or .xlf file. The training process involves:

Custom models are especially useful for companies with proprietary jargon. The exam may test whether you know that Custom Translator requires parallel data, and that it is used for domain-specific translation.

Translator supports multiple language variants, such as 'en-us', 'en-gb', 'pt-br', 'pt-pt'. It also supports transliteration for many languages, converting between scripts (e.g., Arabic to Latin, Japanese Kanji to Romaji). The language list is updated regularly; as of the latest, over 100 languages are supported. The exam may ask about the number of languages or specific language support features.

Azure AI Translator is SOC, ISO, HIPAA, and FedRAMP compliant when used with specific configurations. Data is encrypted in transit and at rest. By default, translation data is not stored; however, if you use Custom Translator, the training data is stored in the region of your resource. The service is available in multiple Azure regions, and you can choose a regional endpoint for data residency requirements.

The AI-900 exam will test your understanding of: