The Center for Data Innovation spoke with Stuart Frankel, chief executive officer and co-founder of Narrative Science, a company in Chicago, Illinois that has built Quill, a natural language generation (NLG) platform, which using artificial intelligence to turn data into written narratives. Frankel discussed how NLG augments the work of human writers, freeing them to focus on more complex analysis.
This interview has been edited.
Eleni Manis: Some NLG software generates narratives from data simply by filling in the blanks of a prewritten template, like a Mad Libs story. How is your product, Quill, different?
Stuart Frankel: Quill is a reporting platform equipped with NLG technology. Unlike templated NLG systems that simply map language onto data, primarily through the use of business rules, Quill performs a number of complex functions in order to produce its output.
This includes complex assessments to characterize events and identify relationships, end-to-end data analysis to understand what pieces of information are relevant, statistical analysis to determine significant reporting events, and lastly, conversion of Quill’s analysis into its NLG plain-English reporting.
Manis: Quill is supposed to be able to analyze, interpret, and communicate information the way its users would do on their own. How does the software adjust to different styles??
Frankel: Narrative Science works closely with our customers to understand their business and content requirements of each use case. Quill is then configured to mimic the decision process that, say, an analyst would go through when drafting a report. The flow, the tone, the order of information, the delivery style, and so on, are all components that are configured within the platform. Equipped with a robust NLG engine, Quill is able to articulate itself in plain-English indistinguishable from that of a human-drafted report.
Manis: Narrative Science says that using NLG decreases “time-to-insight.” What does this mean?
Frankel: Our software frees up employees from data analysis and interpretation activities to focus on higher value, more strategic work by augmenting the work of employees and enabling companies to turn their data into easy-to-understand and insightful stories.
Manis: Does taking the human writer out of the equation come with any trade-offs or benefits in the quality of the final product?
Frankel: Our clients see a lot of benefit in having our software complement the human writer throughout the reporting workflow. Quill is able to perform a series of analytics, data analysis, and complex decision making to ensure the output captures a holistic and precise picture of what is going on in the data. It’s time consuming and often impossible for human writers to look at the different permutations of scenarios that could impact what they are trying to convey. When you pair the heavy lifting of software with the polishing and touch of the human writer, the end output is above and beyond what either of the two entities could do alone.
Manis: How does Narrative Science measure NLG accuracy?
Frankel: Quill is a smart solution that has built-in data validation. There is also a logic trace, or auditing, feature that helps employees go back and see the data that was used to drive every decision that Quill has made.
If there is incorrect data, Quill will perform just like a person would who is using bad data. The benefit of using Quill, however, in this process is the speed at which it operates—users will immediately know if there’s something wrong in the data or the analysis based on what they are reading in the output. If the results are inaccurate, users can quickly investigate and make fixes, drastically reducing the time and effort spent trying to figure out what’s wrong and then reanalyzing, rewriting, and re-communicating.