--- title: "Culturally-Aware Image Captioning for Guaraní — Shi 2026" description: "The paper describes the IUHoosiers entry for the Guaraní track of the AmericasNLP 2026 Shared Task, which asks systems to caption images with culturally…" authors: ["Wenchen Shi", "Phakphum Artkaew", "Luke Gessler"] year: 2026 venue: "Proceedings of the Sixth Workshop on NLP for Indigenous Languages of the Americas (AmericasNLP)" source: "https://aclanthology.org/2026.americasnlp-6.21/" doi: "10.18653/v1/2026.americasnlp-6.21" licence: "CC BY 4.0 (ACL Anthology)" licence_url: "https://creativecommons.org/licenses/by/4.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/papers/shi-2026-culturally-aware-image-captioning-guarani/" --- [Elamur Research](https://md.research.elamur.ai/) · [this page in HTML](https://research.elamur.ai/free/papers/shi-2026-culturally-aware-image-captioning-guarani/) [home](https://md.research.elamur.ai/) / [papers](https://md.research.elamur.ai/free/papers/) / [2026](https://md.research.elamur.ai/free/papers/2026/page/2/) / Culturally-Aware Image Captioning for Guaraní with Multimodal Prompting: IUHoosiers at AmericasNLP 2026 # Culturally-Aware Image Captioning for Guaraní with Multimodal Prompting: IUHoosiers at AmericasNLP 2026 **Authors:** Wenchen Shi, Phakphum Artkaew, [Luke Gessler](https://md.research.elamur.ai/free/authors/luke-gessler/) · **Year:** 2026 · **Venue:** Proceedings of the Sixth Workshop on NLP for Indigenous Languages of the Americas (AmericasNLP) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/) **Links:** [Publisher's page](https://aclanthology.org/2026.americasnlp-6.21/) · [PDF](https://aclanthology.org/2026.americasnlp-6.21.pdf) · [DOI 10.18653/v1/2026.americasnlp-6.21](https://doi.org/10.18653/v1/2026.americasnlp-6.21) · [OpenAlex](https://openalex.org/W7166855943) **Citations:** 0 citations (openalex-snapshot · 2026-10-14) · 1 in our library, of the 140673 papers whose bibliographies we read ## Summary The paper describes the IUHoosiers entry for the Guaraní track of the AmericasNLP 2026 Shared Task, which asks systems to caption images with culturally grounded descriptions in Indigenous languages. Guaraní is morphologically rich and heavily mixed with Spanish (jopara), and only 50 dev captions were available. Because fine-tuning risked overfitting, the authors used no parameter updates and relied on inference-time knowledge injection instead. Gemma 4 31B first describes each image in English or Spanish. That description becomes a BM25 query over four Guaraní resource pools: DAMPY dictionary entries, FLORES-200 parallel sentences, dev-set captions, and a hand-built cultural knowledge base (CUL). The retrieved items go into the system prompt together with a fixed grammar excerpt and interlinear-glossed examples. A variant also supplies DAMPY images as multimodal few-shot demonstrations. The system ranked first for Guaraní in both automatic and human evaluation. This suggests that curated grammatical and cultural context, delivered through prompting, can compete with adapted models in very low-resource settings. ## Key points - The best run (v4, text+RAG) reached 24.67 chrF++ and 0.149 CIDEr on the test set. The official baseline scored 20.14 and the next team (gators) scored 23.10. Human evaluation gave 3.45/5, compared with 3.30 for the runner-up. - Eight of the nine submitted configurations beat every other team's best entry. - Every prompt included two fixed resources, both adapted from Aycock et al. (2025): a 157-line grammar excerpt based on Estigarribia (2020) and 144 IGT blocks. - The retrieval pools were: - DAMPY: 111 food, fauna and flora entries, with captions generated with LLM help. - FLORES-200: 1,012 EN–GN sentence pairs. - DEV: 50 gold captions. - CUL: 22 thematic sections, always included in full. - The visual few-shot run (v7) had the highest dev score (24.17) but the lowest test chrF++ (22.43). Even so, it correctly distinguished mate from tereré in a case where most text+RAG runs did not. *Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.* Fields: [Linguistics & languages](https://md.research.elamur.ai/free/shelves/linguistics/) ## Cite The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{shi2026culturally, title = {{Culturally-Aware Image Captioning for Guaraní with Multimodal Prompting: IUHoosiers at AmericasNLP 2026}}, author = {Shi, Wenchen and Artkaew, Phakphum and Gessler, Luke}, booktitle = {Proceedings of the Sixth Workshop on NLP for Indigenous Languages of the Americas (AmericasNLP)}, year = {2026}, doi = {10.18653/v1/2026.americasnlp-6.21}, url = {https://doi.org/10.18653/v1/2026.americasnlp-6.21} } ``` - [← previous CUHKSZ Simultaneous Speech Translation System for IWSLT 2026](https://md.research.elamur.ai/free/papers/yang-2026-cuhksz-simultaneous-speech-translation-system/) - [next → Culture-Aware Machine Translation in Large Language Models: Benchmarking and Investigation](https://md.research.elamur.ai/free/papers/yuan-2026-culture-aware-machine-translation-large/) - [↑ back to papers of 2026](https://md.research.elamur.ai/free/papers/2026/page/2/) ## References 16 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Seth Aycock, David Stap, Di Wu, Christof Monz, and Khalil Sima’an. 2025. Can LLMs really learn to translate a low-resource language from one grammar book? arXiv preprint arXiv:2409.19151. [find](https://scholar.google.com/scholar?q=Seth+Aycock%2C+David+Stap%2C+Di+Wu%2C+Christof+Monz%2C+and+Khalil+Sima%E2%80%99an%2E+2025%2E+Can+LLMs+really+learn+to+translate+a+low%2Dresource+language+from+one+grammar+book%3F+arXiv+preprint+arXiv%3A2409%2E19151%2E) 2. Minh Duc Bui, David Guzmán, Abteen Ebrahimi, Franklin Morales, Marvin Agüero-Torales, Raquel Insfrán, Cecilia González, Ramón Araujo, Luca Cernuzzi, Carlos Raul Noh Chi, Carlos Eduardo Tec Cahun, Sindi Estrella Poot Cohuo, Daniel Ricardo Benítez Chi, Santos Natanael Palomo Arévalo, Jessica Elizabeth… [find](https://scholar.google.com/scholar?q=Minh+Duc+Bui%2C+David+Guzm%C3%A1n%2C+Abteen+Ebrahimi%2C+Franklin+Morales%2C+Marvin+Ag%C3%BCero%2DTorales%2C+Raquel+Insfr%C3%A1n%2C+Cecilia+Gonz%C3%A1lez%2C+Ram%C3%B3n+Araujo%2C+Luca+Cernuzzi%2C+Carlos+Raul+Noh+Chi%2C+Carlos+Eduardo+Tec+Cahun%2C+Sind) 3. Luis Chiruzzo, Santiago Castro, Mariella Cardenas, Gustavo Gimenez González, Yliana Gimenez, and Dina Wonsever. 2020. Development of a Guaraní-Spanish parallel corpus. In Proceedings of the 12th Language Resources and Evaluation Conference, pages 893–898, Marseille, France. European Language Resourc… [find](https://scholar.google.com/scholar?q=Luis+Chiruzzo%2C+Santiago+Castro%2C+Mariella+Cardenas%2C+Gustavo+Gimenez+Gonz%C3%A1lez%2C+Yliana+Gimenez%2C+and+Dina+Wonsever%2E+2020%2E+Development+of+a+Guaran%C3%AD%2DSpanish+parallel+corpus%2E+In+Proceedings+of+the+12th+Langu) 4. Garrett Tanzer, Mirac Suzgun, Eline Visser, Dan Jurafsky, and Luke Melas-Kyriazi. 2023. A benchmark for learning to translate a new language from one grammar book. arXiv preprint arXiv:2309.16575. [find](https://scholar.google.com/scholar?q=Garrett+Tanzer%2C+Mirac+Suzgun%2C+Eline+Visser%2C+Dan+Jurafsky%2C+and+Luke+Melas%2DKyriazi%2E+2023%2E+A+benchmark+for+learning+to+translate+a+new+language+from+one+grammar+book%2E+arXiv+preprint+arXiv%3A2309%2E16575%2E) 5. Chen Zhang, Jiuheng Lin, Xiao Liu, Zekai Zhang, and Yansong Feng. 2025. Read it in two steps: Translating extremely low-resource languages with code-augmented grammar books. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 3977… [find](https://scholar.google.com/scholar?q=Chen+Zhang%2C+Jiuheng+Lin%2C+Xiao+Liu%2C+Zekai+Zhang%2C+and+Yansong+Feng%2E+2025%2E+Read+it+in+two+steps%3A+Translating+extremely+low%2Dresource+languages+with+code%2Daugmented+grammar+books%2E+In+Proceedings+of+the+63rd) 6. Bruno Estigarribia. 2020. A Grammar of Paraguayan Guaraní. UCL Press. [find](https://scholar.google.com/scholar?q=Bruno+Estigarribia%2E+2020%2E+A+Grammar+of+Paraguayan+Guaran%C3%AD%2E+UCL+Press%2E) 7. A Gemini Team, Petko Georgiev, Ving Ian Lei, Ryan Burnell, Libin Bai, Anmol Gulati, Garrett Tanzer, Damien Vincent, Zhufeng Pan, Shibo Wang, and 1 others. 2024. Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context. Preprint, arXiv:2403.05530. [find](https://scholar.google.com/scholar?q=A+Gemini+Team%2C+Petko+Georgiev%2C+Ving+Ian+Lei%2C+Ryan+Burnell%2C+Libin+Bai%2C+Anmol+Gulati%2C+Garrett+Tanzer%2C+Damien+Vincent%2C+Zhufeng+Pan%2C+Shibo+Wang%2C+and+1+others%2E+2024%2E+Gemini+1%2E5%3A+Unlocking+multimodal+unders) [our card](https://md.research.elamur.ai/free/papers/gemini-2024-team-gemini-1-5-unlocking-multimodal/) 8. The CUL resource was built in three LLM-assisted steps using Claude Opus 4. [find](https://scholar.google.com/scholar?q=The+CUL+resource+was+built+in+three+LLM%2Dassisted+steps+using+Claude+Opus+4%2E) 9. Google DeepMind. 2025. Gemma 4 technical report. [https://blog.google/technology/](https://blog.google/technology/) google-deepmind/gemma-4/. [↗](https://blog.google/technology/) 10. Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. LoRA: Low-rank adaptation of large language models. Preprint, arXiv:2106.09685. [find](https://scholar.google.com/scholar?q=Edward+J%2E+Hu%2C+Yelong+Shen%2C+Phillip+Wallis%2C+Zeyuan+Allen%2DZhu%2C+Yuanzhi+Li%2C+Shean+Wang%2C+Lu+Wang%2C+and+Weizhu+Chen%2E+2022%2E+LoRA%3A+Low%2Drank+adaptation+of+large+language+models%2E+Preprint%2C+arXiv%3A2106%2E09685%2E) 11. Yuchen Huang, Zhiyuan Fan, Zhitao He, Sandeep Polisetty, Wenyan Li, and Yi R. Fung. 2025. Cul-tureCLIP: Empowering CLIP with cultural awareness through synthetic images and contextualized captions. Preprint, arXiv:2507.06210. — [arXiv:2507.06210](https://arxiv.org/abs/2507.06210) [find](https://scholar.google.com/scholar?q=Yuchen+Huang%2C+Zhiyuan+Fan%2C+Zhitao+He%2C+Sandeep+Polisetty%2C+Wenyan+Li%2C+and+Yi+R%2E+Fung%2E+2025%2E+Cul%2DtureCLIP%3A+Empowering+CLIP+with+cultural+awareness+through+synthetic+images+and+contextualized+captions%2E+Pr) 12. Yue Li, Zhixue Zhao, and Carolina Scarton. 2025. It’s all about in-context learning! teaching extremely low-resource languages to LLMs. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 29544–29559, Suzhou, China. Association for Computational Linguisti… [find](https://scholar.google.com/scholar?q=Yue+Li%2C+Zhixue+Zhao%2C+and+Carolina+Scarton%2E+2025%2E+It%E2%80%99s+all+about+in%2Dcontext+learning%21+teaching+extremely+low%2Dresource+languages+to+LLMs%2E+In+Proceedings+of+the+2025+Conference+on+Empirical+Methods+in+Na) 13. Sourab Mangrulkar, Sylvain Gugger, Lysandre Debut, Younes Belkada, Sayak Paul, and Benjamin Bossan. 2022. PEFT: State-of-the-art parameter-efficient fine-tuning methods. [https://github.com/huggingface/peft](https://github.com/huggingface/peft). [↗](https://github.com/huggingface/peft) 14. NLLB Team, Marta R. Costa-jussà, James Cross, Onur Çelebi, Maha Elbayad, Kenneth Heafield, Kevin Heffernan, Elahe Kalbassi, Janice Lam, Daniel Licht, Jean Maillard, Anna Sun, Skyler Wang, Guillaume Wenzek, Al Youngblood, Bapi Akula, Loic Barrault, Gabriel Mejia-Gonzalez, Prangthip Hansanti, and 20 o… [find](https://scholar.google.com/scholar?q=NLLB+Team%2C+Marta+R%2E+Costa%2Djuss%C3%A0%2C+James+Cross%2C+Onur+%C3%87elebi%2C+Maha+Elbayad%2C+Kenneth+Heafield%2C+Kevin+Heffernan%2C+Elahe+Kalbassi%2C+Janice+Lam%2C+Daniel+Licht%2C+Jean+Maillard%2C+Anna+Sun%2C+Skyler+Wang%2C+Guillaume+We) 15. Stephen Robertson and Hugo Zaragoza. 2009. The probabilistic relevance framework: BM25 and beyond. Foundations and Trends in Information Retrieval, 3(4):333–389. [find](https://scholar.google.com/scholar?q=Stephen+Robertson+and+Hugo+Zaragoza%2E+2009%2E+The+probabilistic+relevance+framework%3A+BM25+and+beyond%2E+Foundations+and+Trends+in+Information+Retrieval%2C+3%284%29%3A333%E2%80%93389%2E) [our card](https://md.research.elamur.ai/free/papers/robertson-2009-probabilistic-relevance-framework-bm25/) 16. Construction of the CUL Cultural Knowledge Base [find](https://scholar.google.com/scholar?q=Construction+of+the+CUL+Cultural+Knowledge+Base)