The Navajo Translation Project

This is an effort to explore a combination of technologies under the umbrella of artificial intelligence to create a tool for translating the Navajo (Diné) language into American English.

The project’s natural language processing aspect focuses on Apple’s Core ML word tagger model (a parts-of-speech tagger variant) to assist with interlinear translation. The training data come from custom spreadsheets built with over 12000 (and growing) word-tokenized records containing words, phrases, and long textual passages, totaling more than 54000 tokens. Once the dataset considerably grows, this project will likely follow the interlinear tagging with a second-phase custom polysynthetic tokenizer (for deep verb analysis) and a transformer model for true translation.

The Swift Package for the framework in on GitLab at https://www.twine.net/signin The “Navajo to English Translator” is in the App Store.