United States

While Large Language Models (LLMs) excel in many general NLP tasks, their formal reasoning capabilities are often compromised by content effects, demonstrating a measurable bias towards real-world plausibility. In this paper, we present our system for SemEval-2026 Task 11, which evaluates the ability of models to disentangle formal logic from content across 12 languages with and without distractor premises. We address this challenge using mDeBERTa-v3 networks fine-tuned on a synthetic, rule-based dataset of syllogistic schemes to avoid the semantic noise of LLM-augmented data. To explicitly decouple plausibility from logical structure, our training pipeline employs a multi-objective loss function combining Adaptive Group Distributionally Robust Optimization (DRO), a scheduled differentiable bias penalty, and KL-Divergence consistency regularization. Our system achieved #1 ranks and perfect Ranking Scores (100.0) with 0.00% bias and 100.0% accuracy on Subtask 1 (English), Subtask 2 (Noisy English), and Subtask 3 (Multilingual). On the highly complex Subtask 4 (Noisy Multilingual), the system achieved the 6th rank with 89.06% Accuracy and F1-score, alongside a limited 2.89% Bias and a 37.78 Ranking Score. Our dataset generation engine and codebase are publicly available to facilitate future work on robust logical reasoning.

ACL 2026

HABIB_TAZ at SemEval-2026 Task 11: Disentangling Formal Logic from Content via Synthetic Training and Multi-Objective Optimization

content effect

syllogistic reasoning

synthetic data augmentation

multi-objective optimization

multilingual nlp

It is my great pleasure and honor to welcome you to the 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026), held in San Diego, California, USA, July 2–7, 2026. ACL 2026 continues our fields tradition of excellence in scholarship, innovation, and inclusivity, and I am deeply grateful to the many volunteers who have worked tirelessly to bring this event to life.

It is exciting to see the many developments in the field reflected at our conference, such as the advancement of large language models (born out of our community) into many application directions. This is not without challenges, partly due to unsolved problems, and partly due to misuse of the technology. All this is reflected well at ACL 2026. The conference also reflects that the field is not only about advancing technology, but also the science of fundamentally understanding language and how it is use by many different speaker communities. <br>
[Full Message from General Chair Philipp Koehn](https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing) <br><br><br>
<html><button style="display: inline-flex; align-items: center; justify-content: center; white-space: nowrap; border-radius: 9999px; font-weight: bold; background: #7c3aed; color: white; font-family: 'Space Grotesk', sans-serif; height: 40px; font-size: 16px; padding: 0 20px; border: none; cursor: pointer" onclick="window.open('https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing')">Conference Handbook</button></html>
<br><br><br><br>

To access this event page, you need to log in with the **email address you registered with**. <br>Access credentials will be sent to your email from Underline -  subject line "Welcome to ACL 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://2026.aclweb.org/registration/) first.

Registration Required

The 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026) will take place in San Diego, California, from July 2 - 7, 2026.

workshop paper

This paper describes our system for the MT-RAG (Multi-Turn Retrieval-Augmented Generation) shared task, which addresses the challenge of multi-turn conversational question answering using retrieval-augmented generation. We participated in three sub-tasks of Task 8: Task A (retrieval), Task B (generation with reference passages), and Task C (end-to-end RAG). For Task A, we evaluated multiple retrieval approaches including BM25, BGE, and hybrid methods, achieving best performance with ELSER (Elastic Learned Sparse EncodeR) with nDCG@5 of 0.4018 (Rank 24/38). For Task B, we employed the Mistral-7B-Instruct-v0.2 model via HuggingFace for response generation using gold reference passages, achieving a harmonic mean score of 0.6976 (Rank 13/26). For Task C, we combined ELSER retrieval with Mistral-7B generation, using top-5 retrieved passages as context, achieving a score of 0.4289 (Rank 23/29). Our system demonstrates the effectiveness of learned sparse retrieval methods and instruction-tuned models for multi-turn conversational RAG scenarios.

IITKanBDone at SemEval-2026 Task 8: MTRAGEval - Evaluating Multi-Turn RAG Conversations

This paper describes HYPOLDET, the system submitted by team pfr821 to SemEval-2026 Task 9 (Polarization Detection, Subtask 1), a binary classification task over 22 typologically diverse languages. Our approach combines three complementary contributions. We first extend XLM-RoBERTa-Large with a custom transformer encoder layer and a learned attention-based pooling mechanism (Hybrid Architecture), allowing the model to aggregate token-level signals beyond the [CLS] representation. We then augment training data through a targeted LLM-based synthetic generation pipeline (Grok API), producing culturally grounded examples for low-resource and imbalanced languages. Finally, we address class imbalance at the training level through an imbalance-aware regime combining a per-language balanced batch sampler, weighted focal loss, and label smoothing. Our best single model achieves an unweighted macro-averaged F1 of 0.796, and a lightweight ensemble reaches 0.798, ranking in the top 10 for 7 languages and 2nd place for Hausa.

pfr821 at SemEval-2026 Task 9: Multilingual Polarization Detection via Hybrid XLM-RoBERTa with Targeted Data Augmentation and Imbalance-Aware Training

This submission presents a zero-shot embedding-based approach for SemEval-2026 Task 4 on Narrative Story Similarity. The system employs the pretrained sentence-transformers/all-mpnet-base-v2 model within a bi-encoder architecture to generate 768-dimensional story embeddings. Narrative similarity is modeled using cosine similarity in embedding space for comparative prediction in Track A and representation generation in Track B. The approach does not involve task-specific fine-tuning and treats narrative comparison as a geometric proximity problem. Experimental results and error analysis highlight the strengths of pretrained semantic encoders in capturing thematic similarity, while revealing limitations in modeling deeper narrative structure and causal progression.

Cryptix at SemEval-2026 Task 4: Zero-Shot Bi-Encoder Modeling for Narrative Story Similarity - A Sentence Transformer Approach

The rise of Large Language Models (LLMs)
has increased the need for reliable detection of
machine-generated code. This paper presents
a low-resource, hybrid detection framework
developed for for SemEval-2026 Task 13 ,de
signed to operate efficiently without the com
putational overhead of end-to-end fine-tuning
of large models. Our approach combines
(i) comprehensive feature extraction pipeline
that calculates interpretable software metrics
capturing stylistic and structural properties of
code, and (ii) we leverage the semantic ca
pabilities of GraphCodeBERT by extracting
frozen embeddings from its pre-trained en
coder to model semantic and data-flow infor
mation while preserving generalizability. This
fusion enables efficient detection of machine
generated code across multiple programming
languages (Python, C++, Java, and Go) and
improves robustness under out-of-distribution
settings. This feature-driven fusion offers a
competitive, computation-efficient alternative
to purely LLM-based fully fine-tuned models,
achieving an F1-score of 38.26.

Königsberg at SemEval-2026 Task 13: Beyond Language Models: A Low-Resource Feature-Driven and Data-Flow Embedding Approach for Machine-Generated Code Detection

Aspect-Based Sentiment Analysis (ABSA) has
evolved to capture continuous affective states,
posing challenges for traditional classification
models. We adopt a hybrid approach tailored
to the varying complexities of the subtasks. For
Task 1 (Valence-Arousal Regression), we em-
ploy a discriminative architecture using pre-
trained DeBERTa encoder with a MeanPool-
ing mechanism to directly regress continuous
sentiment scores. For Tasks 2 and 3, which re-
quire complex structural extraction of opinion
triplets and quadruplets, we utilize a generative
approach by fine-tuning the Qwen3-4B-Instruct
large language model via 4-bit QLoRA. Our
system effectively handles both precise numer-
ical regression and complex structural text gen-
eration, achieving competitive results across
the English laptop and restaurant domains.

looploop at SemEval-2026 Task 3: A Dimensional Aspect-Based Sentiment System with DeBERTa Regression and Qwen3 Instruction Fine-Tuning

This paper describes the PFW system for SemEval-2026 Task 6 (CLARITY), which addresses the classification of response clarity and evasion techniques in political interview question-answer pairs. Rather than relying on large language model prompting, we pursue a competitive non-LLM approach based on fine-tuning DeBERTa-xlarge and DeBERTa-v3-large with a multi-seed ensemble strategy: 5-fold cross-validation with 10 random seeds yields 50 models per architecture, combined through simple logit averaging. Our system achieves a macro F1 of 0.76 on Subtask 1 (clarity-level classification) and 0.50 on Subtask 2 (evasion-type classification). We also find that three post-hoc optimization techniques—learned ensemble weights, thresh old calibration, and hierarchical masking— each improve out-of-fold performance yet degrade evaluation scores by 0.02–0.10 F1. This pattern should be interpreted cautiously: the 237-sample evaluation set likely contributes substantial variance, and two of the three degradations fall within the ±0.06 95% CI expected from sampling noise. Still, the consistent directional pattern across all three prediction-level interventions provides suggestive evidence for an optimization paradox, highlighting the risk of overfitting to cross-validation predictions when evaluation data is limited. Our code is publicly available at https://github.com/ Taleef7/semeval-2026-task6.

PFW at SemEval-2026 Task 6: Multi-Seed DeBERTa Ensembles for Political Response Clarity and Evasion Classification

We describe PFW Task 8's system for SemEval 2026 Task 8 (MTRAGEval), a benchmark for multi-turn retrieval-augmented generation across four English-language corpora. Our submission combines BM25, SPLADE-v3, and Jina Embeddings v4 with weighted reciprocal rank fusion for retrieval, plus zero-shot GPT 4o/GPT-4o-mini prompting for generation. Officially, our system ranks 6th of 26 on Task B (H = 0.756), 14th of 29 on Task C (H = 0.533), and 20th of 38 on Task A (nDCG@5 = 0.433). For the camera-ready analysis, we re-run retrieval at the official nDCG@5 cutoff, strengthen the prompt ablation with per-domain statistics and exact tests, and analyze official outputs by answerability and domain. On a balanced 100-example development sample, explicit citation-format instructions—not chain of-thought alone—raise citation use from 4% to 93%, and a fixed-context Task C control improves from H = 0.463 with GPT-4o-mini to H = 0.523 with GPT-4o. Official analytics also show near-perfect UNANSWERABLE handling (H = 0.990) but weak behavior on UNDERSPECIFIED turns, where the system answers or refuses instead of clarifying. Our code is publicly available.

PFW Task 8 at SemEval-2026 Task 8: Lightweight Tri-Fusion Retrieval with Prompt-Engineered Faithful Generation for Multi-Turn RAG

Online polarization, the critical division between social, political, or identity groups, often leads to hate speech and social fragmentation. Detecting polarization, especially across diverse linguistic and cultural contexts, is a critical challenge. This paper presents our submission for SemEval-2026 Task 9, which focuses on detecting online polarization of multilingual, multicultural, and multievent (Naseem et al., 2025). The task is divided into three subtasks: (1) binary polarization detection, (2) multi-label classification of polarization type (e.g., political, racial, religious), and (3) multilabel identification of its manifestation (e.g., stereotype, vilification, dehumanization). For each subtask, we employ fine tune BERT-based transformer models. Model configurations are described in Section 4. The results are evaluated using the F1 macro score. We have achieved scores of 78.6, 55.8, 44.6 on the development
test set for subtasks 1, 2, and 3, respectively. Overall, the results demonstrate the effectiveness of BERT-based models for multilingual polarization detection.

MSqrd at SemEval-2026 Task 9: Detecting Multilingual, Multicultural and Multievent Online Polarization

This paper describes our first-place submission to Task B (generation with reference passages) of the SemEval-2026 Task 8 MTRAGEval shared task on multi-turn retrieval-augmented generation. We propose a heterogeneous ensemble of seven LLMs organised into two groups with distinct prompting strategies, and use a GPT-4o-mini judge to select the best candidate response for each instance. Our system ranked first among 26 teams, achieving a conditioned harmonic mean score of 0.78 and substantially outperforming the strongest 
organiser baseline (0.64). Ablation experiments show that diversity across model families, scales, and prompting strategies is critical: the ensemble consistently outperforms any individual model. We also include Meno-Lite-0.1, a 7B domain-adapted model with a favourable cost--performance trade-off, and present an analysis of MTRAGEval that highlights annotation limitations and directions for benchmark improvement.

RaguTeam at SemEval-2026 Task 8: Meno and Friends in a Judge-Orchestrated LLM Ensemble for Faithful Multi-Turn Response Generation

Large language models suffer from content effects in reasoning tasks, particularly in multilingual contexts. We introduce a novel method that reduces these biases through explicit structural abstraction that transforms syllogisms into canonical logical representations and applies deterministic parsing to determine validity. Evaluated on the SemEval-2026 Task 11 multilingual benchmark, our approach achieves top-5 rankings across all subtasks while substantially reducing content effects and offering a competitive alternative to complex fine-tuning or activation-level interventions.

Content not yet available

Downloads

Next from ACL 2026

IITKanBDone at SemEval-2026 Task 8: MTRAGEval - Evaluating Multi-Turn RAG Conversations

Similar lecture

GheGheGhe at SemEval-2026 Task 11: Decoupling Logic from Belief with Bias-Targeted Fine-Tuning and Neuro-Symbolic Syllogistic Reasoning

Content not yet available

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from ACL 2026

IITKanBDone at SemEval-2026 Task 8: MTRAGEval - Evaluating Multi-Turn RAG Conversations

Similar lecture

GheGheGhe at SemEval-2026 Task 11: Decoupling Logic from Belief with Bias-Targeted Fine-Tuning and Neuro-Symbolic Syllogistic Reasoning

Downloads