Publication

@inproceedings{Hammerla:Mehler:2026:b,
  title     = {Negation in Reasoning Traces: Interpretable Signals of Correctness
               and Provenance},
  author    = {Hammerla, Leon Lukas and Mehler, Alexander},
  editor    = {Yanaka, Hitomi and Abzianidze, Lasha},
  booktitle = {Proceedings of the 6th Workshop on Natural Language Meets Logic
               and Machine Learning ({NALOMA})},
  month     = {aug},
  year      = {2026},
  address   = {Prague, Czechia},
  publisher = {Association for Computational Linguistics},
  url       = {https://aclanthology.org/2026.naloma-1.4/},
  pages     = {19--39},
  isbn      = {979-8-89176-389-0},
  abstract  = {Chain-of-thought (CoT) reasoning is widely used in large language
               models (LLMs), but the resulting reasoning traces remain underexplored.We
               study these traces through the lens of discourse-level negation.Specifically,
               we distinguish between corrective negation, which rejects a prior
               reasoning step, and refining negation, which narrows or qualifies
               it, and introduce metrics to quantify their use in human- and
               LLM-authored reasoning traces.Across multiple benchmarks, we find
               that negation occurs much more frequently in intermediate reasoning
               traces than in final response texts.We then test whether negation-based
               features provide predictive and descriptive signal for correctness,
               model identity, and human-vs.-LLM authorship.For correctness prediction,
               negation-based features consistently outperform simple structural
               baselines and in several settings add complementary signal to
               embedding-based representations, although embeddings remain stronger
               overall.In a controlled comparison on correct human and LLM traces
               from the same dataset, our strongest results arise in human-vs.-LLM
               classification, where negation features outperform both structural
               and embedding baselines.Overall, these findings position discourse-level
               negation as an interpretable feature for reasoning-trace analysis,
               with especially strong utility for provenance-related classification
               and modest but consistent value for correctness prediction.}
}

@inproceedings{Schrottenbacher:et:al:2026:a,
  author    = {Schrottenbacher, Patrick and Mehler, Alexander and Bernhardt, Vivienne
               and Rohe, Leon and Abrami, Giuseppe},
  title     = {ReEmote: Towards Emotion Representation in {VR} Through {Va.Si.Li}-Lab},
  booktitle = {Proceedings of XR Salento 2026},
  year      = {2026},
  publisher = {Springer International Publishing},
  keywords  = {VR, XR, affective computing, virtual humans, emotion detection, FACES},
  abstract  = {Human social interactions are inherently multimodal, shaped not
               only by what speakers convey but also by cues such as facial expressions,
               posture, and gestures. Together, these channels shape both participants'
               perceptions and behaviors, further reinforcing conversational
               feedback loops. This multimodal system extends to VR, where avatars
               serve as proxies for human interaction, making both visual and
               auditory fidelity essential for engaging. To properly utilize
               the emotional expression space that virtual environments allow,
               we introduce ReEmote. ReEmote extends the capabilities of Va.Si.Li-Lab,
               a collaborative, multi-user VR platform built on Ubiq. While Va.Si.Li-Lab
               supports user emotional expression through facial and hand tracking,
               ReEmote extends this by introducing schema-based emotion mappings
               that affect both avatars and their environments. This fosters
               immersive, emotionally aware environments that are beneficial
               for human and chatbot agent interactions, where human users and
               virtual agents share an emotional expression space. By enabling
               richer emotional dynamics, ReEmote opens up new ways of designing
               affective and engaging virtual experiences.In this paper, we describe
               the design choices behind ReEmote and present an evaluation of
               the graphical validity of the emotion representation introduced
               by ReEmote. Our results indicate that emotions can be validly
               represented through avatar facial expressions that users can quickly
               identify as Ekman's basic emotions.This opens up several possibilities
               for extending emotion-related text-to-speech (TTS) applications
               in Extended Reality (XR) with ReEmote. The paper also outlines
               use cases for XR-based TTS applications.},
  note      = {accepted}
}

@article{Chen:et:al:2026,
  doi       = {10.1371/journal.pone.0346096},
  author    = {Chen, Yanran and Zhao, Wei and Breitbarth, Anne and Stoeckel, Manuel
               and Mehler, Alexander and Schlechtweg, Dominik and Eger, Steffen},
  journal   = {PLOS ONE},
  publisher = {Public Library of Science},
  title     = {Syntactic language change in English and German: Metrics, parsers,
               and convergences},
  year      = {2026},
  month     = {04},
  volume    = {21},
  url       = {https://doi.org/10.1371/journal.pone.0346096},
  pages     = {1-33},
  abstract  = {Syntactic language change has gained increasing attention in recent
               years. Previous computational work based on dependency relations
               has focused on diachronic trends in dependency distance, which
               measures the linear distance between dependent words, using dependency
               trees automatically predicted by a dependency parser (mostly the
               Stanford CoreNLP parser). In this work, we introduce a set of
               15 syntax metrics that extend the analysis beyond linear distance
               by incorporating both linear and tree graph properties of dependency
               trees, such as tree height and degree. Besides, we propose a multi-parser
               approach to reduce the impact of using specific parsers, thereby
               increasing the robustness of the detected language changes. Through
               a cross-lingual investigation of English and German in parliamentary
               debates from the last 160 years, using 6 different parsers (CoreNLP
               and five newer alternatives), we demonstrate that: (1) Relying
               on one single parser can be problematic, as the agreement on predicted
               trends can be low across parsers. (2) Our set of metrics can capture
               subtle patterns of syntactic changes. Our analysis shows that
               syntactic change over the time period inspected is largely similar
               between English and German, with only 2.2% of cases yielding opposite
               trends in these metrics. (3) We also show that changes in syntactic
               metrics seem to be more frequent at the tails of sentence length
               distributions and often move in opposite directions for short
               and long sentences. To our best knowledge, ours is the most comprehensive
               computational analysis of syntactic language change using modern
               NLP technology in recent corpora of English and German.},
  number    = {4}
}