Workshop: "Explicit and implicit coherence relations: Different, but how exactly?"
Sprach- und literaturwissenschaftlichen Fakultät
Venue: Humboldt-Universität zu Berlin, Unter den Linden 6, 10099 Berlin, Germany (Room: 2249a)
Date: January 17-18, 2020
Participation and Registration
The workshop is open to everyone. Anyone who is willing to attend (even though they do not have a paper to present) is welcome to drop by and participate in the workshop.
There is NO registration fee (and NO need to register).
Workshop topic and content:
The explicit-implicit divide of discourse relations is of crucial importance in discourse processing, from both a computational and psycholinguistic point of view. Often discourse relations with and without an explicit connective are considered to be equal. Consequently, removing a discourse connective from a given explicit relation (in which the connective originally occurred) is assumed to yield its implicit version (or, at least a relation that is similar or comparable to its ‘true’ implicit version).
Computational studies, aiming to automatically identify the presence and type of coherence relations, often postulate that classifiers trained on texts with naturally occurring connectives would perform satisfactorily for novel texts, even when the relations are not marked by connectives (Lapata & Lascarides, 2004; Marcu & Echihabi, 2002; Sporleder & Lascarides, 2005). However, the results of Sporleder and Lascarides (2008) suggest that these two relation types may differ. In an attempt to develop an automatic classifier for discourse relations, Sporleder and Lascarides observe that a classifier trained on marked relations performs well for identifying relations in the same data set after the connectives have been removed. However, it does not generalize very well to other unmarked data, in which relations occur naturally without connectives. They conclude that marked and unmarked relations might be linguistically too dissimilar, and that, consequently, removing connectives in the automatic labelling process might result in a change of meaning in the relations in question. This can be illustrated by the following examples.
(1) [Talks have broken off between Machinists representatives at Lockheed Corp. and the Calabasas, Calif., aerospace company.] [The union is continuing to work through its expired contract, however.]
(Source: RST Discourse Treebank (Carlson et al., 2002))
The two segments in the above text appear to be linked by an adversative relation (Concession, to be precise, as it is annotated in the corpus). The interpretation is largely influenced by the presence of the connective ‘however’ in the second segment. Now, consider the text without the connective:
(2) [Talks have broken off between Machinists representatives at Lockheed Corp. and the Calabasas, Calif., aerospace company.] [The union is continuing to work through its expired contract.]
The removal of the connective seems to undermine the adversative link between the segments that was previously postulated. The segments, instead, are now more likely to be linked by a causal or even an additive relation in the absence of a connective.
Psycholinguistic studies on relation marking also examine the role of discourse connectives in text comprehension by comparing the subjects’ performances in processing a relation in two circumstances: first, when a relation (naturally) occurs with a connective, and second, when that relation (or a similar one) occurs without it, as a result of removal of the connective from the relation (Britton et al., 1982; Cain & Nash, 2011; Haberlandt, 1982; Kamalski, 2007; Meyer, 1975; Millis & Just, 1994; Mulder, 2008; Sanders & Noordman, 2000). However, this approach is also based on the assumption that marked and unmarked discourse relations are sufficiently comparable.
This workshop intends to discuss whether explicit relations (relations with connectives) differ from implicit ones (relations without connectives), and it aims to explore the phenomenon of relation marking in greater detail. This includes the question of whether/how naturally occurring implicit relations are different from relations from which discourse connectives are deliberately removed. We hope to address these topics from diverse perspectives, and intend to incorporate contributions of researchers from multiple disciplines, such as discourse analysis, pragmatics, corpus linguistics and computational discourse. Specific interests related (but not restricted) to the broad theme of explicit and implicit relations would include:
- Signalling of discourse relations (both by discourse connectives and other relational markers (lexical or syntactic, etc.), and their correlations)
- Semantics and functions of discourse relations (how relations may differ with respect to varying linguistic, semantic or pragmatic parameters)
- Discourse parsing (identification of discourse relations and their arguments, both in the presence and absence of explicit relational signals)
- Psycholinguistic processing of text (both in the presence and absence of discourse connectives and other relational signals)
Invited speakers
Bonnie Webber, University of Edinburgh, UK
Gisela Redeker, University of Groningen, The Netherlands
Vera Demberg, Saarland University, Germany
Caroline Sporleder, University of Göttingen, Germany
Note: Unfortunately, Prof. Gisela Redeker is not being able to participate in the workshop because of her illness. Therefore, her talk had to be cancelled.
Submission of abstracts
We invite submission of extended abstracts on the above-mentioned topics, preferably involving many different languages. For more information, visit the page Call for Papers.
Workshop schedule
The workshop will be held for two days on January 17-18, 2020. The talks of the invited speakers will be accompanied by a selection of talks submitted in response to the Call for Papers. For more information, see Workshop Program.
Venue
Humboldt-Universität zu Berlin
Unter den Linden 6, 10099 Berlin, Germany
Room: 2249a
Organizer
(Wissenschaftlicher Mitarbeiter
Deptartment of English and American Studies
Humboldt-Universität zu Berlin)
Program committee
Maite Taboada, Simon Fraser University, Canada
Manfred Stede, University of Potsdam, Germany
Hannah Rohde, University of Edinburgh, UK
Amir Zeldes, Georgetown University, USA
David Schlangen, University of Potsdam, Germany
Markus Egg, Humboldt University of Berlin