NLG Technologies: Artificial Intelligence vs. Rule-Based Approach
by David Llorente
Here’s a useful image if you want to understand the two routes one can take to cross to the finish line in Natural Language Generation (NLG).
The first and most developed approach is like a slot car that’s locked into a pre-defined track. It runs on a series of business and linguistic rules to translate data into narratives. The second approach (the one used by Narrativa) is more like a self-driving vehicle made by the likes of Tesla or Google. It’s able to make autonomous decisions and drive on almost any road and terrain thanks to Artificial Intelligence (AI) systems that learn from complex data sets and draw on existing output narratives as examples.
Rich or reduced storylines
While there’s no doubt that a system loaded with plenty of rules will be able to produce good narratives, that’s not the real issue with a rule-based approach. Problems arise as soon as the subject matter domain or type of narrative become too complex to really know in advance all the rules that apply.
It’s quite easy to design rules for data sets with 10 or 15 data points, but the use-case scenario for NLG technologies most often looks differently. NLG technologies usually have to deal with hundreds and thousands of data points, and AI has a clear advantage in these scenarios.
Take the automatic generation of summaries of European football games, one of the most common applications for NLG today. It might sound like a fairly simple domain, but providers such as Optasports offer more than 10,000 data points per game, and that’s just an average.
Rule-based approaches are clearly insufficient to manage and analyze these datasets, forcing companies that use this approach to reduce the number of data points to handle the complexity. The result is a slimmed-down game summary covering just the main events but without the ability to analyze rich data and find insights.
Scalability and SEO
When it comes to scalability, AI technologies are ahead again. If you compare a large number of narratives generated with a rule-based approach, many sentences will be, pun intended, repeated repeatedly. The structure of the content will be also repetitive and will often contain irrelevant information. That’s a major problem when it comes to SEO performance because search engines might penalize what they classify as low quality content due to the many duplicates.
Artificial Intelligence, on the other hand, is capable of generating a much richer variety of narratives when it comes to overall structure and individual sentences.
That’s because an AI system determines the structure of the narrative based on the data and on the learning it has acquired during its training. The system then decides what’s more or less relevant and structure the narrative accordingly. The same applies to individual sentences, with the system choosing the best way to describe the data using natural language.
Curating complex narratives
Narratives are not static, and there’s a constant need to update and improve them. There are many possible scenarios. We might, for example, want to add new data points such as weather information to a game summary, or we might want to change the angle of the story, say highlight the performance of certain KPIs in a marketing report.
A rule-based approach makes this a hard task, Identifying, analyzing and codifying the new business rules and integrating them into the system requires a significant investment in terms of time and money.
AI systems have the upper hand here, yet again. Their ability to learn makes maintaining narratives simple and easy. In order to update a story, the system will analyze, weight, and represent the new data automatically. An AI system just needs the new data points and some examples of output narratives to do its job.
Making the transition to AI
The market has begun a clear shift over the last year. Two rule-based NLG companies
(Automated Insights and Ax-Semantics) are trying to make the transition to an AI approach since they’ve realized that the rule-based way doesn’t afford them the technological advantage necessary to crack complex problems.
To be sure, switching from a slot car running along a well-defined track to an autonomous vehicle that keeps learning is a big shift. The move to AI requires considerable effort and goes beyond simply adding some AI specialists. To win this race, you need to be ready to change your mentality, overall approach, processes, and even the company culture.