On the Use of Context for Predicting Citation Worthiness of Sentences in Scholarly Articles

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 397–415 June 6–11, 2021. ©2021 Association for Computational Linguistics 397 Text Generation from Discourse Representation Structures Jiangming Liu Shay B. Cohen Mirella Lapata Institute for Language, Cognition and Computation School of Informatics, University of Edinburgh jiangming.liu@ed.ac.uk, {scohen,mlap}@inf.ed.ac.uk Abstract We propose neural models to generate text from formal meaning representations based on Discourse Representation Structures (DRSs). DRSs are document-level representa- tions which encode rich semantic detail per- taining to rhetorical relations, presupposition, and co-reference within and across sentences. We formalize the task of neural DRS-to-text generation and provide modeling solutions for the problems of condition ordering and vari- able naming which render generation from DRSs non-trivial. Our generator relies on a novel sibling treeLSTM model which is able to accurately represent DRS structures and is more generally suited to trees with wide branches. We achieve competitive perfor- mance (59.48 BLEU) on the GMB benchmark against several strong baselines. 1 Introduction It is not uncommon for text generation systems to produce natural language output from interme- diate semantic representations (Yao et al., 2012; Takase et al., 2016). The literature presents sev- eral examples of generating text from logical forms underlying various grammar formalisms (Wang, 1980; Shieber et al., 1990; Carroll and Oepen, 2005; White et al., 2007), typed lambda calculus (Lu and Ng, 2011), Abstract Meaning Representations (AMR; Flanigan et al. 2016; Konstas et al. 2017; Song et al. 2018; Beck et al. 2018; Damonte and Cohen 2019; Ribeiro et al. 2019; Zhu et al. 2019; Cai and Lam 2020; Wang et al. 2020), Discourse Representation Theory (DRT; Basile and Bos 2011; Basile 2015), and Minimal Recursion Semantics (MRS; Horvat et al. 2015; Hajdik et al. 2019). In this work, we propose neural models to gen- erate high-quality text from semantic representa- tions based on Discourse Representation Structures (DRSs). DRSs are the basic meaning-carrying units in Discourse Representation Theory (DRT; Kamp b5 b6 : x1, b1 : e1, b1 : x2, b2 : t1 b1 b6 : Pred(x1, male.n.02) b1 : Pred(e1, play.v.03) b1 : Agent(e1, x1) b1 : Theme(e1, x2) b1 : Pred(x2, piano.n.01) b2 : Pred(t1, now.n.01) b1 : temp after(e1, t1) b4 b4 : ⋄ : b3 : x3, b3 : e2 b3 b3 : Named(x3, “tom”) b3 : Pred(e2, stop.v.05) b3 : Agent(e2, x3) b1 : Pred(x1, male.n.02) b3 : Patient(e2, x1) b3 : temp before(e2, e1) CONTRAST(b1, b4) The man is going to play the piano. Tom may stop him. Figure 1: DRS representing two-sentence discourse. 1981; Kamp and Reyle 1993; Asher and Lascarides 2003), a formal semantic theory designed to han- dle a variety of linguistic phenomena, including anaphora, presuppositions (Van der Sandt, 1992; Venhuizen et al., 2018), and temporal expressions within and across sentences. DRSs are scoped meaning representations, they capture the seman- tics of negation, modals, and quantification. Figure 1 displays in box format the meaning representation for a discourse consisting of two sentences. …

Content truncated to 500 words