AI-102 Question 235
Select 3You are developing a custom translation model using Microsoft Azure Translator to translate technical documents from English to French. After training the model with domain-specific data, you notice that certain technical terms are being inaccurately translated. What steps should you take to improve the model's performance before publishing it?
- A
Add more domain-specific sentence pairs to the training data and retrain the model.
- B
Use the Translator's dictionary feature to add custom translations for specific technical terms.
- C
Publish the current model and gather user feedback to improve it iteratively.
- D
Evaluate the model using a validation dataset to identify specific areas where it is underperforming.
- E
Adjust the confidence threshold of the model to prioritize more accurate translations over uncertain ones.
Show answer and explanation
Correct answers: A, B, D
Explanation
Improving a custom translation model requires enhancing the quality and relevance of the training data, as well as leveraging features like custom dictionaries to handle domain-specific terms. Additionally, evaluating the model with a validation dataset helps uncover specific issues that need to be addressed. These steps ensure that the model delivers accurate translations before being published.
- A. Correct.
Adding more domain-specific sentence pairs helps the model learn better translations for technical terms as it increases the relevance of the training dataset.
- B. Correct.
Using the Translator's dictionary feature allows you to define specific translations for technical terms that the model might not inherently learn from the dataset.
- C. Incorrect.
Publishing the current model without addressing the known inaccuracies is not recommended, as it may lead to user dissatisfaction and degrade the quality of the translation experience.
- D. Correct.
Evaluating the model with a validation dataset helps identify specific weaknesses and areas of improvement, which can guide the retraining process.
- E. Incorrect.
Adjusting the confidence threshold will not directly address the inaccuracies caused by insufficient training data or lack of term-specific translations.