Source PaperYearLineSentence
P14-1069 2014 15
Recent studies (Zhu et al, 2013;Zhang et al, 2013) show, however, that this approach can also achieve the state-of-the-art perfor mance with improved training procedures and the use of additional source of information as features.However, there is still room for improvement for these state-of-the-art transition-based con stituent parsers
P14-1069 2014 83
Zhu et al (2013) proposed a padding method to align terminal states containing different number of actions
P14-1069 2014 89
POP(beam i ) 6: for t ? {sh-x,rl-x,rr-x} do 7: s new ? apply t to s 8: score s new with E.q (1) 9: insert s new into P 0 10: for all state s in P 0 do 11: for all possible t ? {ru-x} do 12: s new ? apply t to s 13: score s new with E.q (1) 14: insert s new into P 1 15: insert all states of P 1 into P 0 16: beam i+1 ? k best states of P 0 17: return the best state in beam 2n?1We propose a novel method to align states during the parsing process instead of just aligning ter minal states like Zhu et al (2013)
P14-1069 2014 113
Semi-supervised word cluster features have been successfully applied to many NLP tasks (Miller et al, 2004; Koo et al, 2008; Zhu et al., 2013)
P14-1125 2014 98
We use the padding method (Zhu et al, 2013), adding an IDLE action to finished transition action sequences, for better alignments between states in the beam (self citation)