Parsey McParseface: What is it?

Despite the silly name, Parsey McParseface is highly advanced linguistic software. The new language parsing model uses SyntaxNet’s neural network framework and algorithms to analyze the syntax and structure of a sentence. It is essentially like a toddler still learning to speak his native tongue, but Google rates it as being 94 percent accurate. This is much more accurate than previous parsers, but still not as accurate as a human speaker.

Parsey McParseface is rather nifty as it allows Artificial Intelligence to more accurately handle ambiguity. Ambiguity is very common in most languages, especially English, which is the only language Parsey McParseface is currently trained in. A human speaker will often be able to use gestures, emphasis, or intonation to clarify and understand the intended message.

Machines, on the other hand, have in the past had to sort through the potentially dozens of different possible meanings by determining the intended grammatical role of each word in the sentence to find one that might fit the given context. This approach comes from Noam Chomsky’s theories of grammar, and so far has been only mildly successful.

Google has taken a different approach by using a concept known as Dependency Grammar. This approach focuses on pairs of words and their relationships rather than individual words. The only problem is that the quantity of these word-pairs and relationships is vastly greater than the quantity of individual words.

Google maintains that while the parser is highly accurate, it is only the first step in AI language development. One hope for the future is that the software will be able to analyze speech and text for logical and factual errors. While this may be detrimental to the careers of some politicians, lawyers, and journalists, it will be very helpful for the rest of us.

If you are interested in trying  it out, Parsey McParseface and SyntaxNet are open-source and run on TensorFlow. For continued updates on Parsey McParseface, contact us!