Nonstationarity in paired-associate learning as indicated by a second guess procedure