Esther Seyffarth (Heinrich Heine University) presents:
From Text to Frame Generation
Semantic Frames in the sense of Gamerschlag et al. (2014) let us represent concepts in the form of recursive feature-value structures. This can be used to represent the meaning of a sentence in terms of a semantic frame: The frame's top-level attributes are the semantic roles involved in the event described by the sentence's root verb, and the entities that fill the semantic role slots are their values. These frame structures allow frames to be embedded into other frames as attribute values.
The automatic creation of a semantic frame resource, as implemented by Kallmeyer, QasemiZadeh and Cheung (2018), involves the clustering of different verb lemmas into the same semantic frame, as well as the separation of different senses of a verb into different semantic frames. The arguments of each verb, as observed in the corpus, need to be grouped into semantic roles.
In this talk, I discuss the difficulties that arise in this context from verbs that participate in diathesis alternations (Seyffarth, 2018), where semantic roles are encoded in different syntactic positions.
Gamerschlag, T., Gerland, D., Osswald, R., & Petersen, W. (Eds.). (2013). Frames and concept types: applications in language and philosophy (Vol. 94). Springer Science & Business Media.
Kallmeyer, L., QasemiZadeh, B. & Cheung, J. C. K. 2018. Coarse Lexical Frame Acquisition at the Syntax-Semantics Interface Using a Latent-Variable PCFG Model. In Proceedings of the Seventh Joint Conference on Lexical and Computational Semantics, Association for Computational Linguistics, pages 130-141.
Seyffarth, E. 2018. Verb alternations and their impact on frame induction. In Proceedings of the NAACL Student Research Workshop, Association for Computational Linguistics, pages 17–24.