MMPOStaggerTrainingAbig labeled corpusMaximum likelihood estimationC(ti,tk)Vt'etag,thetag,P(t It')C(t')Vt" e tag, w' e word, (w It) - C(w,t)C(t')通大学C(...) is the number of occurring
MM POS tagger Training • A big labeled corpus • Maximum likelihood estimation C(.) is the number of occurring
MMPOStaggerPredictionFinding a global optimal solutionTom'smathandfurtherphysicsscoreimprovenhn9.89交通大学nR0元n = 2.52bn(Tom's) = 7.37
MM POS tagger Prediction • Finding a global optimal solution Tom‘s math and physics score further improve n n nh c p v n n a n v a d v �n = 2.52 �n 𝑇�′� = 7.37 9.89
MMPOStaggerPredictionFinding a global optimal solutionTom'smathandfurtherphysicsscoreimprovenhn9.8920.02交通大学nn0
MM POS tagger Prediction • Finding a global optimal solution n n nh c p v n n a n v a d v 9.89 20.02 Tom‘s math and physics score further improve
MMPoStaggerPredictionFinding a global optimal solutionTom'smathandfurtherphysicsscoreimprove60.02nhn9.8920.02交通大学nn0
MM POS tagger Prediction • Finding a global optimal solution n n nh c p v n n a n v a d v 9.89 20.02 60.02 Tom‘s math and physics score further improve
MMPOStaggerPredictionFinding a global optimal solutionTom'smathandfurtherphysicsscoreimprove60.02nhn25.329.8920.02+交通大学nn0
MM POS tagger Prediction • Finding a global optimal solution n n nh c p v n n a n v a d v 9.89 20.02 60.02 25.32 Tom‘s math and physics score further improve