The predictText app allows the user to type words/phrases, then seeks to (1) predict the next word and (2) if applicable, autocomplete an unfinished word. The app is available here.

Simplicity drives the user interface design. When a user is in the midst of typing a word, a list of autocomplete suggestions are generated:

When a user completes typing a word, a list of next word suggestions are generated. The program is alerted to the completion of a word when a space is placed after the word.

• Model based upon N-gram-Katz Backoff Model concepts
• Corpus from which model is derived converted to lower-case, removed of URLs and special characters, stripped of excess white space and cleaned of numbers and punctuation
• Matrices excluded from 5-grams to unigrams
• Improved speed by excluding least-frequent occurences of all n-gram orders
• Exclude least frequent unigram and bigram to improve speed with little sacrifice to accuracy

• Reactively, the algorithm reads the text input and predicts the next word
• Iteratively reads from longest 5-gram to shortest 2-gram
• Predicts the longest, most frequent, n-gram
• In cases where no match is found, the algorithm selects the most-frequent single word
• For source code and technical notes, visit predictText Github Repository