Discussions (about the course or NLP in general)

Hello,
I'd rather argue that you have to split it and your conclusions are erroneous.
In this case, we have a distance metric that we want to minimize, as a lower distance means that your text is more lexically coherent, which is the inverse of what you claim. Additionally, D(S1) + D(S2) < D(S1, S2), thus it's more beneficial to split it.