Tags: * make a findings decision in each cycle, separate from conference decision * link meta/reviews to conferences * against fully automatic matching
"I agree that each reviewing cycle should include a decision-making round for the next conference. But
my proposal is that after the reviews are written, the meta-reviewer should write a meta-review and
answer the question if the paper should (at least) be accepted to Findings of ACL, with an editor signing
off on that decision (by 2 months after submission). This decision is given to the authors immediately.
If yes by then default it is considered for the next conference, but the authors can then opt out if
they'd rather revise the paper first or have it considered at a later conference with or without revision.
In the following month, the conference acceptance decision is then made by the conference's program
committee (area chairs / programme chairs), who write another metareview to justify the conference decision
(only). If the paper is then not accepted to the conference, the authors can (a) revise and resubmit,
keeping the same reviewers (b) ask that it be reconsidered at a future conference without revising the
submission (with some limitation on how many times this can be done) or (c) say they are happy with
it being published in Findings, which becomes the default if it is not accepted after some time. If
the paper is reconsidered for a future conference without a revised submission, that conferences programme
committee reconsiders it and write a new metareview for their decision (but the original metareview
for the Findings decision is sticky). If the paper is not accepted to Findings, it cannot be considered
for the conference and the paper has to be resubmitted (and the metareviewer can specify a timeframe
for how long they have to wait for resubmission). In such a resubmission there should be the option
to request new reviewers. Regarding tracks: I do think there should be tracks to organize reviewers,
but papers can be assigned to tracks semi-automatically rather than relying on authors to specify them.
Tracks can then also be created and changed dynamically (e.g. create a prompt engineering track at short
notice). If only we had any expertise in unsupervised topic models. But the assignment of reviewers
to papers should be done by action editors, assisted by a system to automatically recommend reviewers.
"