Introduction (Christopher)
---
Reframe problem as Spanish-language image captioning followed by multimodal machine translation
Brief Previous work on Spanish image captioning
Brief Previous work on MMT
Brief Method
State the contributions


Background & Related Work (Christopher)
---
Previous work on Spanish image captioning
Previous work on MMT
Previous work on Retrival and its importance


Datasets & Method (Dzmitry)
---
Ok. Datasets in each language
    Guarani: MultiScript30k
    Maya: Not in 2023 Shared Task

Method:
    Outline the winning submission
    Frame retrival from other datasets and the dev set as hyperparameters r and d


Results (Dzmitry)
---
State the results


Ablations (Both)
---
mBART50, GPT, Gemini, etc.
MultiScript30k (Guarani)
Varying r, d for each language
Different prompts
Morphology (Bribri only)


Discussion (Christopher)
---
Retrival appears to be helping
More data = better results. This is probably why Guarani is so far ahead (as well as morphology)
Synthetic data repeats performance gains shown in EACL (tentative)
Morphology: 
    Bribri & Wixarika morphology. 
    Guarani is morphologically similar to languages that Gemini already knows.

Future Work (Aashish)
---
See hand-off document


Limitations (Aashish)
---
Defer