Particle-based language modelling

Abstract

This paper investigates the use of particle (sub-word) £ -grams for language modelling. One linguistics-based and two datadriven algorithms are presented and evaluated in terms of perplexity for Russian and English. Interpolating word trigram and particle 6-gram models gives up to a 7.5% perplexity reduction over the baseline word trigram model for Russian. Lattice rescoring experiments are also performed on 1997 DARPA Hub4 evaluation lattices where the interpolated model gives a 0.4% absolute reduction in word error rate over the baseline word trigram model.

Morphologically motivated word classes for very large vocabulary speech recognition of Finnish and Estonian

“…We have also evaluated the word boundary modelling with mostly similar results. For Finnish, the dedicated word boundary symbol [47,15] has so far been the most effective approach, whereas for Estonian, using the redundant approach has sometimes resulted in a small improvement. For the experiments in this work, the dedicated word boundary symbol was used because it provided the best or equal results in all cases.…”