Inspiration
There are more than 6,000 languages in the world, but anywhere between 50-90% of them could go extinct within our lifetime. The 20 most common languages are spoken by the majority of the world's population, but most of the world's languages are spoken by only a few thousand people. It is vital that we preserve these languages; when a language dies, the heart of the cultural history it carries dies with it. This project was driven by the urgent need for language preservation and revitalization.
Hack Dearborn takes place in Michigan, on the ancestral homeland of the Anishinaabe people. Of the three Anishinaabeg peoples who predominantly reside in Michigan (Ojibwe, Ottawa, Potawatomi), the Ojibwe population is the largest. The Ojibwe dialect of Anishinaabemowin (sometimes referred to as Ojibwemowin) was used to create this dictionary.
What it does
This is a simple web app implementation of an Ojibwe-English dictionary. Users are able to search in both Ojibwe and English, and the dictionary will return entries that match the query. An entry contains the word in Ojibwe, the English translation, example sentences (if available), and inflectional/morphological data* (if available).
*Anishinaabemowin is a morphologically rich language, a characteristic shared by many indigenous North American languages. The morphological complexity of these languages is often significant when compared to high-resource languages such as English. The structure of a language is an important factor that affects the ease with which the languages can be digitized; models trained on languages like English often simply cannot handle morphologically rich languages like Anishinaabemowin without extensive alteration.
How we built it
All of the dictionary data used in this project was collected from the Ojibwe People's Dictionary, published by the University of Minnesota. This data is available under Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License. Dictionary data was processed using Python in Jupyter Notebook.
The dictionary web implementation was created using React, JavaScript, and CSS in Visual Studio. ChatGPT assisted in troubleshooting coding errors.
Challenges we ran into
My initial goal for this project was actually to create an Ojibwe-English machine translator. I attempted to utilize several models (mBART, MarianMT, T5), with varying degrees of success. After almost 9 hours of trial and error, I ultimately decided building a fully functioning translation model was outside of the scope of what I'd be able to accomplish within a 24-hour hackathon. I instead decided to use the extensive dictionary data I had collected to make a dictionary app.
What we learned
This was actually my first time using React, so I basically had to learn it on the fly. I learned an incredible amount about React, UI/UX, and JavaScript that I might not have learned before this project. I also learned about recognizing my limits and being able to acknowledge when I had bitten off more than I could chew. Overall, it was an exceptionally informative experience.
What's next for Ojibwe Dictionary
I absolutely plan on continuing this project after the conclusion of the hackathon. With more time, I'll be able to research more in-depth about utilizing pre-trained language models to process low-resource languages. My main area of focus will likely be transfer learning. With a quality trained model, I'll be able to implement a machine translation feature (ideally both word-to-word and full sentences).
Log in or sign up for Devpost to join the conversation.