Advanced and bespoke machine translation for low-resource languages.

The largest commercial translation engines target about 130 languages. These engines perform well for some languages; other languages struggle to produce meaningful translations of even general texts. And then there are languages for which MT capability does not yet exist.

In these instances, VoxCroft set out to build MT on models trained on domain-specific bilingual corpora. We also leveraged the similarities between related languages by building multilingual models. When we tested these in a real-world environment, they outperformed the MT engines which are typically trained on much larger corpora.

VoxCroft’s training data is human-translated and verified. For example, a commercial translation engine trained on 4 million pairs received a BLEU score – standard metric of translation performance ranging from 0 to 100 – of 22.6 on English-Tigrinya translations of news content. VoxCroft’s model, trained on only 80,000 sentences from online news content, received a BLEU score of 28.5.

VoxCroft has created over 1.8 million high-quality sentence pairs for 14 low-resource African languages. We crowdsource our linguists by tapping into our academic and industry network. All linguists undergo rigorous vetting, testing, and training.

Our linguists work on a range of projects, such as translation of domain-specific sentence pairs, transcription, text-to-audio recording, image description tasks, terminology creation, and other linguistic tasks.

All tasks undergo QA checks using VoxCroft’s QA methodology. All tasks that trigger a quality issue alert undergo human review and improvement. We use a hybrid human-machine approach to track linguist performance over time, ensuring that we continue to task only the best-performing workers.

VoxCroft’s language footprint is currently in Africa, but we can apply our workflow to any austere language environment.

Proteus is a micro-tasking platform. It is language agnostic and accommodates translation, image identification, transcription, audio recording, and quality assurance tasks. Proteus’ internal quality checks rewards good workers and penalizes poor workers. This incentivizes workers to maintain high quality standards throughout the project.

Proteus also has a manager interface. Managers can load new tasks, download completed tasks, and view other aspects of the linguists’ performance. Tasks that fail to meet VoxCroft’s project quality standards go to another worker for improvement or rework. Proteus produces a monthly worker report that can be integrated with your organization’s billing system.

The platform permits the anonymization of worker information to meet data protection regulations. We can also build specific customer requirements into Proteus.