Analysis of Machine Translators on Sentences Generated by Portuguese Image Captioning Models

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

Summary

This study evaluates the performance of image captioning models trained on datasets translated into Portuguese by various automatic machine translators. The research addresses the prevalent focus on English in computer vision and natural language processing, necessitating adaptation for other languages like Portuguese. Researchers employed a translate-train approach, translating training datasets into Portuguese using different translators. The evaluation was conducted both quantitatively, assessing cost, training time, and test set metrics, and qualitatively, through comparative evaluation forms and error analysis. The findings demonstrate that valid automatic descriptions in Portuguese can be generated by image captioning models trained on translated datasets, with more robust translators yielding more meaningful descriptions.

Key takeaway

For research scientists developing image captioning models for non-English languages, you should prioritize the quality of your machine translation tool when using a translate-train approach. Investing in more robust translators will directly improve the meaningfulness and validity of the generated descriptions, potentially reducing post-processing and error correction efforts.

Key insights

Machine translation quality directly impacts image captioning model performance when adapting to new languages.

Principles

• Translate-train adapts models to new languages.
• Robust translators yield better descriptions.

Method

Image captioning models were trained on Portuguese datasets translated by different machine translators, then evaluated quantitatively (cost, time, metrics) and qualitatively (human review, error analysis).

In practice

• Use translate-train for language adaptation.
• Prioritize robust machine translators.

Topics

• Image Captioning
• Machine Translation
• Portuguese Language
• Translate-Train Approach
• Model Evaluation Metrics

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

Related on AIssential

• Evaluating Small Language Models for English-to-Portuguese Translation: Impact of Model Scale and Quantization — 8-bit quantization maintains translation quality in SLMs, while model scale weakly correlates with performance.
• Bridging Cultural Gaps in Automated Translation of Brazilian Expressions: A Study on Cultural Adaptation — Automated translation systems struggle with cultural adaptation, even with advanced LLMs, requiring human oversight.
• PorTEXTO: A European Portuguese Benchmark for Visual Text Extraction — Specialized multilingual data significantly improves European Portuguese OCR performance over model size or resolution.
• I spent 4 days training an AI model to understand my country. It barely works — Low-resource NLP projects face significant challenges in data collection, translation, and accurate labeling, often yielding modest baselines.
• As Easy as Rocket Science: Assessing the Ability of Large Language Models to Interpret Negation in Figurative Language — LLMs struggle with combined negation and figurative language, with prompt style significantly impacting performance.
• A Dataset for Probing Translationese Preferences in English-to-Swedish Translation — Language models often prefer literal "translationese" over idiomatic phrasing, especially when exposed to the source language.

Open in AIssential →

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