Neural Transcomplier

NCI is building a product that can generate code from requirements written by a human. The traditional approach of using a rules based system does not work. Neural models perform much better than rule-based in natural language translation, but scarcity of parallel data makes their applications difficult. However, it is possible to apply recent approaches in unsupervised machine translation to train a fully unsupervised neural transcompiler.

Tags: No tags

Comments are closed.