Enhanced Universal Dependencies in the Wild: Evaluating Portuguese EUD Parsing in Realistic Scenarios

2026-04-12 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Advanced, quick

Summary

A study evaluated the robustness of Enhanced Universal Dependencies (EUD) conversion rules for Portuguese across diverse text genres and domains, moving beyond previous evaluations limited to journalistic text and gold-standard basic syntactic trees. The research specifically assessed the performance of Portuguese-specific EUD rules within realistic parsing pipelines that utilize automatically generated basic syntax. Findings indicate that the Portuguese-specific rules consistently surpass universal rules in performance. However, the accuracy of the EUD parsing significantly degrades when relying on automatically generated basic syntax, especially when the input text's domain differs from the basic parser's training data. The study also includes a detailed error analysis, pinpointing challenging linguistic phenomena and scenarios.

Key takeaway

For NLP Engineers developing Portuguese dependency parsers, you should prioritize using Portuguese-specific EUD conversion rules over universal ones, as they consistently yield better results. Be aware that relying on automatically generated basic syntax significantly degrades performance, especially with domain shifts, necessitating careful consideration of your basic parser's training data and potential fine-tuning for target domains.

Key insights

Portuguese-specific EUD rules outperform universal rules, but automatic basic syntax severely impacts performance.

Principles

• Domain mismatch degrades parser performance.
• Automatic syntax impacts EUD accuracy.

Method

Evaluated Portuguese EUD conversion rules using diverse text genres and domains, comparing performance with automatically generated basic syntax against gold-standard basic trees to assess robustness "in the wild."

In practice

• Prioritize domain-matched training data.
• Use Portuguese-specific EUD rules.

Topics

• Enhanced Universal Dependencies
• Universal Dependencies
• Syntactic Parsing
• Portuguese Language Processing
• Domain Mismatch

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• Towards a Universal Dependencies Corpus for Portuguese Epidemiological Reports — Document extraction and representation significantly impact parsing epidemiological texts more than parser selection.
• ALBA: A European Portuguese Benchmark for Evaluating Language and Linguistic Dimensions in Generative LLMs — ALBA provides a linguistically-grounded benchmark for European Portuguese LLM evaluation, addressing a critical language variety gap.
• Universal Dependencies v2.5 Benchmarks for spaCy — spaCy v3.2 demonstrates competitive performance against Stanza and Trankit on Universal Dependencies v2.5 benchmarks using CoNLL 2018 evaluation.
• Syntax as a Rosetta Stone: Universal Dependencies for In-Context Coptic Translation — Syntactic augmentation via Universal Dependencies significantly improves low-resource Coptic-to-English machine translation when combined with dictionary glosses.
• AfriSUD: A Dependency Treebank Collection for Evaluating Models on African Languages — AfriSUD provides the first large-scale, native-speaker verified dependency treebanks for nine diverse African languages, revealing a significant syntax gap in current NLP models.
• Parsing English in 500 Lines of Python — Transition-based dependency parsing offers a breakthrough in NLU with a concise, dependency-free Python implementation.

Open in AIssential →

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