According to an article in the current issue of BusinessWeek, a company called Narrative Science (which oddly has only a placeholder website) has figured out how to have computers turn out routine sports stories for, among others, the Big Ten Network. I don’t think Rick Reilly or the ghost of Red Barber need worry.
Actually, what Narrative Science is doing sounds like a more automated version of work I did at the beginning of my career. Way back in the days of the teletype, I worked for the Associated Press in Detroit. Like all rookies, I started working evenings and weekends and one of my jobs was to take reports of high school and small college athletic events and turn them into three or four paragraphs of passable text. The reports came from a variety of sources: newspaper, TV, and radio reporters from around the states, sports information directors, and sometimes what we’d now call citizen journalists. What they had in common was that they provided only the barest bones of information, well short of a box score. (In those pre-computer days, a box score would have had to be dictated, and no one had time for that except for events such as Michigan or Michigan State basketball.) If we ‘d had computers back then, they certainly could have done the job.
But natural language continues to be a struggle for machines. Consider this Narrative Science-written sentence from a report on a Purdue-Indiana baseball game: “Both teams seemed to have the advantage at times as there were five lead changes in the game, eventually the Boilermakers took the lead in the tenth inning and held on for the victory.” Even if we ignore the awful comma splice that wouldn’t have good past a minimally competent copy editor, that just doesn’t sound like anything written by a human being, at least not a native English speaker. As a crude, one-way Turing test, it fails.
On the other hand, Narrative Science’s sportswriting robot probably has good enough rules that it can avoid the sort of embarrassment I once suffered. A reporter phoned in the results of a competition in some sport I knew nothing about–I think it was cross-country running. One of the things I didn’t know was that the low score won, like in golf. So I wrote a brief article about one school’s runaway victory when in fact they had finished far behind the pack in last place.