@inproceedings{tratzsch-etal-2026-steerforce,
    title = "{S}teer{F}orce at {S}em{E}val-2026 Task 11: Reducing Content Effects Using Layered Activation Steering",
    author = "Tratzsch, Noah  and
      Al-Raian, Asmaa  and
      Marreddy, Mounika  and
      Mehler, Alexander",
    editor = "Kochmar, Ekaterina  and
      Ghosh, Debanjan  and
      North, Kai  and
      Komachi, Mamoru",
    booktitle = "Proceedings of the 20th {I}nternational {W}orkshop on {S}emantic {E}valuation (2026)",
    month = jul,
    year = "2026",
    address = "San Diego, California, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl-workshops/2026.semeval-1.143/",
    pages = "1050--1055",
    ISBN = "979-8-89176-414-9",
    abstract = "Large language models exhibit content effects, where surface plausibility interferes with formal logical reasoning. In SemEval-2026 Task 11, this appears as a performance gap between plausibility-aligned and plausibility-conflicting syllogisms, reflecting directional content bias. We address this issue using inference-time activation steering, modeling bias as a geometric deviation between plausibility-driven and validity-driven representations. We introduce a layered steering framework that combines Activation Transport (ACT) with input-adaptive contrastive steering (K-CAST), applied to layers identified through sensitivity analysis. This architecture-aware strategy enables targeted interventions without retraining.On BERT, sequential multi-layer steering improves validity accuracy from 77.1{\%} to 82.3{\%} while reducing bias by 75{\%}. In contrast, for the decoder-only Qwen2.5-1.5B-Instruct, a single mid-to-late layer intervention reduces bias from 0.26 to 0.04 with modest accuracy gains, whereas multi-layer steering offers no additional benefit. These results reveal a fundamental architectural distinction: encoder-based models benefit from distributed low-intensity corrections, while decoder-only instruction-tuned models concentrate reasoning signals within a narrow late-layer band. Our findings demonstrate that effective bias mitigation requires architecture-aware activation steering."
}