A rejection with an average score of 2.83 after rebuttals that went unanswered feels less like a verdict and more like a shrug. For a first solo paper, especially one where the meta-reviewer was positive and the core weaknesses were already addressed in the text, this is the most frustrating kind of outcome. It is not that the work is bad. It is that the process did not engage with it. If you are a master's student trying to get an internship, this is exactly the moment where the system feels like a black box, and the temptation is to either rage-quit or submit the same thing again and hope for a different draw.
Here is the practical truth about the ARR-to-conference pipeline: yes, you can commit to NAACL in December using the same review discussion, but you are not required to, and you should not assume the old reviewers will carry you. The scores you have are decent, not damning, but the silence from the reviewers during rebuttal is a signal. It usually means they have moved on, not that they are convinced. If you resubmit to ARR, you are rolling the dice on new reviewers who will not see the history, which can be freeing but also means you are gambling on their mood. The better move is to treat the existing reviews as a map. Read the 2.5 and the 3s as specific, actionable friction points. If they say the experiments are insufficient, add one more baseline. If they say the writing is dense, tighten it. But do not just resubmit the same file. That is not persistence; that is hoping for amnesia.
What we would tell you directly is this: your goal is not to please EMNLP. Your goal is to build a publication record that signals competence to internship committees. That changes the calculus. If you can revise the paper in a meaningful way in the next month, then yes, submit to ARR again. But if the only change is a rebuttal response that says "we already addressed this," you are not giving new reviewers a reason to advocate for you. New reviewers need a new hook. Maybe that means reframing the contribution around the multimodality track more explicitly, or adding a failure analysis that shows you understand the limits of your method. The Clean Data Starts With Catching AI Slop Before It Skews Your Model post in our publication is a reminder that reviewers are often filtering for signal in noisy outputs. Your job is to make the signal louder.
The deeper issue here is that the current review process rewards perceived novelty and punishes solo authors who cannot lean on a lab's reputation. That is not a reason to stop. It is a reason to be strategic. For your situation, the highest-leverage action is not to chase a single top-tier acceptance but to submit to a venue where the review culture is more dialogic, or to use this draft as the foundation for a workshop paper that you can cite later. The Exploring Real-World Computer Vision: Deployments, Edge Models, and Current Challenges piece we ran highlights how much practical insight gets lost in the gap between academic benchmarks and deployment reality. Your paper, whatever its scores, is part of that conversation. Do not let a 2.83 define its value.
The concrete detail to watch is this: when you resubmit, do not ask "will the old reviewers help?" Ask "what would make a new reviewer read this and think it is obviously an accept?" If you cannot answer that in one sentence, you are not ready to resubmit yet. And if you can, then the rejection was just a detour, not a dead end.