Technology that understands what text really means

Natural Language Understanding (NLU) is a subset of Natural Language Processing (NLP), a technology that enables computers to extract information from human speech or text. While NLP can analyse the structure of a set of words, NLU aims to provide insight into what those words actually mean.

Doing this well is, understandably, difficult. There are problems with slang, irony, different syntax and, especially in spoken language, fragmentary phrasing. In fact, it’s known as an “AI-hard” problem, one which, if solved, would put computers on the way to being as intelligent as people.

However, a new approach to NLU has been developed by Cortical.io based on the way the brain processes information. This approach enables businesses to reduce the time and costs of extracting meaning from documents by analysing the text automatically. The technology is being used by many companies in the Fortune 100 because, unlike systems built on keyword statistical analysis, it can analyse the meaning of whole paragraphs rather than individual words.

Francisco Webber, CEO and co-founder of Cortical.io, explains the firm’s technology is unique. It solves problems other systems fail to do by being able to understand the semantic nuances in any language. For instance, it can recognise sentences that have similar meanings even if they do not use the same words. It avoids vocabulary mismatches which a system that just uses a keyword search can give.

“We have a system that works differently,” Webber says. “It does not work on statistical methods. It works by using computational neuroscience as a starting point. That makes the results better than you would get from other systems.”

The system eliminates repetitive and error-prone manual steps, freeing up employees to focus on higher-value tasks. For example, Cortical.io’s technology helped a manufacturer of network equipment reduced the average handling time of support requests by 70 per cent.

Using Cortical.io’s Contract Intelligence Engine, a big four accounting firm was able to reduce the review time of lease agreements by 80 per cent. Changes to US accounting regulations in the US meant the company needed to put customers’ lease agreements on the balance sheets. As hundreds of thousands of documents needed to be searched, using humans would have been extremely time-consuming and costly.

“The automation we do works better than a human because of the volume of text that needs to be processed,” says Webber. “If you ask humans to extract the same kind of information from 300,000 contracts, it will take forever and the quality will be something around 50 per cent. With an automation tool like our Contract Intelligence Engine, the whole economy becomes more efficient.”

The system works by building a database of the field the client wants to analyse and extract information from. For example, Webber explains, if a medical company wants to search patient classification records, the system would be trained in medical terminology. University textbooks would be taken from different medical departments and be used as training tool for the search engine. The main difference with other systems is that Cortical.io's technology performs very well with little training data and that the training is unsupervised.

Semantic fingerprints are then generated for the words, sentences or documents contained in the database and the meaning of the text data is encapsulated in a numeric form. Through these fingerprints, semantic relationships can be compared, making it easy for information to be extracted quickly, without the company having to spend large sums of money.

The collection of business-critical data by organisations will only continue to grow, and if businesses can get the information they want from it quickly and cheaply, it will certainly give them a competitive advantage. By using technology which understands the nuance of natural language, getting that edge is now one step closer.


Learn more about Cortical.io:

Introduction video “Semantic Folding: a new model for intelligent text processing

Podcast “Francisco Webber: Statistics vs. Semantics for Natural Language Processing

White Paper “Semantic Folding and its Application in Semantic Fingerprinting

Cortical.io Free Tools

Shares

Get our latest features in your inbox

Join our community of business leaders