@inproceedings{steele-2026-annotation,
    title = "Annotation Entropy Predicts Per-Example Learning Dynamics in {L}o{RA} Fine-Tuning",
    author = "Steele, Brady",
    editor = "T.Y.S.S., Santosh  and
      Rodriguez, Juan Diego  and
      de Gibert, Ona",
    booktitle = "Proceedings of the 64th Annual Meeting of the {A}ssociation for {C}omputational {L}inguistics (Volume 4: Student Research Workshop)",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/revision-previews/2026.acl-srw.11/",
    doi = "10.18653/v1/2026.acl-srw.11",
    pages = "129--141",
    ISBN = "979-8-89176-393-7",
    abstract = "Annotator disagreement on tasks like natural language inference (NLI) reflects genuine linguistic ambiguity, yet most fine-tuning recipes treat every example as equally learnable. We ask whether this external signal of ambiguity predicts \textit{per-example} learning dynamics under LoRA, the most widely used parameter-efficient fine-tuning method, and find that it does. Correlating annotation entropy (from ChaosNLI{'}s 100 labels per example) with per-example area under the loss curve (AULC) on SNLI and MNLI, the correlation is positive in all 25conditions tested (Spearman $\rho {=} 0.06$-0.43), with decoder-only models showing stronger correlations than encoders at matched LoRA rank. More strikingly, under LoRA contested examples exhibit \textit{un-learning}: their gold-label loss \textit{increases} during training, a pattern that is largely absent under full fine-tuning and IA$^3$ in the encoder setting where matched comparisons are available, and that we also observe under LoRA on two decoder-only models. The effect survives partial-correlation controls and replicates across seeds and datasets. A preliminary noise-injection experiment is consistent with these findings."
}