Linguistic Variation and Anomalies in Comparisons of Human and Machine-Generated Image Captions