Since: $$pmi(x,y)\quad =\quad \log(\frac{P(X=x,Y=y)}{P(X=x)P(Y=y)})\quad =\quad -t$$ iff $$ 2^{t}P(Y=y | X=x) \quad =\quad P(Y=y)$$ the interpretation of a negative PMI seems very clear to me. So I don't really understand the following comment in the Wikipedia article on PMI (section Positive PMI) explaining why it makes sense to set negative values to 0:
"negative PMI values (which imply things are co-occurring less often than we would expect by chance) tend to be unreliable unless our corpora are enormous" and also by a concern that "it's not clear whether it's even possible to evaluate such scores of 'unrelatedness' with human judgment"
https://en.wikipedia.org/wiki/Pointwise_mutual_information
How should I interpret this?