MACHINE LEARNING / LANGUAGE
Machine Translation & Quality
Applying multilingual machine learning to improve translation quality and reveal discrepancies in existing outputs.
The challenge
A translation workflow needed a way to assess whether machine-generated Tagalog-to-English translations could meet or improve on the existing baseline.
My contribution
- Built a machine translation solution using multilingual models.
- Deployed the solution on AWS SageMaker.
- Compared model outputs with existing human translations through reviewer assessment.
Architecture & approach
High-level conceptual view of the workflow:
01 · Source textTagalog
02 · Multilingual modelML translation
03 · Reviewer evaluationEnglish output quality
Results are described qualitatively because numerical evaluation results were not provided.
Design considerations
- Evaluate translation quality through reviewer comparison rather than model output alone.
- Use a managed deployment environment for serving the ML capability.
Outcome & impact
In reviewer comparisons, the model translations were assessed as more accurate than the existing human translations and surfaced discrepancies in those translations. The résumé does not provide a numerical accuracy uplift.