Benchmarking Models for Low-Resource Nepali Event Extraction with Trigger Phrase Identification and Event Classification

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

Researchers introduce Nepali Event Extraction (NepEE), a new manually annotated corpus designed to advance event extraction in low-resource South Asian languages. Comprising 10,226 Devanagari sentences, NepEE includes annotations for trigger spans and event types, developed through a rigorous three-phase protocol with five expert native speakers, achieving high inter-annotator agreement (Fleiss' kappa = 0.812 for trigger identification, kappa = 0.855 for event classification). The study benchmarks various approaches, including classical feature-based models, five fine-tuned Transformer encoders, and instruction-tuned Large Language Models using zero-shot and few-shot prompting. Analysis reveals that Indic-specialized Transformers achieve superior event classification performance, while traditional methods and few-shot prompting face challenges with exact span extraction in morphologically complex Nepali. The findings establish strong baselines and quantify performance differences between sentence-level and span-level tasks, providing a valuable public resource for future research.

Key takeaway

For NLP engineers developing event extraction systems for low-resource South Asian languages like Nepali, you should prioritize fine-tuning Indic-specialized Transformer models. These models demonstrate superior event classification performance compared to traditional methods and few-shot LLM prompting. Utilize the new NepEE dataset as a robust baseline for your research, but be prepared for challenges in exact span extraction due to morphological complexity. Focus your efforts on improving span identification techniques.

Key insights

Nepali Event Extraction (NepEE) dataset and benchmarks advance low-resource language event understanding, highlighting Indic-specialized Transformers' superior classification.

Principles

• High-quality manual annotation is crucial for LRL datasets.
• Morphological complexity challenges exact span extraction.
• Indic-specialized Transformers excel in LRL event classification.

Method

Developed NepEE via a rigorous iterative three-phase annotation protocol involving five expert native speakers to ensure linguistic precision for trigger spans and event types.

In practice

• Utilize NepEE for Nepali event extraction research.
• Prioritize Indic-specialized Transformers for LRL classification.
• Address span extraction challenges in morphologically complex languages.

Topics

• Event Extraction
• Low-Resource Languages
• Nepali NLP
• NepEE Dataset
• Transformer Models
• LLM Benchmarking

Code references

• SUJAL390/EEUCA-ACL-2026-Trigger-Phrase-Identification-and-Event-Classification-in-Low-Resource-Languages

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• The Future of Indian Language Technologies: Natural Language Processing for Low-Resource Languages — NLP for low-resource Indian languages is crucial for digital inclusion, overcoming data scarcity through advanced models and collaborative efforts.
• A Multimodal Text- and Graph-Based Approach for Open-Domain Event Extraction from Documents — MODEE combines LLMs and graph learning for robust open-domain event extraction, outperforming prior methods.
• A Beginner’s Guide to NLP Token Classification: NER, POS Tagging, Chunking, and BERT — Token classification, including NER, POS tagging, and chunking, is fundamental for machines to understand human language.
• TantraTagger: A Benchmark Dataset for Tantrayukti-Based Discourse Structure Labelling in Sanskrit Śāstra Texts — TantraTagger is a benchmark dataset for discourse structure labelling in Sanskrit Śāstra Texts.
• Data Augmentation for Historical NER: A Systematic Comparison of Lexical and LLM-based Approaches — This paper systematically compares lexical and LLM-based data augmentation for Historical Named Entity Recognition.
• An End-to-End Hybrid Framework for Rumour Detection in Low-Resources Algerian Dialect — A hybrid framework combining transformer embeddings with classical classifiers enables effective rumour detection in low-resource Algerian dialect.

Open in AIssential →

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