Investigating Word-Class Distributions in Word Vector Spaces