Crowd workers are contributors who are bilingual and execute defined translation tasks. Crowd workers translate small snippets of text generated from the ALIA system. Microtasks are used when the task is highly parallelizable and can be divided into smaller pieces down to a ‘micro’ level, which takes only seconds to minutes to complete. Imperfect results from the NMT translation will be farmed out as microtasks. There are many potential scenarios where this would be useful such as rare words, idioms and specific professional terms.
Approved crowd workers can participate in translating in larger batch documents to help improve accuracy levels of the NMT. As an example, participants may be asked to translate pre-generated materials into target languages for specific subjects or genres that the NMT does not have a lot of current source material to generate an acceptable accuracy rating. This is an excellent way to augment weaker language training sets in the VERBIS repository and will continuously improve the accuracy of the NMT output. Every translation in this matter marginally improves the accuracy rating for the specific language. Essentially, this is one way that ALIA’s NMT “learns”.