Inspiration
We are children of immigrants. Growing up, our parents relied on us to translate words for them. This is an issue at large, but especially important in the legal arena. Legalese is incomprehensible and largely inaccessible to immigrants like us. When legal issues are at hand, families, finances, and health are at stake. In order to make the world more equitable through service, we created Legal Translation Hub.
What it does
LTH performs translation on legal documents specifically in three new ways, resulting in greatly increased translation accuracy.
First, we aggregate multiple translation APIs (e.g. Google Translate, DeepL, etc.). We evaluate each translation engine based on its accuracy for the particular language pair for our user, and select the best engine possible.
Second, we provide a human-in-the-loop translation. Volunteers from VT who are certified in language are able to look at documents and certify them. Each document is passed through 2 volunteers. If both volunteers disagree on revision, it's passed to a third tie-breaker volunteer. Volunteers will be incentivized by community service or college credit hours.
Third, we abstract the issue of complex terminology. Often, American law terms don't have direct translations in other languages. But, their definitions, written in simpler language, do. Thus, we use a legal dictionary to flag and provide definitions for any legal terms found in the user's document. Then, those definitions are translated, in tandem with the greater document.
How we built it
We used Angular and Firebase for this project, primarily. Angular's components and Materials library greatly facilitated rapid prototyping. What we have right now is a minimum viable product. In the future, the project also will implement OCR via a Hugging Face library, as well as various translation engine APIs. Also via Hugging Face, the project will perform image segmentation to preserve page text structure. After this Hackathon, we want to continue this project and further integrate more translation engines.
Challenges we ran into
One challenge was a diverse team; given team members studying computer engineering and quantum information science, a software engineering project was a challenge. However, through teamwork (and copious amounts of caffeine and Google), we were able to surmount it.
Accomplishments that we're proud of
We believe above all else, the concept is strong: making legal documents more accessible
What we learned
We learned that selecting the best translation engine can be a challenge. Originally, we planned to use confidence intervals to evaluate strength. However, the APIs we used did not provide those intervals. So, we ended up using a majority-vote approach.
What's next for Legal Translation Hub
After this Hackathon we plan on refining the website, deploying it to a server, and recruiting volunteers/discussing credit hour possibilities with VT faculty. We strongly believe this app embodies Ut Prosim, and we're excited to introduce it to Virginia Tech.
Log in or sign up for Devpost to join the conversation.