What are the criteria for evaluating the quality of machine translation?

The criteria for evaluating the quality of machine translation (MT) typically include the following key aspects:

1. Accuracy (Faithfulness): The translation should correctly convey the meaning of the source text without omissions, additions, or distortions.

   • Example: Translating "The meeting is scheduled for 3 PM" should not become "La réunion est annulée à 15h" (incorrect meaning).

2. Fluency (Naturalness): The output should read smoothly in the target language, adhering to grammar, syntax, and idiomatic expressions.

   • Example: "She go to school" (incorrect) vs. "She goes to school" (fluent).

3. Terminology Consistency: Proper use of domain-specific terms (e.g., medical, legal) without inconsistent translations.

   • Example: In healthcare, "hypertension" should not be translated as both "high blood pressure" and "hypertension" in the same text.

4. Context Awareness: The translation should adapt to the broader context, avoiding word-for-word errors.

   • Example: "Bank" (financial institution vs. riverbank) should be translated correctly based on context.

5. Human Evaluation (BLEU, METEOR, etc.): Automated metrics like BLEU (Bilingual Evaluation Understudy) compare machine output to human references, while METEOR considers synonyms and word order. Human evaluators also assess quality subjectively.

6. Latency & Efficiency: For real-time applications (e.g., live chats), speed and computational efficiency matter.

For businesses needing reliable MT, Tencent Cloud's Machine Translation (MT) service offers high-quality neural machine translation (NMT) with support for multiple languages, customizable terminology, and industry-specific optimizations. It integrates well with applications requiring fast, accurate translations.