Barbara Plank · page 2
Papers
- Neural Text Normalization for Luxembourgish Using Real-Life Variation Data (2025 · 10 citations)
Written Luxembourgish varies widely in spelling. Its official orthography is still being standardized, and most writers never learned it at school. Normalization tools are hard to build because there is almost no annotated or parallel standard/non-standard text. The authors work around this by creating parallel data… - A Rose by Any Other Name: LLM-Generated Explanations Are Good Proxies for Human Explanations to Collect Label Distributions on NLI (2025 · 9 citations)
Human annotators often disagree on natural language inference (NLI) items. That disagreement can be recorded as a human judgment distribution (HJD), but measuring it well takes many annotators per item. Earlier work showed that LLMs can estimate HJDs from a few human labels paired with human-written explanations.… - Evaluating Pixel Language Models on Non-Standardized Languages (2025 · 9 citations)
Pretrained language models perform poorly on dialects, partly because subword tokenizers split non-standard spellings into meaningless fragments. - Circuit Compositions: Exploring Modular Structures in Transformer-Based Language Models (2025 · 8 citations)
Mechanistic interpretability usually locates circuits (minimal subgraphs that drive a model's behaviour) one task at a time. - LiTEx: A Linguistic Taxonomy of Explanations for Understanding Within-Label Variation in Natural Language Inference (2025 · 8 citations)
Research on NLI disagreement usually studies annotators who choose different labels. This paper targets a less visible case: annotators who pick the same label but justify it in different ways. Highlighted spans, the usual structured form of explanation, cannot show this. Identical spans can hide different arguments… - More Labels or Cases? Assessing Label Variation in Natural Language Inference (2024 · 8 citations)
The paper argues that summarizing many crowd votes in NLI, either as a majority label or as raw class frequencies, gives an unreliable picture of how uncertain an instance's label really is. - The Validation Gap: A Mechanistic Analysis of How Language Models Compute Arithmetic but Fail to Validate It (2025 · 6 citations)
LLMs can often solve arithmetic yet frequently miss errors in a given solution. This paper asks why, from an internal-mechanism perspective. Using templates, the authors created math word problems in which two single-digit numbers are summed to yield a two-digit result. Each solution either contains an injected… - Standard-to-Dialect Transfer Trends Differ across Text and Speech: A Case Study on Intent and Topic Classification in German Dialects (2026 · 4 citations)
The paper asks whether the usual approach of training on a standard language and testing on a closely related dialect behaves the same way for text and for speech. Dialects are mostly spoken, and their irregular spellings tend to break subword tokenizers. The authors compare three pipelines: - Agree, Disagree, Explain: Decomposing Human Label Variation in NLI through the Lens of Explanations (2026 · 3 citations)
NLI datasets often contain conflicting labels for the same premise–hypothesis pair. Earlier work with the LiTEx taxonomy studied only one kind of variation: annotators who pick the same label but justify it differently. This paper extends LiTEx to cases where the labels themselves diverge. One trained annotator… - Controlling Reading Ease with Gaze-Guided Text Generation (2026 · 3 citations)
LLM prompting can adjust text readability, but the outputs often miss the target difficulty and give little control over how generation works. - Add Noise, Tasks, or Layers? MaiNLP at the VarDial 2025 Shared Task on Norwegian Dialectal Slot and Intent Detection (2025 · 3 citations)
The paper describes the MaiNLP team's entry to the VarDial 2025 NorSID shared task: slot and intent detection (SID) for Bokmål and three Norwegian dialect groups (North Norwegian, Trønder, West Norwegian). - EVADE: LLM-Based Explanation Generation and Validation for Error Detection in NLI (2026 · 1 citation)
For NLI tasks, where a hypothesis is judged against a premise, the variation among human annotators' labels (HLV) means that several labels can be reasonable for the same premise–hypothesis pair. - Position: From Noise to Signal to Selbstzweck - Reframing Human Label Variation in the Era of Post-training in NLP (2026 · 1 citation)
This position paper addresses how NLP treats legitimate annotator disagreement, known as Human Label Variation (HLV), now that LLMs are post-trained on human preference data. - Improved Statistical Measures to Assess Natural Language Parser Performance across Domains (2010 · 1 citation)
Parsers trained on one kind of text often lose accuracy on another, and it is hard to tell in advance which systems will suffer most. - Bi-Text Mining across German Dialects: On the Role of Synthetic Training Data for Dialect Adaptation (2026)
Multilingual sentence encoders perform well on standardized languages, but their behavior on dialects has received little attention. - Exploring Cognitively Informed Sentence Simplification with Gaze-Guided Text Generation (2026)
Text simplification systems usually learn what counts as "easy" from human judgments or crowdsourced reference simplifications. - I Came, I Saw, I Explained: Benchmarking Multimodal LLMs on Figurative Meaning in Memes (2026)
Memes often carry non-literal meaning through the interplay of picture and caption. However, it is poorly understood how generative multimodal large language models (MLLMs) combine the two channels to recognise that meaning. The authors define a combined task: a model must decide which kinds of figurative meaning a… - If Probable, Then Acceptable? Understanding Conditional Acceptability Judgments in Large Language Models (2026)
The paper asks whether large language models judge "If A, then B" statements the way people do. In humans, such judgments depend on two things: how likely B is once A is assumed, and whether A actually bears on B (supports it, undermines it, or has nothing to do with it). The authors reuse the human-rated materials… - Information Asymmetry across Language Varieties: A Case Study on Cantonese-Mandarin and Bavarian-German QA (2026)
LLMs are increasingly used to look up facts, but they may not know things that are documented only in Wikipedia editions written in regional language varieties. - Linear Script Representations in Speech Foundation Models Enable Zero-Shot Transliteration (2026)
Multilingual ASR systems such as Whisper learn from web data in which one language may be transcribed in several writing systems. - ltzGLUE: Luxembourgish General Language Understanding Evaluation (2026)
Luxembourgish (LTZ) has roughly 400k speakers and few NLU resources. Most existing tasks come from news text and often contain under a thousand instances, so model comparisons for the language are unreliable. The authors address this with ltzGLUE, a GLUE-style benchmark of eight sentence- and token-level… - Too Open for Opinion? Embracing Open-Endedness in Large Language Models for Social Simulation (2026)
Research that uses LLMs to stand in for survey respondents or social agents usually restricts the models to multiple-choice or short-answer formats, because such outputs are easy to score and aggregate. - Variation Is the Norm: Embracing Sociolinguistics in NLP (2026)
NLP pipelines usually treat linguistic variation as noise and normalise it away, even though sociolinguistics regards variation as a core property of language and a carrier of social meaning.
This page in Markdown → https://md.research.elamur.ai/free/en/authors/barbara-plank/page/2/