Computer Programmer Aims To Help Sports Broadcasters

A Q&A with Greg Lee about SCoReS.

Computer Programmer Aims To Help Sports Broadcasters
Download Audio
Audio Playlist
Computer Programmer Aims To Help Sports Broadcasters

Relax sports broadcasters, robots aren't coming for your job. At least not yet.

"The human aspect is important," said Greg Lee, a recent Ph.D. graduate in computer science from the University of Alberta. Dr. Lee recounted how, while watching baseball on TV, he stumbled upon Vin Scully, the Hall of Fame sportscaster now in his 59th season as the voice of the L.A. Dodgers.

"In addition to keeping you up to date on the score of the game, he drops in little tidbits about the players and will tell stories from past games," he said. This got the self-described baseball fan thinking about how he could use his programming background to help rookie sportscasters lacking Mr. Scully's deep knowledge of statistics and anecdotes.

The Sports Commentary Recommendation System (SCoReS) was the result. Dr. Lee said the program monitors game statistics in real time, matching those numbers to a compendium of pre-loaded stories. If a broadcaster finds they are running out of material, SCoReS provides a story related to what's happing in on the field.

Here's an example. Say it's late April and a batter on the Mets just hit two home runs. SCoReS scans its database and pulls up a story about New York Mets All-Star Keith Hernandez, who did the very same thing on April 26, 1988. As the colorful (but probably apocryphal) story goes, following the game Hernandez said, "I should get a divorce every day. I'd be broke, but I'd be in the Hall of Fame."

I spoke with Dr. Lee about the development of SCoReS.

What makes for a good sports commentator?

To me, particularly in baseball, what makes a good commentator is when the action is slow, which is fairly often - especially in regular season games - they fill up that time with stories from baseball’s rich past.

When you start to design a computer program to replicate that, what are some of the first things you do?

It’s capturing what’s important about a game and a story. We call those features. What features of the current game - what numbers describe it succinctly? And the same for a story. That was a big first step - how were we going to to relate the two things? How are we going to model the problem? And that's not a simple task.

For the game features - we mostly went with what was available from Major League Baseball’s online live updated site. For the stories, I must mention one of the books we used was Rob Neyer’s Big Book of Baseball Legends. He does an excellent job of getting more details on the story than what people have usually heard.

How many people did you test this on?

In total, 264 test subjects.

And you showed them Triple-A Baseball games?

Yeah, we were able to get rights to minor league baseball games. The International League provided the 2009 Triple-A All-Star game. Also, the Buffalo Bisons and Syracuse Chiefs were nice enough to let me use footage from one of their games.

Can you describe what the testing process for SCoReS was like? 

We played games with just crowd noise, games with their original commentary, and games with the original commentary plus a SCoReS selected story added in. We tried to see which clip they liked more. So subjects would sit there with a pencil and paper, watch the clips and answer the same questions for each one - how much did they enjoy it? What did they learn? Did it make them want to watch baseball more?

You also demonstrated SCoReS to professional commentators?

I had SCoReS serve up stories to them and just asked, would you tell this story at all?

Did the commentators suggest any tweaks?

Yeah. All of them did hockey commentary at one time or another. I'm in Canada, so that's not too surprising. They said the stories are too long for hockey because the action is so much faster. There is the time between whistles, but we need faster stories, they told me. I designed it for baseball because I thought it was most applicable there, but I don’t see any reason it would’t be applicable to other sports. You just need a few tweaks and then it should work.

Looking forward - talk about the future of automated sports broadcasting. I know some news outlets already use automated reporting systems, but where do you see all this stuff going?

I don’t see play-by-play and color commentators being replaced by computers. I suppose you could get sufficiently good voice generation that it would sound real, but the human aspect is important. Going into this project my goal was never to replace anybody. It was to help. Particularly commentators who didn’t have a lot of stories to tell. I mean, if you’re just starting out - if you’re young - you don’t have as many stories as an 85-year-old who’s been doing it for 50 years. I would hope that soon you could have these stories recommended to color commentators and they could tell them.

In the future, I don’t see much changing in terms of human play-by-play and color commentators. I would personally hope, as a baseball fan, that’s the way it stays. That’s what I enjoy. I assume other people do as well.

THIS INTERVIEW HAS BEEN EDITED AND CONDENSED.