6fanle Submission to the AmericasNLP 2026 Shared Task on Wixarika Image Captioning

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing, Computer Vision · Depth: Advanced, quick

Summary

The 6fanle submission details a Wixarika image captioning system developed for the AmericasNLP 2026 Shared Task on Cultural Image Captioning for Indigenous Languages. This system utilizes Spanish as a pivot language, integrating several key components. It combines CLIP-based image retrieval with Qwen3-VL for generating initial Spanish captions. Subsequently, the official Sheffield-compatible Spanish-to-Wixarika machine translation model translates these captions. To enhance accuracy, the system incorporates character n-gram language-model reranking. The submission includes local 5-fold development results, official test results from the task, and provides an error analysis along with comprehensive implementation details to ensure reproducibility of their methodology.

Key takeaway

For NLP engineers developing image captioning systems for low-resource indigenous languages, this work demonstrates a viable pivot language strategy. You should consider integrating established multimodal models like CLIP and VLMs (e.g., Qwen3-VL) with robust machine translation for the target language. Furthermore, incorporating post-translation reranking, such as character n-gram models, can significantly refine caption quality, offering a reproducible path for similar challenges.

Key insights

A multi-component system pivots through Spanish to generate Wixarika image captions, integrating CLIP retrieval, Qwen3-VL, MT, and n-gram reranking.

Method

The system retrieves images via CLIP, generates Spanish captions with Qwen3-VL, translates to Wixarika using a Sheffield-compatible MT model, then reranks with character n-grams.

In practice

• Use Spanish as a pivot for low-resource languages.
• Combine CLIP and VLM for cross-modal tasks.
• Employ n-gram reranking for MT output refinement.

Topics

• Wixarika Language
• Image Captioning
• AmericasNLP Shared Task
• Multimodal AI
• Machine Translation
• CLIP Model
• Qwen3-VL

Best for: Research Scientist, AI Scientist, NLP Engineer, Computer Vision Engineer

Related on AIssential

• Analysis of Machine Translators on Sentences Generated by Portuguese Image Captioning Models — Machine translation quality directly impacts image captioning model performance when adapting to new languages.
• pplx-embed + Qdrant: Building Production-Grade Semantic Search with Quantization — Perplexity introduced `pplx-embed` on February 26, 2026, a new family of state-of-the-art multilingual text embedding models that utilize diffusion-based pretraining and are built on Qwen3.
• Zero-Shot Captioning for Cultural Heritage: Automated Image Analysis of Traditional Indonesian Clothing — Retrieval-augmented vision-language models effectively generate culturally accurate captions for low-resource heritage data.
• An AI image generator for non-English speakers — Researchers at the University of Amsterdam Faculty of Science, in collaboration with Cohere labs, have developed NeoBabel, an open-source AI image generator designed to address digital inequality by s…
• Hierarchical Multi-Modal Retrieval for Knowledge-Grounded News Image Captioning — Retrieval-augmented image captioning uses hierarchical multi-modal knowledge retrieval and LLMs to generate context-rich news image descriptions.
• G-IdiomAlign: A Gloss-Pivoted Benchmark for Cross-Lingual Idiom Alignment — Cross-lingual idiom alignment challenges LLMs due to non-compositionality, with glosses offering semantic grounding but literal bias remaining dominant.

Open in AIssential →

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