Contextualizing the variation in causal clause ordering in Mandarin Chinese: a multifactorial analysis

1 Introduction

Language variation is ubiquitous and systematic in language use, conditioned by both linguistic factors (e.g., animacy, length of constituents) and social factors (e.g., age, gender). Since the 1960s, language variation has been widely studied across the English-speaking world (e.g., Labov 1963), forming the basis of the variationist paradigm, which has been later extended to research on variation in Spanish (Torres Cacoullos 1999), Portuguese (Bortoni-Ricardo 1985) and French (Sankoff and Cedergren 1971). These languages share typological and genealogical similarities, as they all belong to western branches of Indo-European, especially Germanic and Romance (Adli and Guy 2022). Non-Indo-European languages are investigated only infrequently, such as Chinese, whose grammatical system heavily relies on word order, which displays typological differences with Germanic and Romance languages in many linguistic aspects (Li and Thompson 1981; Wang 1984). As Adli and Guy (2022) pointed out, the generalizability of the variationist paradigm and its theoretical underpinnings should be tested with languages beyond Germanic and Romance (see also Stanford 2016). Another gap in the variationist paradigm is the long-standing focus on phonetics since its inception (cf. Gries 2001; Poplack and Tagliamonte 1999); comparatively fewer studies have examined variation beyond sound structures in languages other than English (although see recent variationist studies of Chinese morphosyntax by Li et al. 2023; Liao et al. 2024; Xu et al. 2024). To further address these gaps, our focus is on one morphosyntactic variable in spoken Mandarin Chinese, causal clause ordering, whose variation and probabilistic constraints have not been systematically examined to date.

Among the frequent types of adverbial clauses in Mandarin Chinese, including temporal, conditional, concessive, and causal clauses, the connectives for the first three types tend to precede the clause material they modify (Wang 1999). In contrast, causal clauses differ notably from other adverbial types in exhibiting considerable variability in the position of their connectives, which can appear either at the beginning or within the sentence (Biq 1995; Tsai 1996; Xing 2001). This positional variability is not unique to Chinese: a typological study of 60 languages, such as Chinese, English, German and Japanese, found that “causal adverbial clauses are structurally more independent of the associated main clause than other semantic types of adverbial clauses” (Diessel and Hetterle 2011: 24), resulting in greater cross-linguistic flexibility in the placement of reason connectives. In Mandarin specifically, causal clauses introduced by reason connectives (e.g., yīnwèi ‘because’, jìrán ‘since’, yóuyú ‘due to’, jiànyú ‘in view of’) may occur in either sentence-initial or sentence-medial positions (Chao 1968; Li and Thompson 1981: 635), as illustrated in Examples (1) and (2), respectively.

Studies from both generative grammar and functional grammar perspectives have not identified truth-conditional differences in the meanings of different causal clause orderings in Mandarin Chinese (e.g., Gasde and Paul 1996; Pan and Zhu 2023; Thompson et al. 2007). Instead, some studies have found that sentence-initial and sentence-medial causal clauses are pragmatically different, making causal clause ordering an interesting linguistic variable to test the semantic-pragmatic contestability (Song and Tao 2009). Furthermore, some linguistic factors have been found to constrain the ordering of causal clauses, such as the type of connective and the presence of the result marker suǒyǐ (‘so’) (Biq 1995; Kirkpatrick 1993; Li et al. 2016). However, existing studies only conducted monofactorial analysis, focusing on one or two linguistic factors at once and ignoring their interactions. Thus, no study has examined the effects of these co-occurring contextual constraints and their interactions on causal clause ordering in Mandarin Chinese.

To bridge these gaps, we adopt a variationist approach. First, variationist studies focus on how language is used by speakers in real-world contexts. Our dataset, consisting of spoken Chinese utterances in social settings, aligns well with the empirical requirements of this paradigm. Second, we aim to understand the variable processes involved in causal clause ordering. A central task of the variationist approach is to define the variable envelope of a linguistic feature, identifying “alternative ways of saying the same thing”, within which meaningful comparisons can be made (Labov 1994: 550). In our study, variation in the ordering of causal clauses represents different ways of expressing causal relationships. Finally, this study employs quantitative reasoning to analyze the probabilistic structure of causal clause ordering. The variationist framework is particularly suited for this purpose because it considers the co-occurrence and distribution of multiple probabilistic constraints to probe the “orderly heterogeneity” in language use (Weinreich et al. 1968: 100).

2 Causal clause ordering in Mandarin Chinese

In expressing causal relations in Chinese, the users have the choices of using or not using connectives (Biq 1995: 49–50). When they do choose to express causal relations explicitly using connectives, there are at least five connectives that are frequently used, including yīnwèi (‘because’), yīn (‘cause’), jìrán (‘since’), yóuyú (‘due to’), and jiànyú (‘in view of’). Previous studies have argued that these reason connectives are substantially similar in their semantics, but some semantic nuances exist among them. Specifically, yīnwèi, yīn, yóuyú, and jiànyú are more similar to each other in terms of semantics compared to jìrán; however, this difference between the two categories is not truth-conditional, as all of them can indicate causal relations (Li et al. 2016; Xing 2003: 199). Thus, the locus of the variation in expressing causal relations through these connectives lies in the ordering of causal clauses that involve them. That is, these connectives may precede as well as follow the clause to which they are explicitly linked, signalling either forward or backward links (e.g., 我没办法去上学 (wǒ méi bànfǎ qù shàngxué; ‘I can’t go to school’) [result], 因为我肚子疼 (yīnwèi wǒ dùzi téng; ‘because my stomach hurts’) [reason] versus 因为我肚子疼 [reason], 我没办法去上学 [result]). In contrast to English where the preferred clause sequence in complex sentences is main to subordinate (Quirk et al. 1985), in Chinese, the modifier-modified principle of sequencing is widely accepted (Chao 1968; Tai 1975), indicating that “the main component comes at the end, and the subordinate component comes at the beginning” (Wang 1984: 96). This principle also applies to the causal clauses where reason connectives, such as yīnwèi, “exclusively signal forward links” (Kirkpatrick 1993: 426). Thus, forward linking has been accepted as the canonical ordering of causal clauses in Chinese (Chao 1968).

However, some other studies have found that forward linking is not necessarily the preferred order for expressing the causal relation in Chinese. For example, Biq (1995) examined the ordering of clauses involving yīnwèi that express an ideationally determined causal relation. The results revealed that in conversation data, only 18 % (n = 12/67) of the tokens demonstrated forward linking, placing yīnwèi in the sentence-initial position, while the majority (82 %, n = 55/67) showed backward linking, placing yīnwèi in the sentence-medial position (Biq 1995: 53). In newspaper data, the rate of sentence-initial causal clauses is also lower compared to that of sentence-medial ones, but the distributional difference between these two variants is smaller (45 % vs. 55 %). Thus, in both spoken and written contexts, sentence-initial causal clauses are less favored, a finding that contrasts with the results of earlier studies (e.g., Chao 1968; Kirkpatrick 1993; Tai 1975; Wang 1984). In some follow up studies along the similar line, sentence-initial causal clauses have been consistently found to be less frequent than sentence-medial ones in Chinese (e.g., Song and Tao 2009; Wang 1999; Xiao et al. 2021). Thus, these studies suggest the existence of variation in causal clause ordering, but its patterning is not consistent across studies.

Nevertheless, examining the frequency differences between sentence-initial and sentence-medial causal clauses does not explain the mechanism of variation in causal clause ordering. To address this gap, some studies have tried to explain such variation by looking at the pragmatic functions of sentence-initial and sentence-medial causal clauses. For example, Song and Tao (2009: 95) found that sentence-initial causal clauses are more likely to function as an information-sharing device to provide background information for the main argument, while sentence-medial causal clauses tend to function as an interactional device to “mitigate dispreferred responses and address anticipated recipient doubts, surprises, or confusion”. Therefore, the pragmatic function of causal clauses has often been the reason used to explain the ordering of their connectives (Wang 1999, 2002).

Focusing on the frequency distribution of sentence-initial and sentence-medial causal clauses, along with their semantic and pragmatic connotations, risks oversimplifying these variants by categorizing them into rigid groups. This approach overlooks the variable processes underlying causal clause ordering and the nuanced contextual factors that shape the semantic, syntactic, and pragmatic relationships between causal clauses and their associated main clauses. Therefore, we shift our focus from merely analyzing frequency and function to adopting a variationist approach, which considers the influence of broader contextual factors on these two variants.

For contextual factors, it has been found that two factors – the type of connective and the presence of a paired result marker suǒyǐ – play important roles in influencing the ordering of causal clauses in Chinese. Specifically, anecdotal evidence suggests that yīnwèi and yóuyú are typically used to mark descriptive causality based on facts, while jìrán often conveys inferential causality based on reasoning (Xing 2001). A corpus-based analysis of three text categories – news reports, novels, and opinion pieces – further reveals that jìrán and sentence-medial yīnwèi consistently exhibit subjective profiles across genres. In contrast, yóuyú and sentence-initial yīnwèi primarily express objective relations (Li et al. 2016). These findings suggest that the type of connective influences both the semantic meaning and the pragmatic functions of causal clauses. Additionally, the ordering of causal clauses may signal different types of causal relationships (e.g., Xiao et al. 2021). For the result marker suǒyǐ, it has been found that its presence generally promotes the use of sentence-initial causal clauses (Biq 1995). When suǒyǐ appears in the main clause, it tends to form a summary statement alongside reason connectives (e.g., yīnwèi, yóuyú) (Kirkpatrick 1993), which introduce background information in the subordinate clause, thus motivating the sentence-initial placement of these reason connectives. Although less frequent, suǒyǐ in the main clause can also appear in the sentence-initial position, followed by reason connectives in the subordinate clauses. This syntactic structure is often used to facilitate the progression of the main course of action, transitioning from a suspended state to the next action (Wang 2020: 328).

Another factor examined to explain the variation in causal clauses is the emotional valence of the main and subordinate clauses, although its impact on the ordering of causal clauses has yet to be explored (Yao 2007). Emotional valence here refers to “the speaker or writer’s attitude or stance towards, viewpoint or feelings about the entities and propositions that he or she is talking about” (Hunston and Thompson 2000: 5).^[2] Adopting a three-way categorization of emotional valence in English (e.g., Partington 2004), Yao (2007, p. 53) annotated the attitudinal meanings of causal clauses with suǒyǐ and jiěguǒ in Chinese as positive (e.g., 所以深受市民支持 (suǒyǐ shēn shòu shìmín zhīchí; ‘so it is widely supported by the public’)), negative (e.g., 结果还是半途而废, 不了了之 (jiéguǒ háishì bàntú érfèi, bùliǎoliǎo zhī; ‘so it was left unfinished and ended in failure’)), and neutral (e.g., 所以史书又叫汗青 (suǒyǐ shǐshū yòu jiào hànqīng; ‘so historical records are also called ‘汗青’ (an ancient term for historical records in Chinese)’)). The results revealed that 74 % of suǒyǐ clauses exhibited a neutral valence, while 14 % and 12 % displayed negative and positive valence, respectively. In contrast, 73 % of jiěguǒ clauses indicated a negative valence, while 8 % and 19 % showed neutral and positive valence, respectively. These patterns suggest that emotional valence is an important factor distinguishing between different types of causal clauses. This raises a new question, which we will address here: Does the emotional valence of clauses influence the ordering of causal clauses?

Additionally, we introduce two new factors that have not yet been examined in studies of causal clause ordering in Chinese: the length and complexity of main and subordinate clauses. There are two key reasons for considering these factors. First, length and complexity have been extensively studied in English causal clauses and found to strongly influence their ordering (Diessel 2005). Second, given the substantial differences between Chinese and English syntax (Li and Thompson 1981; Wang 1984), it is likely that the effects of length and complexity on clause ordering in Chinese may follow distinct patterns. Thus, examining these factors in Chinese provides an opportunity to investigate potential cross-linguistic, typological universals in how syntactic features influence causal clause ordering.

For the length of main and subordinate clauses in English, it has often been measured by the number of words in each clause (including the subordinator, e.g., because). For example, Diessel (2005) looked at the ordering of multiple English adverbial clauses, including causal clauses, in three sources: conversational discourse, fiction, and scientific writing. Diessel found that when the adverbial clause is in sentence-initial position, it tends to be shorter than the main clause; conversely, in cases where the adverbial clause is in sentence-medial position, it tends to be longer than the main clause (see similar findings in Gries and Wulff 2021 regarding L1 English speakers). Diessel interprets this distribution of lengths as evidence for a general adherence to the short-before-long principle in English (Behaghel 1909), which reflects speakers’ general attempts to minimize cognitive effort in utterance planning, according to parsing efficiency theories (Hawkins 1994, 2004).

For syntactic complexity, it focuses on the embeddedness of each clause which indicates whether the examined main or subordinate clause contains any embedded clauses (Diessel 2008), which has been widely used as a measure of complexity in studies of English. In English, syntactically complex constituents are more likely to appear towards the end of a sentence, as this allows the parser to quickly identify the high-level structure, thereby facilitating the analysis of subsequent constituents (Hawkins 1994, 2004; Wasow 1997). Given that causal clauses can vary in complexity, it is reasonable to assume that sentence-initial causal clauses are structurally less complex than sentence-medial ones. A corpus-based analysis of English “because” sentences revealed that subordinate causal clauses tend to be more complex with a greater degree of embeddedness, making them more likely to occur in sentence-medial positions (Xu and Kang 2022). These findings suggest that syntactic complexity influences the positioning of causal clauses, with more complex structures favoring sentence-medial positions, highlighting an intersection of clause complexity and causal clause ordering in English. Even though syntactic length and complexity have strong effects on the ordering of causal clauses in English, no studies have yet examined how these factors influence the positioning of causal clauses in Chinese, leaving a gap in our understanding of cross-linguistic patterns of clause ordering.

Furthermore, previous research on causal clause ordering in Chinese has shown that genre can influence the distribution of sentence-initial and sentence-medial variants. For example, Biq (1995) compared conversations and newspaper texts and found that sentence-medial causal clauses were more frequent in both genres, but occurred at a much higher rate in conversations than in newspapers (82 % vs. 55 %). Similarly, Song and Tao (2009), using the written Lancaster Corpus of Mandarin Chinese and the spoken CallFriend telephone conversation corpus, observed the same trend: sentence-medial causal clauses were more frequent in both genres, with a higher proportion in spoken than in written data (83 % vs. 68 %). These findings suggest that genre is a relevant factor in shaping causal clause positioning. These earlier studies were not grounded in a variationist framework and did not consider the role of contextual or probabilistic constraints, such as clausal complexity or emotional valence. Instead, they focused primarily on the functions of the two variants. This leaves open the question of how genre might interact with other probabilistic factors to condition clause ordering. Variationist studies in English morphosyntax, such as those on future temporal reference (will vs. be going to) (Engel and Szmrecsanyi 2023) and the dative alternation (e.g., Tim gives Amy a book vs. Tim gives a book to Amy) (Gan 2024), have often shown that genre interacts with broader probabilistic constraints. In this study, we test whether genre not only affects the overall distribution of sentence-initial and sentence-medial causal clauses but also interacts with other constraints identified in the literature above.

To summarize, variation in the ordering of causal clauses in Chinese is widely observed, with reason connectives such as yīnwèi and yóuyú either preceding or following the associated main clause, resulting in sentence-initial and sentence-medial positioning. This variation is not random but is systematically influenced by genre and contextual factors at syntactic, semantic, and discourse-pragmatic levels. However, no research has yet considered all these factors within a single multifactorial model, which would allow for concurrent measurement of the effect sizes and interactions among the different factors. In this study, we employ a large spoken corpus of Mandarin Chinese and adopt a variationist approach to examine the variation in causal clause ordering, focusing on the probabilistic influence of genre and multiple contextual constraints.

3 Methodology

4 Results

4.2 Summary of effects

To facilitate interpretation of the results, we summarize the main and interaction effects from the fitted model in Table 1. Statistically significant effects are highlighted in grey. It is important to note that predictors such as MainClauseEmotionalValence, MainClauseLength, PresenceSuoyi, and ConnectiveType may appear to exhibit main effects but are, in fact, involved in interactions with other predictors. As such, they are not discussed here (see Gries 2021: 291 for details on this statistical decision). The predicted levels for all categorical variables in the model are shown in brackets, while the reference levels are as follows: Dialogue for Genre, Neutral for both MainClauseEmotionalValence and SubClauseEmotionalValence, Complex for MainClauseComplexity, Absence for PresenceSuoyi, and Yinwei for ConnectiveType.

Table 1:

Output of logistic mixed-effects model predicting Medial. OR, odds ratio; SE, standard error.

	OR	SE	z	p-Value
(Intercept)	2.66	0.16	5.96	<0.001
Genre (Monologue)	0.64	0.23	-1.94	0.053
MainClauseEmotionalValence (Negative)	0.58	0.19	-2.86	<0.01
MainClauseEmotionalValence (Positive)	0.83	0.20	-0.93	0.353
MainClauseComplexity (Simple)	0.8	0.16	-1.40	0.162
MainClauseLength	0.35	0.20	-5.18	<0.001
PresenceSuoyi (Presence)	0.1	0.19	-11.73	<0.001
ConnectiveType (Non-yinwei)	0.03	0.61	-5.91	<0.001
SubClauseLength	1.84	0.14	4.39	<0.001
SubClauseEmotionalValence (Negative)	1.1	0.16	0.62	0.537
SubClauseEmotionalValence (Positive)	1.52	0.19	2.24	<0.05
Genre (Monologue): MainClauseEmotionalValence (Negative)	2.64	0.32	3.04	<0.01
Genre (Monologue): MainClauseEmotionalValence (Positive)	1.77	0.33	1.71	0.087
MainClauseComplexity (Simple): MainClauseLength	5.8	0.40	4.40	<0.001
PresenceSuoyi (Presence): ConnectiveType (Non-yinwei)	19.88	1.25	2.39	<0.05

Our analysis focuses on both the directions and magnitudes of these factors. In Table 1, positive z-values for each predictor indicate a preference for sentence-medial causal clauses, given that the predicted variant is sentence-medial causal clauses, while negative z-values suggest a preference for sentence-initial causal clauses. To aid interpretation, we also report odds ratios for each predictor to convey effect sizes. An odds ratio greater than 1 indicates a higher likelihood of sentence-medial causal clauses, whereas a value less than 1 signifies a greater likelihood of sentence-initial causal clauses. For example, the predictor SubClauseLength has an odds ratio of 1.84, indicating that for every one-character increase in subordinate clause length, the odds of a causal clause being sentence-medial (as opposed to sentence-initial) increase by 84 %, holding all other predictors constant.

To further interpret the directions and effect sizes of significant predictors, all effects are visualized in Figures 1–4. Partial effects plots were utilized, offering insights into how individual predictors influence the outcome while holding other predictors constant or at their average levels. The x-axes display the predictor names, while the y-axes, ranging from 0 to 1 in each figure, represent the predicted probability of sentence-medial causal clauses as estimated by the model. Higher values on the y-axis indicate a greater probability of sentence-medial causal clauses compared to sentence-initial causal clauses.

Figure 1:

Plots of the main effects of subordinate clause length and emotional valence in the model.

Figure 2:

Plot of the interaction effect between genre and main clause emotional valence in the model.

Figure 3:

Plot of the interaction effect between main clause length and complexity in the model.

Figure 4:

Plot of the interaction effect between the presence of suǒyǐ and connective type in the model.

As shown in Figure 1, the main effect of SubClauseLength, displayed on the left panel, reveals that as the length of the subordinate clause increases, the probability of sentence-medial causal clauses also increases. This suggests that longer subordinate clauses tend to favor sentence-medial causal clauses compared to shorter ones, as illustrated in Example 6 above. In the right panel, the main effect of SubClauseEmotionalValence shows that the probability of sentence-medial causal clauses is highest when the emotional valence is positive, compared to when it is neutral or negative. Therefore, when the emotional valence of subordinate clauses is positive, sentence-medial causal clauses are most favored over sentence-initial ones, as demonstrated in Example 8 above.

Figure 2 illustrates the interaction effect between MainClauseEmotionalValence and Genre. Three types of emotional valence – neutral, negative, and positive – are represented by blue, red, and green lines, respectively. In dialogues, neutral valence of main clauses is the most favorable for sentence-medial causal clauses, whereas negative valence is the least favorable, with a significant difference between them, as confirmed by post-hoc comparisons using emmeans in R (estimate = 0.54, p = 0.04). Positive valence falls between neutral and negative valence in terms of favorability for sentence-medial clauses. In monologues, both positive and negative valence tend to favor sentence-medial causal clauses when compared to neutral valence; however, these differences do not reach statistical significance (ps > 0.05).

When comparing the effect of emotional valence between dialogues and monologues, the results show that with neutral valence, the probability of sentence-medial causal clauses is higher in dialogues than in monologues (estimate = 0.44, p = 0.38). In contrast, with negative valence, the probability of sentence-medial causal clauses is greater in monologues than in dialogues (estimate = −0.53, p = 0.25). Similarly, with positive valence, the probability of sentence-medial causal clauses is higher in monologues than in dialogues (estimate = −0.13, p = 1.00). It should be noted that none of these differences reach statistical significance.

Figure 3 illustrates the interaction effect between main clause length and main clause complexity. The red line shows that when main clauses are syntactically simple, an increase in main clause length corresponds to a higher probability of sentence-medial causal clauses. This indicates that longer but simple main clauses favor sentence-medial over sentence-initial causal clauses. In contrast, the blue line shows that when main clauses are syntactically complex, an increase in main clause length corresponds to a lower probability of sentence-medial causal clauses. This suggests that shorter but complex main clauses favor sentence-medial over sentence-initial causal clauses.

Figure 4 illustrates the interaction effect between the presence of suǒyǐ in the main clause and the type of connective. When suǒyǐ is absent in the main clause and yīnwèi is used (blue line), sentence-medial causal clauses are most favored. In contrast, when suǒyǐ is absent in the main clause and non-yīnwèi connectives are used (red line), sentence-medial causal clauses are least favored. A post-hoc comparison confirms that these two contexts differ significantly (estimate = 3.58, p < 0.0001). When both yīnwèi and suǒyǐ are present, the probability of sentence-medial causal clauses is higher than in contexts where suǒyǐ is present but non-yīnwèi connectives are used. However, this difference is not statistically significant (estimate = 0.59, p = 0.95).

When connective types are compared, new patterns emerge. For yīnwèi, sentence-medial causal clauses are significantly more favored when suǒyǐ is absent in the main clause compared to when it is present (estimate = 2.27, p < 0.0001). In contrast, for non-yīnwèi connectives, the probability of sentence-medial causal clauses remains consistently low, regardless of whether suǒyǐ is absent or present. This suggests that non-yīnwèi connectives generally favor sentence-initial causal clauses over sentence-medial ones (estimate = −0.72, p = 0.94).

5 Discussion

Adopting a variationist approach to analyze a large speech corpus of Mandarin Chinese, the data reveals that sentence-medial causal clauses occur slightly more frequently than sentence-initial ones (51 % vs. 49 %). While this difference is smaller than reported in some previous studies (Biq 1995; Li et al. 2016; Song and Tao 2009), it challenges earlier claims that sentence-initial positioning represents the canonical word order for causal clauses in Chinese (Chao 1968; Kirkpatrick 1993; Young 1994). This finding aligns with arguments from studies that emphasize the variation of word order in causal clauses (e.g., Biq 1995; Song and Tao 2009; Wang 1999). Moreover, this variation mirrors patterns observed in many other languages, such as English, German, Japanese, and Santali (Munda, India). In these languages, formally, causal clauses lack signs of syntactic embedding, and functionally, they act as independent assertions that explain or support the semantically associated main clause (Diessel and Hetterle 2011). Their discourse-pragmatic role contrasts with that of other adverbial clauses, which typically favor sentence-initial positions to establish broader discourse coherence. Thus, causal clauses are more likely to occur sentence-medially, where they remain largely independent of the preceding main clause (Diessel 2005).

We also compared the usage rates between dialogues and monologues, finding that sentence-medial causal clauses occur slightly more frequently in monologues than in dialogues (55 % vs. 49 %). Sentence-medial causal clauses in Chinese have been argued to serve as an information-sharing device, adding context to the preceding discourse, whereas sentence-initial causal clauses function as an interactional device, setting a discourse frame for the upcoming discussion (Song and Tao 2009; Wang 1999). Therefore, in monologues, where speakers have more time and control over the flow of information, allowing for more detailed explanations, elaborations, or arguments, they tend to use more sentence-medial causal clauses.

Based on the multifactorial inferential analysis of the data, which considered multiple factors within a single statistical model, we observed several patterns regarding the syntactic factors influencing causal clause ordering in Chinese, which have been overlooked in previous research. Subordinate clause length exhibited a significant main effect, with longer subordinate clauses favoring sentence-medial positions over sentence-initial ones. This finding aligns with parsing efficiency theories (Hawkins 1994, 2004), which propose that longer constituents tend to appear towards the end of a sentence that facilitate efficient processing. However, subordinate clause complexity did not demonstrate either a main effect or an interaction effect with subordinate clause length. This suggests that, in processing Chinese causal clauses, subordinate clause length carries relatively greater cue weighting and serves as a more valid and salient cue for their positioning than complexity, according to the Competition Model in sentence processing (Bates and MacWhinney 1987; MacWhinney et al. 1984). The distinction between these two dimensions underscores the need for future studies to treat subordinate clause length and complexity as separate factors, not only in causal clauses but also in other types of adverbial clauses in Chinese. Furthermore, there is an interaction between main clause length and main clause complexity, which provides further insights into the variation of causal clause ordering. Specifically, when main clauses are syntactically simple, increased length correlates with a higher likelihood of sentence-medial causal clauses. This pattern likely reflects a cognitive preference for balancing the information load within a sentence. A long but simple main clause may ease processing demands, particularly when followed by a subordinate causal clause in a sentence-medial position. Conversely, when main clauses are syntactically complex, increased length decreases the probability of sentence-medial causal clauses. This suggests that the combination of complexity and length introduces a heavier cognitive load, making sentence-initial placement of subordinate causal clauses a preferable strategy to enhance comprehensibility.

These findings on syntactic length and complexity in Chinese only partially align with previous research on English (Diessel 2008; Wasow 1997; Wasow and Arnold 2003). In English, longer and more complex clauses tend to appear sentence-finally to facilitate parsing by allowing the main structure of the sentence to be processed first (Hawkins 1994; Wasow 1997). In contrast, Chinese demonstrates a distinct strategy: while longer subordinate clauses are associated with sentence-medial positioning, complexity – particularly in subordinate clauses – does not significantly influence placement. This divergence between English and Chinese suggests that the latter employs unique strategies for managing parsing and processing demands, likely shaped by its typological characteristics, such as its topic-prominent structure and flexible word order (Gasde and Paul 1996; Li and Thompson 1976, 1981; Thompson et al. 2007; Wang 1984). Furthermore, the observed preference for sentence-medial positioning of causal clauses with long but simple main clauses, and sentence-initial positioning of causal clauses with short but complex main clauses in Chinese, highlights this distinctive approach to balancing syntactic and cognitive processing requirements. These findings underscore the importance of considering language-specific constraints when examining syntactic phenomena such as clause ordering. They also demonstrate that generalizations based on English may not always apply to Chinese, particularly regarding the interplay of length and complexity.

For the semantic factor, emotional valence, the regression model underscores its important role in modulating the positioning of causal clauses in Chinese, a factor also overlooked in previous studies. Specifically, the main effect of subordinate clause emotional valence indicates that positive valence favors sentence-medial positioning of causal clauses, suggesting that positive, favorable semantic content integrates more smoothly into the ongoing discourse than neutral or negative valence (Hunston 2007; see also Hunston and Thompson 2000; see Partington 2004 for similar functions of emotional valence in English). Additionally, an interaction between main clause emotional valence and genre further influences clause positioning. In dialogues, the preference for neutral valence of main clauses in sentence-medial positioning of causal clauses reflects the cooperative nature of conversational exchanges, where less emotive content promotes clarity and balance. In contrast, monologues tend to favor both positive and negative valence in sentence-medial positioning of causal clauses, reflecting a narrative style that foregrounds evaluative content within extended discourse. These findings highlight the nuanced effects of genre and emotional valence on the variation between sentence-medial and sentence-initial causal clauses across diverse communicative and emotional contexts, aligning with arguments in the register variation framework (Biber and Conrad 2019; Goulart et al. 2020).

For the factors at the discourse-pragmatic level, there is an interaction between the type of connective and the presence of the result marker suǒyǐ. Specifically, when suǒyǐ is absent and yīnwèi is used, sentence-medial causal clauses are most favored. This finding suggests a strong integration of yīnwèi with sentence-medial positioning, corroborating findings from other studies (Wang 1999; Xiao et al. 2021). This could reflect the cooperation of yīnwèi and the absence of suǒyǐ allows for greater flexibility in clause positioning, facilitating a smoother connection between the main and causal clauses in ongoing discourse. As found in previous studies, when suǒyǐ appears in the main clause (which may mark a stronger or more explicit causal relationship), it tends to form a summary statement, alongside yīnwèi clauses, which introduce background information in the subordinate clause, motivating sentence-initial rather than sentence-medial positioning of yīnwèi (Biq 1995; Kirkpatrick 1993). However, when suǒyǐ is absent and non-yīnwèi connectives such as jìrán, yóuyú, and jiànyú, are used, sentence-medial clauses are least favored. This pattern may suggest that certain non-yīnwèi connectives are more likely to occur in sentence-initial position, possibly reflecting a preference for more explicit or rhetorically foregrounded causal structures (Li et al. 2016). While some of these connectives, such as yóuyú and jiànyú, have been associated with more objective causal relations (as demonstrated in Example 11 above and in Xiao et al. (2021)), this does not uniformly apply to all non-yīnwèi forms. For example, jìrán tends to encode subjective causal relations, indicating that the relationship between connective type and clause position is not strictly determined by the objective-subjective distinction (Xiao et al. 2021). Moreover, because non-yīnwèi connectives are generally less semantically prototypical for expressing causality than yīnwèi (Li et al. 2016; Xing 2003), they may be less integrated into the discourse flow, contributing to their tendency to appear sentence-initially. Overall, these results point to a complex interplay between connective type, the presence of suǒyǐ, and clause ordering preferences in spoken discourse.

It is worth noting that this study focuses exclusively on utterances containing explicit causal connectives (e.g., yīnwèi, yóuyú, jìrán) and does not include instances of implicit causal relations – those in which causality is inferred from context rather than overtly marked. While implicit causal relations are common in spoken Chinese discourse (cf. Biq 1995), their identification requires close, manual reading of each utterance and contextually grounded interpretation, which differs fundamentally from the automated extraction approach adopted in this study. The probabilistic constraints that govern clause ordering in implicit causal constructions may also diverge from those in explicitly marked ones. For instance, features such as double marking with yīnwèi and suǒyǐ, which are relevant to the analysis of explicit constructions, do not occur in implicit contexts (cf. Xiao 2020). Therefore, utterances with implicit causal relations are not directly comparable to those containing explicit connectives in terms of both extraction method and probabilistic constraints. An investigation of implicit causality in spoken Mandarin would require a separate, dedicated study using different theoretical frameworks and analytical procedures.

6 Conclusions

This corpus-based multifactorial analysis of spoken Mandarin Chinese provides compelling evidence for the variation in the sequential ordering of causal clauses, with reason connectives (e.g., yīnwèi, yóuyú) appearing either before or after the main clause (sentence-initial vs. sentence-medial positions). The analysis reveals that this variation is not random but is systematically constrained by syntactic, semantic, and discourse-pragmatic factors, aligning with the fundamental “orderly heterogeneity” argument within the variationist framework (Weinreich et al. 1968: 100). Notably, robust interactions were identified between factors. For instance, the interaction between two discourse-pragmatic factors, the type of connective and the presence of suǒyǐ, demonstrates that the preference for sentence-medial clauses with yīnwèi in the absence of suǒyǐ reflects a flexible, discourse-driven process of clause positioning. In contrast, the preference for sentence-initial clauses with non-yīnwèi connectives points to a more rigid structure governing certain types of causal relationships. Furthermore, there is an interaction between main clause length and its complexity, underscoring the nuanced role of syntactic factors.

Together, these findings contribute new evidence on the multifactorial conditioning of causal clause ordering in Mandarin Chinese, offering novel insights into sentence structure and discourse cohesion in the language. As this study focuses exclusively on utterances with explicit connectives, future research should explore implicit causality in Chinese, which is a complementary area for deepening our understanding of causal expression in discourse.