As pointed out by Aaron Beardslee, random.sample() will generate collisions if the source dictionary is not uniq'd. This is a possible way of handling it.

words = list(set(word.strip() for word in words))