arXiv Computation and Language

Zero-shot narrative detection in social messaging

The paper explores how large language models can detect hidden narratives in social messages without training data. By feeding the models human-written narrative descriptions, performance improves markedly, while automatically generated descriptions or few-shot examples can hurt accuracy. Ensemble techniques, especially majority voting, further boost robustness, and larger models show the best results with less sensitivity to prompts.

arXiv Computation and Language
Sep 11

Characterizing Narrative Content in Web-scale LLM Pretraining Data

The paper presents a detailed examination of narrative elements—agency, setting, and events—within the Dolma web-scale pretraining corpus. Using a framework of 11 interpretable dimensions, the authors hand‑annotated 400 passages, expanded this to a 25,000‑passage LLM‑labeled dataset, and trained NarraBERT models to predict narrative features across 13 million passages, producing the NarraDolma dataset. The study reveals that narrative structure is measurable at scale and that narrative qualities vary unevenly across different data sources, topics, and formats, highlighting gaps in current data curation practices.

By Teagan Johnson, Elliott Ash, Andrew Piper, Maria Antoniak
arXiv AI
Aug 25

ProBel: Propaganda Detection with Techniques, Spans, and Explanations

ProBel is a bilingual Arabic and English resource for propaganda detection that aligns binary labels, multi-label annotations for 23 propaganda techniques grouped into six categories, technique-labeled spans, and reference explanations for news sentences. The dataset supports matched binary, coarse-grained, multi-label, and span-level tasks in both languages, and the authors evaluate zero‑shot prompting, task‑specific fine‑tuning, and joint training. A single bilingual multi‑task model achieves the best overall performance, with cross‑task analysis revealing that joint classification preserves binary performance while span‑only training can weaken sentence‑level prediction, and that joint bilingual training yields the most stable results.

By Mohamed Bayan Kmainasi, Ali Ezzat Shahroor, Elisa Sartori, Giovanni Da San Martino, Firoj Alam
arXiv Computation and Language
4d ago

SWARM: A Multilingual Human-Annotated Dataset for Russian Propaganda Detection in Search Engine Results

The paper introduces SWARM, a multilingual dataset of 2,183 search engine results in nine languages, annotated for support of Russian propaganda narratives. It evaluates a source-based blocklist, supervised classifiers, and zero‑shot large language models, finding that blocklists miss most propaganda and that content‑level models vary in performance, with the best LLM achieving an F1 of 0.73. The study highlights the need for per‑language, content‑level detection of search‑borne propaganda.

By Manuel Tonneau, Abhinav Dubey, Farhan Shaikh, Ilaria Vitulano, Martha Stolze, Hale Dedeoglu, Clara Riechert, Ella Kuka, Maryna Sydorova, Mykola Makhortykh, Elizaveta Kuznetsova
arXiv Machine Learning
Jul 31

Selecting Open-Weight Language Models for Zero-Shot Intent Classification: A Systematic Evaluation of 41 Models

arXiv:2607. 27421v1 Announce Type: cross Abstract: Intent classification is a core component of task-oriented dialogue systems, yet practitioners have limited systematic guidance for selecting deployable open-weight language models under compute, latency, and robustness constraints.

By Parishruthi Ganesh, Gerry Dozier, Cheryl Seals
arXiv Computation and Language
Aug 27

Just Pass Twice: Efficient Token Classification with LLMs for Zero-Shot NER

Just Pass Twice (JPT) is a method that allows causal large language models to perform token classification for zero‑shot named entity recognition by concatenating the input with itself, giving each token full bidirectional context without architectural changes. The approach combines these representations with definition‑guided entity embeddings to enable flexible zero‑shot generalization. JPT achieves state‑of‑the‑art results, outperforming prior methods by an average of +7.9 F1 on CrossNER and MIT benchmarks and running over 20× faster than comparable generative approaches.

By Ahmed Ewais, Ahmed Hashish, Amr Ali