Thanks, Matt, for thinking carefully about these issues.

It seems to me that the hardest suggestion to implement would be making conference deadlines continuous. Even without the issue of arXiv, I agree that the current system of December–May deadlines for the major conferences makes little sense. But I don’t know if it’s safe to assume that we can transition instantly to a TACL-like model for all submissions. The advantage of batching submissions in a few deadlines per year is that reviewing can be organized by areas, with bidding and so on, rather than requiring an action editor to recruit a custom set of reviewers appropriate to each paper as it arrives.

An incremental step in that direction could be to institute an early joint deadline and review process in September. Authors with a paper accepted in that round of review would then choose which conference to present it at. Strong submissions would get accepted early; most others would receive a revise-and-resubmit decision, possibly with a constraint like “you may only submit once more this cycle” to keep the reviewing burden in check.

The incentive to submit early is that the paper gets two bites at the apple for a conference. And work that becomes ready for submission in, say, June does not have to wait 6 months to actually be submitted for a major conference. This approach gives greater flexibility to those who have constraints (e.g., visa restrictions) limiting them to certain conferences, and gives conference organizers a way to estimate relative popularity in advance of the regular deadlines so reviewing effort can be allocated accordingly. For the community, it spreads out the reviewing process so some of it happens in the fall.