Automatic Detection of Causal Relations in German Multilogs

Abstract

This paper introduces a linguisticallymotivated, rule-based annotation system for causal discourse relations in transcripts of spoken multilogs in German. The overall aim is an automatic means of determining the degree of justification provided by a speaker in the delivery of an argument in a multiparty discussion. The system comprises of two parts: A disambiguation module which differentiates causal connectors from their other senses, and a discourse relation annotation system which marks the spans of text that constitute the reason and the result/conclusion expressed by the causal relation. The system is evaluated against a gold standard of German transcribed spoken dialogue. The results show that our system performs reliably well with respect to both tasks.

Interactive Visual Analysis of Transcribed Multi-Party Discourse

“…As a consequence, the annotation system does not take into account relations that are split up between utterances of one speaker or utterances of different speakers. For causal relations (reason and conclusion spans), we show in Bögel et al (2014) that the system performs with an F-score of 0.95.…”

“…While a large amount of work is for English and based on landmark corpora such as the Penn Discourse Treebank (Prasad et al, 2008), the parsing of discourse relations in German has only lately received attention (Versley and Gastel, 2012;Stede and Neumann, 2014;Bögel et al, 2014).…”

Rhetorical strategies in German argumentative dialogs

“…1 We further use the transcripts of experimentally controlled discussions on whether or not to allow fracking in Germany (2,000 utterances, ∼282,000 tokens) and on establishing a hypothetical African government (3670 utterances, ∼363,000 tokens). 2 For the investigation we take the 20 most frequent particles (plus their combinations) in the three corpora (among them ja 'yes', halt 'stop', doch 'indeed', eben 'even', wohl 'probably' -for an overview see Table 4) and calculate their relative frequency in premises, conclusions, contrasts, concessions and conditions -units that constitute core building blocks of argumentative discourse in our type of data. These argumentative units are marked explicitly by the 38 discourse connectives given in Table 1 which are taken from the Potsdam Commentary Corpus [46], complemented with our own curated list of items.…”

“…These argumentative units are marked explicitly by the 38 discourse connectives given in Table 1 which are taken from the Potsdam Commentary Corpus [46], complemented with our own curated list of items. 3 To extract the information from the dialogs, we use the VisArgue pipeline, a parsing pipeline that reliably annotates the spans triggered by discourse connectives and several rhetorical devices [2,12,15, for more details on the annotation of argumentative units and the performance of the system see Section 5.1]. The relative frequencies of argumentative units that contain one or more particles are shown in Table 2.…”

“…We make crucial use of the VisArgue pipeline [2,12,15], a linguistically motivated, high-quality parsing pipeline that automatically identifies, disambiguates and annotates discourse information in transcribed natural speech, among them information coming from discourse particles.…”