Dokument: Computing the semantics of verb alternations: A constructional approach
| Titel: | Computing the semantics of verb alternations: A constructional approach | |||||||
| Weiterer Titel: | Computing the semantics of verb alternations: A constructional approach | |||||||
| URL für Lesezeichen: | https://docserv.uni-duesseldorf.de/servlets/DocumentServlet?id=69643 | |||||||
| URN (NBN): | urn:nbn:de:hbz:061-20250520-110141-0 | |||||||
| Kollektion: | Dissertationen | |||||||
| Sprache: | Englisch | |||||||
| Dokumententyp: | Wissenschaftliche Abschlussarbeiten » Dissertation | |||||||
| Medientyp: | Text | |||||||
| Autor: | Seyffarth, Esther [Autor] | |||||||
| Dateien: |
| |||||||
| Beitragender: | Prof. Dr. Kallmeyer, Laura [Gutachter] | |||||||
| Stichwörter: | verb alternations, semantics, syntax-semantics interface, construction grammar, tree-adjoining grammar, alternation identification, frame semantics | |||||||
| Dewey Dezimal-Klassifikation: | 400 Sprache » 410 Linguistik | |||||||
| Beschreibungen: | This thesis explores the syntactic and semantic properties of verbs, their arguments, and the argument structure constructions they appear in, and investigates in what ways these properties matter for a successful computational treatment of verb alternations. Verbs that participate in a verb alternation are able to instantiate an alternationspecific set of constructions in which certain syntactic argument positions are linked to certain semantic roles (Levin 1993).
A construction is defined by the combination of its syntactic form and semantic meaning (Goldberg 1995: 1). As the syntactic form of a construction is not necessarily unique, ambiguity is possible: constructions can share a syntactic form and differ in their meaning. This can pose issues for computational systems when they encounter sentences involving alternating verbs. A reliable distinction of constructions is a prerequisite for a successful interpretation of such sentences. This thesis approaches these issues from two angles. First, it presents a metagrammar for a lexicalized tree adjoining grammar (Joshi & Schabes 1997) with semantic representations in the form of typed, recursive frames (Kallmeyer & Osswald 2013). The metagrammar models the syntax and semantics of a selection of English verb alternations, as well as a number of argument structure constructions whose syntactic form is identical to that of constructions in the alternations. Second, the thesis presents a set of classifiers to determine whether a given verb participates in a specific alternation or not, based on corpus attestations of the verb in question and of known alternating and non-alternating verbs. The classification features are designed to approximate the overlap in each verb's selectional preferences on arguments in the relevant syntactic positions in different constructions. Based on the findings reported in this thesis, an optimal approach to a computational treatment of alternating verbs would be one that combines the advantages of the different approaches presented here. Handcrafted models like the metagrammar implemented in this thesis do not scale well, but are promising with respect to a reliable handling of ambiguous sentences. Instead of classifying verbs based on features approximating selectional restriction overlap, it could be beneficial to predict exactly the properties that are modeled in the metagrammar. This would involve learning a type hierarchy of a suitable granularity, including incompatibility and subtype relationships, as well as learning type requirements imposed on specific arguments both by argument structure constructions and by verbs. The implementation of a system that learns these properties and the corresponding type hierarchy is a promising direction for future work.This thesis explores the syntactic and semantic properties of verbs, their arguments, and the argument structure constructions they appear in, and investigates in what ways these properties matter for a successful computational treatment of verb alternations. Verbs that participate in a verb alternation are able to instantiate an alternationspecific set of constructions in which certain syntactic argument positions are linked to certain semantic roles (Levin 1993). A construction is defined by the combination of its syntactic form and semantic meaning (Goldberg 1995: 1). As the syntactic form of a construction is not necessarily unique, ambiguity is possible: constructions can share a syntactic form and differ in their meaning. This can pose issues for computational systems when they encounter sentences involving alternating verbs. A reliable distinction of constructions is a prerequisite for a successful interpretation of such sentences. This thesis approaches these issues from two angles. First, it presents a metagrammar for a lexicalized tree adjoining grammar (Joshi & Schabes 1997) with semantic representations in the form of typed, recursive frames (Kallmeyer & Osswald 2013). The metagrammar models the syntax and semantics of a selection of English verb alternations, as well as a number of argument structure constructions whose syntactic form is identical to that of constructions in the alternations. Second, the thesis presents a set of classifiers to determine whether a given verb participates in a specific alternation or not, based on corpus attestations of the verb in question and of known alternating and non-alternating verbs. The classification features are designed to approximate the overlap in each verb's selectional preferences on arguments in the relevant syntactic positions in different constructions. Based on the findings reported in this thesis, an optimal approach to a computational treatment of alternating verbs would be one that combines the advantages of the different approaches presented here. Handcrafted models like the metagrammar implemented in this thesis do not scale well, but are promising with respect to a reliable handling of ambiguous sentences. Instead of classifying verbs based on features approximating selectional restriction overlap, it could be beneficial to predict exactly the properties that are modeled in the metagrammar. This would involve learning a type hierarchy of a suitable granularity, including incompatibility and subtype relationships, as well as learning type requirements imposed on specific arguments both by argument structure constructions and by verbs. The implementation of a system that learns these properties and the corresponding type hierarchy is a promising direction for future work. | |||||||
| Lizenz: |  Dieses Werk ist lizenziert unter einer Creative Commons Namensnennung 4.0 International Lizenz | |||||||
| Fachbereich / Einrichtung: | Philosophische Fakultät » Institut für Sprache und Information » Computerlinguistik | |||||||
| Dokument erstellt am: | 20.05.2025 | |||||||
| Dateien geändert am: | 20.05.2025 | |||||||
| Promotionsantrag am: | 03.05.2023 | |||||||
| Datum der Promotion: | 08.08.2023 |
