Tags: * need new reviewing criteria
"Compared to reviewing in the 90s and 00s, the field has dramatically expanded, and I find it alarming
that now people who lack even the most basic linguistic knowledge and readily admit that they never
look at the data seem to constitute the majority of the reviewers. Without tracks this makes it a crapshoot
to publish papers with linguistic insight in a main ACL conference. Reviewers systematically look solely
for progress in machine learning. A typical comment like ""I don't see what the contribution of this
paper is supposed to be - this kind of machine learning model has been used before"" or ""The paper
does not use transformers, which currently is the best model, so reject"" to me indicates giving up
comp ling as a field separate from machine learning. Perhaps it is time to consider splitting the field
into an engineering and machine learning main stream and an academic field also valuing insight into
languages."