Gaussian Process Classification for Segmenting and Annotating Sequences
Yasemin Altun - Brown University
Thomas Hofmann - Brown University
Alexander Smola - Australian National University
Many real-world classification tasks involve the prediction of multiple, inter-dependent class labels. A prototypical case of this sort deals with prediction of a sequence of labels for a sequence ofobservations. Such problems arise naturally in the context of annotating andsegmenting observation sequences. This paper generalizes Gaussian Processclassification to predict multiple labels by taking dependencies betweenneighboring labels into account. Our approach is motivated by the desire toretain rigorous probabilistic semantics, while overcoming limitations ofparametric methods like Conditional Random Fields, which exhibit conceptualand computational difficulties in high-dimensional input spaces. Experimentson named entity recognition and pitch accent prediction tasks demonstrate thecompetitiveness of our approach.