Walker Translation is a research-oriented company working toward improved methods in low-resource neural machine translation.   Specifically, we are working with Arabic vernaculars and English to arrive at a Linguist-Developer Approach  to low-resource neural machine translation.   Our approach seeks to maximize the demonstration of grammatical patterns by use of minimal pairing and supplementing datasets with confident variations of natural language data.

In addition to conducting low-resource NMT research, we are also exploring practical applications of NMT technology.   Successful NMT models will be made available by license for use in software development projects.

Walker's NMT models are created using OpenNMT, an open source neural machine translation system started in 2016 by machine translation company SYSTRAN and a natural language processing research group at Harvard. OpenNMT is currently maintained by SYSTRAN and language service company Ubiqus, enabling researchers to create unique NMT models from custom datasets.