Quantitative evaluation of a portuguese word predictor

Abstract

This paper reports the results of a first evaluation of a Portuguese statistical word predictor. For a quantitative evaluation we developed a simulator that acts as an ideal user of the system. Simulations using texts from three distinct sources presented similar values of efficacy (e.g. 46%, 47% and 50% keystroke saving rate). We also studied the effect of changing the weight of each statistical factor. In general the introduction of each factor improved the system efficacy, however the variation of each factor’s coefficient did not produced significant changes to that efficacy.