Skip to content
santhoshtr edited this page Sep 14, 2010 · 2 revisions

Welcome to the ibus-sulekha wiki!

Introduction

ibus-sulekha is an ibus input method for predictive text entry in applications. Have a look at the initial screenshots:

Screenshot1

Screenshot2

Screenshot3

Algorithm



  • An sqllite database of ngrams: trained initially from an arbitrary corpus.

  • ngrams are of two types are stored: word ngrams – for word completion, letter ngrams – for letter completion.

  • We use bigrams -ngrams where n=2.

  • When user type a letter, find out the most probable next letter using the ngram frequency saved in the database. Based on frequency scores list the candidates – the number candidates can be limited. The first candidate being selected by default in cursor position

  • When a new word being started, look at the possible words based on the previous word. This is again based on word ngrams.

  • The input method should also learn from what user types and should update the database.

Related links

Clone this wiki locally