DeepCx: A transition-based approach for shallow semantic parsing with complex constructional triggers

Abstract

This paper introduces the SURFACE CONSTRUCTION LABELING (SCL) task, which expands the coverage of Shallow Semantic Parsing (SSP) to include frames triggered by complex constructions. We present DeepCx, a neural, transition-based system for SCL. As a test case for the approach, we apply DeepCx to the task of tagging causal language in English, which relies on a wider variety of constructions than are typically addressed in SSP. We report substantial improvements over previous tagging efforts on a causal language dataset. We also propose ways DeepCx could be extended to still more difficult constructions and to other semantic domains once appropriate datasets become available.

Guided Generation of Cause and Effect

“…Of various efforts in guided generation [Ammanabrolu et al, 2019;Tang et al, 2019;Clark et al, 2018;Hu et al, 2019b], Sentential Causal Resource # CE Pairs TCR [Ning et al, 2018] 172 SemEval-2007Task4 [Girju et al, 2007 220 Causal-TimeBank [Mirza et al, 2014] 318 CaTeRS [Mostafazadeh et al, 2016] 488 EventCausalityData [Do et al, 2011] 580 RED [O'Gorman et al, 2016] 1,147 SemEval2010 Task8 [Hendrickx et al, 2009] 1,331 BECauSE 2.0 [Dunietz et al, 2017b] 1,803 EventStoryLine [Caselli and Vossen, 2017] 5,519 PDTB 2.0 [Prasad et al, 2008] 8,042 Altlex [Hidey and McKeown, 2016] 9,190 PDTB 3.0 [Webber et al, 2019] 13 K DisSent [Nie et al, 2019] 167 K CausalBank (Ours) 314 M…”

“…[Chambers and Jurafsky, 2008;Riaz and Girju, 2010;Do et al, 2011;Luo et al, 2016]. Others have mined patterns via discourse patterns in the form of 'A led to B', 'if A then B', etc., e.g., [Khoo et al, 2000;Girju, 2003;Zhao et al, 2017]). See Asghar [2016] for review.…”

“…We do not rely on a supervised text extractor to pick out specific sub-spans of a sentence that represent a cause-effect pairing between propositions. 2 We instead curate a series of patterns from previous studies [Mirza et al, 2014;Luo et al, 2016;Girju, 2003]. These patterns can be classified into two categories, according to how they are mostly used in language to convey a causal relation: 1.…”

“…Already mentioned are recent crowdsourcing efforts [Rashkin et al, 2018;Sap et al, 2019]. More related are PDTB [Prasad et al, 2008] and BECauSE [Dunietz et al, 2017b], but where our resource goal is a much larger corpus, for the purpose of training a neural text generation model. Most related would be the extractive approach of DisSent [Nie et al, 2019], but where we focus specifically on causality, and derive a much larger corpus.…”

“…Force dynamics theory [Talmy, 1988] from cognitive psychology posits three primary kinds of causal semantics [Wolff, 2007] -CAUSE, ENABLE and PREVENT -which were lexicalized as causal verbs [Wolff and Song, 2003]. The annotation scheme of Dunietz et al [2017b] distinguishes three types of causal semantics: CONSEQUENCE, MOTI-VATION, and PURPOSE. In PDTB 2.0 [Prasad et al, 2008], "CONTINGENCY" has two subtypes ("Cause" and "Condition").…”