Open ymym3412 opened 5 years ago
Generalized chart constraints for efficient PCFG and TAG parsing Stefan Grunewald ¨ and Sophie Henning and Alexander Koller
NNを使ったPCFGとTAGのパーサ
先行研究の対数線形モデルより高速で高精度
Constituentの開始位置と終了位置を予測するchart constraints。 従来は対数線形モデルによる予測が主だったが、2層のBi-LSTMを使って特徴抽出を行い開始位置と終了位置を予測するモデルを提案した。pruningを行うことで高速に動作
PCFGとTAGのparsingタスクで精度と計算速度の比較を実施
特になし
Finite-State Chart Constraints for Reduced Complexity Context-Free Parsing Pipelines Brian Roark, Kristy Hollingshead, and Nathan Bodenstab
0. 論文
Generalized chart constraints for efficient PCFG and TAG parsing Stefan Grunewald ¨ and Sophie Henning and Alexander Koller
1. どんなもの?
NNを使ったPCFGとTAGのパーサ
2. 先行研究と比べてどこがすごい?
先行研究の対数線形モデルより高速で高精度
3. 技術や手法のキモはどこ?
Constituentの開始位置と終了位置を予測するchart constraints。 従来は対数線形モデルによる予測が主だったが、2層のBi-LSTMを使って特徴抽出を行い開始位置と終了位置を予測するモデルを提案した。pruningを行うことで高速に動作
4. どうやって有効だと検証した?
PCFGとTAGのparsingタスクで精度と計算速度の比較を実施
5. 議論はある?
特になし
6. 次に読むべき論文は?
Finite-State Chart Constraints for Reduced Complexity Context-Free Parsing Pipelines Brian Roark, Kristy Hollingshead, and Nathan Bodenstab