Errors in coreference resolution in German: Effects of modality, simplification and heterogeneous training data

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Advanced, quick

Summary

The paper "Errors in coreference resolution in German: Effects of modality, simplification and heterogeneous training data" investigates mention detection errors within the CorPipe coreference resolver. Researchers Sarah Jablotschkin, Ekaterina Lapshinova-Koltunski, and Heike Zinsmeister evaluated CorPipe's performance across various German text variants, including written, spoken, original, and simplified forms. Their analysis considers that CorPipe was trained on a combination of German coreference corpora, some of which had conflicting annotation guidelines. A significant finding indicates that text simplification notably impacts mention detection accuracy, a phenomenon observed independently of the text's modality (written or spoken). This highlights challenges in building robust coreference systems for diverse linguistic inputs.

Key takeaway

For NLP Engineers developing or deploying German coreference resolution systems, you should rigorously test your models against simplified text variants, regardless of whether the input is written or spoken. Be aware that training data derived from multiple corpora with differing annotation guidelines can introduce performance inconsistencies. Prioritize harmonizing annotation schemes or employing robust normalization techniques to mitigate errors in mention detection for improved system reliability.

Key insights

Text simplification significantly impacts German coreference mention detection, independent of modality, especially with heterogeneous training data.

Principles

• Coreference errors stem from mention detection.
• Text simplification impacts mention detection.
• Heterogeneous training data affects resolvers.

Method

The study evaluated CorPipe's mention detection errors across written, spoken, original, and simplified German variants, analyzing performance against heterogeneous training data.

In practice

• Account for text simplification effects.
• Scrutinize coreference training data.
• Test resolvers on diverse German variants.

Topics

• Coreference Resolution
• German NLP
• Mention Detection
• Text Simplification
• CorPipe
• Training Data

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• Coreference Resolution in spaCy — spaCy's experimental "coref" component enables robust coreference resolution for NLP applications.
• On the Limits of LLM Adaptability: Impact of Model-Internalized Priors on Annotation Task Performance — LLM annotation reliability hinges on definition alignment, not just text memorization, as prompt-based corrections are largely ineffective.
• Event and Entity Coreference Across Five Languages: Effects of Context and Referring Expression — Crosslingual coreference biases show personal pronouns favor entities, demonstratives favor events, influenced by antecedent complexity.
• Plug-and-Adapt: Multimodal Coreference Resolution at First Sight with a Pretrained Alignment Model — Plug-and-Adapt repurposes a pre-trained vision-language alignment model for zero-shot multimodal coreference resolution, outperforming existing methods.
• When Surface Form Changes Moderation Decisions: A Paired Study of Code-Mixed Workflow Instability — Code-mixed inputs cause significant moderation workflow instability, increasing review burden and non-hate false-flags.
• Towards a Universal Dependencies Corpus for Portuguese Epidemiological Reports — Document extraction and representation significantly impact parsing epidemiological texts more than parser selection.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.