Retrieval-Augmented Long-Context Translation for Cultural Image Captioning: Gators submission for AmericasNLP 2026 shared task

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

Summary

The University of Florida Gators submitted a two-stage system for the AmericasNLP 2026 shared task, focusing on cultural image captioning for Indigenous languages. Their pipeline initially generates Spanish captions from images using a vision-language model. Subsequently, these Spanish captions are translated into target Indigenous languages through retrieval-augmented many-shot prompting, specifically utilizing the Gemini 2.5 Flash model. This system demonstrated strong improvements over the shared task baseline across multiple Indigenous languages. The submission includes a detailed analysis of the roles played by retrieval mechanisms, synthetic exemplars, and morphology-aware prompting in enhancing translation quality. It also discusses critical limitations related to dev-set exemplars, potential cascade errors within the two-stage process, and the effectiveness of chrF++ based evaluation.

Key takeaway

For NLP engineers developing systems for low-resource language translation, consider a cascaded approach. Your team should explore retrieval-augmented many-shot prompting with powerful LLMs like Gemini 2.5 Flash to improve translation quality. Be mindful of potential cascade errors and the limitations of evaluation metrics like chrF++ when designing your pipeline. This method offers a viable path for cultural image captioning.

Key insights

Retrieval-augmented many-shot prompting with a powerful LLM improves Indigenous language translation for image captions.

Principles

• Two-stage pipelines can bridge modality gaps.
• Retrieval augmentation enhances few-shot translation.
• Morphology-aware prompting is beneficial.

Method

A two-stage pipeline: first, a vision-language model generates Spanish captions; then, Gemini 2.5 Flash translates them to target Indigenous languages using retrieval-augmented many-shot prompting.

In practice

• Use a VLM for initial caption generation.
• Employ retrieval for low-resource translation.
• Consider synthetic exemplars for prompting.

Topics

• Retrieval-Augmented Generation
• Image Captioning
• Indigenous Languages
• Machine Translation
• Gemini 2.5 Flash
• AmericasNLP 2026

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• AI-assisted cultural heritage dissemination: Comparing NMT and glossary-augmented LLM translation in rock art documents — Glossary-augmented LLM prompting significantly improves terminology accuracy in specialized translation without sacrificing overall quality.
• CIAN: Multi-Stage Framework for Event-Enriched Image Captioning via Retrieval-Augmented Generation — CIAN enriches image captions with external event context using retrieval-augmented generation and linguistic refinement.
• Supervision versus Demonstration-Based In-Context Learning for Multiword Expression Classification — In-context learning for Turkish LVCs is highly prompt-sensitive, with few-shot demonstrations improving LLM calibration to match supervised baselines.
• Machines Have Been Trying to Understand Us for Decades. They’re Finally Getting Close. — The transformer architecture, combined with massive scale, fundamentally reshaped NLP, enabling emergent language understanding capabilities previously deemed impossible.
• Multilingual Coreference Resolution via Cycle-Consistent Machine Translation — Coreference resolution in low-resource languages can be significantly improved by cycle-consistent machine translation data augmentation.
• CulturALL: Benchmarking Multilingual and Multicultural Competence of LLMs on Grounded Tasks — CulturALL benchmarks LLMs' multilingual, multicultural, and grounded reasoning on real-world tasks, revealing significant performance gaps.

Open in AIssential →

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