AI-102 exam dumps

AI-102 practice question 238 of 493

Designing and Implementing a Microsoft Azure AI Solution. Professional level, Microsoft. Free question with the correct answer and a full explanation.

AI-102 Question 238

Select 3

You are tasked with implementing a custom translation model to support a niche language pair in your organization. After training the model using the Microsoft Translator custom feature, you notice that the translation quality is not meeting your expectations. What steps should you take to improve the model before publishing it?

  1. A

    Provide additional bilingual sentence pairs that cover a broader range of vocabulary and grammar.

  2. B

    Use the 'Test' feature in the Custom Translator portal to evaluate translations with a test dataset.

  3. C

    Retrain the model with the same dataset without making any changes to the data.

  4. D

    Manually edit the translation outputs in the Custom Translator portal to fix errors.

  5. E

    Analyze the model quality using BLEU scores and focus on improving poorly performing areas.

Show answer and explanation

Correct answers: A, B, E

Explanation

Improving a custom translation model involves providing high-quality, diverse training data, evaluating the model's performance using tools like the 'Test' feature, and analyzing metrics like BLEU scores to identify weak areas. Retraining without addressing data quality or manually editing outputs does not contribute to meaningful improvements in the model.

  • A. Correct.

    Providing additional bilingual sentence pairs helps enhance the model's ability to understand complex vocabulary and grammar, ultimately improving translation quality.

  • B. Correct.

    Using the 'Test' feature allows you to evaluate the model's performance on a test dataset, providing insights into areas where the model needs improvement.

  • C. Incorrect.

    Retraining the model with the same dataset without changes will not improve the translation quality, as the underlying data issues remain unaddressed.

  • D. Incorrect.

    Manually editing translation outputs in the portal is not a supported feature for improving the underlying model. Instead, you need to improve the training data.

  • E. Correct.

    Analyzing BLEU scores helps identify specific weaknesses in the model, such as grammar or vocabulary gaps, allowing you to focus your efforts on targeted improvements.

Timed practice exam

Take a AI-102 practice test under exam conditions

40 questions in 120 minutes, drawn from this bank, with a score report and a per-question review when you finish.

Start timed exam