A Data-Driven Semi-Automatic Framenet Development Methodology

Proceedings of Recent Advances in Natural Language Processing, pages 1471–1479 Sep 1–3, 2021. https://doi.org/10.26615/978-954-452-072-4_165 1471 A Data-Driven Semi-Automatic Framenet Development Methodology Shafqat Mumtaz Virk Språkbanken Text, Dept. of Swedish University of Gothenburg 405 30 Gothenburg, Sweden virk.shafqat@gmail.com Dana Dannélls Språkbanken Text, Dept. of Swedish University of Gothenburg 405 30 Gothenburg, Sweden dana.dannells@svenska.gu.se Lars Borin Språkbanken Text, Dept. of Swedish University of Gothenburg 405 30 Gothenburg, Sweden lars.borin@svenska.gu.se Markus Forsberg Språkbanken Text, Dept. of Swedish University of Gothenburg 405 30 Gothenburg, Sweden markus.forsberg@svenska.gu.se Abstract FrameNet is a lexical semantic resource based on the linguistic theory of frame seman- tics. A number of framenet development strategies have been reported previously and all of them involve exploration of corpora and a fair amount of manual work. De- spite previous efforts, there does not exist a well-thought-out automatic/semi-automatic methodology for frame construction. In this paper we propose a data-driven methodology for identification and semi-automatic construc- tion of frames. As a proof of concept, we re- port on our initial attempts to build a wider- scale framenet for the legal domain (LawFN) using the proposed methodology. The con- structed frames are stored in a lexical database and together with the annotated example sen- tences they have been made available through a web interface. 1 Introduction Frame semantics is a theory of meaning in natural languages proposed by Charles Fillmore (Fillmore, 1977, 1982). The theory stipulates that meanings of words can be best understood with reference to the situations they invoke in the minds of the speakers. The concrete manifestation of frame semantics is the computational lexical resource FrameNet, first constructed within the English Berkeley FrameNet (BNF) project (Baker et al., 1998). The resource has inspired work on framenets for many other lan- guages and domains, including Chinese (You et al., 2007), French (Candito et al., 2014), German (Bur- chardt et al., 2009), Hebrew (Hayoun and Elhadad, 2016), Korean (Kim et al., 2016), Italian (Lenci et al., 2010), Japanese (Saito et al., 2008), Por- tuguese (Torrent, 2013), Spanish (Subirats, 2009), and Swedish (Borin et al., 2010a; Dannélls et al., Forthc.a). FrameNet is built around the notion of frames. A frame in this context is a schematic represen- tations of events, objects, situations, institutions, etc. The primary components of a frame are the frame definition along with the frame elements and the lexical units documented with corpus evidence (Ruppenhofer et al., 2016). There are at least three major framenet devel- opment strategies that have been reported in the literature (Candito et al., 2014) namely: (1) lexico- graphic frame-by-frame (2) corpus driven lemma- by-lemma (3) the full-text strategy. Of these three, the most commonly applied strategy is frame-by- frame, where the frame is defined along with its frame-elements first. Example sentences are then chosen from a corpus for a disambiguated anno- tation of lexical units and frame-elements. In the lemma-by-lemma strategy, a set of lemmas (lexical units) are chosen first and then all their occurrences are annotated in a given corpus together with the annotation of frame-elements. In …

Content truncated to 500 words