A UID-based rational account of coherence relation expression in Mandarin Chinese

This corpus study presents a novel application of a rational account of language production, Uniform Information Density (UID), to the marking of coherence relations in Mandarin discourse, specifically through conjunctions. Illustrated in (1) and (2), from CCL corpus (2003), are cases of double conjunction marking (DM), where either one or even both two conjunctions can be omitted, resulting in a case of single conjunction marking (SM) or no marking (implicit). (1) 精神分裂症型患者的妄想虽然具体内容多种多样，但其基本模式都差不多。 Although the specific contents of delusions of schizophrenic patients are diverse, but their basic patterns are all similar. (2) 今年过年的时候，因为我也在美国，所以你来美国开演唱会的时候，我有去参加。 During the Chinese New Year this year, because I was also in the US, so when you came to the US to hold a concert, I attended. The UID principle (Levy and Jaeger, 2006) highlights that people structure language production (e.g., syntactic planning by including or excluding conjunctions) to maximize the uniformity of information density across utterances, thereby optimizing efficient communication. For a comprehensive view, we also take into account the cognitive bias (causal vs. contrastive) (Sanders, 2005), and potential medium effects (written vs. spoken). Unlike the few studies on coherence relations, we approach UID from two perspectives: (1) relational information, such as implicit causality verbs for causals (Asr and Demberg, 2012), antonyms or parallelism for contrasts (Crible and Demberg, 2020), and adverbs that express more directly the relational meaning, eg., jiu ‘then’, que ‘however’, hai ‘still’; (2) information density profile of the context, namely, the complexity of semantic, syntactic, and pragmatic (e.g., statement, judgment, augmentation) information encoded in the relation. Our hypotheses are : 1. UID, operationalized through the two perspectives above, is the overarching principle restricting conjunction presence in coherence expression. 2. Cognitive bias is a secondary factor: while UID is maintained, causal relations present a lesser degree of explicitness (e.g., less DM cases), and require fewer relational signals. 3. Medium effects are present: DM occurs more frequently in spoken discourse, due to the real-time processing pressure imposed by the dynamic nature of speech. A descriptive analysis of 1,200 explicit instances marked by yinwei/youyu ‘because’, suoyi ‘so’ and suiran/jinguan ‘although’, danshi ‘but’ reveals that DM usage is more frequent in spoken discourse and contrastive relations, partially supporting hypotheses 2&3. In-depth coding and analysis are in progress to thoroughly test these hypotheses. Preliminary observations indicate that lexical chains often contribute to the inference or interpretation of specific relations: those encoding conceptual associates in causal contexts while conceptual conflicts in contrastive context, such as “在美国in USA, 去参加 attended” in example (2), and “多种多样diverse, 都差不多all similar” in example (1). However, such lexical link appears more dominant in contrastive than causal contexts, implying causality being default assumption. Further validation of hypotheses and extensive discussions will be presented based on final results.