Google open sources Parsey McParseface, so machines can speak English

Parsey McParseface is the most accurate natural language understanding parser ever created, and Google just open sourced it.


Parsey McParseface might seem like it is made by a raging teenager. However, this nomenclature format has emerged from the hivemind. The name "Boaty McBoatface"  recently won an internet poll for naming a sea research vehicle in the UK, and Parsey McParseface follows this format. However trollish the nomenclature,  Parsey McParseface addresses one of the fundamental problems of human computer interaction. Understanding speech can be difficult for humans, and it is much more difficult for machines, considering that the objective is to do it fast.

McParseface is a parser based on SyntaxNet, a neural network framework. The SyntaxNet framework enables natural language understanding (NLU) in machines. SyntaxNet is implemented on TensorFlow, a software library for machine learning applications. Google recently upgraded TensorFlow with distributed computing technology.

Parsey McParseface is a pre-trained parser. This means that Google has a functional and advanced system in place for explaining the role of each word in a sentence, and how a statement comes together. Machine learning applications that use Neural Networks learn as they operate, and get better as they are used. McParseface is the most advanced system of its type in the world, and is ready to be deployed into various applications that need natural language understanding.

This could mean implementation in Facebook's Messenger chatbots, any Bots made on Microsoft's Bot Framework or the next generation intelligent digital assistants similar to Viv. Viv uses its own superfast and complex syntax modelling, but McParseface can be used on more evolved and powerful voice based digital assistants like Cortana or Siri.

Google open sources Parsey McParseface, so machines can speak English

Google Research Blog

For humans, it is possible to intuitively understand complex sentences. There are many ways a sentence can be parsed, good examples are "garden path" sentences. These are difficult for humans as well, but the ambiguity can be resolved with a re-read. Consider this sentence, The complex houses married and single soldiers and their families. Or, this one: The rat the cat the dog chased killed ate the malt. McParseface reads the sentence from left to right, and makes a list of possible interpretations. These interpretations are given scores based on plausibility. As it goes through the sentence, the interpretations with the lowest scores are discarded. McPraseface is the most accurate such parser ever.

It is possible to train your own parser using Universal Treebanks. You can also start using Parsey McParseface itself.

Find our entire collection of stories, in-depth analysis, live updates, videos & more on Chandrayaan 2 Moon Mission on our dedicated #Chandrayaan2TheMoon domain.