Towards Identifying Alternative-Lexicalization Signals of Discourse Relations

Proceedings of the 29th International Conference on Computational Linguistics, pages 837–850 October 12–17, 2022. 837 Towards Identifying Alternative-Lexicalization Signals of Discourse Relations René Knaebel and Manfred Stede Applied Computational Linguistics Department of Linguistics University of Potsdam Germany {rene.knaebel,manfred.stede}@uni-potsdam.de Abstract The task of shallow discourse parsing in the Penn Discourse Treebank (PDTB) framework has traditionally been restricted to identifying those relations that are signaled by a discourse connective ("explicit") and those that have no signal at all ("implicit"). The third type, the more flexible group of "AltLex" realizations has been neglected because of its small amount of occurrences in the PDTB2 corpus. Their number has grown significantly in the recent PDTB3, and in this paper, we present the first approaches for recognizing these "alternative lexicalizations". We compare the performance of a pattern-based approach and a sequence labeling model, add an experiment on the pre- classification of candidate sentences, and pro- vide an initial qualitative analysis of the error cases made by both models. 1 Introduction The view that discourse relations serve to model central aspects of the coherence of a text is widely accepted, and several approaches with different theoretical commitments have been developed (e.g. Mann and Thompson, 1988; Prasad et al., 2008a; Lascarides and Asher, 2007; Sanders et al., 1992). Our work is situated in the framework of Shallow Discourse Parsing and thus grounded in the Penn Discourse Treebank (PDTB) corpus (Prasad et al., 2008a). Here, some distinctions are commonly being made regarding the surface realization of discourse relations; most importantly: 1. A relation can be signalled by a connective, i.e., a lexical item from a closed class (con- junctions, certain adverbials). 2. A relation can be signalled by a different lex- ical form, which the PDTB calls Alternative lexicalization or AltLex for short. 3. A relation can also be stated without any lexi- cal signal; in this case it is called implicit. In the PDTB corpus, (1) and (3) are by far the most frequent cases, and accordingly, they have re- ceived much attention in shallow discourse parsing. As for (2), Lin et al. (2014) had developed a first, relatively simple approach; to our knowledge there have not been any follow-up proposals (including all the parsers presented in the CoNLL shared tasks in 2015 and 2016 (Xue et al., 2015, 2016) and in the recent DISRPT tasks (Zeldes et al., 2019, 2021)). With the introduction of the PDTB corpus ver- sion 3.0 (Prasad et al., 2018), amongst some other changes, the number of annotated AltLex instances has grown from 624 (in version 2.0) to 1632. With the corpus now being considerably richer in Alt- Lex signals, we believe that their role in shallow discourse parsing now needs to be strengthened. Besides, also for advancing the theoretical descrip- tion of the AltLex category and its role in coher- ence marking, it is important to perform empirical studies. Essentially, alternatively–lexicalized discourse relations are signalled by an open set of phrases that verbalizes the connection between two discourse arguments. In the PDTB, …

Content truncated to 500 words