End-to-end development of natural language generation application for the PGA Tour
Due to limited writing staff, Professional Golf Association Tour articles were predominately created for the top handful of players in a tournament. Logistically, it is impossible to provide human-generated coverage to all players.
Company
Narrative Science was a technology company specializing in Natural Language Generation (NLG) software. The proprietary software, Quill, once coded by a solution consultant takes in structured data and provides English prose to generate reports or add analysis to data.
Role: Solution Consultant
Solution consultants were tasked to Implement our intelligent automation platform, Quill; own project management; solution sales; Quill developer; relationship management; manage data inputs and innovation process.
Shot link captures incredibly detailed data including club used, distance struck, distance remaining, the surface the ball starts and ends at and more.
The Challenge
The PGA tour shot link system captures data to an incredible degree of detail. Multiple variables are tracked for every shot a player takes during a round.
That data in the hands of even a moderately competent writer who knows the game of golf could easily be used to write engaging content. The challenge is how do you code a program to write that for you. How can you add the complexity and nuance so the writing isn't overtly computer writen, stale, and boring?
The Outcome
Fully configured product launched live and generated hundreds of articles in the opening weekend and many thousands since. Leveraging this technology has provided the Tour with the scalability to generate content and fulfill organizational mission.
The goal of this effort was to create a system so that any player on the tour famous or not, could have published content for their family, friends, and fans.
Logical Structure
HOW TO TEACH A COMPUTER TO SPEAK GOLF IN QUILL CONFIGURATION
Teaching a computer to recognize interesting content that makes a worthy golf article means codifying building blocks of golf within the data and teaching the resulting language that is associated - starting with a taxonomic breakdown of a round of golf looking at defining types of shots, types of holes, types of available scores and other entities of the game.
I developed a lexicon of available entities that are defined in the data. Setting up Boolean triggers on these entities allows for the generation of related content. If you build in variation and various degrees of complexity and relation, you craft a report generator capable of writing many hundreds of thousands of unique stories.