Papers by Wessel Poelman
What is ”Typological Diversity” in NLP? (2024.emnlp-main)
Copied to clipboard
| Challenge: | linguistic typology is commonly used to motivate language selections, but there are no set definitions or criteria for such claims. |
| Approach: | They propose to use linguistic typology to motivate language selections on the basis that a broad typological sample ought to imply generalization across a wide range of languages. |
| Outcome: | The proposed measures show that skewed language selection can lead to overestimated multilingual performance. |
Transparent Semantic Parsing with Universal Dependencies Using Graph Transformations (2022.coling-1)
Copied to clipboard
| Challenge: | Existing semantic parsers are based on deep learning, but rule-based approaches offer advantages . a drawback of neural semantic parses is that their output lacks explainability . |
| Approach: | They propose a method that maps a syntactic dependency tree to a formal meaning representation using a series of graph transformations. |
| Outcome: | The proposed method outperforms neural parsers in English, German, Italian and Dutch. |
Confounding Factors in Relating Model Performance to Morphology (2025.emnlp-main)
Copied to clipboard
| Challenge: | morphological differences between languages are unclear, but are often considered unimportant . confounding factors make it hard to compare results and draw conclusions, authors argue . |
| Approach: | They propose to use token bigram metrics to predict difficulty of causal language modeling . they argue that confounding factors are contributing to the conflicting evidence . |
| Outcome: | The proposed metrics better capture the relation between morphology and tokenization compared to word-based models. |
TIPA: Typologically Informed Parameter Aggregation (2026.findings-eacl)
Copied to clipboard
| Challenge: | Massively multilingual language models enable cross-lingual generalization but underperform on low-resource and unseen languages. |
| Approach: | They propose a typologically informed framework that constructs proxy language adapters by aggregating existing ones, weighted by typological similarity. |
| Outcome: | The proposed framework outperforms baselines on five NLP tasks and over 230 languages. |
How Good is Your Wikipedia? Auditing Data Quality for Low-resource and Multilingual NLP (2026.acl-long)
Copied to clipboard
Kushal Tatariya, Artur Kulmizev, Wessel Poelman, Esther Ploeger, Marcel Bollmann, Johannes Bjerva, Jiaming Luo, Heather Lent, Miryam de Lhoneux
| Challenge: | Wikipedia’s perceived high quality and broad language coverage have established it as a fundamental resource in NLP. |
| Approach: | They propose a data filtering procedure which removes a large percentage of Wikipedia's data and a 4-level quality ranking of the site. |
| Outcome: | The results show that the proposed filtering procedure outperforms the raw Wikipedia models in three language modelling scenarios. |
Form and Meaning in Intrinsic Multilingual Evaluations (2026.eacl-long)
Copied to clipboard
| Challenge: | Existing metrics for conditional language models are not universally comparable . a number of assumptions are made about their quality in multilingual settings . |
| Approach: | They make assumptions about the quality of conditional language models and their semantic meanings that are not consistent with current metrics. |
| Outcome: | The results show that current evaluation metrics are not universally comparable . they use six metrics on two multi-parallel corpora with mono- and multilingual models . |