Jan Philip Wahle

Jan Philip Wahle, Terry Ruas, Frederic Kirstein et al.

The recent success of large language models for text generation poses a severe threat to academic integrity, as plagiarists can generate realistic paraphrases indistinguishable from original work. However, the role of large autoregressive transformers in generating machine-paraphrased plagiarism and their detection is still developing in the literature. This work explores T5 and GPT-3 for machine-paraphrase generation on scientific articles from arXiv, student theses, and Wikipedia. We evaluate the detection performance of six automated solutions and one commercial plagiarism detection software and perform a human study with 105 participants regarding their detection performance and the quality of generated examples. Our results suggest that large models can rewrite text humans have difficulty identifying as machine-paraphrased (53% mean acc.). Human experts rate the quality of paraphrases generated by GPT-3 as high as original texts (clarity 4.0/5, fluency 4.2/5, coherence 3.8/5). The best-performing detection model (GPT-3) achieves a 66% F1-score in detecting paraphrases.

Jan Philip Wahle, Terry Ruas, Saif M. Mohammad et al.

DBLP is the largest open-access repository of scientific articles on computer science and provides metadata associated with publications, authors, and venues. We retrieved more than 6 million publications from DBLP and extracted pertinent metadata (e.g., abstracts, author affiliations, citations) from the publication texts to create the DBLP Discovery Dataset (D3). D3 can be used to identify trends in research activity, productivity, focus, bias, accessibility, and impact of computer science research. We present an initial analysis focused on the volume of computer science research (e.g., number of papers, authors, research activity), trends in topics of interest, and citation patterns. Our findings show that computer science is a growing research field (approx. 15% annually), with an active and collaborative researcher community. While papers in recent years present more bibliographical entries in comparison to previous decades, the average number of citations has been declining. Investigating papers' abstracts reveals that recent topic trends are clearly reflected in D3. Finally, we list further applications of D3 and pose supplemental research questions. The D3 dataset, our findings, and source code are publicly available for research purposes.

Jonas Becker, Jan Philip Wahle, Terry Ruas et al.

The growing prominence of large language models, such as GPT-4 and ChatGPT, has led to increased concerns over academic integrity due to the potential for machine-generated content and paraphrasing. Although studies have explored the detection of human- and machine-paraphrased content, the comparison between these types of content remains underexplored. In this paper, we conduct a comprehensive analysis of various datasets commonly employed for paraphrase detection tasks and evaluate an array of detection methods. Our findings highlight the strengths and limitations of different detection methods in terms of performance on individual datasets, revealing a lack of suitable machine-generated datasets that can be aligned with human expectations. Our main finding is that human-authored paraphrases exceed machine-generated ones in terms of difficulty, diversity, and similarity implying that automatically generated texts are not yet on par with human-level performance. Transformers emerged as the most effective method across datasets with TF-IDF excelling on semantically diverse corpora. Additionally, we identify four datasets as the most diverse and challenging for paraphrase detection.

Frederic Kirstein, Jan Philip Wahle, Terry Ruas et al.

Despite the recent success of multi-task learning and pre-finetuning for natural language understanding, few works have studied the effects of task families on abstractive text summarization. Task families are a form of task grouping during the pre-finetuning stage to learn common skills, such as reading comprehension. To close this gap, we analyze the influence of multi-task learning strategies using task families for the English abstractive text summarization task. We group tasks into one of three strategies, i.e., sequential, simultaneous, and continual multi-task learning, and evaluate trained models through two downstream tasks. We find that certain combinations of task families (e.g., advanced reading comprehension and natural language inference) positively impact downstream performance. Further, we find that choice and combinations of task families influence downstream performance more than the training scheme, supporting the use of task families for abstractive text summarization.

Terry Ruas, Jan Philip Wahle, Lennart Küll et al.

This paper presents CS-Insights, an interactive web application to analyze computer science publications from DBLP through multiple perspectives. The dedicated interfaces allow its users to identify trends in research activity, productivity, accessibility, author's productivity, venues' statistics, topics of interest, and the impact of computer science research on other fields. CS-Insightsis publicly available, and its modular architecture can be easily adapted to domains other than computer science.

Jan Philip Wahle, Terry Ruas, Mohamed Abdalla et al.

Natural Language Processing (NLP) is poised to substantially influence the world. However, significant progress comes hand-in-hand with substantial risks. Addressing them requires broad engagement with various fields of study. Yet, little empirical work examines the state of such engagement (past or current). In this paper, we quantify the degree of influence between 23 fields of study and NLP (on each other). We analyzed ~77k NLP papers, ~3.1m citations from NLP papers to other papers, and ~1.8m citations from other papers to NLP papers. We show that, unlike most fields, the cross-field engagement of NLP, measured by our proposed Citation Field Diversity Index (CFDI), has declined from 0.58 in 1980 to 0.31 in 2022 (an all-time low). In addition, we find that NLP has grown more insular -- citing increasingly more NLP papers and having fewer papers that act as bridges between fields. NLP citations are dominated by computer science; Less than 8% of NLP citations are to linguistics, and less than 3% are to math and psychology. These findings underscore NLP's urgent need to reflect on its engagement with various fields.

Florian Valentin Wunderlich, Lars Benedikt Kaesberg, Jan Philip Wahle et al.

Advances in inference methods have enabled language models to improve their predictions without additional training. These methods often prioritize raw performance over cost-effective compute usage. However, computational efficiency is key for real-world applications with resource constraints. We provide a systematic analysis of the inference scaling strategies self-consistency, self-refinement, multi-agent debate, and mixture-of-agents, to study their computational performance tradeoffs. We evaluate methods on two reasoning benchmarks (MMLU-Pro, BBH) and include extensive parameter configurations (e.g., scaling the number of parallel predictions, agents, and debate rounds) across different model sizes. Across 34 configurations and over 100 evaluations, we compute the Pareto-optimal front to select methods that achieve the best accuracy with the lowest computational budget. Notably, inference scaling improves accuracy by up to +7.1% points over chain-of-thought at the highest evaluated budgets (20x the CoT compute budget) on MMLU-Pro. With an equal computing budget, debate and mixture-of-agents outperform self-consistency by 1.3% and 2.7% points, respectively. While self-consistency saturates earlier, multi-agent gains persist, particularly on more complicated tasks. We identify a simple multi-agent design guideline: mixture-of-agents is most efficient when the number of parallel generations exceeds the number of sequential aggregations.

Finn Schmidt, Jan Philip Wahle, Terry Ruas et al.

Automatic evaluation metrics are central to the development of machine translation systems, yet their robustness under domain shift remains unclear. Most metrics are developed on the Workshop on Machine Translation (WMT) benchmarks, raising concerns about their robustness to unseen domains. Prior studies that analyze unseen domains vary translation systems, annotators, or evaluation conditions, confounding domain effects with human annotation noise. To address these biases, we introduce a systematic multi-annotator Cross-Domain Error-Span-Annotation dataset (CD-ESA), comprising 18.8k human error span annotations across three language pairs, where we fix annotators within each language pair and evaluate translations of the same six translation systems across one seen news domain and two unseen technical domains. Using this dataset, we first find that automatic metrics appear surprisingly robust to domain-shifts at the segment level (up to 0.69 agreement), but this robustness largely disappears once we account for human label variation. Averaging annotations increases inter-annotator agreement by up to +0.11. Metrics struggle on the unseen chemical domain compared to humans (inter-annotator agreement of 0.78-0.83 vs. 0.96). We recommend comparing metric-human agreement against inter-annotator agreement, rather than comparing raw metric-human agreement alone, when evaluating across different domains.

Jan Philip Wahle, Bela Gipp, Terry Ruas

Current approaches in paraphrase generation and detection heavily rely on a single general similarity score, ignoring the intricate linguistic properties of language. This paper introduces two new tasks to address this shortcoming by considering paraphrase types - specific linguistic perturbations at particular text positions. We name these tasks Paraphrase Type Generation and Paraphrase Type Detection. Our results suggest that while current techniques perform well in a binary classification scenario, i.e., paraphrased or not, the inclusion of fine-grained paraphrase types poses a significant challenge. While most approaches are good at generating and detecting general semantic similar content, they fail to understand the intrinsic linguistic variables they manipulate. Models trained in generating and identifying paraphrase types also show improvements in tasks without them. In addition, scaling these models further improves their ability to understand paraphrase types. We believe paraphrase types can unlock a new paradigm for developing paraphrase models and solving tasks in the future.

Lars Benedikt Kaesberg, Terry Ruas, Jan Philip Wahle et al.

We present CiteAssist, a system to automate the generation of BibTeX entries for preprints, streamlining the process of bibliographic annotation. Our system extracts metadata, such as author names, titles, publication dates, and keywords, to create standardized annotations within the document. CiteAssist automatically attaches the BibTeX citation to the end of a PDF and links it on the first page of the document so other researchers gain immediate access to the correct citation of the article. This method promotes platform flexibility by ensuring that annotations remain accessible regardless of the repository used to publish or access the preprint. The annotations remain available even if the preprint is viewed externally to CiteAssist. Additionally, the system adds relevant related papers based on extracted keywords to the preprint, providing researchers with additional publications besides those in related work for further reading. Researchers can enhance their preprints organization and reference management workflows through a free and publicly available web interface.

Jonas Becker, Liang-Chih Yu, Shamsuddeen Hassan Muhammad et al.

Stance detection is an established task that classifies an author's attitude toward a specific target into categories such as Favor, Neutral, and Against. Beyond categorical stance labels, we leverage a long-established affective science framework to model stance along real-valued dimensions of valence (negative-positive) and arousal (calm-active). This dimensional approach captures nuanced affective states underlying stance expressions, enabling fine-grained stance analysis. To this end, we introduce DimStance, the first dimensional stance resource with valence-arousal (VA) annotations. This resource comprises 11,746 target aspects in 7,365 texts across five languages (English, German, Chinese, Nigerian Pidgin, and Swahili) and two domains (politics and environmental protection). To facilitate the evaluation of stance VA prediction, we formulate the dimensional stance regression task, analyze cross-lingual VA patterns, and benchmark pretrained and large language models under regression and prompting settings. Results show competitive performance of fine-tuned LLM regressors, persistent challenges in low-resource languages, and limitations of token-based generation. DimStance provides a foundation for multilingual, emotion-aware, stance analysis and benchmarking.

Jan Philip Wahle, Krishnapriya Vishnubhotla, Bela Gipp et al.

Work in Computational Affective Science and Computational Social Science explores a wide variety of research questions about people, emotions, behavior, and health. Such work often relies on language data that is first labeled with relevant information, such as the use of emotion words or the age of the speaker. Although many resources and algorithms exist to enable this type of labeling, discovering, accessing, and using them remains a substantial impediment, particularly for practitioners outside of computer science. Here, we present the ABCDE dataset (Affect, Body, Cognition, Demographics, and Emotion), a large-scale collection of over 400 million text utterances drawn from social media, blogs, books, and AI-generated sources. The dataset is annotated with a wide range of features relevant to computational affective and social science. ABCDE facilitates interdisciplinary research across numerous fields, including affective science, cognitive science, the digital humanities, sociology, political science, and computational linguistics.

Lung-Hao Lee, Liang-Chih Yu, Natalia Loukashevich et al.

Aspect-Based Sentiment Analysis (ABSA) focuses on extracting sentiment at a fine-grained aspect level and has been widely applied across real-world domains. However, existing ABSA research relies on coarse-grained categorical labels (e.g., positive, negative), which limits its ability to capture nuanced affective states. To address this limitation, we adopt a dimensional approach that represents sentiment with continuous valence-arousal (VA) scores, enabling fine-grained analysis at both the aspect and sentiment levels. To this end, we introduce DimABSA, the first multilingual, dimensional ABSA resource annotated with both traditional ABSA elements (aspect terms, aspect categories, and opinion terms) and newly introduced VA scores. This resource contains 76,958 aspect instances across 42,590 sentences, spanning six languages and four domains. We further introduce three subtasks that combine VA scores with different ABSA elements, providing a bridge from traditional ABSA to dimensional ABSA. Given that these subtasks involve both categorical and continuous outputs, we propose a new unified metric, continuous F1 (cF1), which incorporates VA prediction error into standard F1. We provide a comprehensive benchmark using both prompted and fine-tuned large language models across all subtasks. Our results show that DimABSA is a challenging benchmark and provides a foundation for advancing multilingual dimensional ABSA.

Dominik Meier, Jan Philip Wahle, Terry Ruas et al.

Paraphrases represent a human's intuitive ability to understand expressions presented in various different ways. Current paraphrase evaluations of language models primarily use binary approaches, offering limited interpretability of specific text changes. Atomic paraphrase types (APT) decompose paraphrases into different linguistic changes and offer a granular view of the flexibility in linguistic expression (e.g., a shift in syntax or vocabulary used). In this study, we assess the human preferences towards ChatGPT in generating English paraphrases with ten APTs and five prompting techniques. We introduce APTY (Atomic Paraphrase TYpes), a dataset of 800 sentence-level and word-level annotations by 15 annotators. The dataset also provides a human preference ranking of paraphrases with different types that can be used to fine-tune models with RLHF and DPO methods. Our results reveal that ChatGPT and a DPO-trained LLama 7B model can generate simple APTs, such as additions and deletions, but struggle with complex structures (e.g., subordination changes). This study contributes to understanding which aspects of paraphrasing language models have already succeeded at understanding and what remains elusive. In addition, we show how our curated datasets can be used to develop language models with specific linguistic capabilities.

Jan Philip Wahle

A recent trend in Natural Language Processing is the exponential growth in Language Model (LM) size, which prevents research groups without a necessary hardware infrastructure from participating in the development process. This study investigates methods for Knowledge Distillation (KD) to provide efficient alternatives to large-scale models. In this context, KD means extracting information about language encoded in a Neural Network and Lexical Knowledge Databases. We developed two methods to test our hypothesis that efficient architectures can gain knowledge from LMs and extract valuable information from lexical sources. First, we present a technique to learn confident probability distribution for Masked Language Modeling by prediction weighting of multiple teacher networks. Second, we propose a method for Word Sense Disambiguation (WSD) and lexical KD that is general enough to be adapted to many LMs. Our results show that KD with multiple teachers leads to improved training convergence. When using our lexical pre-training method, LM characteristics are not lost, leading to increased performance in Natural Language Understanding (NLU) tasks over the state-of-the-art while adding no parameters. Moreover, the improved semantic understanding of our model increased the task performance beyond WSD and NLU in a real-problem scenario (Plagiarism Detection). This study suggests that sophisticated training methods and network architectures can be superior over scaling trainable parameters. On this basis, we suggest the research area should encourage the development and use of efficient models and rate impacts resulting from growing LM size equally against task performance.

Jan Philip Wahle

Language enables humans to share knowledge, reason about the world, and pass on strategies for survival and innovation across generations. At the heart of this process is not just the ability to communicate but also the remarkable flexibility in how we can express ourselves. We can express the same thoughts in virtually infinite ways using different words and structures - this ability to rephrase and reformulate expressions is known as paraphrase. Modeling paraphrases is a keystone to meaning in computational language models; being able to construct different variations of texts that convey the same meaning or not shows strong abilities of semantic understanding. If computational language models are to represent meaning, they must understand and control the different aspects that construct the same meaning as opposed to different meanings at a fine granularity. Yet most existing approaches reduce paraphrasing to a binary decision between two texts or to producing a single rewrite of a source, obscuring which linguistic factors are responsible for meaning preservation. In this thesis, I propose that decomposing paraphrases into their constituent linguistic aspects (paraphrase types) offers a more fine-grained and cognitively grounded view of semantic equivalence. I show that even advanced machine learning models struggle with this task. Yet, when explicitly trained on paraphrase types, models achieve stronger performance on related paraphrase tasks and downstream applications. For example, in plagiarism detection, language models trained on paraphrase types surpass human baselines: 89.6% accuracy compared to 78.4% for plagiarism cases from Wikipedia, and 66.5% compared to 55.7% for plagiarism of scientific papers from arXiv. In identifying duplicate questions on Quora, models trained with paraphrase types improve over models trained on binary pairs. Furthermore, I demonstrate that...

Jonas Becker, Lars Benedikt Kaesberg, Niklas Bauer et al.

Multi-agent debate (MAD) has demonstrated the ability to augment collective intelligence by scaling test-time compute and leveraging expertise. Current frameworks for multi-agent debate are often designed towards tool use, lack integrated evaluation, or provide limited configurability of agent personas, response generators, discussion paradigms, and decision protocols. We introduce MALLM (Multi-Agent Large Language Models), an open-source framework that enables systematic analysis of MAD components. MALLM offers more than 144 unique configurations of MAD, including (1) agent personas (e.g., Expert, Personality), (2) response generators (e.g., Critical, Reasoning), (3) discussion paradigms (e.g., Memory, Relay), and (4) decision protocols (e.g., Voting, Consensus). MALLM uses simple configuration files to define a debate. Furthermore, MALLM can load any textual Hugging Face dataset (e.g., MMLU-Pro, WinoGrande) and provides an evaluation pipeline for easy comparison of MAD configurations. MALLM enables researchers to systematically configure, run, and evaluate debates for their problems, facilitating the understanding of the components and their interplay.

Tianyu Yang, Terry Ruas, Yijun Tian et al.

Vision-language models (VLMs) excel at interpreting text-rich images but struggle with long, visually complex documents that demand analysis and integration of information spread across multiple pages. Existing approaches typically rely on fixed reasoning templates or rigid pipelines, which force VLMs into a passive role and hinder both efficiency and generalization. We present Active Long-DocumEnt Navigation (ALDEN), a multi-turn reinforcement learning framework that fine-tunes VLMs as interactive agents capable of actively navigating long, visually rich documents. ALDEN introduces a novel fetch action that directly accesses the page by index, complementing the classic search action and better exploiting document structure. For dense process supervision and efficient training, we propose a rule-based cross-level reward that provides both turn- and token-level signals. To address the empirically observed training instability caused by numerous visual tokens from long documents, we further propose a visual-semantic anchoring mechanism that applies a dual-path KL-divergence constraint to stabilize visual and textual representations separately during training. Trained on a corpus constructed from three open-source datasets, ALDEN achieves state-of-the-art performance on five long-document benchmarks. Overall, ALDEN marks a step beyond passive document reading toward agents that autonomously navigate and reason across long, visually rich documents, offering a robust path to more accurate and efficient long-document understanding.

Jan Philip Wahle, Terry Ruas, Tomáš Foltýnek et al.

Employing paraphrasing tools to conceal plagiarized text is a severe threat to academic integrity. To enable the detection of machine-paraphrased text, we evaluate the effectiveness of five pre-trained word embedding models combined with machine-learning classifiers and eight state-of-the-art neural language models. We analyzed preprints of research papers, graduation theses, and Wikipedia articles, which we paraphrased using different configurations of the tools SpinBot and SpinnerChief. The best-performing technique, Longformer, achieved an average F1 score of 81.0% (F1=99.7% for SpinBot and F1=71.6% for SpinnerChief cases), while human evaluators achieved F1=78.4% for SpinBot and F1=65.6% for SpinnerChief cases. We show that the automated classification alleviates shortcomings of widely-used text-matching systems, such as Turnitin and PlagScan. To facilitate future research, all data, code, and two web applications showcasing our contributions are openly available at https://github.com/jpwahle/iconf22-paraphrase.

Shamsuddeen Hassan Muhammad, Nedjma Ousidhoum, Idris Abdulmumin et al.

People worldwide use language in subtle and complex ways to express emotions. Although emotion recognition--an umbrella term for several NLP tasks--impacts various applications within NLP and beyond, most work in this area has focused on high-resource languages. This has led to significant disparities in research efforts and proposed solutions, particularly for under-resourced languages, which often lack high-quality annotated datasets. In this paper, we present BRIGHTER--a collection of multi-labeled, emotion-annotated datasets in 28 different languages and across several domains. BRIGHTER primarily covers low-resource languages from Africa, Asia, Eastern Europe, and Latin America, with instances labeled by fluent speakers. We highlight the challenges related to the data collection and annotation processes, and then report experimental results for monolingual and crosslingual multi-label emotion identification, as well as emotion intensity recognition. We analyse the variability in performance across languages and text domains, both with and without the use of LLMs, and show that the BRIGHTER datasets represent a meaningful step towards addressing the gap in text-based emotion recognition.

Lars Benedikt Kaesberg, Jonas Becker, Jan Philip Wahle et al.

Much of the success of multi-agent debates depends on carefully choosing the right parameters. The decision-making protocol stands out as it can highly impact final model answers, depending on how decisions are reached. Systematic comparison of decision protocols is difficult because many studies alter multiple discussion parameters beyond the protocol. So far, it has been largely unknown how decision-making influences different tasks. This work systematically evaluates the impact of seven decision protocols (e.g., majority voting, unanimity consensus). We change only one variable at a time - the decision protocol - to analyze how different methods affect the collaboration between agents and measure differences in knowledge and reasoning tasks. Our results show that voting protocols improve performance by 13.2% in reasoning tasks and consensus protocols by 2.8% in knowledge tasks compared to other decision protocols. Increasing the number of agents improves performance, while more discussion rounds before voting reduce it. To improve decision-making by increasing answer diversity, we propose two new methods, All-Agents Drafting (AAD) and Collective Improvement (CI). Our methods improve task performance by up to 3.3% with AAD and up to 7.4% with CI. This work demonstrates the importance of decision-making in multi-agent debates beyond scaling.

Jonas Becker, Jan Philip Wahle, Bela Gipp et al.

Text generation has become more accessible than ever, and the increasing interest in these systems, especially those using large language models, has spurred an increasing number of related publications. We provide a systematic literature review comprising 244 selected papers between 2017 and 2024. This review categorizes works in text generation into five main tasks: open-ended text generation, summarization, translation, paraphrasing, and question answering. For each task, we review their relevant characteristics, sub-tasks, and specific challenges (e.g., missing datasets for multi-document summarization, coherence in story generation, and complex reasoning for question answering). Additionally, we assess current approaches for evaluating text generation systems and ascertain problems with current metrics. Our investigation shows nine prominent challenges common to all tasks and sub-tasks in recent text generation publications: bias, reasoning, hallucinations, misuse, privacy, interpretability, transparency, datasets, and computing. We provide a detailed analysis of these challenges, their potential solutions, and which gaps still require further engagement from the community. This systematic literature review targets two main audiences: early career researchers in natural language processing looking for an overview of the field and promising research directions, as well as experienced researchers seeking a detailed view of tasks, evaluation methodologies, open challenges, and recent mitigation strategies.

Shamsuddeen Hassan Muhammad, Nedjma Ousidhoum, Idris Abdulmumin et al.

We present our shared task on text-based emotion detection, covering more than 30 languages from seven distinct language families. These languages are predominantly low-resource and are spoken across various continents. The data instances are multi-labeled with six emotional classes, with additional datasets in 11 languages annotated for emotion intensity. Participants were asked to predict labels in three tracks: (a) multilabel emotion detection, (b) emotion intensity score detection, and (c) cross-lingual emotion detection. The task attracted over 700 participants. We received final submissions from more than 200 teams and 93 system description papers. We report baseline results, along with findings on the best-performing systems, the most common approaches, and the most effective methods across different tracks and languages. The datasets for this task are publicly available. The dataset is available at SemEval2025 Task 11 https://brighter-dataset.github.io

Andreas Stephan, Lukas Miklautz, Kevin Sidak et al.

Image clustering divides a collection of images into meaningful groups, typically interpreted post-hoc via human-given annotations. Those are usually in the form of text, begging the question of using text as an abstraction for image clustering. Current image clustering methods, however, neglect the use of generated textual descriptions. We, therefore, propose Text-Guided Image Clustering, i.e., generating text using image captioning and visual question-answering (VQA) models and subsequently clustering the generated text. Further, we introduce a novel approach to inject task- or domain knowledge for clustering by prompting VQA models. Across eight diverse image clustering datasets, our results show that the obtained text representations often outperform image features. Additionally, we propose a counting-based cluster explainability method. Our evaluations show that the derived keyword-based explanations describe clusters better than the respective cluster accuracy suggests. Overall, this research challenges traditional approaches and paves the way for a paradigm shift in image clustering, using generated text.

Jonas Becker, Lars Benedikt Kaesberg, Andreas Stephan et al.

Multi-agent debate - multiple instances of large language models discussing problems in turn-based interaction - has shown promise for solving knowledge and reasoning tasks. However, these methods show limitations when solving complex problems that require longer reasoning chains. We analyze how multi-agent debate over multiple turns drifts away from the initial problem, thus harming task performance. We define this phenomenon as problem drift and quantify its presence across ten tasks (i.e., three generative, three knowledge, three reasoning, and one instruction-following task). To identify the reasons for this issue, eight human experts analyze 170 multi-agent discussions suffering from problem drift. We find the most common issues related to this drift are the lack of progress (35% of cases), low-quality feedback (26% of cases), and a lack of clarity (25% of cases). To address problem drift, we propose DRIFTJudge, an LLM-as-a-judge method, to detect problem drift at test-time. We also propose DRIFTPolicy, a method that mitigates problem drift cases to improve task performance. Our study is a step toward understanding a key limitation of multi-agent debate, highlighting why longer debates can harm task performance and how problem drift could be addressed.

Frederic Kirstein, Jan Philip Wahle, Terry Ruas et al.

Meeting summarization has become a critical task considering the increase in online interactions. While new techniques are introduced regularly, their evaluation uses metrics not designed to capture meeting-specific errors, undermining effective evaluation. This paper investigates what the frequently used automatic metrics capture and which errors they mask by correlating automatic metric scores with human evaluations across a broad error taxonomy. We commence with a comprehensive literature review on English meeting summarization to define key challenges like speaker dynamics and contextual turn-taking and error types such as missing information and linguistic inaccuracy, concepts previously loosely defined in the field. We examine the relationship between characteristic challenges and errors by using annotated transcripts and summaries from Transformer-based sequence-to-sequence and autoregressive models from the general summary QMSum dataset. Through experimental validation, we find that different model architectures respond variably to challenges in meeting transcripts, resulting in different pronounced links between challenges and errors. Current default-used metrics struggle to capture observable errors, showing weak to mid-correlations, while a third of the correlations show trends of error masking. Only a subset reacts accurately to specific errors, while most correlations show either unresponsiveness or failure to reflect the error's impact on summary quality.

Tomáš Horych, Martin Wessel, Jan Philip Wahle et al.

Media bias detection poses a complex, multifaceted problem traditionally tackled using single-task models and small in-domain datasets, consequently lacking generalizability. To address this, we introduce MAGPIE, the first large-scale multi-task pre-training approach explicitly tailored for media bias detection. To enable pre-training at scale, we present Large Bias Mixture (LBM), a compilation of 59 bias-related tasks. MAGPIE outperforms previous approaches in media bias detection on the Bias Annotation By Experts (BABE) dataset, with a relative improvement of 3.3% F1-score. MAGPIE also performs better than previous models on 5 out of 8 tasks in the Media Bias Identification Benchmark (MBIB). Using a RoBERTa encoder, MAGPIE needs only 15% of finetuning steps compared to single-task approaches. Our evaluation shows, for instance, that tasks like sentiment and emotionality boost all learning, all tasks enhance fake news detection, and scaling tasks leads to the best results. MAGPIE confirms that MTL is a promising approach for addressing media bias detection, enhancing the accuracy and efficiency of existing models. Furthermore, LBM is the first available resource collection focused on media bias MTL.

Liang-Chih Yu, Jonas Becker, Shamsuddeen Hassan Muhammad et al.

We present the SemEval-2026 shared task on Dimensional Aspect-Based Sentiment Analysis (DimABSA), which improves traditional ABSA by modeling sentiment along valence-arousal (VA) dimensions rather than using categorical polarity labels. To extend ABSA beyond consumer reviews to public-issue discourse (e.g., political, energy, and climate issues), we introduce an additional task, Dimensional Stance Analysis (DimStance), which treats stance targets as aspects and reformulates stance detection as regression in the VA space. The task consists of two tracks: Track A (DimABSA) and Track B (DimStance). Track A includes three subtasks: (1) dimensional aspect sentiment regression, (2) dimensional aspect sentiment triplet extraction, and (3) dimensional aspect sentiment quadruplet extraction, while Track B includes only the regression subtask for stance targets. We also introduce a continuous F1 (cF1) metric to jointly evaluate structured extraction and VA regression. The task attracted more than 400 participants, resulting in 112 final submissions and 42 system description papers. We report baseline results, discuss top-performing systems, and analyze key design choices to provide insights into dimensional sentiment analysis at the aspect and stance-target levels. All resources are available on our GitHub repository.

Lars Benedikt Kaesberg, Tianyu Yang, Niklas Bauer et al.

Spatial reasoning is central to navigation and robotics, yet measuring model capabilities on these tasks remains difficult. Existing benchmarks evaluate models in a one-shot setting, requiring full solution generation in a single response, unlike humans, who work in interactive environments step-by-step. We introduce Spatial-Gym, a Gymnasium environment that isolates spatial constraint reasoning by testing pathfinding in 2D-grid puzzles as a sequential decision task with optional backtracking. We evaluate eight models in three settings (one-shot, step-by-step, step-by-step with backtracking) against human, random, and A* baselines on 500 episodes. The best model, GPT-OSS 120B, achieves a solve rate of 16.0%, 82 points below the human baseline (98.0%). Step-by-step format helps weaker models (up to +5.4%) by removing formatting errors, but hurts stronger models (up to 5.6%) by constraining global planning. Backtracking improves episode completion, but increases solve rate only for weaker models; stronger models rarely backtrack and do not benefit from it. Our experiments have three key findings: (1) models fail to scale reasoning effort with difficulty, (2) vision models receiving images of the spatial environment reduce solve rate by 73%, and (3) extended chain-of-thought reasoning retains a 3-5x accuracy advantage over standard inference even in the step-by-step setting. Spatial-Gym enables diagnosis of model limitations and provides a framework for improving spatial reasoning through reinforcement learning.

Lars Benedikt Kaesberg, Jan Philip Wahle, Terry Ruas et al.

Existing reasoning datasets saturate and fail to test abstract, multi-step problems, especially pathfinding and complex rule constraint satisfaction. We introduce SPaRC (Spatial Pathfinding Reasoning Challenge), a dataset of 1,000 2D grid pathfinding puzzles to evaluate spatial and symbolic reasoning, requiring step-by-step planning with arithmetic and geometric rules. Humans achieve near-perfect accuracy (98.0%; 94.5% on hard puzzles), while the best reasoning models, such as o4-mini, struggle (15.8%; 1.1% on hard puzzles). Models often generate invalid paths (>50% of puzzles for o4-mini), and reasoning tokens reveal they make errors in navigation and spatial logic. Unlike humans, who take longer on hard puzzles, models fail to scale test-time compute with difficulty. Allowing models to make multiple solution attempts improves accuracy, suggesting potential for better spatial reasoning with improved training and efficient test-time scaling methods. SPaRC can be used as a window into models' spatial reasoning limitations and drive research toward new methods that excel in abstract, multi-step problem-solving.

Sophie Wu, Jan Philip Wahle, Saif M. Mohammad

This paper is the first investigation of the connection between emotion, embodiment, and everyday language in a large sample of natural language data. We created corpora of body part mentions (BPMs) in online English text (blog posts and tweets). This includes a subset featuring human annotations for the emotions of the person whose body part is mentioned in the text. We show that BPMs are common in personal narratives and tweets (~5% to 10% of posts include BPMs) and that their usage patterns vary markedly by time and %geographic location. Using word-emotion association lexicons and our annotated data, we show that text containing BPMs tends to be more emotionally charged, even when the BPM is not explicitly used to describe a physical reaction to the emotion in the text. Finally, we discover a strong and statistically significant correlation between body-related language and a variety of poorer health outcomes. In sum, we argue that investigating the role of body-part related words in language can open up valuable avenues of future research at the intersection of NLP, the affective sciences, and the study of human wellbeing.

Jan Philip Wahle, Terry Ruas, Mohamed Abdalla et al.

This study examines the tendency to cite older work across 20 fields of study over 43 years (1980--2023). We put NLP's propensity to cite older work in the context of these 20 other fields to analyze whether NLP shows similar temporal citation patterns to these other fields over time or whether differences can be observed. Our analysis, based on a dataset of approximately 240 million papers, reveals a broader scientific trend: many fields have markedly declined in citing older works (e.g., psychology, computer science). We term this decline a 'citation age recession', analogous to how economists define periods of reduced economic activity. The trend is strongest in NLP and ML research (-12.8% and -5.5% in citation age from previous peaks). Our results suggest that citing more recent works is not directly driven by the growth in publication rates (-3.4% across fields; -5.2% in humanities; -5.5% in formal sciences) -- even when controlling for an increase in the volume of papers. Our findings raise questions about the scientific community's engagement with past literature, particularly for NLP, and the potential consequences of neglecting older but relevant research. The data and a demo showcasing our results are publicly available.

Max Martin Gnewuch, Jan Philip Wahle, Terry Ruas et al.

Over the past four decades, artificial intelligence (AI) research has flourished at the nexus of academia and industry. However, Big Tech companies have increasingly acquired the edge in computational resources, big data, and talent. So far, it has been largely unclear how many papers the industry funds, how their citation impact compares to non-funded papers, and what drives industry interest. This study fills that gap by quantifying the number of industry-funded papers at 10 top AI conferences (e.g., ICLR, CVPR, AAAI, ACL) and their citation influence. We analyze about 49.8K papers, about 1.8M citations from AI papers to other papers, and about 2.3M citations from other papers to AI papers from 1998-2022 in Scopus. Through seven research questions, we examine the volume and evolution of industry funding in AI research, the citation impact of funded papers, the diversity and temporal range of their citations, and the subfields in which industry predominantly acts. Our findings reveal that industry presence has grown markedly since 2015, from less than 2 percent to more than 11 percent in 2020. Between 2018 and 2022, 12 percent of industry-funded papers achieved high citation rates as measured by the h5-index, compared to 4 percent of non-industry-funded papers and 2 percent of non-funded papers. Top AI conferences engage more with industry-funded research than non-funded research, as measured by our newly proposed metric, the Citation Preference Ratio (CPR). We show that industry-funded research is increasingly insular, citing predominantly other industry-funded papers while referencing fewer non-funded papers. These findings reveal new trends in AI research funding, including a shift towards more industry-funded papers and their growing citation impact, greater insularity of industry-funded work than non-funded work, and a preference of industry-funded research to cite recent work.

André Greiner-Petter, Maik Fröbe, Jan Philip Wahle et al.

The generative plagiarism detection task at PAN 2025 aims at identifying automatically generated textual plagiarism in scientific articles and aligning them with their respective sources. We created a novel large-scale dataset of automatically generated plagiarism using three large language models: Llama, DeepSeek-R1, and Mistral. In this task overview paper, we outline the creation of this dataset, summarize and compare the results of all participants and four baselines, and evaluate the results on the last plagiarism detection task from PAN 2015 in order to interpret the robustness of the proposed approaches. We found that the current iteration does not invite a large variety of approaches as naive semantic similarity approaches based on embedding vectors provide promising results of up to 0.8 recall and 0.5 precision. In contrast, most of these approaches underperform significantly on the 2015 dataset, indicating a lack in generalizability.

Frederic Kirstein, Muneeb Khan, Jan Philip Wahle et al.

Meeting summarization suffers from limited high-quality data, mainly due to privacy restrictions and expensive collection processes. We address this gap with FAME, a dataset of 500 meetings in English and 300 in German produced by MIMIC, our new multi-agent meeting synthesis framework that generates meeting transcripts on a given knowledge source by defining psychologically grounded participant profiles, outlining the conversation, and orchestrating a large language model (LLM) debate. A modular post-processing step refines these outputs, mitigating potential repetitiveness and overly formal tones, ensuring coherent, credible dialogues at scale. We also propose a psychologically grounded evaluation framework assessing naturalness, social behavior authenticity, and transcript difficulties. Human assessments show that FAME approximates real-meeting spontaneity (4.5/5 in naturalness), preserves speaker-centric challenges (3/5 in spoken language), and introduces richer information-oriented difficulty (4/5 in difficulty). These findings highlight that FAME is a good and scalable proxy for real-world meeting conditions. It enables new test scenarios for meeting summarization research and other conversation-centric applications in tasks requiring conversation data or simulating social scenarios under behavioral constraints.

Jan Philip Wahle, Terry Ruas, Yang Xu et al.

Much of the success of modern language models depends on finding a suitable prompt to instruct the model. Until now, it has been largely unknown how variations in the linguistic expression of prompts affect these models. This study systematically and empirically evaluates which linguistic features influence models through paraphrase types, i.e., different linguistic changes at particular positions. We measure behavioral changes for five models across 120 tasks and six families of paraphrases (i.e., morphology, syntax, lexicon, lexico-syntax, discourse, and others). We also control for other prompt engineering factors (e.g., prompt length, lexical diversity, and proximity to training data). Our results show a potential for language models to improve tasks when their prompts are adapted in specific paraphrase types (e.g., 6.7% median gain in Mixtral 8x7B; 5.5% in LLaMA 3 8B). In particular, changes in morphology and lexicon, i.e., the vocabulary used, showed promise in improving prompts. These findings contribute to developing more robust language models capable of handling variability in linguistic expression.

Frederic Kirstein, Jan Philip Wahle, Bela Gipp et al.

Abstractive dialogue summarization is the task of distilling conversations into informative and concise summaries. Although reviews have been conducted on this topic, there is a lack of comprehensive work detailing the challenges of dialogue summarization, unifying the differing understanding of the task, and aligning proposed techniques, datasets, and evaluation metrics with the challenges. This article summarizes the research on Transformer-based abstractive summarization for English dialogues by systematically reviewing 1262 unique research papers published between 2019 and 2024, relying on the Semantic Scholar and DBLP databases. We cover the main challenges present in dialog summarization (i.e., language, structure, comprehension, speaker, salience, and factuality) and link them to corresponding techniques such as graph-based approaches, additional training tasks, and planning strategies, which typically overly rely on BART-based encoder-decoder models. We find that while some challenges, like language, have seen considerable progress, mainly due to training methods, others, such as comprehension, factuality, and salience, remain difficult and hold significant research opportunities. We investigate how these approaches are typically assessed, covering the datasets for the subdomains of dialogue (e.g., meeting, medical), the established automatic metrics and human evaluation approaches for assessing scores and annotator agreement. We observe that only a few datasets span across all subdomains. The ROUGE metric is the most used, while human evaluation is frequently reported without sufficient detail on inner-annotator agreement and annotation guidelines. Additionally, we discuss the possible implications of the recently explored large language models and conclude that despite a potential shift in relevance and difficulty, our described challenge taxonomy remains relevant.

Mohamed Abdalla, Jan Philip Wahle, Terry Ruas et al.

Recent advances in deep learning methods for natural language processing (NLP) have created new business opportunities and made NLP research critical for industry development. As one of the big players in the field of NLP, together with governments and universities, it is important to track the influence of industry on research. In this study, we seek to quantify and characterize industry presence in the NLP community over time. Using a corpus with comprehensive metadata of 78,187 NLP publications and 701 resumes of NLP publication authors, we explore the industry presence in the field since the early 90s. We find that industry presence among NLP authors has been steady before a steep increase over the past five years (180% growth from 2017 to 2022). A few companies account for most of the publications and provide funding to academic researchers through grants and internships. Our study shows that the presence and impact of the industry on natural language processing research are significant and fast-growing. This work calls for increased transparency of industry influence in the field.

Jan Philip Wahle, Nischal Ashok, Terry Ruas et al.

A drastic rise in potentially life-threatening misinformation has been a by-product of the COVID-19 pandemic. Computational support to identify false information within the massive body of data on the topic is crucial to prevent harm. Researchers proposed many methods for flagging online misinformation related to COVID-19. However, these methods predominantly target specific content types (e.g., news) or platforms (e.g., Twitter). The methods' capabilities to generalize were largely unclear so far. We evaluate fifteen Transformer-based models on five COVID-19 misinformation datasets that include social media posts, news articles, and scientific papers to fill this gap. We show tokenizers and models tailored to COVID-19 data do not provide a significant advantage over general-purpose ones. Our study provides a realistic assessment of models for detecting COVID-19 misinformation. We expect that evaluating a broad spectrum of datasets and models will benefit future research in developing misinformation detection systems.

Jan Philip Wahle, Terry Ruas, Norman Meuschke et al.

We present two supervised (pre-)training methods to incorporate gloss definitions from lexical resources into neural language models (LMs). The training improves our models' performance for Word Sense Disambiguation (WSD) but also benefits general language understanding tasks while adding almost no parameters. We evaluate our techniques with seven different neural LMs and find that XLNet is more suitable for WSD than BERT. Our best-performing methods exceeds state-of-the-art WSD techniques on the SemCor 3.0 dataset by 0.5% F1 and increase BERT's performance on the GLUE benchmark by 1.1% on average.

Jan Philip Wahle, Terry Ruas, Norman Meuschke et al.

The rise of language models such as BERT allows for high-quality text paraphrasing. This is a problem to academic integrity, as it is difficult to differentiate between original and machine-generated content. We propose a benchmark consisting of paraphrased articles using recent language models relying on the Transformer architecture. Our contribution fosters future research of paraphrase detection systems as it offers a large collection of aligned original and paraphrased documents, a study regarding its structure, classification experiments with state-of-the-art systems, and we make our findings publicly available.