Keywords

1 Introduction

This paper takes a fresh look at the different exhaustivity levels of wh-interrogatives embedded under the veridical and distributive predicate wissen ‘know’, and under the cognitive-emotive and non-distributive überraschen ‘surprise’, cf. (1), (2).Footnote 1

figure a

The discussion will be based on two novel empirical diagnostics regarding the interaction of embedded wh-interrogatives with the concessive particle combination SCHON…aber ‘alright…but’ and the Q-adverb teilweise ‘partially’, as shown in (3).

figure b

A highly debated issue in question semantics is which of the observable surface readings of varying exhaustivity (strongly exhaustive [SE], intermediate exhaustive [IE], weakly exhaustive [WE]) are underlying semantic interpretations, and which ones are mere pragmatic inferences, if any. To this end, we will investigate the interpretive effect of particle combinations and Q-adverbs on the interpretation of interrogatives under know and surprise. We will show that insertion of the particle combination SCHON…aber blocks the generation of some pragmatic implicatures. From this, we conclude that exhaustivity inferences of wh-interrogatives that are blocked by the presence of the particle combination are pragmatic inferences. The Q-adverb teilweise ‘partially’, by contrast, operates on truth-conditional semantic content proper. We conclude that exhaustivity inferences targeted by teilweise must be part of the truth-conditional semantic content of embedded wh-interrogatives. Applying the two diagnostics to wh-interrogatives embedded under wissen ‘know’ and überraschen ‘surprise’, we find the following: First, SE-readings under wissen ‘know’ are pragmatic inferences that are derived from a weaker semantic interpretation [19, 40] under an internal subject perspective [13, 39]. This internal perspective follows from the novel general pragmatic Principle of Attitude Report Verification (PARV). Second, the observable distributive readings with überraschen ‘surprise’ result from pragmatic strengthening of a relatively weak underlying semantic interpretation, which can be cast in terms of an existential WE-semantics [14, 34, 35], or by analyzing cognitive-emotive attitude verbs like surprise as predicates operating on facts/situations rather than propositions/questions [12].

The article is structured as follows: Sect. 2 provides background information on the exhaustivity of wh-interrogatives and the interpretive effects of SCHON…aber and teilweise. Section 3 presents the novel empirical findings for wh-interrogatives embedded under wissen ‘know’ (henceforth: know + wh), and it sketches an event-based analysis of know as operating over the plural sum of knowledge sub-events, effectively giving rise to a semantic WE-interpretation. Section 4 presents the novel empirical findings and a preliminary analysis of wh-interrogatives embedded under überraschen ‘surprise’. Section 5 concludes.

2 Background: Exhaustive Force, Particles, and Q-Adverbs

This section provides background information on the variable interpretation of embedded wh-interrogatives as weakly or strongly exhaustive (Sect. 2.1), on the interpretive effects of the discourse particle SCHON ‘alright’ in combination with concessive aber ‘but’ (Sect. 2.2), and on the semantic import of the Q-adverb teilweise ‘partially’ (Sect. 2.3).

2.1 Different EXH-Force Under Know and Surprise: SE vs. WE

The surface interpretation of sentences with embedded wh-interrogatives can vary in the exhaustive force of the embedded interrogative, depending in part on the meaning of the embedding predicate. Consider (1) with wissen ‘know’ in a scenario with four individuals, Mary, Alex, Paul and Anna. Of these four, Mary and Alex danced, and Paul and Anna did not. The two readings of (1) of interest differ in how much information Nino must have regarding who did and who did not dance. On the strongly exhaustive reading [13], she must have complete information regarding the entire answer space, namely that Mary and Alex danced, and Paul and Anna did not. On the weakly exhaustive reading [19], it suffices for (1) to be true that Nino’s information state is complete with respect to the positive answer space: She would only need to know that Mary and Alex danced. Moreover, non-exhaustive readings [41] with know are blocked by the inherent distributivity or homogeneity of this predicate [4, 24].Footnote 2

The cognitive-emotive attitude predicate überraschen ‘surprise’ differs semantically from wissen ‘know’ in several ways. This has repercussions for the interpretation of embedded wh-interrogatives. For one, surprise is not obligatorily distributive [24]. So, (2) could still be true if Nino did not expect both Mary and Alex to dance at the same party (because they are rivals and never dance if the other does) even though she is not surprised by Mary’s dancing per se, nor by Alex’s. Given non-distributivity, surprise + wh may give rise to different readings than know + wh in the above scenario. The different readings will crucially depend on Nino’s prior expectations. On the distributive WE-reading (WE_dist), Nino didn’t expect Mary nor Alex to dance, so that her surprise is complete with respect to the positive answer space of Who danced?. A non-distributive WE-reading (WE_nondist; cf. [14, 34, 35]) obtains if it’s just Alex that Nino didn’t expect to dance. Now her surprise is directed at the positive answer space in a non-distributive manner. In addition, there may be two SE-readings with surprise, which make reference to the full logical answer space including the non-dancers: the non-distributive SE-reading (SE_nondist) obtains if Nino is not surprised by the actual dancers, but she did expect Anna to dance as well, contrary to fact. Finally, the distributive SE-reading (SE_dist) would require Nino to be surprised by everybody who danced and by everybody who didn’t (= complete counter-expectation).

Notice that know and surprise also exhibit different entailment patterns [32, 40], i.a. Know is upward entailing so that SE entails WE: If Nino knows who was and who was not at the party (SE), it follows that she knows who was at the party (WE). The same entailment does not hold for surprise: If it surprises Nino who did and who did not dance (SE), it does not follow that she is surprised by who actually danced (WE). The surprise may be directed exclusively at the non-dancers.

The literature offers different views on the available interpretations of wh-interrogatives under know. In [13], all embedded wh-interrogatives denote propositions inducing a full partition of the entire logical space. In this partitioning question semantics, all embedded wh-interrogatives are predicted to be strongly exhaustive. For [17], the SE-reading with interrogatives under know follows from the lexical partitioning semantics of the matrix predicate, such that (1) will be true iff Mary knows the complete answer to who was at the party, and that this is the complete answer. Differences aside, both accounts only predict SE-readings for know + wh. This strong position is problematic on at least two counts: Firstly, whereas SE-readings are indeed prominent with know, other embedding verbs such as predict, tell, or announce allow for weaker interpretations, which cannot be modelled in a partition semantics [4, 17]. Secondly, recent experimental work has found the weaker IE/WE-readings (i.e. to know the complete positive answer and nothing more) to be readily available with an acceptability rate of >90% even with English know and French savoir ‘know’ [6, 8]. In addition, [7] provide experimental evidence for both WE_nondist and SE_nondist-readings with interrogatives under surprise. The experiments in [6] and [7] involved picture matching and acceptability judgments with an external, participant-centered perspective.

Novel Experimental Evidence.

In two experiments with novel setups, we were able to replicate to some extent these findings for German wissen ‘know’ and überraschen ‘surprise’. In a contradiction experiment [10], we tested for the obligatoriness of SE-readings with wissen ‘know’ predominantly from the internal perspective of the attitude holder. Participants had to judge the contradictoriness of discourse sequences, such as (4) (in italics), in which the SE-reading is explicitly negated in the final clause.

figure c

If only SE-readings were available under wissen, such sequences should be systematically judged as contradictory. Conversely, if participants judge them as non-contradictory, this constitutes evidence for the WE/IE-reading. The results show that more than 25% of all cases were judged as non-contradictory, indicating that WE/IE-readings are available to some extent.

The second experiment was carried out for a range of matrix predicates in German, including wissen ‘know’ and überraschen ‘surprise’ [11]. Target sentences were objects of bets, and compensation was performance based, so that participants were actively engaged through a financial incentive. Again, the linguistic items and contexts were designed such that target sentences had to be judged from the internal perspective of the attitude holder, while the external perspective of the addressee had to be taken into account as well. This design targets the optimal reading from a communication-oriented perspective; see [11] for details on the experimental setup. The descriptive results for the two predicates of interest are as follows: For wissen, there was evidence for a WE/IE-reading in 46% of all cases, as opposed to a ceiling 100% for SE. For überraschen, there was evidence for the two WE-readings (WE_dist: 100%, WE_nondist: 96%), but, interestingly, also for the SE_nondist-reading at a robust level of 58%. The availability of SE_nondist will play a crucial role in the analysis of überraschen in Sect. 4.

Previous Analyses of Flexible SE/WE-Interpretations.

There is ample evidence from introspection and experiments that the interpretation of wh-interrogatives is flexible between SE and WE under wissen ‘know’, and variable between three surface interpretations under überraschen ‘surprise’. The literature offers different ways to account for this flexibility, with different sources for the observed variability in exhaustive force. [3] derive the variability from two covert answer operators ANS1 (giving rise to WE) and ANS2 (deriving SE), which both operate on an unconstrained interpretation of the interrogative in terms of Hamblin-alternatives [16]. [22] postulate covert EXH-operators either in the embedded interrogative (deriving SE) or in the matrix clause (deriving WE/IE). [40] derives an IE-interpretation as the only available semantic reading by placing covert EXH in the matrix clause. SE-readings are derived as a pragmatic enrichment via a hearer-based (excluded middle) competence assumption. Finally, [39] posit a lexical ambiguity in the attitude verb know as expressing an internal perspective (SE-reading) or external perspective (WE/IE-reading), respectively. Our analysis of know + wh in Sect. 3 will incorporate core ingredients from the last two accounts.

2.2 The Interpretive Effect of SCHON…Aber: Implicature Blocking

According to [42], the German discourse particle schon ‘alright’ is a modal comparative degree operator that commits the speaker to the truth of the prejacent proposition p, after weighing the circumstantial evidence in favor of p against the evidence for its polar counterpart . In general, the presence of schon indicates that there may be some reason to doubt the validity of p. Because it expresses polar comparison, schon, and accented SCHON in particular, are commonly found in verum focus contexts [18]. In combination with the (implicit) concessive particle aber ‘but’ in a subsequent clause, accented SCHON has an additional effect on interpretation: It consists in the blocking of pragmatic implicatures based on prototypicality or relevance.Footnote 3 Consider (5A), which gives rise to the relevance-based implicature that Levan is not hungry in the absence of SCHON. With SCHON, this implicature is blocked. Likewise, B’s implicit question in (6) is whether she can get petrol, so that A’s response without SCHON would give rise to the relevance-based conversational implicature that the petrol station is open and sells petrol. This implicature is blocked in the presence of SCHON, thereby indicating that the implicit question is answered in the negative: no petrol available.

figure e

Crucially, SCHON…aber does not block scalar implicatures. In (7), its presence does not rescue the impending contradiction between the implicature (not all) and its contradiction in the subsequent clause (all).

figure f

We speculate that the insensitivity to scalar implicatures follows from the polar comparative nature of modal SCHON [42], and from the fact that the scalar alternative (C ate all the cakes) logically entails the literal meaning p (C ate some cakes): Adding implicature-blocking SCHON to a proposition p normally constitutes a reason for doubting p, but in the case of the scalar not-all implicature in (7) the validity of the unblocked alternative (all the cakes) casts no doubt on the entailed p (some cakes). For this reason, the presence of SCHON is unmotivated as there is no contradiction.

Moreover, modal SCHON does not resolve lexical ambiguities, as shown for the German homonym Bank (‘bench’ or ‘bank’) in (8). (8) can only be understood in jest ( ) as a play of words, i.e., at a meta-linguistic level.

figure h

Finally, SCHON…aber does not affect truth-conditional semantic content. Its presence in (9) does not lead to a rejection of the claim that at least five beers were drunk:

figure i

The insensitivity of SCHON…aber to semantic content will play an important role in our semantic analysis of wh-interrogatives under know and surprise. In particular, we can conclude that any inferences blocked by the presence of SCHON…aber are not semantic entailments, but mere pragmatic implicatures triggered by considerations of prototypicality or relevance. Pragmatically, the presence of SCHON…aber indicates that a prototypical default does not obtain, which in turn casts doubt on the truth of the prejacent p by the semantic meaning of SCHON as a modal degree operator.

2.3 The Meaning of Teilweise ‘Partially’: Quantifying Over Pluralities

In contrast to SCHON…aber, the Q-adverb teilweise ‘partially, in parts’ is a quantificational modifier operating on truth-conditional semantic content. For the purposes of this paper, there are three important aspects to the meaning of teilweise:

Firstly, teilweise affects the truth-conditions. Whereas Nino must have eaten all of the (contextually salient) Khachapuris for (10) to be true, (11) will already be true if Nino ate only a subset of them. More generally, sentences with teilweise are true if a subpart of the theme-related eventualities in question are instantiated.

figure j

Secondly, we assume that teilweise only excludes maximal eventualities in the pragmatics, as (11) will also be true in situations in which Nino ate all of the Khachapuris. Having said this, we concede that it is quite misleading to use teilweise in a situation in which Nino ate all of the Khachapuris, but there is good evidence for the assumption that the partiality associated with teilweise is a pragmatic effect. (12) will be true if Ninos granny is pleased with Nino eating some or (even better!) all of the Khachapuris. Moreover, there is a clear contrast between teilweise vs. nur teilweise, as shown in (13).

figure k

Thirdly, teilweise only operates on pluralities of discrete eventualities, which must be tied to atomic entities in the individual domain, as expressed by plural count NPs. As a result, teilweise in (14) cannot be used to express that Nino ate only part of the soup, unlike the part-whole modifier zum Teil ‘in part’. The only felicitous reading of (14) is one in which Nino ate the soup in discrete portions (possibly together with others).

figure l

(11) and (14) show that teilweise is not lexically connected to question embedding, but to plural eventualities. Most importantly, all these requirements can only be fulfilled if the atomic pluralities are targeted by teilweise in the process of semantic composition.

There are several conceivable ways to implement the semantics of teilweise. An obvious possibility would be to follow [4] or [24] in assuming that teilweise is a run-of-the-mill adverbial quantifier that takes individuals, propositions, or eventualities as its arguments. An alternative would be to implement teilweise as a quantifier that takes a plurality as argument and returns a part of that plurality for the further compositional procedure, [2]. Here, we opt for an event-semantic analysis, though, in which teilweise operates on mereological part-whole structures. Providing arguments in support of our analysis goes beyond the scope of this paper. For its core arguments, not much hinges on the particular choice of analysis for teilweise, as long as it accounts for the three main empirical observations. For this reason, the present analysis should be considered a mere handy tool for formally implementing the essential points viz. the semantics of embedded questions. For the same reason, we refrain from a detailed compositional analysis. Most of what follows could be restated in any analysis that takes questions to denote a Hamblin/Karttunen style set of alternatives.

For explicitness, we analyze the Q-adverb teilweise as a quantificational part-whole modifier of a verbal projection that operates over plural mereological sub-event structures. We assume that a clause with teilweise will be true iff there exists some sub-event e of a complex plural event e’. In (11), this plural event is the maximal eating event of a contextually given maximal set of khachapuris, which is formally derived by the sum-formation operator ⨁. The Neo-Davidsonian event-semantic representation of (11) is shown in (15), with \(TH\) = Theme; see also [2, 26], i.a.

figure m

To conclude, the meaning parts targeted by teilweise constitute semantic content proper. Any inferences that are not affected by teilweise must be considered pragmatic implicatures. In Sect. 3, we will employ this diagnostic to show that SE-readings with know + wh must be pragmatic implicatures, and not semantic entailments!

3 Wissen ‘Know’ + Wh: Data and Analysis

This section presents novel empirical data on the interpretation of wh-interrogatives embedded under wissen ‘know’. In Sect. 3.1, we present evidence from the interpretation of such interrogatives in combination with SCHON…aber and with teilweise that shows that their basic semantic interpretation is the WE-reading. We will put forward an event-based semantic analysis of know + wh in Sect. 3.2. The SE-reading, in turn, is not an independent semantic reading, but derived from the WE-reading by way of pragmatic enrichment. Our analysis in Sect. 3.3 will take up ideas by [40] and [39], but we will put the ingredients together in a different manner.

3.1 Novel Evidence on Know + Wh: IE is Semantic, but SE is Pragmatic!

Looking first at the interpretive effect of teilweise, we find that this Q-adverb only ranges over the positive alternatives in the question, i.e. the complete set of true answers constituting the WE-reading [19]. The semantic effect of teilweise is to turn this WE-interpretation into a non-exhaustive question interpretation. Consider (16) and recall from Sect. 2.3 that teilweise only operates on truth-functional semantic content. (16) will be true if Nino knows for only part of the dancers that they danced, i.e., her knowledge is non-exhaustive regarding the WE-interpretation. As a result, the follow-up in (16a) is licit. Crucially, the alternative follow-up in (16b), in which Nino’s knowledge is shown to be incomplete regarding the entire answer space including negative answers (= SE), is NOT felicitous. But it should be if SE-readings were bona fide semantic entailments, thus making (16) semantically ambiguous. The infelicity of (16b) thereby constitutes negative evidence against the analysis of SE as a semantic entailment.

figure n

Next, consider the effect of SCHON…aber in (17). Here, the particle combination indicates that the SE-inference blocked. This is compatible with the felicitous follow-up in (17b), which is directed at the negative answer space (= part of the SE-denotation), and which improves significantly in the presence of SCHON…aber as opposed to its counterpart without. In contrast, as SCHON…aber cannot operate on the semantic content of the clause, cf. (9), it cannot be used to turn the underlying WE-reading into a non-exhaustive reading, viz. the infelicity of (17a), which marks Nino’s knowledge as incomplete regarding the WE-denotation.

figure o

In sum, the infelicity of (17a) constitutes negative evidence that the WE-reading is the underlying semantic interpretation of know + wh, whereas the felicity of (17b) constitutes positive evidence that SE is a mere pragmatic implicature. The data in (18) and (19) illustrate the same point (follow-ups in English for reasons of space):

figure p

3.2 An Event-Semantic Analysis of WE-Readings with Wissen ‘Know’

In our event-semantic account of the basic semantic WE-reading of know + wh, completeness of the answer is aspectually derived via event summation. We suggest the lexical entry in (20) for wissen ‘know’, using event composition with knowledge events and content arguments, as suggested by [29] and [26]. According to (20), for x to know (the answer to) Q means that x is in an attitudinal state e that is composed of the maximal sum of K(nowledge) substates \(e^{\prime}\) that have the individual positive answers p to Q as their content.Footnote 4

figure q

We also assume that the denotation Q of wh-interrogative clauses is the set of Hamblin-alternatives [16]. Given the veridicality and factivity of the knowledge attitude, we moreover assume that only true propositions can be known, i.e., that only true propositions in w can form the content of a knowledge eventuality; in other words \(K_{w} \left( {e^{\prime}} \right) \wedge Content_{w} \left( {e^{\prime},p} \right)\) can only be true iff \(w \in p\). Finally, we assume that the ⨁-operator is part of the lexical aspect of wissen, making e the maximal possible knowledge eventuality concerning the question Q. This derives weak exhaustiveness for Q as an aspectual phenomenon, thereby eliminating the need for a covert ANS-operator [3, 17]: ⨁ sums in e the sub-states of knowledge of all true propositions in Q. For (1), this results in an event predication over the stative eventuality of x knowing the complete list of dancers, or rather the complete list of true propositions of the form y danced, as shown in (21) for the world of evaluation w. Further application of (21) to the denotation of Nino and subsequent existential closure over events will yield the complete meaning of (1).

figure r

The analysis in (20) and (21) directly extends to know + that when that-CPs are modelled as singleton sets of sets of worlds (〈〈s, t〉, t〉) [5]. Notice, too, that the event maximality imposed by ⨁ makes the eventuality bounded, which explains the old puzzle of why stative verbs of knowledge are crosslinguistically marked as perfective/telic, such as e.g. in Finnish [21] or in Hausa [27].

Applying the Q-adverb teilweise ‘partially’ to (21), and following the logic from Sect. 2.3, we derive the meaning of (18) in (22). (22) specifies a sub-event e of the maximal knowledge eventuality e’ regarding the question Who danced?, and x is the attitude holder of this knowledge sub-eventuality e.

figure s

Feeding in the subject meaning and existential closure over events yields the correct meaning for (18). In sum, combining teilweise and know + wh results in a non-exhaustive semantic interpretation. We turn to pragmatic strengthening from WE to SE next.

3.3 Pragmatics: Strengthening to SE

As mentioned in Sect. 2.1, the SE-reading of know + wh does not only entail knowledge of the complete answer to the question, but also the knowledge that this is the complete answer [17]. In other words, to know-SE entails not only that the attitude holder knows the complete answer, but also that she knows that this is the complete answer, cf. [17]. In the event-semantic reformulation of know + wh in (23), this is represented in terms of two conjoined knowledge eventualities, where the second eventuality e’’ captures the missing component that turns the formula into a valid representation of SE-knowledge.

figure t

In view of the evidence for the blocking of pragmatic implicatures and SE-readings with SCHON…aber presented in Sect. 2.3 and Sect. 3.1, we propose to analyze the strengthened SE-reading in (23) as a pragmatic enrichment of (22). This enrichment follows from a hearer-based pragmatic preference for interpreting 3rd person attitude reports from the internal 1st person-perspective of the attitude holder. To capture this preference, we propose the novel general pragmatic principle PARV in (24).

figure u

With PARV, the SE-reading of (1) (Nino weiß, wer getanzt hat ‘Nino knows who danced’) is derived from its underlying semantic WE-interpretation in (21) by the defeasible assumption that Nino is able to confirm (1) by uttering (25), i.e. that she knows that she knows the WE-reading, and not just part of it. Crucially, such 1st person knowledge reports are always SE, as is evidenced by the infelicity of the subsequent follow-up, which contradicts the 1st person SE-knowledge. The obligatory SE-construal with 1st person attitude reports follows from the fact that the reporting 1st person attitude holder must know that the summed (WE) knowledge eventuality is the complete knowledge state regarding Q, for else she cannot rule out that her knowledge is incomplete. In the formula in (24), this is captured in the occurrence of the second event e’’.

figure v

Notice that (24) is mute on negative embedders, such as keine Ahnung haben ‘be unaware’ in (26), in which case the speaker cannot commit to the embedded content. Such predicates trigger logical scale reversal, such that the SE-interpretation is no longer an independent and logically stronger entailment, but rather entailed by semantic WE. If Nino is already unaware of the complete list of dancers in w (WE), it follows that she is also unaware of the complete list of dancers and non-dancers (SE).

figure w

Although not semantic in nature, the PARV-driven SE-reading is the default surface interpretation of know + wh in the absence of further evidence, which makes it difficult to cancel in the absence of context information or explicit discourse marking. This is evidenced, for instance, by the fact that SE-violations with know + wh were rated as contradictory in almost 75% of all cases in the contradiction experiment in [10] reported in Sect. 2.1. However, same as other prototypicality-based implicatures (Sect. 2.2), the default pragmatic SE-enrichment can be blocked by the particle combination SCHON…aber, as illustrated in (17) in Sect. 3.1.

More generally, PARV captures the implicit hearer-based assumption that attitude holders will normally be reported to have an attitude X if they are de se aware of having X. In such cases, they could explicitly commit to X in the form of a 1st person report. Presumably, the PARV-driven preference for evaluating attitude reports from the internal perspective of the attitude holder is due to the fact that attitudes are mental objects located in the holder’s mind, for which the best or most reliable kind of evidence is a commitment by the attitude holder in the form of a 1st person report. If so, PARV would be connected to more general cognitive mechanisms associated with Theory of Mind [30]. Importantly, PARV in (24) is best considered a general interpretive principle that is not tied to questions per se, but which is also active, for instance, in the resolution of de re/dicto-ambiguities: In full parallel to SE-readings with know + wh, DPs contained in 3rd person attitude reports receive de dicto readings by default, and they must be de dicto in 1st person reports, cf. (27). In particular, the de re reading of (27a) is verified by a situation in which Rico owns a ruby which he falsely believes to be a worthless glass stone. The speaker may use the term a ruby to refer to that ruby, and correctly report that Rico knows that he owns that object. Crucially, Rico cannot report of himself that he owns a ruby, as long as he is not aware of the fact that this stone is in fact a ruby, cf. (27b). The contrast can be replicated with definite descriptions, too.

figure x

Likewise, de se-pronouns as commonly found with logophoric construals [20, 28] are also tied to the internal perspective of the attitude holder. Given these observations, the pragmatic SE-enrichment with wh-interrogatives under know appears to be just another instance of perspective-dependent interpretation in natural language.

Our proposal to derive SE-readings by way of pragmatic enrichment is similar in spirit to the account in [40], but it differs in how the enrichment is triggered. [40] derives SE-readings from IE-readings via a hearer-based (excluded middle) competence assumption (CA). However, this is problematic, as the exact content of CA is unclear. On the formulation in (28a), CA is already equivalent to the SE-reading of (1), resulting in circularity. The formulation in (28b) does not generalize to other SE-compatible verbs, such as the verbs of saying predict or tell, as predictions or statements do not follow from beliefs. Another issue with the analysis of [40] is that it assumes a neg-raising like property of predicates like know,Footnote 5 even though no evidence exists to this assumption – in fact, it seems that exactly the contrary is supported by facts.

figure y

This being said, there are some valid concerns as to whether PARV can also handle speech act verbs correctly. After all, PARV is limited to verbs of propositional attitude. For speech act verbs, it no longer holds true that the main evidence for their truth is in fact in the mind of the subject, as speech act verbs have public effects. But then again, (i.) speech act verbs tend to have less of a bias towards SE-readings; (ii.) even with speech act verbs it is essential what the subject meant when making her utterance, cf. the de re/de dicto ambiguities in (29); and (iii.) there are no sufficient empirical data for teasing apart the attitude component and the quotational aspects of speech act verbs [37] as would be necessary for an in-depth evaluation of PARV.

figure z

In deriving SE-readings as an effect of assuming an internal perspective, we adopt a core idea of [39], first traces of which are already found in [13]. [39] also link the weaker WE- (for them: IE) and the SE-reading to the external and internal perspectives of speaker and attitude holder, respectively. They do so, however, by treating the attitude predicate know as semantically ambiguous between [± internal perspective]. Their account in terms of a lexical ambiguity clashes with the above argument against semantic SE, though, and in particular with the observation that SCHON…aber cannot be exploited for disambiguating semantic ambiguities, cf. (8). Moreover, the availability of WE- and SE-readings with other question-embedding verbs (predict, tell…) [6, 11, 22]) would necessitate the assumption of a systematic WE/SE-ambiguity in the lexicon of such verbs, an undesirable consequence. In view of these findings, and given the observable parallels to other perspective-dependent phenomena in 1st and 3rd person reports, we consider our pragmatic account superior.

3.4 Conclusion on Wissen ‘know’ + Wh

In this section, we presented two novel diagnostics shedding light on the underlying semantic interpretation of wh-interrogatives under the veridical and homogeneous/distributive attitude verb wissen ‘know’. The combination of such interrogatives with the Q-adverb teilweise ‘partially’ and the particle combination SCHON…aber ‘alright…but’ shows that their underlying semantic interpretation is WE, whereas the SE-reading is a pragmatic enrichment. In Sect. 3.3, we argued that this pragmatic enrichment is triggered by a default tendency to interpret 3rd person attitude reports from the attitude holder’s 1st person internal perspective. We also suggested that the same enrichment process is at work in the derivation of de dicto readings and logophoricity effects.

4 Überraschen ‘Surprise’ + Wh: Data and Analysis

This section presents novel empirical data on the interpretation of wh-interrogatives embedded under the cognitive-emotive attitude verb überraschen ‘surprise’. In Sect. 4.1, we consider the interpretation of surprise + wh in combination with SCHON…aber and teilweise. Our findings provide novel evidence for the claim in [34, 35], and [14] that they come with a fairly weak non-distributive, or non-homogeneous semantic WE_non-dist interpretation, which can be pragmatically strengthened to WE_dist. Again, such pragmatic strengthening is blocked in the presence of SCHON…aber. Sect. 4.2 discusses the interpretation of surprise + wh from a theoretical perspective. We discuss a shortcoming of the existential WE-interpretation à la [14], and we end by sketching a tentative analysis of überraschen ‘surprise’ and other cognitive-emotive attitude verbs as denoting a cognitive-emotive attitude towards a fact, or a proposition-dependent or proposition-exemplifying situation à la [12, 25], and [1].

4.1 Novel Evidence: WE_nondist is Semantic, but WE_dist is Pragmatic!

Recall from Sect. 2.1 that wh-interrogatives under überraschen ‘surprise’ allow for two WE-construals of different logical strength. In (2), the attitude holder Nino may be surprised by each and every individual in the positive answer space of dancers (= WE_dist). Alternatively, she may be surprised by just some of the dancers (WE_nondist), cf. [14, 34]. WE_dist logically entails WE_nondist.

If we add the concessive particle combination SCHON…aber, we find that it blocks the logically stronger WE_dist interpretation, which involves surprise at each individual answer. This is evidenced by the felicitous follow-up in (30a) vs. (30b), in which the presence of SCHON…aber does not serve to cancel a semantic entailment.

figure aa

Secondly, the Q-adverb teilweise ‘partially’ is difficult to interpret with surprise + wh, if not outright degraded, in the absence of other suitable plural expressions, cf. (31). This combination is also not readily attested in corpora:

figure ab

As teilweise operates over plural events only, cf. (14), it is conceivable that the deviant status of (31) is due to the absence of such event pluralities with surprise + wh.

4.2 Towards a Non-propositional Analysis of Surprise + Wh

A classic way of deriving WE_nondist-readings for surprise + wh would consist in adopting an existential analysis with weak exhaustive force à la [34, 35], and [14]. Überraschen ‘surprise’ would take the WE-set of minimal (believed to be) true answers Q as its complement and map these to true iff there is at least one proposition p in this set such that the attitude holder did not expect this proposition to be true in w, cf. [14]:

figure ac

Pragmatic strengthening to WE_dist would formally amount to replacing the existential quantifier in (32) with the universal quantifier. Informally, such pragmatic strengthening is licit as the strengthened readings still entail the truth of the underlying semantic entailment. They just depict particular ways of making (32) true. This is entirely parallel to what we find in the domain of adnominal quantifier scope in (33), in which the surface ∀∃-reading (all the students watched a movie) can be pragmatically strengthened to an inverse ∃∀-pseudoscope reading (there is a movie that all the students watched), which is again just a specific way of making the semantic ∀∃-reading true [31]:

figure ad

Finally, the deviant status of (31) with teilweise may simply follow from semantic redundancy, as the underlying WE_nondist semantics already captures the incompleteness or subpart requirement of teilweise.

Alternatively, the deviant status of (31) may also follow from the inability of teilweise to access the subparts of individual situations with complex non-atomic substructure [23]. And indeed, there is some reason to believe so, as surprise can also give rise to SE_nondist-readings, which are not accounted for at all on the WE-analysis in (32) [7, 9, 11]. There is indeed some experimental evidence that the target of the surprise in cases of SE_nondist is not from the set of positive true answers that are accessed in (32). For illustration, consider the following example from the betting experiment [9, 11]. In the betting experiment, participants could decide to cash in a betting slip, or not, depending on how they interpreted the meaning of a sentence with a wh-interrogative embedded under surprise, cf. (34a). The truth-value judgment underlying participants’ choices is made on the basis of a 1st person report of the attitude holder (here: Tiffany), cf. (34b), and of information about the circumstantial facts, cf. (34c).

figure ae

In the setting in (34bc), the surprise of Mary is directed at the negative answer space: What is unexpected is that Carlo and Sophie did NOT eat the grasshopper. Crucially, the WE-based lexical entry for surprise in (32) predicts the bet to be false in this SE_nondist-setting, so that participants should not cash it in. This prediction stands in stark contrast to participants’ behavior, who opted for cashing in in 58% of all cases, where cashing in is equivalent to judging (34a) true in the SE_nondist setting (34bc).

The availability of SE_nondist readings for surprise + wh casts some serious doubt on the adequacy of the WE-meaning representation in (32). For this reason, we would like to raise the possibility that überraschen ‘surprise’, and other cognitive-emotive attitude verbs, such as be glad, be happy, be worried etc., differ from know (and other epistemic attitude verbs) in a more fundamental way. Following [12], we would like to propose that such predicates do not select for a set of propositions (a question meaning), but rather for – what [12] call – a fact, or an exemplified or situated proposition [1, 25]. On this line of thought, the attitude of surprise may be conceptualized as a psychological state that is caused by potentially complex situations and their overall constitution or make-up, including missing subparts.Footnote 6 Put differently, we think of the meaning of surprise and of other emotive-cognitive factives as lexically decomposable into a causing eventuality and a primitive emotional state (here: surprisal) caused by the eventuality.

It is important to see that this means that the actual states of surprisal or happiness or worry etc. are primitive neuropsychological or emotional states, as typically assumed in language processing [15]. They are not phenomenologically intentional in that they do not have a propositional attitude argument. The impression of intentionality, i.e., the directedness towards a proposition or situation, is the result of associating the causing propositional attitude or cognitive attitude towards a situation with the resulting state. The general pattern for the meaning of cognitive-emotive factives is formally captured in (35), where the causing stimulus s could stand for a situation or a fact; see above.

figure af

The surprisal is then not caused by a belief in the truth of a proposition, but more directly by becoming acquainted with some situation or fact. In this vein, surprise can also be triggered non-verbally, e.g., by the content of pictures and photographs, or by the absence of content on such pictures, which are visual representations of complex situations. The famous picture of Lenin giving a speech in front of a revolutionary crowd in Sverdlov Square, Moscow, which was later purged of Trotzki’s presence, constitutes a striking example of surprise by the absence of content. As a result, there are different ways of making (36) true:

figure ag

There are other kinds of evidence pointing towards a different semantic status of epistemic and cognitive-emotive attitude verbs. Surprise can take situation-referring DPs or depictive DPs as arguments (37a), whereas wissen ‘know’ cannot (37b).

figure ah

Secondly, the situation argument is directly expressed with the mandatory pronoun es ‘it’ with überraschen in (38a), whereas such a pronominal reference is at best optional with wissen ‘know’ in (38b).

figure ai

The empirical differences in (37) and (38) motivate a different semantic analysis for überraschen and other cognitive-emotive verbs in which they do not operate directly on the propositional content of the wh-interrogative. Following ideas in [12], and in particular [1] on the cognitive-emotive attitude predicate interesting, überraschen ‘surprise’ can be analyzed as directly selecting for a situation s such that s is a stimulus situation or fact that is part of a larger situation s’ that (fully) resolves the wh-interrogative meaning Q, and s causes a surprisal e of x in w, as tentatively shown in (39). For a situation to resolve a wh-question meaning, the situation must contain sufficient information for allowing at least for a partial answer to the wh-question.

figure aj

Importantly, our theory of surprise naturally predicts that surprise has both a stative and an achievement reading, as shown in (40). For the stative reading (40a), the aspectual modification targets the resulting surprisal state whereas the achievement reading (40b) focuses on the causation event.

figure ak

Given that a situation can cause surprisal by its size or by its general make-up or constitution [38], the denotation in (39) is general enough to be compatible with WE_dist, WE_nondist and SE_nondist readings alike. In the default case, this underspecified interpretation will be pragmatically enriched to the strongest logical reading, namely WE_dist, which expresses surprisal at all relevant subparts of the situation. Same as with wissen ‘know’, such pragmatic enrichment is blocked in the presence of SCHON…aber. Finally, the Q-adverb teilweise can only operate on semantically plural sums of eventualities, but not on the internal subparts (or lumps, [23]) of a complex situation, cf. the soup-eating situation by Nino in (14) above. This accounts for the observed infelicity of teilweise in combination with surprise + wh, where the surprise is directed at a complex situation. In order to express partial surprise, i.e., surprise at the subparts of a complex situation, we require the part-whole modifier zum Teil ‘in part’, which CAN operate on the material subparts of individual situations:

figure al

We postpone a more detailed situation-based analysis of überraschen ‘surprise’ to another occasion, and we conclude by pointing the interested reader to a recent analysis in [25] of depictive verbs like imagine as taking proposition-dependent situations as complements. As imagine can select for wh-interrogatives, too, it is tempting to aim at a unified analysis of different situation-selecting attitude verbs.

5 Conclusions and Theoretical Implications

In this paper, we investigated the interpretation of wh-interrogative clauses embedded under the attitude predicates wissen ‘know’ and überraschen ‘surprise’ in interaction with the particle combination SCHON…aber ‘alright…but’ and the Q-adverb teilweise ‘partially’. We have shown that SCHON…aber does not operate on semantic content, but rather blocks the emergence of pragmatic implicatures based on considerations of relevance or prototypicality. The Q-adverb teilweise, by contrast, operates on semantic content by presenting an event as a mereological subpart of some plural sum event. Applying these novel empirical diagnostics to know + wh, we found that SE-inferences with know + wh are pragmatic in nature, whereas the logical weaker WE-inferences are semantic in nature. Applying the same diagnostics to surprise + wh, we found that the WE_dist reading under surprise is pragmatic and the result of default pragmatic strengthening. We also saw that the existence of both WE_dist and WE_nondist readings with surprise is accounted for on an existential WE-analysis à la [19] and [14], but the unexpected emergence of SE_nondist-readings is not! This led us to tentatively propose a fact- or situation-based reanalysis of cognitive-emotive attitude verbs like überraschen ‘surprise’ à la [12], on which the denotation of surprise does not operate on a set of propositions, i.e. the set of true answers in w, but on a fact that is situated or exemplified by the Karttunen-meaning of the wh-interrogative.

The general theoretical repercussions of our endeavor are as follows. We have presented novel empirical evidence that the meaning of embedded wh-interrogatives is indeed underspecified in the form of a set of Hamblin-alternatives, cf. [3]. Moreover, the observation that there is no inherent distributivity or homogeneity component built into the meaning of such wh-clauses argues against the obligatory presence of a max-operator in wh-clauses, pace [33]. Likewise, we have argued that the exhaustivity effects frequently observed with embedded questions are not located in the denotation of the wh-interrogatives themselves, for instance in the form of covert ANS(wer)- or EXH-operators. Instead, they follow from the aspectual semantics of the embedding attitude predicates. As a result, some attitude verbs such as cognitive-emotive surprise only come with very weak exhaustivity requirements, whereas the complete WE-interpretation with epistemic know is the result of sum formation over knowledge sub-events. The corresponding SE-inferences are not semantically derived. Finally, we tentatively suggested that cognitive-emotive attitude verbs may express a relation not to sets of propositions, but to proposition-dependent situations or facts, which may also be expressed in the form of plain nominal DPs.