A computer on a desk with a bullhorn producing a thought bubble.
PHOTO: Shutterstock

Natural language generation is exactly what it sounds like — it is the use of artificial intelligence and machine learning to create and generate language. In other words, it is a computer that can write. There are various levels of advancements in natural language generation, explained Tim Lynch, founder of Psychsoftpc. Previous incarnations have been as simple as a merge function in a form letter — i.e. filling in a blank in a letter with a name or a number — to more recent iterations in which complex sentences have been generated from data, he said.

It is these later, more recent evolutions that has the industry’s attention, said Juan Jose Lopez Murphy, AI & Big Data Tech Director at Globant. “The first wave was about being able to make sense of 'dark data' — information hidden within complex forms that humans perceive naturally but computers struggle with, like natural text, images, sounds,” he said. This next stage will be about producing them in a way that they sound human and can be consumed by humans — that is, creating text that reads naturally.

How It’s Being Used Today

Computers can do highly-templated natural language generation or writing very easily, according to Christopher Penn, co-founder of BrainTrust Insights by using stock formats. Some examples include financial press releases, stock reports, earnings calls, fantasy football and fantasy baseball league. 

“The computer can substitute in different adjectives so that the text doesn’t read exactly the same,” Penn said. “This is already being done — The Associated Press uses machine learning to do natural language generation on many of its services.”

Related Article: Will Natural Language Processing Change Search as We Know It?

Where Is It Going?

Penn said that the technology is improving to the point where computers will be able to write more freeform content, such as blogs, news articles and so on. “It’s getting better,” he said — “but it’s still not great. It’s not a substitute for a human being because there is too much variability in a completely freeform environment.” So if a computer were asked to write a 1,500 word article on marketing it would struggle with it, Penn said. “We saw that watching IBM Watson trying to debate a human opponent. Under the surface, under the hood, you could tell that the computer was not aware of the context and the depth of the content in the same way that a human is.”

What Does This Mean For Marketers?

If content is very repetitious it’s a good candidate for natural language generation, Penn said. “There are tools in the market that can help you do that right now,” he said. “Some of them are open source and have no financial cost. You just need a developer to bring them to life.” Also, just because the freeform capability is not here yet doesn’t mean it isn’t en route, Penn added. He estimated that it will probably be another year or two before the market sees good, solid applications that are close to production ready and then another two to five years for these applications to become commercially available and inexpensive.

Related Article: How Machine Learning Will Tame the Explosion of Unstructured Data

A Customer Service Application

The customer service space will also be a beneficiary when the technology reaches that state, according to Murphy. There will be succinct abstracts of long email threads that read in a pleasurable way, he said, “non-alienating interactions with customer service platforms, actual answers to complex questions rather than a set of links and so on.” The technology will also affect chatbots making the interactions feel very natural, he continued; likewise voice interactions with Alexa and Google Home. “Recommendations could be attached to a simple explanation of why it’s a relevant choice for us, which would be a game changer for content marketing.”

Indeed, conversational AI technology will allow companies to re-imagine the entire customer engagement process — even the most basic of customer service transactions — and collect valuable data that can inform future transactions, said Andy Peart, CSO at Artificial Solutions. It will be a change long in coming, he added. “For too long, customer service has been relegated to a formulaic, question-and-answer scenario that rarely leaves the customer satisfied and often doesn’t solve the problem at hand.”