This presentation demonstrates an interactive next-word English text prediction application built using Shiny in R. The application presents an exciting opportunity to automate user text communication.

• The Coursera/Swiftkey Text Predictor application is a simple Shiny application built using R.
• The application user interface includes brief instructions on using the app, a field for user text entry, and a submission button.
• The user enters English text into the field, presses submit, and the most likely word predicted by the algorithm is displayed on the screen.
• The application can be easily modified to show more than one prediction at a time (e.g., the top five predicted words).

• The text-prediction algorithm is a statistical n-gram model built using English-language text data sourced from SwiftKey’s corpus of news, blog, and Twitter text.
• The algorithm is trained on a random sample of 50,000 lines of text per source file.
• When a user enters text, the model searches for previously observed word patterns and predicts the most probable next word using a backoff strategy that prioritizes longer phrase matches.

• The Courserea/Swiftkey Text Predictor prioritizes the user experience and aims to balance predictive accuracy with processing speed:
• The model performs approximately 1,000 predictions in 2.5 seconds (about 2.5 milliseconds per prediction), prioritizing responsiveness and speed.
• These results highlight the model’s utility as a lightweight application that provides near-instant predictions after loading while maintaining useful next-word suggestions.

• The Coursera/Swiftkey Text Predictor is built entirely in R using Shiny.
• The predictive algorithm is trained on real-world English text from blogs, news, and social media, using an efficient n-gram backoff model for next-word prediction.
• The app demonstrates how statistical language models can provide responsive predictive text with modest computational requirements.