Grammar, More Features. Proceedings of the 2014 Association for Computational Linguistics. Abstract: We present a parser that relies primarily on extracting information directly from surface spans rather than on propagation information through enriched grammar structure. For example, instead of creating separate grammar symbol to mark the definiteness of an NP, our parser might instead from the first word of the NP. Moving the context out of grammar and onto surface features can greatly simplify the structural component of the parser: because so many deep syntactic cues have surface reflexes, our system can still parse accurately with context-free backbones. Keeping the structural backbone and moving features to new languages and even to new tasks.(Seddah et al., 2013) On the SPMRL 2013 multilingual constituency parsing shared task(Seddah et al.,2013) our system outperforms the top single parser system of Bjorkelund et al.(2013) on a range of languages. In addition, despite being designed for syntactic analysis, our system also achieves state-of-the-art numbers on the structural sentiment task of Socher et al.(2013). Finally, we show that, in the both syntactic analysis and sentiment analysis, many broad linguistic trends can be captured vi surface features. 要旨: 文法規則が少なくとも、表層系(surface)から得られる情報をしっかり使 うことで、文法規則を抑えつつも多くのタスクで大きい情報源となる、と 主張. 本研究ではそれをPCFGを中心に検証している。 2014/8/1 長岡技術科学大学 自然言語処理研究室 2014年度 文献紹介 Less Grammar, More Features.
→ V NP) 0.3, (PP → P NP) 1.0 – Head Lexicalization(Eisner’96; Collins’97; Charniak’97) – Structual annotation(Johonson’98; Klein and Maning’03) – State-splitting(Matsuzaki et al.’98; Petrov et al’06) • Berkeley Parser(F値 90.2 %) • 文法規則の増加 → 精度の向上 [言語依存, 計算量の増加] 2014/8/1 長岡技術科学大学 自然言語処理研究室 2014年度 文献紹介 1. 導入 S NP VP VP NP V NP P PP astronomers saw stars with ears S NP VP NP V NP P PP astronomers saw stars with ears NP
• 本研究で提案されたような文法規則最小限の構造解析器 – 表層形のみの情報(だけ)でも、強力な素性が作り出せる – 言語依存も少なく、拡張性が高い – 素性を変更することで、多タスクへの応用が可能 • Epic Parser(https://github.com/dlwh/epic)として公開 David Hall, Greg Durrett, and Dan Klein. 2014. Less Grammar, More Features. Proceedings of the 2014 Association for Computational Linguistics. 奥村 学. 2010. 自然言語処理の基礎. 能地宏. 2014. ACL読み会2014. Less Grammar , More Features. Recursive Deep Models for Semantic Compositionality. 参考文献