The State of Natural Language Generation in Content Creation

Using Natural Language Generation (NLG) tools to create content has been getting quite a bit of buzz in the past few years. (NLG is a subset of the general discipline of artificial intelligence.) The first content creation engines struggled with relatively basic writing tasks, producing content that was robotic in every sense. Many of these engines have since advanced to the point that even experts find it hard to distinguish machine-generated content from that created by professional writers.

We’ve even come to the point where NLG plus Machine Translation can provide real value by producing content and then instantly translating it into multiple languages. Not only that, but the content can be tailored to individuals, pulling out key features according to their profiles so that each user is served with content that’s relevant to them.

Whether it’s a set of sports scores and statistics or a company’s financial data, NLG engines can create a narrative around the numbers, and can even ensure that the content follows a brand’s tone of voice and style. When writing a review of a football game, for example, it can embed a bias for a particular team, lauding its modest achievements while downplaying the successes of an opponent.

A number of content creation engines listed here include Quill, Wordsmith and textengine.io. Their websites feature case studies of AI-based NLG descriptions for a range of topics including real estate, weather forecasts and financial reports.

Thanks to machine learning, the quality of NLG content is likely to keep improving. Systems learn from vast amounts of data and continually refine their output, and it’s this self-improvement capability that has enabled enterprises to create increasingly higher-quality content without needing a team of writers. As the volume of data consumed by these systems increases, the wrinkles are gradually ironed out and complex language can be transformed into meaningful, clear content that’s also easy on the ear.

Andy Jarosz, Content Strategist at RWS Moravia, is bullish about the trend of automated content creation:

“One of the most exciting uses of AI in content creation is in taking huge, dry data sets and presenting them in the form of natural language. Very soon you’ll be able to read or listen to content that has been curated and customized according to your known interests or preferences, and in the language you choose. The idea of one-size-fits-all content might soon look outdated.”

Writers step aside?

Talk always turns to the fear that machines will replace humans. Whether or not the use of NLG is a sign of impending doom for human writers is a topic we’ve tackled before. It’s more relevant than ever as NLG capabilities mature. So where does this development in content generation leave the professional writer? Should they start planning an alternative career?

While freelancers are still largely able to find enough writing work to pay the bills, there’s no doubt that the advances in NLG are affecting their opportunities. For example, the days of employing an army of remote writers to conjure up large volumes of descriptive content from structured data may be numbered. (The same goes for freelance translators: not only is a machine faster and often cheaper than its human rival, but MT can remove the time lags and complexities associated with providing content in multiple languages.)

All this said, the reduction in relying on professional writers may not be imminent, as different companies and markets will need time and budget to adopt these new technologies. Skilled writers and editors will still be needed to manage the quality of the data and provide sanity checks on the machine-generated output—skills different from pure writing, but tasks required of writers nonetheless.

The rise of fake reviews

As is often the case, there’s a downside. The ability of NLG-driven content to fool people into believing it was created by humans has inevitably been exploited for less-than-pure motives, and nowhere more so than in the field of user-generated reviews. So many of us check what our peers think of a product or service before we decide to buy, and often we put these user reviews on par with or even above the official product descriptions. Being able to attract 100 positive reviews of a product organically might take a lot of time and patience, but it’s possible to create these fake reviews with NLG in a similar way to generating descriptions. With machine-generated reviews increasingly hard to distinguish from real ones, it’s a relatively low-risk misdemeanor. While the concept of fake reviews might date back over a hundred years, the ability to create them on an industrial scale is a new problem that many sites are facing.

Companies have in turn used AI to spot fake reviews, as this Guardian story about Amazon’s efforts shows. We see the intriguing scenario where on one side, AI is used to create fake content and on the other, AI is deployed to detect those rogue entries in an attempt to preserve the credibility of the site. As an added twist, there are even sites that help users identify fake reviews so that they can tip off the retailer.

Embrace and extend

No matter the risks, the development of NLG is neither slowing down nor going away. There’s just too much content out there that needs to be created and not enough humans to do it. NLG is continuing to become more sophisticated and is rapidly adapting to meet global demands. As Andy concludes:

“Machine learning is raising content quality to human standards and beyond, and the time may soon come when the quality expectations of NLG content are so high that an AI engine will be measured not by how well it fools readers into believing it’s human, but by making it clear that it’s not.”

There are some issues, but there is promise that they will be resolved over time. It’s a space all global marketers and content development managers should keep their eyes on.