Text this: Evaluating automatic model selection