Abstract
Corpus studies from European languages have shown that some causal connectives are used preferentially to express subjective versus objective meanings (e.g., omdat vs. want in Dutch). However, there is not much empirical work on non-European languages on this cognitive perspective of causal connectives. In this study, we explored whether two Turkish causal connectives which belong to different lexical categories, namely, cunku (a conjunction) and icin (a postposition functioning as a complex subordinator) are sensitive to propositional attitudes and whether such sensitivity varies according to genre (academic vs. narrative). Consistent with previous findings from European language corpus studies, our logistic mixed regression models offered new insights into subjectivity in the distribution of Turkish causal connectives cunku and icin: there seems to be a division of labour between the two connectives, in that cunku has a preference for expressing subjective relations, whereas icin mainly expresses objective relations. An exception is that speech act relations (e.g., a question, advice, command, or promise) are mainly expressed by icin. All preferences hold over genres.1 Connectives are prototypical linguistic markers of coherence relations in discourse and can be grouped according to the type of relation they express, including: additive, temporal, causal or contrastive (e.g., Knott and Sanders 1998; Mann and Thompson 1988; Martin 1992; Pander Maat and Sanders 2006; Prasad, Webber, and Joshi 2014). In addition, different annotation schemes (e.g., Penn Discourse Treebank [PDTB], Rhetorical Structure Theory [RST], and Segmented Discourse Representation Theory [SDRT]) have been used to annotate discourse relations in corpora, enabling the investigation of connective distribution over various types of relations. Recently, a unified annotation scheme (UniDim account) was proposed, which incorporates elements from the above-described annotation schemes (Sanders et al. 2018). A common point of agreement in all these annotation schemes is that many languages have connectives that express causal relations in discourse (Diessel and Hetterle 2011). Various languages - especially Germanic languages - have subtle systematic distinctions within the same class of causal connectives. For instance, in order to express a sequence S1, because S2, language users often prefer one causal connective (e.g., want in Dutch) over another (e.g., omdat, in Dutch), although the connectives are within the same causal connective class (Sanders and Spooren 2013; 2015). In (1), the state of affairs The neighbors are not at home is rendered as a factual explanation, while in (2), the use of probably indicates that John probably won’t come to the meeting is not a fact, but the speaker’s judgment or conclusion, based on circumstances. In (1) the interpretation of a consequence statement is based on a fact (i.e., the statement is objective), while in (2) it is based on a personal assumption (i.e., the statement is subjective). To express the propositional attitude (i.e. objectivity/fact) in (1), omdat is preferred, whereas in (2) want is preferred to express the speaker’s conclusion. (1) The neighbors are not at home because they were at the office. (2) John probably won’t come to the meeting because he is ill. In this situation, several questions arise, including: (1) What is the mechanism behind the causal connective selection process? (2) Do propositional attitudes affect the selection of Turkish causal connectives, as well? (3) If Turkish causal connectives are sensitive to the type of propositional attitudes, does this sensitivity vary according to genre?
