Fluency API: Score the quality of your translations

Step 1. Create an API key

1. Sign in at platform.algebras.ai.

2. In the sidebar, open API Keys (under Organization).

3. Click Create new API Key, enter a label (for example, "QA Pipeline"), and click Create.

4. Copy the key right away. It's shown only once. If you lose it, you'll need to create a new one.

   

Step 2. Send a translation for scoring

curl -X POST "https://platform.algebras.ai/api/v1/translation/fluency" \
     -H "X-Api-Key: your_api_key_here" \
     -H "Content-Type: application/json" \
     -d '{
           "sourceLang": "en",
           "targetLang": "de",
           "sourceText": "That makes sense.",
           "translatedText": "Das macht Sinn."
         }'

To score up to 20 translations in one request, use /translation/fluency-batch.

Step 3. Read the result

{
  "fluency": 6.4,
  "idiomatic": 7,
  "collocational": 6,
  "discourse": 7,
  "pragmatic": 6,
  "calque": 6,
  "issue_type": "collocation",
  "severity": "minor",
  "problematic_phrase": "macht Sinn",
  "suggested_fix": "ergibt Sinn",
  "main_issue": "The phrase is a calque from English and unnatural in German."
}

• fluency is the average of five dimensions (idiomatic, collocational, discourse, pragmatic, calque), each scored from 1 to 10.

• issue_type, severity, problematic_phrase, suggested_fix, and main_issue tell you what's wrong and how to fix it.

Also available: GEMBA score

/translation/gemba scores a translation from 0 to 100, where 0 means no meaning is preserved and 100 means perfect meaning and grammar. You can optionally add a human reference translation to calibrate the score.

Pricing

Each evaluation (fluency or GEMBA) is billed as a flat 10 words, regardless of text length. In batch mode, 10 words per item.

Learn more

• Metrics documentation: all fields, batch mode, and code examples in Python, Node.js, and Go

• API reference: all endpoints