Natural Language Understanding Wiki

Konstas et al. (2014)[1]:

Current approaches rely primarily on syntactic

features (such as path features) in order to identify and label roles. This has been a mixed blessing as the path from an argument to the predicate can be very informative but is often quite complicated, and depends on the syntactic formalism used. Many paths through the parse tree are

likely to occur infrequently (or not at all), resulting in very sparse information for the classifier to learn from [...] There is previous SRL work employing Tree Adjoining Grammar, albeit in a non-incremental setting, as a means to reduce the sparsity of syntax- based features. Liu and Sarkar (2007) extract a rich feature set from TAG derivations and demonstrate that this improves SRL performance.

Syntax features[]

Daniel & Palmer (2002)[2] demonstrate the importance of syntax features.

See also: [1]

Word embeddings[]

Hermann et al. (2014)[3]: "represent the syntactic context of runs as a vector with blocks for all the possible dependents warranted by a syntactic parser;" [...] "At inference time, the predicate-context is mapped to the low dimensional space, and we choose the nearest frame label as our classification"


From [2]:

"In a typical feature-based approach (Johansson, 2009; Che et al., 2009), feature templates give rise to rich feature descriptions of the semantic structure. The score Ssem(x, ysyn, zsem) is then defined as the inner product between the parameter vector and the feature vector. In the first-order arc-factored case

Screen Shot 2015-06-02 at 19.59.02.png"


  1. Konstas, I., Keller, F., Demberg, V., & Lapata, M. (2014). Incremental Semantic Role Labeling with Tree Adjoining Grammar.
  2. Daniel Gildea and Martha Palmer. 2002. The necessity of syntactic parsing for predicate argument recognition. In Proceedings of the 40th Annual Conference of the Association for Computational Linguistics (ACL-02), Philadelphia, PA.
  3. Hermann, M. K., Das, D., Weston, J., & Ganchev, K. (2014). Semantic Frame Identification with Distributed Word Representations. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (pp. 1448–1458). Association for Computational Linguistics.