Skip to main content

What is machine translation?

The AI Glossary

In our classes, we often pause to explain the ideas behind AI and the language used to describe it. These posts grew out of those conversations. They’re for anyone who would like to understand AI better, with no technical background assumed.

Machine translation uses software to convert language from one language into another. It may work on written text or form part of a spoken-language system. In the latter case, recognizing the speech and translating its meaning are separate tasks, even when the application presents them as one smooth interaction.

Translation is not simply replacing each word with a dictionary equivalent. Languages arrange information differently and use expressions whose meaning cannot be recovered word by word. The system must preserve the intended meaning while producing suitable wording in the target language.

Earlier systems used explicit linguistic rules or statistical patterns learned from translated material. Neural translation systems use models trained on examples, and large multilingual language models can also translate through prompting. These methods can produce fluent results, but fluency does not establish that every detail has been preserved.

Important errors include omissions, additions, changed negation and incorrect numbers. “Do not stop this medication” must not become “Stop this medication.” A specialist term may have several possible translations, and the appropriate choice can depend on the reader’s location and familiarity with the subject.

Performance varies by language pair, subject matter, input quality and communication style. An evaluation in one language does not establish performance in another. Translating the output back into the original language can reveal some errors, but it is not independent proof of accuracy: the second translation may reproduce or conceal the first mistake.

For a clinic, translating parking instructions is a different task from translating consent information or urgent treatment advice. The level of review should reflect the consequences of misunderstanding and the applicable language-access requirements. Appropriate bilingual reviewers and qualified interpreters remain important for clinical communication. Evaluation should include whether patients understand the intended action, not only whether the grammar sounds natural. Machine translation can make information more accessible, provided the organization knows which messages it can handle, which need expert review and how a person can obtain help when the translation is unclear.

Sources: Hugging Face — Translation; Ji and colleagues — Survey of Hallucination in Natural Language Generation (2023)

To learn more, take a look at the AI in Health Care Certificate of Specialization from the Advanced Learning Academy at Harvard T.H. Chan School of Public Health. You can take an individual course or work toward the full certificate.


AI is a fast-moving field. This explanation reflects our class discussions and related material at the time of writing. Terminology, capabilities and tools may change.

For suggested additions or corrections, please email me at tpanch@hsph.harvard.edu.

Dr. Trishan Panch
Program Director, AI in Health Care
Harvard T.H. Chan School of Public Health — Advanced Learning Academy

Lead with AI in Health Care

The AI Pragmatist is a monthly newsletter from ALA faculty, with new explainers & essays from the Advanced Learning Academy