Request for Specific Technical Feedback on AKBC Shared Task Submission #16

[Sachin Gupta]

Dear AKBC Shared Task Organizers,

Thank you for your decision. We understand the concerns regarding scientific coherence and rigor, and we are not writing to dispute the outcome.

However, as the feedback outlines substantial problems without identifying representative passages, we would appreciate further clarification to help us learn from this submission. 

We have rechecked the reported numerical results and were able to reproduce them. However, we recognize that this may not address the broader concerns raised in the feedback.

Could you please identify two or three specific claims or sections that motivated the decision?

Thanks,
Submission # 16

[Simon Razniewski]

Dear Submission 16,

The paper is full of incomprehensible GenAI jargon, we could not comprehend it.

Examples

 - cache-only audit

 - many-valued roster completion

 - manifests

 - scorer-parity implementation

 - no-finetuning attestation

We are not against GenAI per se, but you should make strong effort to make your paper accessible to the relevant community. The community has developed a shared language for many years, if you write, without an obvious reason, completely different, that's not useful for the community.

Best wishes,

Simon

(on behalf of the AKBC shared task organizers)

--
You received this message because you are subscribed to the Google Groups "akbc2026-shared-task" group.
To unsubscribe from this group and stop receiving emails from it, send an email to akbc2026-shared-...@googlegroups.com.
To view this discussion, visit https://groups.google.com/d/msgid/akbc2026-shared-task/1acb2ec1-62b4-45a2-8821-115b2389ab2dn%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Sachin Gupta]

Dear Simon,

Thank you for sharing the examples. This helps us understand the concern much more clearly.

Those phrases came from the terminology we used internally while building and documenting the system. We now realize that we should not have assumed they would be familiar to everyone. For example, “cache-only audit” simply meant evaluating previously saved model outputs, and “manifests” referred to files recording the model version, configuration, and checksums. Similarly, “scorer-parity implementation” meant that we checked our local evaluation against the official script, while “no-finetuning attestation” was only intended to say that we used the published model weights without updating them.

Our goal was to make the work transparent and reproducible, but we can see that this terminology made the paper harder to understand. We should have described the method using simpler and more familiar terms such as candidate generation, filtering, aggregation, and one-to-many relations.

We accept the decision and appreciate you taking the time to clarify the issue. This is useful feedback, and we will apply it carefully in our future work.

Best regards,
Submission #16