Low-resource language
A low-resource language is one for which little digital text and few parallel translations exist to train language technology. Because machine translation learns from data, low-resource languages get weaker automatic quality than high-resource ones like English or Spanish. Many community and Indigenous languages are low-resource, which raises the value of human expertise for them.
How it works
Machine translation quality tracks the amount and quality of training data for a language pair. High-resource pairs have billions of translated sentences; low-resource pairs may have very few, so engines generalise poorly and the metrics themselves become less reliable.
The practical consequences are fewer engine options, rougher raw output, and heavier reliance on human translators, which affects turnaround and cost for exactly the communities that language access most needs to reach.
How SourceTarget uses it
For low-resource pairs, SourceTarget leans more on human translation and review and less on raw machine output, and is candid about it in a quote. The workflow flexes: more machine drafting where the data supports it, more human effort where it does not.
Low-resource language compared with High-resource language
| Low-resource language | High-resource language | |
|---|---|---|
| Training data | Little digital text, few parallel translations | Abundant text and translations |
| Machine quality | Rougher, less reliable | Strong and fluent |
| Relies on | Human expertise more heavily | Machine drafting with light editing |
| Examples | Many community and Indigenous languages | English, Spanish, French |