14.6 Language Translation

Language translation, or machine translation, automatically converts text from one language to another — for example Amharic to English. Modern systems use neural networks, especially the transformer architecture, which reads the whole source sentence and generates a fluent translation rather than swapping words one by one.

Attention lets the model focus on the most relevant source words when producing each translated word.

A tourist visiting Lalibela can point their phone at an Amharic menu and read it instantly in English. Behind the scenes, a translation model encodes the Amharic words and decodes them into natural English, handling word order differences between the two languages.

Scenario

Your Amharic-to-English translator produces fluent but sometimes wrong translations for rare farming terms. What is the most effective improvement?

Check your understanding

1/4 · 40 XP

Which architecture powers most modern machine translation systems?