Negation-Aware Data Augmentation for Portuguese Natural Language Inference

2026-04-12 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

This research investigates the effect of targeted data augmentation, specifically focusing on negation cues, within Portuguese Natural Language Inference (NLI) datasets. The study utilized InferBR, ASSIN, and ASSIN2 datasets, synthetically generating new instances by negating hypotheses to enhance training and test set diversity. A BERT-based model was fine-tuned and evaluated on these combined and augmented datasets. The findings indicate that the model's performance was significantly affected by existing biases in negation usage within the original data. The introduction of increased data diversity through negation-aware augmentation demonstrably improved the model's ability to process and understand negation.

Key takeaway

For research scientists developing NLI models for Portuguese, you should consider implementing negation-aware data augmentation. This approach can mitigate biases present in current datasets and significantly improve your model's capacity to handle complex logical reasoning involving negation, leading to more robust and accurate NLI systems.

Key insights

Targeted negation-aware data augmentation improves BERT-based NLI model performance on Portuguese datasets by reducing negation bias.

Principles

• Negation is critical for logical reasoning.
• NLI datasets often underrepresent negation.
• Data diversity improves model robustness.

Method

Synthetically generate new NLI instances by negating hypotheses in existing datasets (InferBR, ASSIN, ASSIN2) to create more diverse training and test sets for BERT-based model fine-tuning.

In practice

• Augment NLI datasets with negated hypotheses.
• Test models for negation bias.
• Apply to other low-resource languages.

Topics

• Negation
• Data Augmentation
• Natural Language Inference
• Portuguese NLI
• BERT Models

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

Related on AIssential

• Data Augmentation for Named Entity Recognition in Domain-Specific Scenarios in Portuguese — Data augmentation significantly improves NER performance in low-resource, domain-specific Portuguese datasets.
• How Dialect Variation Challenges Natural Language Processing Systems — NLP systems struggle with dialects due to biased training data, leading to technical and social inequities.
• Datasets for Verb Alternations across Languages: BLM Templates and Data Augmentation Strategies — This work addresses the underexplored ability of large language models (LLMs) to capture cross-sentence paradigmatic patterns, specifically verb alternations.
• Textual Inference in Portuguese: Comparing Language Models — LLMs struggle with logic-sensitive semantic reasoning in Portuguese, highlighting the need for specialized benchmarks.
• PiDA: Phonetically-Informed Data Augmentation for Robust Vietnamese Speech Translation — Phonetically-informed data augmentation significantly improves Vietnamese speech translation robustness against ASR errors.
• Revisiting the Systematicity in Negation in the Era of In-Context Learning — LLMs partially grasp negation via in-context learning, but struggle with robust scope recognition and function vector composition for it.

Open in AIssential →

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