Colag domain object should store sentences in a list

[pfeyz]

instead of a set.

https://github.com/pfeyz/colag-irrel-and-ambig/blob/master/colag/colag.py#L112

By storing them in a set we end up taking

• a sentence that appears in colag with 20 different parses
• a sentence that appears in colag with 1 parse

and representing both of them only once. We don't want to lose that frequency information (I think?)

[pfeyz]

hm... I think this is debatable.