Google’s own language parsing neural network framework will soon be available to everyone as the company has made SyntaxNet open source today.
Available for download here, SyntaxNet is a neural network framework implemented in TensorFlow that provides a foundation for Natural Language Understanding (NLU) systems. Inlcuded in this release includes all the code needed to train new SyntaxNet models on your own data, as well as an English parser that google have trained called Parsey McParseface.
Based on a benchmark done by Google, Parsey McParseface is able to recover individual dependencies between words with over 94% accuracy. According to Google, the result is encouraging as the in-house annotation projects that linguist trained for this task agree in 96-97% of the cases.
“The way we evaluate technologies internall is actually pretty different. We care much less about benchmarks and much more about how they impact performance of downstream systems. Our goal is to improve user experiences,” says Dave Orr, product manager for Google Research.
Source: VentureBeat, Google Research