Adding Semantic Relations to a Large-Coverage Connective Lexicon of German

Adding Semantic Relations to a Large-Coverage Connective Lexicon of German Tatjana Scheffler and Manfred Stede FSP Cognitive Science / Applied Computational Linguistics University of Potsdam Karl-Liebknecht-Str. 24-25, 14476 Potsdam, Germany tatjana.scheffler|manfred.stede@uni-potsdam.de Abstract DiMLex is a lexicon of German connectives that can be used for various language understanding purposes. We enhanced the coverage to 275 connectives, which we regard as covering all known German discourse connectives in current use. In this paper, we consider the task of adding the semantic relations that can be expressed by each connective. After discussing different approaches to retrieving semantic information, we settle on annotating each connective with senses from the new PDTB 3.0 sense hierarchy. We describe our new implementation in the extended DiMLex, which will be available for research purposes. Keywords: connectives, lexicon, semantics 1. Introduction Discourse connectives are closed-class lexical items that are known to provide very useful information for tasks like discourse parsing in the style of RST (e.g., (Hernault et al., 2010)) or PDTB (e.g., (Lin et al., 2014)), relation ex- traction (e.g., finding causal statements in biomedical text, (van der Horn et al., 2008)), or argumentation mining (e.g., (Peldszus and Stede, 2015)), because they indicate the kind of coherence relation holding between adjacent text spans (e.g., because: causal; although: concessive). While there can be a non-trivial ambiguity problem,1 connectives are generally regarded as highly predictive cues. Having access to a list of connectives for a language, along with informa- tion on their syntactic and semantic/pragmatic behavior, is thus an important asset. In addition to the “straightforward” use in a discourse parser, connective lists can moreover help to generate training data for identifying unsignalled (‘im- plicit’) coherence relations, as shown by (Ji et al., 2015). In this paper, we report on our recently completed exten- sion of a German connective lexicon, which has grown both in the number of entries and in the amount of information per connective. In particular, we focus here on the diffi- cult problem of selecting the coherence relations that are to be associated with lexical entries. First, Section 2. pro- vides background on the lexicon, then Section 3. presents our current solution. We discuss results of the procedure in Section 4., and Section 5. concludes. 2. A Lexicon of Discourse Connectives Our work extends DiMLex, the German Discourse Marker Lexicon (Stede and Umbach, 1998; Stede, 2002). We have now partially restructured the lexicon and added almost 100 new connectives. Our underlying definition of discourse connectives in German is based on (Pasch et al., 2003, p. 331): (1) Def.: A discourse connective is a lexical item x that exhibits each of the following five properties: 1Some words have non-/connective uses; some connectives can signal more than one relation; see (Stede, 2014). (M1) x cannot be inflected. (M2) x does not assign case features to its syntactic environment. (M3) The meaning of x is a two-place relation. (M4) The arguments of the relation (the meaning of x) are propositional structures. (M5) The expressions of the arguments of the …

Content truncated to 500 words