in an effort to increase the representation of low resource languages which a lot of indigenous languages . Many of those languages do not have enough easy available data to train a model using a large dataset. We wanted to develop a translator which would not depend on a lot of data but that would instead learn from the features of a language and words formation.
What it does
Given a word or sentence from a source language (within the list of languages provided), the machine learning algorithm translates the input and outputs the translated result
How we built it
Used Byte-pairing encoding(BPE) for translation method using python. Python to build the machine learning algorithm
Challenges we ran into
Training the model Time constraints when working on the web-app front end and backend processing.
Accomplishments that we're proud of
- Gaining comprehension of alternative models for translation algorithms.
What we learned
BPE, server-client data transmission using REST, using python for data preparation
What's next for Machine Translation for Low Resource Languages
Improving the training model to better incorporate the use of language-specific features (syntax, grammatical rules, ect) . Developing a more comprehensive webapp that is user-friendly .