@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://aclanthology.org/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."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="tratzsch-etal-2026-steerforce">
<titleInfo>
<title>SteerForce at SemEval-2026 Task 11: Reducing Content Effects Using Layered Activation Steering</title>
</titleInfo>
<name type="personal">
<namePart type="given">Noah</namePart>
<namePart type="family">Tratzsch</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Asmaa</namePart>
<namePart type="family">Al-Raian</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mounika</namePart>
<namePart type="family">Marreddy</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alexander</namePart>
<namePart type="family">Mehler</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 20th International Workshop on Semantic Evaluation (2026)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Ekaterina</namePart>
<namePart type="family">Kochmar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Debanjan</namePart>
<namePart type="family">Ghosh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kai</namePart>
<namePart type="family">North</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mamoru</namePart>
<namePart type="family">Komachi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California, USA</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-414-9</identifier>
</relatedItem>
<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.</abstract>
<identifier type="citekey">tratzsch-etal-2026-steerforce</identifier>
<location>
<url>https://aclanthology.org/2026.semeval-1.143/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>1050</start>
<end>1055</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T SteerForce at SemEval-2026 Task 11: Reducing Content Effects Using Layered Activation Steering
%A Tratzsch, Noah
%A Al-Raian, Asmaa
%A Marreddy, Mounika
%A Mehler, Alexander
%Y Kochmar, Ekaterina
%Y Ghosh, Debanjan
%Y North, Kai
%Y Komachi, Mamoru
%S Proceedings of the 20th International Workshop on Semantic Evaluation (2026)
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, USA
%@ 979-8-89176-414-9
%F tratzsch-etal-2026-steerforce
%X 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.
%U https://aclanthology.org/2026.semeval-1.143/
%P 1050-1055
Markdown (Informal)
[SteerForce at SemEval-2026 Task 11: Reducing Content Effects Using Layered Activation Steering](https://aclanthology.org/2026.semeval-1.143/) (Tratzsch et al., SemEval 2026)
ACL