SORNLERTLAMVANICH Virach, INUI Kentaro, SHIRAI Kiyoaki, TANAKA Hozumi, TOKUNAGA Takenobu, TAKEZAWA Toshiyuki
情報処理学会研究報告. NL,自然言語処理研究会報告 119(53) 61-68 1997年5月26日
We propose a probabilistic model for incorporating probability into a (G)LR parser. The model is formalized based on stack transition during parsing distinguishing it from the existing models proposed by Wright and Wrigley, and Briscoe and Carroll. Our model produces a remarkable improvement in statistical parsing with probability. Associating probabilities directly to actions in an LR parsing table, and theoretically requiring only one probability for each action promise model trainability and potential extension to other related tasks.