Learning to Complete Sentences

Authors: Steffen Bickel Peter Haider Tobias Scheffer

Publish Date: 2005/10/3

Volume: , Issue: , Pages: 497-504

PDF Link

Abstract

We consider the problem of predicting how a user will continue a given initial text fragment Intuitively our goal is to develop a “tabcomplete” function for natural language based on a model that is learned from text data We consider two learning mechanisms that generate predictive models from collections of applicationspecific document collections we develop an Ngram based completion method and discuss the application of instancebased learning After developing evaluation metrics for this task we empirically compare the modelbased to the instancebased method and assess the predictability of callcenter emails personal emails and weather reports

Keywords: