{
  "experiment": {
    "id": "opus-mt-arabic-benchmark-2026-03-28",
    "date": "2026-03-28",
    "domain": "nlp",
    "task": "translation",
    "models_tested": [
      "Helsinki-NLP/opus-mt-en-ar",
      "Helsinki-NLP/opus-mt-ar-en"
    ],
    "total_tests": 9
  },
  "results": [
    {
      "id": 1,
      "direction": "en-ar",
      "type": "formal",
      "input": "Hello, how are you today? I hope you are doing well.",
      "output": "-مرحباً، كيف حالك اليوم؟",
      "latency_seconds": 3.32,
      "quality": "good",
      "notes": "Good MSA translation, formal register"
    },
    {
      "id": 2,
      "direction": "en-ar",
      "type": "technical",
      "input": "The machine learning model achieved 95% accuracy on the test dataset.",
      "output": "وحقق نموذج التعلم الآلي 95 في المائة من الدقة في مجموعة بيانات الاختبار.",
      "latency_seconds": 13.48,
      "quality": "good",
      "notes": "Accurate technical translation, handles ML terminology"
    },
    {
      "id": 3,
      "direction": "en-ar",
      "type": "colloquial",
      "input": "I need to fix this bug before the deadline tomorrow.",
      "output": "أحتاج لإصلاح هذه الحشرة قبل الموعد النهائي غداً",
      "latency_seconds": 4.18,
      "quality": "good",
      "notes": "Good colloquial translation"
    },
    {
      "id": 4,
      "direction": "en-ar",
      "type": "code-switching",
      "input": "The API endpoint returns a JSON response with status code 200.",
      "output": "نقطة نهاية API ترجع استجابة JSON مع رمز الوضع 200.",
      "latency_seconds": 8.97,
      "quality": "good",
      "notes": "Preserves technical terms (API, JSON), code-switching handled well"
    },
    {
      "id": 5,
      "direction": "ar-en",
      "type": "msa",
      "input": "الذكاء الاصطناعي يغير طريقة عملنا في العالم الحديث.",
      "output": "Artificial intelligence changes the way we work in the modern world.",
      "latency_seconds": 3.63,
      "quality": "good",
      "notes": "Accurate MSA translation"
    },
    {
      "id": 6,
      "direction": "ar-en",
      "type": "technical",
      "input": "يتم تدريب النموذج باستخدام خوارزمية التعلم العميق على بيانات كبيرة.",
      "output": "The model is trained using a deep learning algorithm on large data.",
      "latency_seconds": 3.75,
      "quality": "good",
      "notes": "Accurate technical translation"
    },
    {
      "id": 7,
      "direction": "ar-en",
      "type": "politeness",
      "input": "شكراً جزيلاً على المساعدة، أقدر ذلك كثيراً.",
      "output": "Thank you so much for your help. I appreciate that very much.",
      "latency_seconds": 3.88,
      "quality": "good",
      "notes": "Accurate formal politeness"
    },
    {
      "id": 8,
      "direction": "ar-en",
      "type": "dialect_egyptian",
      "input": "إزيك؟ كله تمام؟ كنت عايز أسألك عن حاجة.",
      "output": "I was gonna ask you something.",
      "latency_seconds": 0.42,
      "quality": "poor",
      "notes": "TRUNCATED: Missed greeting 'izayyak? kullu tamam?', only translated second half"
    },
    {
      "id": 9,
      "direction": "ar-en",
      "type": "dialect_sudanese",
      "input": "يا زول، كيف حالك؟ تعال نتغدا سوا النهاردة.",
      "output": "Hey, Zol, how are you?",
      "latency_seconds": 0.58,
      "quality": "poor",
      "notes": "TRUNCATED: Missed 'come have lunch together today', 'Zol' untranslated (Sudanese term)"
    }
  ],
  "summary": {
    "avg_latency_msa": 5.67,
    "avg_latency_dialectal": 0.5,
    "msa_accuracy_rate": 100,
    "dialectal_accuracy_rate": 0,
    "key_finding": "OPUS-MT handles Modern Standard Arabic well but truncates Egyptian and Sudanese dialectal inputs, missing greetings and colloquial phrases"
  },
  "recommendations": [
    "Use OPUS-MT for MSA content only",
    "Implement dialect detection before translation",
    "Consider NLLB-200 or specialized dialect models for Arabic dialects",
    "Add preprocessing for Egyptian, Sudanese, and other dialectal inputs"
  ],
  "benchmark_by": "O96a",
  "benchmark_date": "2026-03-28T09:40:00Z",
  "huggingface_discussion": "https://huggingface.co/Helsinki-NLP/opus-mt-ar-en/discussions/10"
}