Future of Reviewing Survey: Analysis

Take-aways

Take-aways

Preprocessing

Fewer than 10 comments were shortened by removing personally identifiable information. Semicolons were replaced with SEMICOLON. Empty comments are not included.

Data

raw data (comments) plus assigned codes

(high-level) topics

mapping of codes to topics

Codes

general

against ARR (20): 008 014 032 045 046 069 092 125 132 162 163 166 168 188 205 222 224 239 258 266

thank you to ARR team (12): 015 031 062 093 099 105 106 134 172 203 220 232

return to pre-ARR system (7): 053 076 092 166 184 188 237

for ARR (5): 007 088 106 131 262

for rolling review (3): 070 199 252

for kummerfeld proposal (3): 035 087 263

against rolling review (3): 041 191 253

adopt ICLR model (3): 163 199 208

against findings (2): 068 168

proposal: 2 week review, two week author response, two week final decision (1) * increase acceptance rate for findings (1) * in favor of infrastructure, but what exactly is part of it? (1) * for integrated system (1) * for infrastructure (1) * for audit of ARR (1) * for ARR and for pre-ARR system (1) * Covid probable reason for reviewer behavior (1) * concrete proposal for how to proceed with ARR: review, analysis, proposal, community consultation, decision (1) * against integrated system (1) * against "permanent" workshops like sigmorphon (1) * adopt TACL model (1) * adopt ICWSM model (1) *

rolling review

against revise-and-resubmit (2): 162 168

utility of revise-and-resubmit unclear (1) * shorten review period for resubmissions (1) * punish new submissions that are resubmissions (1) * Martin Haspelmath on revise-and-resubmit: href=https://dlc.hypotheses.org/2333 (1) * for single-conference rolling review (1) * for revise-and-resubmit (1) * for multi-conference rolling review (1) * consider frequent deadlines without revise-and-resubmit (1) * ban resubmission of bad papers (1) *

openreview

link all versions in openreview (8): 067 113 129 146 160 204 208 268

against openreview (4): 056 070 210 252

for openreview (2): 144 168

improve openreview interface (1) *

reviewing (general)

poor ARR meta/review quality (24): 001 004 009 027 032 054 060 063 068 096 115 136 143 159 166 179 199 219 235 251 253 265 267 268

against year-round reviewing (7): 002 023 031 043 113 156 166

need new reviewing criteria (6): 112 115 149 195 243 265

against junior reviewers (6): 095 149 154 190 209 239

against 4 reviews (5): 006 085 116 128 219

pay reviewers (4): 011 057 227 239

fix late reviews (4): 015 087 118 234

prohibit authors from reviewing for a conference they submit to (3): 066 095 178

incentivize reviewers (3): 019 109 128

for subreviewers (3): 052 147 149

senior reviewers are often bad reviewers (2): 056 145

release reviews earlier (2): 167 175

publish meta/reviews (2): 094 130

penalize bad reviewers (2): 128 219

make reviewer stats available (2): 007 048

improve interface for reviewers (2): 037 166

for reviewer diversity (2): 009 184

for authors as mandatory reviewers (2): 097 229

appeals process for bad reviews (2): 139 146

against structured review forms (2): 202 245

video conference for new (meta)reviewers (1) * use same scale for meta/reviews (1) * too much discrepancy between reviews and metareview (1) * the community does not have enough competent reviewers for all submissions (1) * short/long papers should have short/long review forms (1) * senior people do not have time for reviewing (1) * reviewing record should be factor in paper recommendation (1) * reviewer management too aggressive (1) * remove reviewers who guess author identity (1) * prohibit authors from reviewing competing papers (1) * previous reviews are often ignored (1) * predict need for emergency reviewers and recruit accordingly (1) * no reviewers from outside NLP (1) * no editing of reviews after release (1) * NLP reviewers do not have sufficient ML expertise (1) * max of 4 papers per reviewer per cycle (1) * many authors are not qualified reviewers (1) * make clear how to reach ARR (for authors, reviewers etc) (1) * let meta/reviewers rate each other (1) * increase reviewer diversity (1) * improve management of junior reviewers (1) * improve interface for reviewer assignment (1) * improve emergency reviewing (1) * force senior people to review (1) * force authors to review (1) * for paper updates during review process (1) * for more informative metareviews: potential for major conference acceptance, long or short etc (1) * for 4 reviews (1) * fix reviewer signup process (1) * fix random removal of reviewers (1) * first time authors should not review (1) * fire late (meta)reviewers (1) * financial rewards for good reviewers and authors (1) * favor/disfavor papers by good/bad reviewers for acceptance (1) * ensure each paper has senior reviewer (1) * enforce reviewer diversity (1) * emergency reviewing is too stressful (1) * do not force non-authors to review (1) * do not change score definitions (1) * disconnect between review scores and acceptance decision (1) * create reviewer certification program (1) * concerns about commitment of partially reviewed papers (1) * change "commitment" into "consideration-for", i.e., revision and reviewing in parallel to commitment (1) * ARR should sign up reviewers for a specific conference (1) * ARR bad fit for interdisciplinary reviewers (1) * against commitment of partially reviewed papers (1) * ACL reviewing is too complex (1) *

meta/reviews specific to conferences

link meta/reviews to conferences (9): 047 050 082 093 111 122 197 235 242

link meta/reviews to acceptance (7): 025 080 089 144 146 186 236

for decoupling reviews and conferences (2): 103 259

separate reviewing for each conference (1) * for per-conference reviewing (1) * against decoupling of meta/reviews (1) *

opt-out/opt-in

for opt-out (4): 012 015 084 245

for easy opt-out (3): 019 039 077

fix opt-out (2): 012 212

sell opt-out better by contacting each author right after submission (1) * opt-out unfair to interdisciplinary and junior researchers (1) * improve interface for opt-in / opt-out (1) * for opt-in because of interdiscipinary authors (1) * for opt-in (1) * against opt-out for all authors (1) * against easy opt-out (1) *

sticky reviews

for sticky meta/reviews (5): 010 050 119 173 231

soften sticky meta/reviews (4): 036 138 145 240

for stricter sticky meta/reviews (3): 058 170 223

against sticky meta/reviews (3): 180 235 259

for sticky meta/reviews (keep old reviews even when reviewers are changed) (1) *

communication

improve communication with reviewers (15): 012 016 021 024 031 049 055 113 161 166 200 216 219 228 248

improve communication (5): 015 098 226 252 258

improve communication, reviewer-(S)AC (1) * improve communication with reviewers (scores) (1) * improve communication with authors (1) * improve communication between area chair and reviewers (1) * improve communication about ARR problems (1) * improve communication (use less email) (1) *

discussion/interactivity

for author response (7): 001 096 160 175 196 213 254

for discussion, author-reviewer (6): 165 175 176 184 196 239

give reviewers time to revise based on author response (1) * for reviewer discussion (1) * for meta/reviewer discussion (1) * for discussion, reviewer-AC (1) * for discussion, author-reviewer and/or reviewer-meta/reviewer (1) * for discussion, author-reviewer and/or reviewer-AC (1) * for discussion, AC-SAC (1) * for discussion (1) * better AC/SAC-PC interaction (1) * against SAC interactions with ACs and reviewers to avoid inefficiency and confusion (1) *

submission-reviewer matching

improve matching (21): 009 020 027 049 063 077 095 128 151 152 155 158 174 175 176 179 194 245 248 255 258

against fully automatic matching (2): 111 179

measure quality of matching (1) * improve interface for reviewer matching (1) *

cycles

against 8 week cycle (4): 016 069 089 187

for longer cycles (3): 043 065 200

for 12 week cycle (2): 075 118

short cycle length means to much load on meta/reviewers (1) * make a findings decision in each cycle, separate from conference decision (1) * for short review cycle (1) * for 8 week cycle or longer (1) * for 8 week cycle (overlapping) (1) * for 8 week cycle (1) * for 6 week cycle (1) * cycle too short (1) *

reviewer load

review load proportional to submissions (7): 021 026 082 128 168 214 221

ARR review load too high (6): 092 097 108 156 177 249

improve interface for load management (4): 023 108 128 230

against review load proportional to submissions (3): 088 184 240

increase number of / load of senior reviewers (2): 149 219

ARR increases review load (2): 101 266

ARR does not decrease review load (2): 176 253

reduce review load for junior reviewers (1) * need statistics on (increased?) load due to ARR to make any decisions (1) * monthly reviewing increases review load? (1) * make ARR review load predictable (1) * load management important (1) * increase load of senior reviewers (1) * equal spacing of community review load (1) *

reviewer performance

reviewer performance: rate, track, act upon (5): 000 026 127 137 184

reviewer performance: track how much authors review (1) * reviewer performance: flag bad reviewers (1) *

reviewer pool

good reviewers do not get invited (4): 022 083 098 152

senior people do not get invited (1) *

mentoring

opinion about mentoring depends on implementation details (1) * mentor area chairs (1) * for reviewer training (1) * for mentoring (1) * against reviewer training (1) * against ARR mentoring (1) *

blinding

blind reviewers to each other (2): 026 086

introduce new completely open, "no-blinding" conference (1) * for double blind (1) * blind (S)ACs to reviewers (1) * allow reviewers to unblind themselves (1) *

ACs/SACs (other)

how to become AC? (2): 185 255

AC should read paper (2): 086 245

no ACs, just SACs (1) * increase AC seniority (1) * increase (S)AC diversity (1) * define roles of (S)ACs (1) * decrease arbitrariness / increase diversity of AC/SAC selection (1) *

submissions and acceptance (other)

distribute submissions over year (2): 001 199

charge submission fees (2): 042 227

time from submission to acceptance is too long (1) * shorten time between submission and acceptance (1) * several rounds of acceptance (1) * make ARR the sole submission system (1) * make all acceptance decisions for the next conference (1) * limit submissions per person (1) * incentivize early submission (1) * for workshop submissions to ARR (1) * enforce higher submission standards, e.g., punish authors who hide information (1) * EMNLP submission modalities were poorly communicated (1) * double submissions to softconf / ARR confusing (1) * allow double submissions both within ARR and ARR/softconf (1) *

ARR (other)

ARR too complex (3): 008 104 239

publish ARR 'decisions' on papers (1) * pre-ARR system for EACL/AACL (1) * improve ARR tech support (1) * hire more staff, not only for ARR (1) * fully test ARR system before launch (1) * communicate plan for ARR (1) * ARR too stressful (1) * ARR not transparent (1) * ARR not timely (1) * ARR is not a helpful resource for PCs (1) * ARR harms diversity (1) * ARR good for authors (1) * ARR excludes interdisciplinary research (1) * ARR decisions improperly influenced (1) * ARR bad for diversity (1) * ARR alienates from ACL (1) * any decision on ARR should be democratic (1) * against new ARR roles (1) *

tracks

for tracks (12): 005 018 027 054 122 128 143 174 181 194 209 252

against tracks (2): 068 192

use keywords instead of tracks (1) * for tracks, but only for a subset (1) * for partial tracks (1) * do not use ARR for special themes/tracks (1) * against current set of tracks used at conferences (1) *

desk rejects

expand desk rejects (3): 079 081 151

for paper triage (2): 202 220

soften desk reject (1) * against expanding desk rejects (1) *

COIs

allow authors to enter COIs (2): 038 238

make COIs private (1) * improve COI detection (acknowledgments) (1) * ensure that COIs are uptodate (1) *

conferences (other)

publish papers before conference (2): 135 146

equal spacing of conferences (2): 143 146

reduce the number of conferences (1) * reduce in limbo times by increasing number of conferences (1) * make conferences more inclusive (LREC model) (1) * make conference PC honorary role (1) * increase number of ACL conferences (1) * if ARR continues, there should be only one conference, potentially with several instances (1) * fundamental flaw in ARR : papers are submitted for conferences (1) * for diversity of conferences (1) * differentiate tier 1 and tier 2 conferences (1) * ARR only for conferences (1) * ARR is the only way to scale our conferences (1) * ARR and legacy should have same deadline for a conference (1) *

anonymity

abolish anonymity policy (4): 028 158 182 267

for anonymous open review (1) * ban arxiv (1) *

data collection

improve data collection interface (1) * for data collection (1) * data collection should be opt-in, not opt-out (1) * concerned about terms of use of data collection (1) *

ethics

track ethical breaches in the infrastructure (for both authors and reviewers) (1) * make handling ethics easier (1) * incorporate community feedback on ethics/reproducibility (1) *

journals

put resources into TACL (3): 061 114 171

turn ACL anthology into a journal (1) * move ACL's focus from conferences to journals (1) * invest in journals not in ARR (1) *

survey/future-of-reviewing document

survey confusing (3): 008 062 242

the survey should have explained what an action editor is (1) * survey unclear about partially reviewed papers (1) * survey question about extending ARR functionality poorly designed (1) * survey does not include masters students (1) * problems with announcement of survey (1) * future of reviewing document should have defined the roles of (senior) area chair / action editor (1) * decisions should be made based on evidence and research not based on surveys (1) *

miscellaneous

march delay unacceptable (4): 004 093 164 234

for FTE (4): 026 042 056 221

for bidding (3): 143 155 181

use open research knowlege graph (1) * reduce online registration cost (1) * one FTE not enough (1) * maximize agency (1) * increase ACL paper length (1) * improve website (1) * horrible experience with special theme (1) * get Nihar Shah to give a tutorial (1) * focus ACL on "computation and language", encourage linguists to leave (1) * change citation style (1) * anywhere on earth time is confusing (1) * against emphasis on career concerns (1) * ACL is welcoming to new researchers (1) *