The quality of NMT systems is directly related to the amount of parallel text available for the language pairs. For many languages, there is little bilingual training data available in sufficient quantities. ALIA relies upon human enrichment to address these problems directly by employing native speakers from every language group to solve this problem. ALIA leverages a global crowd-enabled distributed computing platform that organizes people around the globe to quickly amass NMT training materials for low resource languages, as well as illuminate rare words, idiomatic expressions, and slang terms.
ALIA’s crowdsourcing platform provides dynamic scalability and global reach through a flexible, cloud-based workforce model. Cryptocurrency payments now make it possible for global crowd workers to efficiently participate in the process. ALIA language workers will earn rewards for completing microwork tasks via the ALIA interface. Rewards consist of a fractional percentage of 1 ALIA token represented as an ALIAbits.
Human corrections to the machine translation are made continuously and fed back into the neural network to achieve increasingly higher accuracy scores. In essence, ALIA creates smarter algorithms and guarantees quality by creating feedback loops with a trusted workforce validating the output from the NMT system.
Fortunately, English is the most learned language in the world and the degree of Internet penetration in many parts of the world make it straightforward to find and hire non-expert translators. Cryptocurrency payments will allow us to tap into a skilled pool of in-country native speakers who can quickly deliver accurate, culturally nuanced translations that costs a fraction of the industry average.