Supervised training suffers from spurious ambiguity and makes parsers prone to error propagation (McDonald & Nivre, 2007) since the model is under-trained on errorneous states. This has induced a line of research on dynamic oracles (Goldberg & Nivre, 2012; Goldberg & Nirve, 2013; Goldberg et al., 2014; Gómez-Rodríguez et al., 2014; Björkelund & Nivre, 2015).
A rare paper about dynamic oracle which doesn't use averaged perceptron: Goldberg (2013). The probabilistic nature of their parser enables them to do run probabilistic exploration. They found that "sampling-based exploration proposed in this work is indeed superior to the error-based exploration which was used in Goldberg and Nivre (2012, 2013)". Code included.
Goldberg & Nirve (2013): "arc-standard system (Nivre, 2004) is not arc-decomposable, and so deriving a dynamic oracle for it remains an open research question."
A static oracle gives an optimal sequence of actions to reach the gold tree. A dynamic oracle gives an optimal transition for every valid configuration, including those that the gold parse tree is not reachable.
From Goldberg & Nivre (2013): "a general characterization of dynamic oracles as oracles that are non-deterministic, in that they return sets of transitions, and complete".
Zero-cost transition: the decrease in minimum loss of all reachable trees after applying the transition is zero.
Goldberg & Nivre (2012)
- The first dynamic oracle
- Applied to arc-eager transition system
- Averaged perceptron
- Goldberg, Y., & Nivre, J. (2012). A Dynamic Oracle for Arc-Eager Dependency Parsing. Proceedings of the 24th International Conference on Computational Linguistics (COLING), 2(December), 959–976.
- Goldberg, Y., & Nivre, J. (2013). Training Deterministic Parsers with Non-Deterministic Oracles. Transactions of the Association of Computational Linguistics -- Volume 1, 403–414.
- Goldberg, Y., Sartorio, F., & Satta, G. (2014). A tabular method for dynamic oracles in transition-based parsing. Transactions of the Association for Computational Linguistics, 2, 119–130.
- Gómez-Rodríguez, C., Sartorio, F., & Satta, G. (2014). A Polynomial-Time Dynamic Oracle for Non-Projective Dependency Parsing. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) (pp. 917–927). Association for Computational Linguistics.
- Björkelund, Anders, and Joakim Nivre. "Non-Deterministic Oracles for Unrestricted Non-Projective Transition-Based Dependency Parsing." Proceedings of the 14th International Conference on Parsing Technologies. 2015.
- Rasooli, M. S., & Tetreault, J. (2015). Yara Parser: A Fast and Accurate Dependency Parser. arXiv preprint arXiv:1503.06733.
- Goldberg, Y. (2013). Dynamic-oracle Transition-based Parsing with Calibrated Probabilistic Output. In IWPT-2013 (pp. 82–90).