Beyond Clean Text: Evaluating Encoder and Decoder Robustness for Bangla Event Detection in Noisy Text
Quick Answer
This study evaluates the robustness of Bangla event detection systems using a benchmark of 9,979 sentences across clean and noisy text.
Quick Take
This study evaluates the robustness of Bangla event detection systems using a benchmark of 9,979 sentences across clean and noisy text. Encoder models like BanglaBERT excel in clean conditions but falter under noise, while decoder-only models like Llama 3 show greater resilience, especially with corrupted event triggers. Combining training on clean and noisy data improves encoder performance and narrows the robustness gap.
Key Points
- Introduced a Bangla news event ontology with 9,979 annotated sentences across 40 event types.
- Encoder models like BanglaBERT perform better on clean text but degrade significantly in noisy conditions.
- Decoder-only models such as Llama 3 are more robust against noise, especially with corrupted triggers.
- Instruction tuning with embedding guidelines improves performance on noisy text but inconsistently.
- Model scaling enhances decoder robustness, while mixed training benefits encoder architectures.
Paper Resources
Article Content
From source RSS / original summaryarXiv:2606. 30914v1 Announce Type: new Abstract: Event detection (ED) systems are typically evaluated on clean, curated text, leaving their robustness to real-world noise largely unexplored, particularly for low-resource languages such as Bangla. We introduce a generalized Bangla news event ontology and a benchmark comprising 9,979 annotated sentences across 40 event subtypes, spanning clean news text, real-world Automatic Speech Recognition (ASR) transcripts, and orthographically corrupted text.
We systematically evaluate fine-tuned encoder-only models (BanglaBERT and XLM-R) alongside instruction-tuned decoder-only large language models (Llama 3 and Gemma 3). Our results reveal a clear architectural trade-off: encoder models achieve higher performance on clean text but degrade substantially under noise, whereas decoder-only LLMs are markedly more robust, particularly when event triggers are corrupted.
We further show that embedding annotation guidelines during instruction tuning establishes a higher performance baseline on noisy text but yields inconsistent reductions in performance degradation across noisy conditions. Finally, model scaling consistently improves the robustness of decoder-only LLMs, while combined training on clean and noisy data serves as an effective regularization strategy that disproportionately benefits encoder architectures, significantly narrowing the robustness gap.
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CL
See more →Quantifying Prior Dominance in Systems
The study introduces the Normalized Context Utilization (NCU) metric to evaluate Retrieval-Augmented Generation (RAG) systems, revealing that Small Language Models (SLMs) outperform larger models in factual extraction. The findings indicate that traditional scaling laws yield diminishing returns, with a commercial API frequently failing against adversarial evidence due to systemic confidence collapse.