Publication

@inproceedings{Schaaf:et:al:2026,
  title     = {GhostWriter: Hidden AI-Generated Texts over Multiple Languages,
               Domains and Generators},
  author    = {Schaaf, Manuel and Bönisch, Kevin and Mehler, Alexander},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation
               Conference (LREC 2026)},
  month     = {May},
  year      = {2026},
  pages     = {10497--10516},
  keywords  = {Corpus, Natural Language Generation; Validation of LRs, AI-generated Text Detection, core, core_b05},
  address   = {Palma, Mallorca, Spain},
  publisher = {European Language Resources Association (ELRA)},
  editor    = {Piperidis, Stelios and Bel, Núria and van den Heuvel, Henk and Ide, Nancy
               and Krek, Simon and Toral, Antonio},
  doi       = {10.63317/57fd7juh5zek},
  abstract  = {The advent of Transformer-based Large Language Models (LLMs) has
               led to an unprecedented surge of AI-generated text (AIGT) across
               online platforms and academic domains. While these models exhibit
               near-human fluency and stylistic coherence, their widespread adoption
               has raised concerns about authorship integrity, research quality,
               and the recursive contamination of training corpora with synthetic
               data. These developments underscore the need for reliable AIGT
               detection methods and benchmark datasets, particularly for malicious
               or deceptive *ghostwriting* scenarios where AIGT is intentionally
               crafted to evade detection. To address this, we present **GhostWriter**,
               a large-scale, bilingual (German and English), multi-generator,
               and multi-domain dataset for AIGT detection. The dataset comprises
               human- and AI-authored texts produced under domain-specific *ghostwriting*
               conditions, including examples intentionally embedded within otherwise
               human-written texts to obscure their AI origin. With **GhostWriter**,
               we (i) aim to expand the resources available for German AIGT datasets,
               (ii) emphasize mixed or fused synthesizations—since most existing
               corpora are limited to the document level—and (iii) introduce
               specifically crafted malicious ghostwriting scenarios across multiple
               domains and generators.}
}

@inproceedings{Dahmann:et:al:2026,
  title     = {Towards the Generation and Application of Dynamic Web-Based Visualization
               of UIMA-based Annotations for Big-Data Corpora with the Help of
               Unified Dynamic Annotation Visualizer},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation
               Conference (LREC 2026)},
  year      = {2026},
  pages     = {6695--6705},
  author    = {Dahmann, Thiemo and Schneider, Julian and Stephan, Philipp and Abrami, Giuseppe
               and Mehler, Alexander},
  address   = {Palma, Mallorca, Spain},
  publisher = {European Language Resources Association (ELRA)},
  editor    = {Piperidis, Stelios and Bel, Núria and van den Heuvel, Henk and Ide, Nancy
               and Krek, Simon and Toral, Antonio},
  doi       = {10.63317/5ce2aaity4yz},
  keywords  = {NLP, UIMA, Annotations, dynamic visualization, uce},
  abstract  = {The automatic and manual annotation of unstructured corpora is
               a routine task in many scientific fields and is supported by a
               variety of existing software solutions. Despite this variety,
               few solutions currently support annotation visualization, especially
               for dynamic generation and interaction. To bridge this gap and
               visualize annotated corpora based on user-, project-, or corpus-specific
               aspects, we developed Unified Dynamic Annotation Visualizer (UDAV).
               UDAV is a web-based solution that implements features not supported
               by comparable tools, enabling a customizable and extensible toolbox
               for interacting with annotations and allowing integration into
               existing big-data frameworks. We exemplify UDAV through a range
               of visualizations and also provide an evaluation of corpus import
               and processing performance.},
  pdf       = {http://www.lrec-conf.org/proceedings/lrec2026/pdf/2026.lrec2026-1.533.pdf},
  video     = {https://www.youtube.com/watch?v=LFBiGlmEDog}
}

@inproceedings{Verma:Mehler:2026,
  title     = {Predicting Topic (Co-)Occurrence Using Topic Networks Built from
               the Project Gutenberg Corpus},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation
               Conference (LREC 2026)},
  pages     = {860--869},
  address   = {Palma, Mallorca, Spain},
  publisher = {European Language Resources Association (ELRA)},
  editor    = {Piperidis, Stelios and Bel, Núria and van den Heuvel, Henk and Ide, Nancy
               and Krek, Simon and Toral, Antonio},
  year      = {2026},
  author    = {Verma, Bhuvanesh and Mehler, Alexander},
  doi       = {10.63317/58x3h7gjbpb4},
  keywords  = {Topic Evolution, Topic Network,Time-aware Networks, Temporal Autocorrelation, Project Gutenberg, satek},
  abstract  = {Although temporal topic modeling has been widely applied to scientific
               and legal texts, literary corpora have largely been overlooked
               in this regard. To address this issue, we analyze topic evolution
               in a subset of the Project Gutenberg (PG) corpus. We model this
               subset as a sequence of topic networks that capture the emergence,
               persistence, and interaction of thematic structures over decades.
               Using supervised topic representations, we predict nodes (topics)
               and edges (topic pairings) to forecast future topics and their
               co-occurrence. Our experiments demonstrate moderate to strong
               temporal persistence in topic connectivity patterns across three
               topic systems, with ROC-AUC and AP values consistently above 0.85.
               We find that the temporal span of topic networks significantly
               impacts predictive performance: longer spans improve the stability
               and recall of topic presence, while shorter spans better capture
               evolving topic relationships. Overall, our findings demonstrate
               the predictability of topics in literary texts over time.} pdf
               = {http://www.lrec-conf.org/proceedings/lrec2026/pdf/2026.lrec2026-1.65.pdf}
}

@inproceedings{Abusaleh:et:al:2026,
  title     = {Learning to Detect Cross-Modal Negation: An Analysis of Latent
               Representations and an Attention-Based Solution},
  author    = {Abusaleh, Ali and Hammerla, Leon and Mehler, Alexander},
  booktitle = {2026 8th International Conference on Natural Language Processing (ICNLP)},
  eventdate = {2026-03-20/2026-03-22},
  location  = {Xi'an,China},
  year      = {2026},
  keywords  = {Vision language model, Natural language processing, Cross-modal retrieval, negation detection, video analysis, Multimodal analysis, Political Communication, neglab, new-data-spaces, circlet},
  abstract  = {Detecting high-level semantic concepts like negation across modalities
               remains a challenge for current multimodal systems. We analyze
               this as a fundamental representation learning problem, providing
               the first evidence that negation does not form a linearly or non-linearly
               separable class in the latent spaces of standard vision-language
               models (VLMs). We demonstrate that pretrained embeddings primarily
               encode modality-specific features, lacking a generalizable negation
               signal. To overcome this, we propose a novel cross-modal attention
               architecture that explicitly models inter-modal dependencies,
               achieving performance gains of up to +7.03% F1 over unimodal baselines.
               Our analysis reveals a key asymmetry: while textual negation often
               appears independently, visual negation is semantically dependent
               on linguistic context, a finding validated through our statistical
               analysis of 3,222 political video-text pairs automatically annotated
               via Qwen2.5-VL. By combining this analysis with self-supervised
               video representations (JEPA2), we advance the modeling of temporal
               negation. This work provides new methods and insights for learning
               robust, semantically-aligned representations in multimodal systems.},
  note      = {accepted}
}