3

I saw that SyntaxNet was released with Parsey McParseface and then not long after the Google Cloud Natural Language API came out which has some similar capabilities but I do not see any mention of SyntaxNet in the Cloud Natural Language API documentation.

Both of SyntaxNet and Google-Cloud-NL do Syntactic Analysis. Is there a difference? Can I use Cloud-NL instead of having to set up my own tensor-flow & SyntaxNet?

new name
  • 15,861
  • 19
  • 68
  • 114

2 Answers2

7

Google Research NLP PM here.

Just to answer specifically about SyntaxNet: we use an internal version of Parsey, not exactly the same thing that we open-sourced. In particular, it's quite a bit faster, and trained on a somewhat different set of data. Otherwise it's very similar.

In terms of quality, Parsey is a tiny bit better on news data, which is what it's trained on. The NL API is better on everything else, like web data.

As Sara says, you should definitely be able to use the NL API rather than install and run Syntaxnet yourself. It should be much easier to use, and will scale up quite well to large amounts of data.

Dave Orr
  • 1,082
  • 1
  • 9
  • 18
1

With the Cloud NL API you don’t need to build your own TensorFlow models or use SyntaxNet. You would use SyntaxNet if you want to build or train your own models, whereas the NL API is a pre-packaged and pre-trained solution for analyzing text (so you don’t need to worry about the underlying models).

Sara
  • 2,729
  • 7
  • 22
  • 30
  • So does Cloud NL use Parsey McParseface for it's syntactic analysis or not? If not, which is the more accurate? –  Jul 25 '16 at 17:13
  • See my other answer for more detail, but we use something very similar to Parsey. – Dave Orr Jul 25 '16 at 20:53