2009 Spring Interactive Innovation Project Machine News
Imagine a publishing tool that would help ensure that even with shrinking budgets for news agencies, small-town little league games would get coverage in local papers and online. Medill students Nicholas Allen, Tian Huang, John Templon and computer science student Thu Cung have developed a way to take raw sports scores (box scores and play-by-plays) and turn them into written news stories. All in a matter of seconds.
And while the thought of machine generated news stories might set off alarms for some journalists, according to the students, the goal is not to replace live sports reporters.
“There are a tremendous number of sporting events that go uncovered, even at the college level,” Allen says. “So we think that there is great potential for our product in that space.”
Instead, by giving sports writers a tool to easily generate simple stories from data, it frees the writers up to delve deeper and spend more time on their stories.
“Our project is a way to keep journalism thriving,” Huang says. “By expanding coverage and freeing reporters from writing the basic game stories, we're giving journalists the time to really focus on the features and analysis that drive readers to their publications.”
To make the tool work, the students created a computer program that can evaluate what was news during the game (using raw sports score data) and then write the components of a baseball game story (using pre-written templates).
As part of the story templates, the students created sample story narratives with language such as Comeback Win – Team X comes from behind in the 9th inning, Blowout Win – Team X takes an early lead and never looks back and Held Off – Team X tries to rally late, but doesn’t make it.
What the writing tool can’t do is incorporate any activity outside of what the raw scores indicate, that is, what makes games unpredictable, such as a fan-caught ball or a on-field altercation. Nor can it add historical context to its stories.
That comes next, say the students, and there are media companies already interested in seeing what they’ve done so far.
The Northwestern Center for Innovation in Technology hired Allen and Templon in June 2009 to continue developing the software, which has been renamed StatsMonkey.