Innovation Watch: The neural network designed to kill jargon

Imagine scientific studies translated to plain English. A neural network has been created to do just that. We picture a world where even complicated legal documents are easily understood by everyone.

A team of researchers have created a neural network designed to produce plain English summaries of complicated texts, such as scientific research papers.

The work actually came from developing new AI approaches for an entirely different project, which focused on particularly tricky physics problems. The team soon realised the approach were there using could be used for other tasks computers find difficult – such as processing natural language.

Neural networks are excellent at pattern recognition. They can recognise faces in photos, forecast sales and detect abnormalities in data. What they’re not so hot at is correlating information from long strings of data – exactly the sort that’s required to summarise a complicated paper or essay.

This new network uses a concept called a rotational unit of memory (RUM), which helps it ‘remember’ better.

The results are eye-opening. Here are the summaries of a research paper, first using a traditional neural network:

“Baylisascariasis,” kills mice, has endangered the allegheny woodrat and has caused disease like blindness or severe consequences. This infection, termed “baylisascariasis,” kills mice, has endangered the allegheny woodrat and has caused disease like blindness or severe consequences. This infection, termed “baylisascariasis,” kills mice, has endangered the allegheny woodrat.

And then using the scientists’ new RUM neural network:

Urban raccoons may infect people more than previously assumed. 7 percent of surveyed individuals tested positive for raccoon roundworm antibodies. Over 90 percent of raccoons in Santa Barbara play host to this parasite.

The difference is remarkable. The first is repetitive and packed with technical language. The RUM neural network makes it clear and concise.

The implications are big for almost anything. Legal documents can become legible to normal people! Terms & Conditions will suddenly make sense and people might even read End User License Agreements!

It could also help communication between technical and non-technical staff. Or make sure that pitch-decks we send to clients don’t contain jargon familiar to creatives but confusing to everyone else! Perhaps it could even get to a point where it’ll tell copywriters if their lines are going to resonate?

Whatever the results are, we’re looking forward to reading what it produces.