...train a machine to recognize a spin down in academic debates...

Off the top of my head, having a program track a Google Alerts feed on the topic (with filtering via Yahoo Pipes if Alerts gives too many false positives) and let you know when it's gone a certain amount of time without getting any input seems like it would be a reasonable first approximation of that.