Presuppositions cross-linguistically: A comparison of soft and hard triggers in Chinese and German

1 Introduction

In everyday communication, speakers take a lot of information for granted. In example (1), for instance, it is assumed to be common knowledge that Karim used to smoke.

In our example, the use of the change of state predicate stop gives rise to the inference presented as ≫. We call such requirements of a specific linguistic expression presuppositions and the expressions that induce such preconditions presupposition triggers: by uttering (1), thus, the speaker presupposes that Karim used to smoke. If the precondition is not met, the utterance is odd or even not interpretable truth-conditionally.

The list of expressions that have been observed to trigger presuppositions is long and it includes, to mention just some examples, definite and possessive expressions like the cat or Peter’s cat, factive predicates such as regret, additive particles like too or even some specific syntactic structures like clefts as in example (2) (see Beaver et al. 2021; Levinson 1983b for a more extensive overview):

Typically, presupposed content is characterized by its ability to project, i.e. to survive under the embedding of an entailment-canceling operator. This can be seen in example (3), where the presupposed content remains unchanged, whereas the assertion of the utterance is cancelled (negated or questioned):

Projection, that is the persistence under negation, a question operator, in conditionalization (if S) and modalization (perhaps S) (the so-called family of sentences following Chierchia and McConnell-Ginet 2000), is a key piece of our investigation and we will come back to it later.

Another characteristic property of presupposed content is that it cannot be challenged directly using No. A direct challenge can be used to target the asserted content of the utterance but not the presupposition, as can be seen in (4) (see also Shanon 1976; von Fintel 2004):

Concerning projection and deniability, it has been observed that not all presupposition triggers behave uniformly – whereas the presuppositions of some triggers are very robust and project with a very high probability, the presuppositions triggered of other expressions do not always arise. This has led to various taxonomic proposals concerning the classification of presupposition triggers, the most important of these will be discussed in Section 3.

As Simons (2001) observes, the presupposition of a lexical trigger is not dependent on a specific linguistic form, but is also triggered by expressions with the same truth-conditional content. This nondetachability of presuppositions can be seen in the following example, where the substitution of the expression stopped through semantically equivalent expressions does not influence the presupposition of the utterance (example slightly modified from Simons 2001, p. 435):^[1]

The example shows that an expression with the same semantic content will generate the same presupposition, presuppositions are, thus, nondetachable – they “attach to the content expressed, and not to any lexical item” (Simons 2001, p. 435). But what if we consider the same utterance in another language? Does the German translation in (6a) or the Chinese equivalent in (6b) induce the same presuppositional content as the English examples? While this is often tacitly assumed, there is very little research on the cross-linguistic comparison of presupposition triggers.

In this article, we are interested in two issues concerning presupposition triggers, namely (a) their universality and (b) their heterogeneity.

1. The universality of presupposition triggers: Most of the time, it is tacitly assumed that two expressions x and y from two different languages X and Y have the same presuppositional requirements, provided their meaning is the same. However, recent research also suggests that pragmatic strategies can be influenced by language specific and cultural aspects. In particular, differences between Western and non-Western societies have been observed. However, there is little to no systematic research on presuppositions in Western and non-Western languages. Thus, we will discuss the results of an experiment in which we investigate the projective behaviour of selected triggers from German in comparison to their equivalents in Chinese.

2. The heterogeneity of presupposition triggers: Previous research has shown that the group of presupposition triggers is more heterogeneous than originally assumed. More specifically, differences can be observed concerning their projective behavior, amongst others, which has led to various subclassifications. In this experiment, we are going to verify the classical distinction between hard and soft triggers (see e.g. Abusch 2002, 2010; Simons 2001), together with the split of factive verbs (Hooper and Thompson 1973; Karttunen 1971).

Of course, these two issues are interrelated: By investigating whether selected expressions in German and their counterparts in Chinese share the same projective behaviour, we also intend to explore the classification of these triggers in both languages.

To investigate these two issues, we will first review previous studies on the cross-linguistic comparison of presupposition triggers in Section 2 and then give a short overview of the most important results concerning the heterogeneity of triggers in Section 3. In Section 4, we will discuss some reflections and difficulties that arise when equivalences in both languages have to be chosen. Our own experiment is presented in Section 5, which will be followed by a detailed discussion of the results in Section 6.

2 Presupposition triggers cross-linguistically: Previous investigations

Recall our example (5) and its translations to German and Chinese. These examples raise the question whether two expressions from two different languages generate the same presuppositional content, provided their meaning is the same. Without being systematically investigated, usually, this is tacitly assumed – as an example consider the translations of Levinson’s (1983a) textbook Pragmatics, in which his list of English presupposition triggers is translated to different languages, such as German and Italian, without further notice. The thinking behind these tacitly assumed universality of presuppositions might be rooted in the observed nondetachability (see example 5), which suggests that presuppositions not so much dependent on a specific linguistic form, but more so on the semantic contribution of an expression: if two formally distinct expressions have the same semantic contribution, they will trigger the same presupposition. Extending this nondetachability beyond the scope of one single language to the contrastive perspective on different languages, we might be allowed to expect that if a trigger has a counterpart in other languages with a similar semantic meaning and perhaps also similar syntactic features, then the generation of their presuppositions and their projective behavior should also be stable across languages. Note, however, that this assumption is neither theoretically nor empirically fully investigated.

Theoretically, neither Gricean accounts (see e.g. Grice 1989; Horn 2004; Levinson 2000) nor Relevance Theory (see e.g. Sperber and Wilson 1996; Wilson and Sperber 2004) do explicitly address the question whether presuppositions are universal. But since in both trends pragmatic principles are considered as general and rooted in human cognition, we are allowed to assume that both pragmatic currents more or less predict the universality of pragmatic strategies and hence also imply the universality of presuppositions, although in Relevancy Theory, some language and cultural specific pragmatic aspects are acknowledged.

Empirically, according to the overview in Ameka and Terkourafi (2019), pragmatic investigations focusing on cross-linguistic comparisons have mainly favored topics such as (im)politeness (see for instance the summaries in Chen 2010; Jia and Yang 2021), conversational strategies (see e.g. Ameka 2006; Ayodele 2016; Boone et al. 2022; Keenan 1976; Kondowe et al. 2014; Obeng 1994, 1999, 2003), speech acts (see e.g. Ameka 2017; Maíz-Arévalo and Méndez-García 2023; Wierzbicka 1985), and conversation analysis (see e.g. Dingemanse and Enfield 2015; Dingemanse et al. 2014; Holler et al. 2016; Rossi et al. 2020). What is important to us is that these investigations of different pragmatic phenomena conclude that cross-linguistic variation exists: concerning the so-called East-West divide, for instance, differences have been observed at least since the 90s with regard to politeness, amongst others (see e.g. Gu 1990; Ide 1989; Matsumoto 1989), and the East-West divide remains a popular topic nowadays in studies on multi-party conversations and online communication (see e.g. Li et al. 2020; Xia and Lan 2019). This suggests that pragmatic principles and strategies are, at least to some degree, subject to cultural and/or language-specific differences, what has to be taken into account by future research (see e.g. also Chen 2019; Ran and Zhao 2018). Moreover, as summarized and discussed in Ameka and Terkourafi (2019), in their majority, pragmatic theories are based mainly on observations from Western societies, therefore, there is a need for verifying these concepts for non-Western languages.

Although all empirical studies mentioned above are doubtlessly relevant for universal pragmatics, they offer hardly any direct evidence for or against the universality of presuppositions. Narrowing the scope to presuppositions, there are only very few systematic cross-linguistic investigations.^[2] To the best of our knowledge, the first experimental study in this question is conducted by Amaral and Cummins (2015), who compare Spanish presupposition triggers with their functional equivalents in English, relying on results from Cummins et al. (2012). In their experiments, two kinds of triggers were compared with respect to deniablity (see example 4 above). The results of the investigation show that in both languages, presuppositions can hardly be addressed by a direct challenge.Additionally, Amaral and Cummins (2015) observe a difference between lexical triggers and resolution triggers, which is also stable for both English and Spanish. The authors therefore propose that “the differences between presupposition triggers are rooted in general logical or cognitive principles, rather than arising as a language-specific feature of English (p. 158)”, supporting the universality of presuppositions.

Another series of experiments using deniability as a hallmark for presuppositions is presented in Djärv et al. (2018) and Schwarz et al. (2020). Comparing Italian presupposition triggers with their equivalents in English, the authors observe slight differences between both languages: in comparison to their English counterparts, ‘no’-answers were judged significantly less appropriate for cognitive factive triggers such as be aware or realize in Italian.

However, a potential weakness of these investigations is pointed out by Snider (2017a, 2017b), who argues that differences in deniability might not result from backgroundedness but rather from differences in anaphoric availability.

More recently, Reins et al. (2021) investigate presuppositions in the context of their research on lying and misleading. In their experiment, they compare presuppositions in English to their Russian equivalents by using presuppositions as one of four possibilities for lying indirectly. After reading a vignette with an untruthful presupposition, participants were asked to judge whether the examples constitute cases of lying or misleading. Overall, the results are very similar for Russian and English speakers, nevertheless, we can observe at least a numerical difference between the two languages when we consider misleading-ratings.

Finally, Chen (2022) presents an empirical investigation on presuppositions involving native German speakers and Chinese learner of German. In this experiment, presuppositions are used to target the Question Under Discussion directly, which leads to pragmatic infelicities. Strikingly, in this study, the Chinese learner are more tolerant to these at-issueness violations than the native German speakers. A follow-up question, then, is whether this tolerance of the Chinese learner of German is a result of the influence of their first language Chinese, or if it solely reflects their insecurity when confronted with a German set-up. Moreover, in this experiment, the difference between groups is mainly significant with regard to hard triggers, but not for soft triggers. It would be, therefore, very interesting to investigate the stability of projective strength in the context of the soft-hard-classification for German and Chinese – we will come back to this in the next section.

It can be seen that previous investigations of the universality of presuppositions mostly used deniability to measure the backgroundedness of the presupposed content. However, as has already been pointed out, differences in the results might be due to anaphoric availability. Moreover, the majority of the experiments discussed above investigate only Western languages, whereas the only investigation that includes non-Western speaker (Chen 2022) can observe differences. Thus, what can be said is that the cross-linguistic stability of presuppositions cannot just be assumed but has to be verified through experimental investigations, especially if we consider Western and non-Western languages. In the context of the results of Chen (2022), who observes different ratings of native German speaker and Chinese learner of German, in our experiment, we will compare presuppositions in German to their Chinese counterparts. In order to avoid the problem linked to deniability, for this comparison, we will use projection as a touchstone.

We can conclude this short review of the most important previous investigations by stating that while pragmatic principles are traditionally considered general, rooted in human cognition, and cultural-neutral, experiments on politeness, conversational strategies, and speech acts, among others, reveal the impact of social and language specific perspectives. As for presuppositions, cross-linguistic experiments, although only a few, illustrate not only similarities but also some potential differences in their behavior. Therefore, the cross-linguistic stability of their projective strength cannot be taken for granted either, but need to be investigated experimentally. Additionally, note that the experiments on presuppositions discussed in this section explore mainly Western languages. Considering the observations on related phenomena, especially the East-West divide, a comparison of presuppositions in Western and non-Western languages is overdue.

3 The heterogeneity of presupposition triggers

Recall that we are interested in two issues concerning presupposition triggers, namely their universality and their heterogeneity. After having reviewed the most important results on the cross-linguistic comparison of presuppositions, we will now summarize the crucial research on the heterogeneity of presupposition triggers, focusing on their projective strength. We will start with the heterogeneity within the class of factive predicates before proceeding to a more general perspective on soft and hard triggers.

It has been known for a long time that factive predicates, which were first investigated by Kiparsky and Kiparsky (1970), do not behave uniformly. According to Hooper and Thompson (1973), predicates that select a that-complements can be classified into five groups. Among them, two groups are classified as being factive, meaning that they presuppose their sentential complement as being true:

Emotive factives such as regret express the subjects’ emotional attitude towards the proposition of their complement, whereas cognitive factives such as discover illustrate the subject’s epistemic status or its cognitive manner towards the embedded proposition. It has been observed that the two groups differ in terms of projection. According to Karttunen (1971, p. 63), the complement of a cognitive factive verb like discover can lose its presupposed status once the main clause is questioned or conditionalized, while the complement of a verb like regret remains projective, as shown in the following examples:

Embedded under a question operator, both sentences can be interpreted as giving rise to the presupposition that the addressee did not tell the truth. However, for (8b), there is also another interpretation available, as Karttunen (1971) observes: in this interpretation, the speaker is ignorant about the truth of the complement – he or she does not know whether the addressee has told the truth. Such an interpretation is hardly possible for (8a). In other words, comparing (8b) with (8a), it can be seen that, although both discover and regret are classified as factive verbs, the projective strength of their complements is very different. Due to this observation, cognitive predicates are also classified as semi-factives (see e.g. Hooper and Thompson 1973; Karttunen 1971; Stalnaker 1977). About half a century later, Karttunen (2016) revises the classification proposed by Hooper and Thompson (1973): the emotive-cognitive contrast is abandoned and predicates are categorized into five groups. In contrast to the classification based on Hooper and Thompson (1973), Karttunen (2016) assumes a separate class for verbs of discovery, including discover. He argues that members of this class are not presupposition triggers at all, as they do not project in examples like (8): “negative sentences, questions, and conditionals with coming-to-know verbs are in principle noncommittal” (p. 713). Therefore, he clearly distinguishes them from a predicate like regret, which is categorized as a propositional attitude verb.

In sum, the classification of factive verbs is still controversial, while the distinction between regret and discover in non-affirmative utterances is generally accepted. Both accounts, that is, the emotive-cognitive contrast by Hooper and Thompson (1973) and the five-group classification proposed by Karttunen (2016), predict that the presupposition of regret should be more stable and projective than the presupposition of discover, if it triggers one at all. For our experiment in Section 5, we will rely on the concepts of emotive and cognitive factive verbs, as this classification is widely accepted and since the difference between these two types of predicates has seen experimental support (see e.g. Djärv et al. 2018; Egré 2008; Schwarz et al. 2020).

The classifications discussed so far mainly concentrate on predicates that select a presupposed complement clause. A taxonomy of presupposition triggers with a broader perspective is the soft-hard split, first proposed by Abusch (2002, 2010. The classification of presupposition triggers into hard and soft ones is also based on the observation that presuppositions differ in their projective behaviour. Consider the following examples: Here, the difference between soft and hard triggers can be illustrated by means of epistemically deficient contexts, first discussed by Simons (2001):

According to Abusch (2002, 2010, these differences in projection can be traced back to a fundamental difference of the triggering process: a hard trigger like too in (9b) generates a semantic based presupposition that projects globally and leads to a contradiction between the two clauses as in (10b). On the other hand, soft triggers like win in (9a) give rise to a pragmatically based presupposition: pragmatic presuppositions of soft triggers can receive a local interpretation as shown in (10a) and are, therefore, less projective:

Note that within this account, the difference between emotive and cognitive factive predicates as described above is subsumed under the more general division of presupposition triggers into soft and hard ones – where emotive factive predicates are considered as typical hard triggers, while cognitive predicates, including discover, are assumed to be soft (see e.g. Abbott 2006; Abusch 2010; Simons 2001, 2007, among others).

The varying projective strength has also seen experimental support. Tonhauser et al. (2018), for instance, use a certainty judgment task to measure projection. In their experiment, presuppositions are considered as commitments of the speaker, therefore, participants were asked to judge how certain a speaker is about the (assumed) presupposed content of the utterance. One example from Tonhauser et al. (2018: 502) is given below.^[3] We will adopt this design in our experiment in Section 5.^[4]

After this short review of the two core elements of our investigation the heterogeneity within factive predicates and the heterogeneity of presupposition triggers in general, we are now able to motivate the selection of presupposition triggers for our experiment.

For our investigation of the universality and the heterogeneity of presuppositions, we chose four triggers: cleft sentences, to win, to regret, and to discover, or rather, their counterparts in German and Chinese. Clefts are considered as a typical hard trigger, that projects strongly out of embedding and resists a local interpretation of its presupposition. The change of state predicate win, on the other hand, is generally assumed to be a typical soft trigger. Together, clefts and win will function as anchors in our rating task. Regret is an emotive factive predicate – this class of factives is typically subsumed under the class of hard triggers. With discover we chose a cognitive factive which behaves like a soft trigger. We are, therefore, allowed to expect that the projective behavior of the emotive factive verb regret patterns with that of the typical hard trigger (cleft structure), while the cognitive factive predicate discover is predicted to show a similar projection behaviour as the soft trigger win. But before we turn to our experimental investigation of these four presupposition triggers, we have to say a few words about their translations to German and Chinese.

4 Trigger selection and their translation

In order to investigate the issues formulated in Section 1, we chose four triggers, namely win, discover, regret and cleft structures. For a cross-linguistic investigation, it is important that the language-specific versions of these four triggers are equally frequent in both languages. Furthermore and even more importantly, their German and Chinese counterparts have to be both semantically and syntactically as parallel as possible. Thus, in this section, we will give some insights on the reasons why we chose a particular translation for each trigger.

We will start with the emotive factive regret. In German, this expression can be translated as ‘bereuen’ (see for instance Antomo 2012; Büring and Hartmann 2001; Ito 2017). In Chinese, the counterpart of regret is assumed to be ‘后悔 (houhui)’ (see for instance Ji 2009; Xu 2014). Besides their correspondence in semantic meaning, syntactically, ‘bereuen’ in German and ‘后悔 (houhui)’ in Chinese can both take a clausal complement. Furthermore, for both languages, it has been argued that the content of the complemet clause is presupposed, as shown below:

The second trigger of our experiment, the verb discover, is a very well investigated presupposition trigger (see for instance Abrusán 2011, 2016; Abusch 2010; Djärv et al. 2018; Simons 2001, 2007, among others). We chose it for two reasons: First of all, it is a typical cognitive factive predicate. If cognitive factives (or semifactives) can be subsumed under the more general category of soft triggers, then discover has to be the best candidate to illustrate this and its projection behaviour should pattern with that of the typical soft trigger win. The second advantage is that the counterparts of discover in German (‘entdecken’) and Chinese (‘发现 faxian’) are syntactically parallel: both ‘entdecken’ and ‘发现 (faxian)’ can trigger presuppositions when taking clausal complements, but are ambiguous when taking simple objects. Since in our experiment, we will investigate only clausal complements, the former property is shown in the following examples, using projection out of negation as a test for presupposition:

As a baseline, we will use win and cleft constructions. There are many theoretical and empirical investigations concerning win, which show that this predicate behaves like a typical soft trigger (see e.g. Abusch 2002, 2010; Bill et al. 2016). Additionally, the syntax of its counterparts in German (‘gewinnen’) and Chinese (‘赢 ying’) can be parallel, furthermore, intuitively, both give rise to the same presupposition:

There are several reasons why we chose clefts as a baseline for hard triggers: First, as all other triggers are lexical items, adding a structural trigger can be interesting in terms of a crosslinguistic comparison. Second, other classical hard triggers like again or too/also do not qualify for a cross-linguistic comparison since they are characterized by a non-parallel behavior in German and Chinese. Take again for example: According to Xu (2014), the counterpart of again in Chinese is ‘又 (you)’, a word with six meanings in the Contemporary Chinese Dictionary (现代汉语词典, The Commercial Press 2016), and not all of them do trigger presuppositions. Even when it expresses the meaning of repetition or iteration, the usage of ‘又 (you)’ is still different from that of ‘wieder’, the counterpart of again in German: as shown in Xu (2014), ‘又 (you)’ can also be used in a context with different subjects or different predicates as shown in the following example (adopted from Xu 2014, pp. 97–99), while ‘wieder’ can only be used with the same subject and the same predicate, similar to again in English:

In short, while ‘wieder’ in German mainly triggers the presupposition that the same subject has done the same thing before (just like again in English), ‘又 (you)’ can trigger several different kinds of presuppositions, some of which even overlap with too and also in English. There are similar problems with too/also in German, especially embedded under a question operator. Depending on the context, they can also give rise to different presuppositions. For these reasons, we refrained from using again, too and also as an example for a hard trigger in our experiment and chose cleft-sentences as a typical hard trigger, since these are not-ambiguous concerning their presupposition.

So how do cleft-constructions look like in German and in Chinese? In German, it-clefts are semantically and syntactically very similar to English it-clefts, although less frequent. They are biclausal constructions consisting of a main clause and a subordinate clause with the construction pattern Es (’it’) copula X + subordinate clause. As in English, they express identity of two descriptions and have been argued to give rise to an existence presupposition:

As for Chinese, it is generally assumed that the counterpart of an it-cleft is the ‘是 … 的 (shi … de)’-structure as in the following example (see e.g. Huang 1982, 1988; Liu and Kempson 2018; Zhan and Sun 2013; Zhou 2020):^[5]

Note that in our experiment, we will investigate the projection of presuppositions out of the scope of a question operator. Now it happens that, embedded under a question operator, 的 (de) is typically omitted in ‘是 … 的 (shi … de)’-structures or, more precisely, ‘的 (de)’ is omitted if the sentence ends with a question particle like ‘吗 (ma)’, amongst others (Zhan and Sun 2013). In other words, a question like (18a) sounds more natural and is used more frequently than (18b):

However, ‘是 … 的 (shi … de)’-structures with omitted 的 (de) are ambiguous: they can be interpreted as clefts or as bare ‘是 (shi)’-sentences. According to Paul and Whitman (2008:420), “shi … de patterns with cleft constructions, while bare shi behaves like association with focus”. However, they also propose that sentences with initial bare ‘是 (shi)’ with subject focus can also be considered as clefts. Therefore, in order to guarantee a cleft-reading, we decided to use the ‘是 … 的 (shi … de)’-structure with the initial ‘是 (shi)’, subject focus and omitted ‘的 (de)’.^[6]

In sum, the four selected triggers used in the experiment are listed below in Table 1 with their traditional classification and translations in both Chinese and German.

Combining the theoretical background discussed in Sections 2 and 3 with the considerations concerning translation and our choice of triggers, we are now able to articulate our particular research questions:

5 Experiment

5.6 Results

For the analysis of the judgments of the critical items, all ratings of the control items and fillers were excluded. The data then encompassed 1,848 judgments taken from 77 individuals out of 2 groups for 24 critical items. The mean judgments with 95 % confidence interval are presented in Figure 1.

Figure 1:

Mean judgments ±95 % confidence interval for the factors Language and Trigger.

To estimate the effects of Trigger, Language and their interactions, the data was analyzed using R (version 4.0.3, R Core Team 2020). Since the 7-points-scale that we used in the test can be understood as a kind of ordinal scale (for more discussion about ordinal data and metric models, see Liddell and Kruschke 2018), and since the residual of the LMM fitting did not seem to be as ideally symmetric as required (see Figure 2 in the Appendix), the data was analyzed with the Cumulative Link Mixed Model (CLMM) using the clmm function from the ordinal package (Christensen 2019). The CLMM full model included Trigger, Language and their interaction as fixed effects and the maximal random effects structure^[9] and used logit as the link function. The factor Language was centered with the sum-contrast, coded with (−0.5, 0.5), while the factor trigger was dummy coded. The results of the full CLMM model are listed in Table 3.

Table 3:

Overview of the CLMM full model, fitted with the Laplace approximation, with estimates, standard errors, z- and p-values of fixed effects and threshold coefficients.

Effect	Estimate	SE	z-Value	p-Value
Triregret	0.48	0.38	1.28	0.20
Tridiscover	0.78	0.47	1.67	0.09
Triwin	−2.03	0.29	−6.92	<0.001
Lan.con1	−0.22	0.42	-0.52	0.60
Triregret × lan.con1	−0.93	0.53	−1.74	0.08
Tridiscover × lan.con1	−0.28	0.65	−0.43	0.67
Triwin × lan.con1	−0.48	0.50	−0.97	0.33
Threshold	Estimate	z-Value
1|2	−4.44	−15.44
2|3	−4.14	−14.69
3|4	−3.89	−14.01
4|5	−3.52	−12.92
5|6	−3.17	−11.83
6|7	−2.23	−8.73

Regarding our first main research question on the cross-linguistic stability, the full model was firstly compared with the restrictive model which comprised only the fixed effect Trigger and the random effects,^[10] that is, the effect of Language and the interaction were removed. The comparison between these two CLMM models was not significant (LR.stat = 6.46, df = 4, p = 0.17), indicating that statistically, the effect of Language and the interaction is not significant.^[11]

Considering our second main research issue, the heterogeneity of presupposition triggers, what we can state is that, according to both LMM and CLMM analyses, Trigger is the only main effect that significantly influences ratings. Applying the CLMM full model to the emmeans function from the emmeans package (Lenth 2021), the four triggers were compared pairwise with each other within Language using z test.^[12] Results that are relevant to our research questions are listed in Table 4 below.

Table 4:

Comparison of triggers based on the CLMM full model with z test, p-value adjusted with tukey method.

Language	Comparison of triggers	Estimate	z-Ratio	p-Value
Chinese	cleft − win regret − discover	1.79	4.58	<0.001
		0.02	0.03	1
	cleft − regret	−0.94	−1.88	0.23
	discover − win	2.72	4.47	<0.001
German	cleft − win regret − discover	2.28	5.99	<0.001
		−0.63	−1.03	0.73
	cleft − regret	−0.02	−0.05	1
	discover − win	2.92	4.61	<0.001

The comparisons show that the similarity and the difference between the individual triggers are also stable across languages. In both Chinese and German, the difference between the typical hard trigger (cleft-sentences) and the typical soft trigger win is highly significant. Additionally, as expected, we can observe a clear similarity between clefts and the typical emotive trigger regret. However, in contrast to our predictions, the typical cognitive trigger discover does not pattern with the soft trigger win, as most theories on the soft-hard classification would predict. Quite on the contrary, the results for discover are very similar to those of clefts, our typical hard trigger, and to those of the emotive trigger regret.

6 General discussion

The aim of our experimental investigation was to answer the research questions (19a) and (19b). Our first research question addresses the universality issue. Through the example of a comparison of presuppositions in Chinese to their equivalents in German, we wanted to investigate whether presupposition triggers are cross-linguistically stable. As a touchstone, we exploited projection out of the scope of a question operator. Of course, it would be presumptuous to extrapolate from our results, collected from only two languages, to a statement about the universality of presupposition triggers. Nevertheless, our data from two typologically very distinct languages allow for a first tentative step towards a broader understanding of presuppositions in language comparison.

Our results show that there is no significant effect of Language nor a significant interaction. Concerning the similarity between patterns, we are allowed to assume that the projective behavior of presuppositions is stable between the two languages investigated, when triggers and their equivalents are carefully selected. This observation is consistent with the main findings in previous investigations like Amaral and Cummins (2015) and Schwarz et al. (2020), who also observe a similar behaviour of presuppositions in some selected Western languages. Additionally, the ratings illustrate that not only the triggering of presuppositions by certain expressions can be cross-linguistically stable, but also their projective strength and their interaction with contextual effects. Recall that Simons (2001) notices the nondetachability of presuppositions within one language, i.e. the observation that expressions with the same semantic content give rise to the same presupposition (see example 5 above). Now our results suggest that this observation can be extended to a cross-linguistic perspective: if a trigger has a semantic and syntactic equivalent expression (or construction as in the case of clefts) in another language, it leads to the same presupposition with the same projective behaviour. For Simons (2001), this nondetachability is proof for a pragmatic source of the presupposition. However, in our opinion, this does not follow straightforwardly. Especially if we consider the observed cross-linguistically stable projective behaviour of clefts, which have to be considered a hard trigger, it is not convincing that the presuppositions of these constructions in German and Chinese arise via pragmatic reasoning. Nevertheless, what we can say is that our results suggest that the triggering process, semantic or pragmatic, seems to be the same in both languages we investigated. It is more than likely that the presupposition triggered by an expression like win, regret or discover is closely related to the semantic content of its trigger. Thus, an expression in another language with the same semantic content should typically lead to the same presupposition. However, this line of reasoning cannot be applied to cleft-structures: without semantic content, the arising presupposition cannot be a result of this content. Instead, giving rise to a presupposition seems to be the raison d’être of cleft-constructions, their specific function that distinguishes them from a mere main clause. As a consequence, we predict that if a language offers the possibility to use a cleft, this construction should give rise to a presupposition. We will leave the question on how presuppositions are triggered for future research – contrastive investigations, however, seem to be a promising way to shed light onto the triggering process.

However, we cannot exclude that cross-linguistic or cross-cultural differences exist. First of all, although German and Chinese are typologically two very different languages, they can of course not represent all languages. Secondly, we selected the triggers for our experiment with great care, excluding some triggers with potential differences already before the test. It is possible that there are expressions that do trigger presuppositions in German, whereas their counterpart in Chinese (or another language) does not give rise to the same or to any presupposition (or vice versa). Furthermore, in this experiment, features that have been said to depend on cultural or social dimensions, such as politeness or lying, for instance, have been reduced to a minimum. Interestingly, recall that in the experiments run by Reins et al. (2021) on lying and misleading, at least numerically, presuppositions were judged differently by English and Russian speakers. Combining our results with their observations, we may assume that this cross-linguistic difference more likely results from a different understanding of the concept of misleading in different cultures or societies, rather than from differences concerning presuppositions or commitment attribution. The interaction between presuppositions and those cultural or social features should be investigated in more depth by future research in contrastive pragmatics. The stability observed in our experiment can serve as a baseline for future studies that compare presuppositions or related phenomena in different languages and cultures, or for investigations of presuppositions involving a L2-perspective, especially when differences are observed.

Our second research aim is repeated in (20):

Regarding these questions, our observations are only partially in line with our predictions. First of all, as predicted, there is a significant difference between clefts and win, both in German and Chinese, showing that these two triggers can function as baselines for a typical hard and a typical soft trigger, respectively. Additionally, in both languages, the emotive factive verb regret patterns with clefts as predicted. Both triggers were judged to involve a very high level of certainty, which distinguishes them from the soft trigger win and is in line with most assumptions made in the context of the traditional classifications discussed in Section 3 (e.g. Abbott 2006; Abusch 2010; Simons 2001, 2007).

However, in contrast to our predictions, the cognitive factive verb discover did not pattern with the soft trigger win, but with clefts, our typical hard trigger. Additionally, the difference between regret and discover, which is expected by researchers like Karttunen (1971), Hooper and Thompson (1973), and Karttunen (2016), was also absent in this experiment. This finding also differs from the results of previous empirical studies such as Djärv et al. (2018).

A question that arises, then, is why did discover differ from win? In most previous theoretical analyses, the softness of win and discover is typically explained with the same mechanism (see e.g. Abusch 2002, 2010; Klinedinst 2016; Romoli 2012, 2014; Zehr and Schwarz 2016, 2020) and cannot straightforwardly explain the difference between the two classical soft triggers that we observe in our test. For illustration, take the items given in Table 2, repeated below as (21) and (22).

The questions then can be specialized as follows: What is it that boosted the projective strength (i.e. the factive interpretation) of the complement embedded by the soft trigger discover? And why is the typical soft trigger win impervious to these factors? How did the questions involving discover become more committal with respect to the embedded clause than the questions involving win?

One might argue that the complement clause of the question (21) is about something that can be observed by the guests at the party, i.e. a change in the bride’s hair ornament. Therefore, the truth of the embedded clause can be considered to be part of the common ground. However, there are two arguments against this assumption: first, there were also critical discover-items where the content of the complement clause cannot be considered to be common knowledge (see items in Table 5). However, the ratings of these items did not decline or differ from the other items. Second, the same observation can be made for some of the items involving win: example (22), for instance, describes an event that happened during the party and should also be widely known among the guests. Nevertheless, ratings for this item were not higher than those of the other three win-items, which described competitions that took place in the last year (for more about mean ratings splited by Trigger and Language, see Figure 5 in Appendix). So this should only be a minor effect.

Another explanation is that the difference between the two soft triggers is due to the events: in comparison, the competition results are perhaps less interesting or less relevant to Mr. Müller’s death than the discovery status of some guests. However, there are several events mentioned in other items that are not relevant to the death either, for instance the best man’s feeling about the four pieces of carrot cake (see Table 2). Nevertheless, these ratings were still different from the win-items. In short, the possible lack of relations between the competitions and Mr. Müller’s death should not be the main reason for the difference between win and discover.

A more plausible explanation is, then, that the soft triggers are not similarly soft. That is, the softness of win and discover is due to different reasons. According to Karttunen (1971), Simons (2007), and Karttunen (2016), the softness of discover is due to the fact that the verb can receive more than one reading: a presuppositional interpretation, where the embedded proposition is not-at-issue and hence projects, and a non-presuppositional interpretation which can serve as a sincere question about the truth status of the embedded proposition. According to Simons (2007), a discover-sentence can lose its presupposition when it is the complement that answers the current Question Under Discussion (QUD), not the main clause. In this case, the complement is at-issue and cannot project, while the verb discover has a purely evidential function, as in the following example (see Abusch 2010; Simons 2007; Simons et al. 2010 for more discussion):

In our test, on the contrary, with such a thrilling background story, expressing a question using discover can effectively boost the relevance of the cognitive status itself and thus that of the matrix clause. As for the question in (21), this means that the discovery status of the cameraman is marked as highly relevant to the QUD and the main clause becomes clearly at-issue, while the embedded proposition gets pushed into the background. With regard to the verb win on the other hand, we are allowed to assume that its softness is due to different reasons. One possibility might be that, in contrast to discover, the presupposition of the achievement verb win is part of the process it describes, so that the presupposition participate and the assertion win are temporally and logically extremely closely related and stored with a strong connection in our mental lexicon. Then, our results suggest that the assumed heterogeneity of presuppositions is more fine-grained than previously assumed. While hard triggers seem to be alike, at least if we consider the projective behaviour of the two hard triggers in our experiment, soft triggers seem to be much more heterogeneous, as their softness is due to different mechanisms.

7 Conclusion

We conducted an empirical investigation involving four presupposition triggers in German and their semantic and syntactic counterparts in Chinese: cleft sentences as an instance of a typical hard trigger and win as a typical soft trigger served as a baseline for the emotive factive verb regret and the cognitive factive verb discover. Our aims were to measure the projective strength of their arising presuppositions out of the scope of a question operator a) to investigate the cross-linguistic stability of these triggers and b) to verify the heterogeneity of these triggers in both languages.

Concerning our first research question, our experiment shows that the projective behavior and the heterogeneity of presuppositions can be considered cross-linguistically stable, as long as suitable equivalences can be found. Of course, these results need further confirmation from other languages and more triggers. Still, with our results in mind, it seems probable that the underlying triggering process of presuppositions functions in a similar way across different languages. We think that our results give us a hint towards a better understanding of presuppositions and projection cross-linguistically, while our data might serve as a baseline for future research.

With regard to our second aim, our data suggest that the group of soft triggers has to be more heterogeneous than theoretical research has expected so far. In other words, if hard triggers are alike, it is possible that each soft trigger might be soft in its own way. That is, their softness or less typicality might result from different reasons or mechanisms.

In sum, our results add a piece to the puzzle to yield a better understanding of the basic but crucial questions on presuppositions: what are they, why are they projective and why do they project differently. Again, further research involving further triggers and other languages has to be done.