Introduction to the 311 App and Other Requirements - monum/311-translation GitHub Wiki

Overview

In 2010, the Mayor's Office of New Urban Mechanics (City of Boston) launched the City of Boston’s 311 app (one of the first in the nation). The app allows residents to report an array of non-emergency issues (such as potholes) with their smartphones. Historically, the app has only been offered in English. The aim of this project is to create bi-directional machine translation models for two language pairs (English - Spanish & English - Vietnamese) on data collected from the app.

Data Requirements

To create a model that is better than the readily-available models present on the internet. The requirements are as follows:

  1. 1100 pairs of English - Spanish human-translated sentences taken from the 311 app.
  2. 2000 pairs of English - Vietnamese human-translated sentences taken from the 311 app.
  3. Human evaluators to evaluate the performance of the created Machine Learning Model.

Libraries Required

  1. Transformer Model a) Pathlib b) Random c) String d) Re e) Numpy f) Tensorflow g) Pandas h) Keras
  2. Google Translate API Library a) GoogleTrans b) Pandas
  3. Request Generation a) Time b) Requests c) Pandas

Accessing the Models

To run the Transformer Model, you can use the link. Make sure that you are using a GPU runtime for faster executions.

To run the translations using the Google Translate API library, make sure you execute the file google_translate.ipynb.

Note: Use this command to install Google Trans - pip install googletrans==4.0.0-rc1