United States

We present NarSiL (Narrative Similarity Learners), our system for SemEval-2026 Task 4 Track A on Narrative Story Similarity. NarSiL employs a two-stage architecture: a Mixture-of-Experts (MoE) initial classifier that also leverages supermajority voting across three large language models (Gemma-3-12B, GPT-3.5-turbo-instruct, and Gemini-2.5-Flash) over multiple runs, followed by a structured three-pathway fallback for ambiguous cases. The three pathways correspond directly to the task&#39;s three core similarity components, abstract theme, narrative outcome, and course of action. Each path yields a similarity score corresponding to its respective component, and the scores are then combined through a weighted aggregation step. NarSiL achieves 64.25% accuracy on the official test set. An improved score of 70.25% is obtained by considering only the supermajority voting of GPT, followed by the previously described fallback.

ACL 2026

NarSiL at SemEval-2026 Task 4: A Multi-Expert, Multi-Pathway System for Narrative Story Similarity

mixture of experts

abstract meaning representation

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.

We present NarSiL (Narrative Similarity Learners), our system for SemEval-2026 Task 4 Track A on Narrative Story Similarity. NarSiL employs a two-stage architecture: a Mixture-of-Experts (MoE) initial classifier that also leverages supermajority voting across three large language models (Gemma-3-12B, GPT-3.5-turbo-instruct, and Gemini-2.5-Flash) over multiple runs, followed by a structured three-pathway fallback for ambiguous cases. The three pathways correspond directly to the task's three core similarity components, abstract theme, narrative outcome, and course of action. Each path yields a similarity score corresponding to its respective component, and the scores are then combined through a weighted aggregation step. NarSiL achieves 64.25% accuracy on the official test set. An improved score of 70.25% is obtained by considering only the supermajority voting of GPT, followed by the previously described fallback.

workshop paper

This paper presents a multi-paradigm approach to the first two subtasks of SemEval-2026 Task 11. For the first subtask, we explore two complementary strategies: a Llama-3 8B PEFT Majority Vote Ensemble, trained with bias-targeted augmented data, and a hybrid approach that separates LLM processing from logical reasoning, converting sentences into canonical logical forms for deterministic analysis. The hybrid approach is further extended to the second subtask. Official results placed us 17th in the first subtask and 15th in the second. Post-evaluation analysis indicates that our best model achieved perfect accuracy on the first subtask and revealed several errors in the ground truth data. After identifying certain implementation issues in the second subtask approach, the F1 retrieval score increased to over 98%, which would place us within the top 5 on the leaderboard.

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

This submission investigates efficient multi-turn retrieval under constrained computational settings. We analyze how passage granularity and conversational query rewriting affect retrieval effectiveness across four benchmark domains. Using compact, locally deployable components, we show that smaller passage segmentation improves early-rank performance and that lightweight keyword-oriented query reformulation substantially enhances dense retrieval quality.

Importantly, we observe that rewriting interacts differently with encoder backbones: some compact models benefit significantly from increased query specificity, while others degrade, indicating sensitivity to rewrite-induced distribution shifts. Our findings demonstrate that competitive multi-turn retrieval does not require large proprietary models, but can emerge from principled structural and preprocessing design choices. The results highlight the importance of aligning chunking strategy, rewriting policy, and encoder characteristics in resource-efficient MT-RAG systems.

AKCIT-UFG at SemEval-2026 Task 8: Structured Chunking and Optimized Query Reformulation for Efficient Multi-Turn Retrieval

The widespread use of LLMs in software development has made the detection of machine-generated code a pressing challenge, particularly when models must generalize across programming languages and domains. We present a lightweight, LLM-free pipeline that combines stylometric feature extraction with a LightGBM classifier and explicitly prioritizes structural generalization over deep semantic modeling. Despite its simplicity, the method achieves a Macro F1 of 0.70–0.72, more than doubling the CodeBERT baseline (0.30) in SemEval-2026 Task 13 Subtask A, while operating without GPUs or any fine-tuning.

AKCIT at SemEval-2026 Task 13: A Lightweight LightGBM Baseline for Cross-Language Detection of LLM-Generated Code

This paper describes our system submitted to SemEval-2026 Task 11: Disentangling Content and Formal Reasoning in Large Language Models. We present an efficient modular neuro-symbolic approach, combining a symbolic prover with small reasoning LLMs (4B parameters). The system consists of an LLM-based parser that translates natural language syllogisms to a first-order logic (FOL) representation, an automated theorem prover, and two optional modules: machine translation for multilingual inputs and a symbolic retrieval component for the identification of relevant premises. The system achieves competitive accuracy and relatively low content effect on most subtasks. Our ablations show that this approach outperforms LLM-based zero-shot baselines in this parameter size range, but also reveal limited multilingual capabilities of small LLMs. Finally, we include a discussion of the task's main ranking metric and analyze its limitations.

UFAL-CUNI at SemEval-2026 Task 11: An Efficient Modular Neuro-symbolic Method for Syllogistic Reasoning

We present the shared task on narrative similarity and narrative representation learning — NSNRL (pronounced "nass-na-rel").
The task operationalizes narrative similarity as a binary classification problem: determining which of two stories is more similar to an anchor story.
We introduce a novel definition of narrative similarity, compatible with both narrative theory and intuitive judgment.
Based on the similarity judgments collected under this concept, we also evaluate narrative embedding representations.
We collected at least two annotations each for more than 1,000 story summary triples, with each annotation being backed by at least two annotators in agreement.
This paper describes the sampling and annotation process for the dataset; further, we give an overview of the submitted systems and the techniques they employ.
We received a total of 71 final submissions from 46 teams across our two tracks.
In our triple-based classification setup, LLM ensembles make up many of the top-scoring systems, while in the embedding setup, systems with pre- and post-processing on pretrained embedding models perform about on par with custom fine-tuned solutions.
Our analysis identifies potential headroom for improvement of automated systems in both tracks.
The task website includes visualizations of embeddings alongside instance-level classification results for all teams.

SemEval-2026 Task 4: Narrative Story Similarity and Narrative Representation Learning

We present the GUIR system for SemEval-2026 Task 7, Everyday Knowledge Across Diverse Languages and Cultures, which probes the extent to which general-purpose LLMs encode cultural knowledge without any culture-specific supervision or fine-tuning. Our system addresses two tracks built on the BLEnD benchmark. For the short-answer question (SAQ) track, we employ zero-shot prompting with gpt-4.1, achieving 55.5% accuracy across 61 language locales. For the multiple-choice question (MCQ) track, we propose a three-stage pipeline: (1) zero-shot chain-of-thought inference with gpt-5-mini, (2) cross-locale majority voting to correct inconsistent predictions, and (3) a multi-agent debate protocol in which three LLM instances argue and adjudicate over residual errors. This pipeline achieves 97.47% overall accuracy across 30 locales, ranking first among all submitted systems on the MCQ track. We further conduct a targeted human evaluation on the Persian locale, revealing that BLEnD's lemma-matching scorer systematically underestimates model performance, with human annotators scoring the system 18 percentage points higher than the lemma-matching evaluation. This reveals the need for better evaluation of morphologically rich languages like Persian.

GUIR at SemEval-2026 Task 7: Probing Cultural Knowledge in LLMs via Multi-Agent Debate

We present the results and findings from SemEval Task 8: MTRAGEval. MTRAGEval measures three Retrieval Augmented Generation (RAG) subtasks: A. Retrieval, B. Generate, and C. Retrieve+Generate (full RAG) on multi-turn conversations. The task is evaluated using MTRAG-UN, a new benchmark for Multi-Turn RAG focusing on Unanswerable, Underspecified, Non-Standalone, and Unclear Questions. The MTRAGEval task attracted strong participation with 107 registered teams and 92 submissions across all subtasks, and yielded several interesting findings on effective retrieval and query rewriting techniques, the use of ensemble models, and the compounding costs of retrieval errors on downstream generation quality.

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

We introduce SemEval-2026 Task 5 on "Rating Plausibility of Word Senses in Ambiguous Stories through Narrative Understanding". 
The dataset for this task consists of 4-5 sentence English short stories. In each story, one sentence includes a lexical ambiguity and different senses are to be judged in terms of plausibility on a Likert scale. The task is intentionally constructed to be challenging by stories only providing sparse contextual cues. We give an overview of well-performing, frequent and interesting approaches used by participating systems. From a total of 175 registered participants and 27 submitted system description papers, the best system achieved an "accuracy within standard deviation" score of 93.3%.

SemEval-2026 Task 5: Rating Plausibility of Word Senses in Ambiguous Stories through Narrative Understanding

SemEval-2026 Task 11 evaluates the ability of Large Language Models (LLMs) to perform content-independent reasoning through a novel multilingual syllogistic dataset designed to measure the "content effect" — the tendency to conflate semantic plausibility with logical validity. The competition featured four subtasks, covering English and multilingual settings with both standard and noisy premise sets. Evaluations of zero-shot baselines reveal that the content effect is pervasive in open models, whereas newer versions demonstrate a significant shift in performance. Across the subtasks, findings indicate that introducing distracting premises can challenge the models' ability to filter misleading information, while multilingual settings amplify their susceptibility to content biases compared to English. Participants proposed diverse approaches, including neuro-symbolic decomposition, fine-tuning and distillation methods, data augmentation, and activation steering. While explicit symbolic verification remains the most reliable strategy, activation-level interventions and fine-tuning methods offer promising pathways for internalising formal logic within neural architectures. Ultimately, the task reinforces the efficacy of neuro-symbolic approaches and emerging architectural trends for logical reliability, while also highlighting that multilingual setups and longer contexts still pose significant challenges to be investigated in future research.

SemEval-2026 Task 11: Disentangling Content and Formal Reasoning in Large Language Models

We present our shared task on predicting variation in emotional valence and arousal over time from ecological essays. The shared task uses a longitudinal dataset collected over 7 data collection phases of 14-day each spanning from 2021 to 2024, consisting of real-time essays and feeling words (e.g., happy, calm, sad, etc.) written in English by U.S. service-industry workers about ``how they are feeling''. Each text is associated with self‑reported valence (V) (0 - 4, highly negative to highly positive affect) and arousal (A) (0 - 2, low to high energy) scores. The shared task consists of three parts, Subtask (1): Longitudinal Affect Assessment, Subtask (2): Forecasting Variation in Affect as a (2a): \textit{state change}, and (2b): \textit{disposition change}.

The task attracted over 200 member registrations on Codabench, receiving official system submissions from 31 teams (total 104 team members), of which 28 teams (with 90 team members) submitted system description papers making it to our leaderboard. We discuss baseline results along with findings from 28 systems, highlighting the best-performing systems, a deeper analysis of performance on essays versus feeling words, and assessments for authors seen versus unseen during training. The datasets for this task are publicly available.

Downloads

Next from ACL 2026

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

Similar lecture

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

.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

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

Similar lecture

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

Downloads