The interaction of discourse markers and prosody in rhetorical questions in German

Abstract

Recent research has shown that rhetorical questions (RQs) have certain prosodic characteristics in terms of voice quality, tempo, and intonation, which distinguish them from genuine, information-seeking questions (ISQs). This paper focuses on the interaction between prosodic cues to rhetorical meaning on the one hand, and lexical and morpho-syntactic means, on the other, in German. The production experiment reported on here addresses three research questions, in short: (i) do speakers prefer a specific syntactic construction for an RQ, (ii) do they make use of specific lexical and morpho-syntactic means to signal rhetorical meaning, and (iii) what is the interaction between those means and prosodic cues. The answers are: (i) yes (wh-questions), (ii) yes (especially discourse markers (DiPs)), and (iii) we find an additive effect enforcing the rhetorical message. When lexical (or morpho-syntactic) cues to rhetorical meaning are used, we do not observe a reduction in or lack of prosodic means at the same time. For example, when a DiP is present, an RQ will still have a typical nuclear accent and edge tone, i.e., cues are used in an additive, rather than an exclusive way. There are, however, RQs that are marked only in the prosody, without any lexical or morpho-syntactic cues present.

1. Introduction

This paper studies the interplay between the prosodic and the morpho-syntactic and lexical make-up of rhetorical questions (RQs) and information-seeking questions (ISQs) in German. Specifically, we address the questions of which lexical and morpho-syntactic means speakers use to signal whether a question is rhetorical or information-seeking (Research Questions I and II, see below), and whether the prosody of RQs remains the same regardless of the simultaneous use of lexical or morpho-syntactic cues (Research Question III). The current section briefly surveys semantic/pragmatic, lexical, morpho-syntactic, and prosodic characteristics of RQs (in comparison to ISQs) with a focus on German and formulates the research questions addressed by the experiment reported on in the next section.

RQs are interrogatives that do not elicit an answer. Instead, the answer is already entailed in the discourse and thus obvious to and known by all discourse participants, including the speaker (Caponigro & Sprouse 2007; Biezma & Rawlins 2017); this includes a possible accommodation of the common ground by the interlocutors in cases where the answer is not known by all addressees. The purpose of an RQ is to make a point and to extract commitment from the interlocutors (Biezma & Rawlins 2017). A genuine ISQ, in contrast, is used to elicit information, i.e., to receive an answer that resolves a knowledge gap on the part of the speaker. Lexically as well as morpho-syntactically, German RQs may be identical to ISQs, both within the interrogative mode (Altmann 1987, 1993). They may generally occur as verb-first (V1) interrogative sentences (polar questions, see (1)), wh-interrogatives (constituent questions, see (2)), or other interrogative structures. Semantically, the meaning of both ISQs and RQs is defined by the set of propositions that pose possible answers to the question, with one proposition being its true answer (e.g., Hamblin 1973; Karttunen 1977). This means that both polar and wh-questions may have either an RQ or an ISQ interpretation, depending on their situational (and linguistic) context.

The rhetorical meaning of a question is thus not incorporated in its morpho-syntactic make-up. Instead, rhetoricity is added on the pragmatic level (see Brandt et al. 1993). Pragmatically, RQs are commonly described to behave similar to assertions (e.g., Bartels 1999; Han 2002) or have an assertive-like feel or function (Caponigro & Sprouse 2007; Biezma & Rawlins 2017). However, while ISQs and RQs may be string-identical, there are certain lexical and morpho-syntactic means that facilitate RQ interpretation. These include negative polarity items (NPIs; e.g., zur Hölle ‘to the hell’: Wer zur Hölle soll das alles essen? ‘Who the hell should eat all this?’; einen Finger krümmen, ‘lift a finger’: Hat er je einen Finger gekrümmt um dir zu helfen? ‘Did he ever lift a finger to help you?’) and discourse particles (DiPs). Relevant DiPs include schon (Zaefferer 1984; Thurmair 1989; Bayer & Obenauer 2011; lit.: temporal ‘already’) and auch (Thurmair 1989; lit.: additive ‘also’). In their non-literal meaning, certain DiPs have been shown to trigger an RQ interpretation (Zaefferer 1984; Meibauer 1986; Brandt et al. 1993), e.g., Wer zahlt schon gerne Steuern? ‘Who likes to pay taxes?’ (Bayer & Obenauer 2011: 457), Wer geht auch sonntags einkaufen? ‘Who goes shopping on a Sunday, after all?’. ² Morpho-syntactic means to facilitate RQ interpretation further include the combination of the imperfect subjunctive and the negation nicht (‘not’), see (3), from Grésillon (1980: 284), and comparative structures as in (4), from Meibauer (1986: 142).

Naturally, since ISQs and RQs may be string-identical, the occurrence of lexical or morpho-syntactic means including DiPs is not obligatory in German RQs. Along with lexical and morpho-syntactic means, prosody distinguishes RQs from ISQs (Braun et al. 2019; Braun et al. 2020 for German; see the cross-linguistic overview in Dehé et al. 2022). Relevant prosodic parameters include temporal cues (constituent duration, speaking rate), voice quality, and intonational markers (nuclear accents, edge tones). Specifically, (i) constituent duration is longer and speaking rate slower in RQs than in ISQs, (ii) sentence-initially there is more breathy voice quality in RQs than in ISQs, and (iii) RQs and ISQs differ in terms of the intonational contour. Polar RQs often terminate in a mid-high plateau contour (H−% in GToBI), i.e., the nuclear part of the intonation contour rises to a mid-high level where it stays until the end of the utterance without rising further (see Figure 1). ³ Polar ISQs, however, typically have a high-rising edge tone (H−^H%), i.e., the contour rises continually until the end of the utterance. ⁴ Wh-RQs almost exclusively terminate in a fall (L−%; see Figure 2), whereas a large number of wh-ISQs also allow rising movements. In terms of nuclear accents, wh-RQs often have a specific rising pitch accent, (L+H)*, where both the low (L) and the high (H) targets are aligned within the accented syllable; see Figures 1, 2. (L+H)* has not been observed in ISQs at all. Wh-ISQs, however, mainly have nuclear L+H*, in which L is aligned either with the pre-nuclear syllable or at the very beginning of the stressed syllable. See Braun et al. (2019) and Zahner-Ritter et al. (2022) for details and further illustrations. ⁵

Figure 1 Spectrogram and f0 contour of the polar RQ Denkst du etwa ich studiere Algebra? (‘Do you really think I study algebra?’) with an H−% edge tone, produced by a female speaker.

Figure 2 Spectrogram and f0 of the wh-RQ Wer kennt denn schon Mangold? (‘Who is likely to know mangel?’) with an (L+H)* nuclear accent (target syllable: man), produced by a female speaker.

While Braun et al.’s (2019) results stem from a carefully designed production experiment, they receive support from German spontaneous speech data (Braun et al. 2020). Moreover, Braun et al.’s (2020) results already suggest that the use of lexical material signalling rhetorical interpretation does not go along with a reduction or loss of prosodic cues, but that instead different cues co-occur in the same utterance. To consolidate these results, we designed the experiment reported on here. The experimental materials are borrowed from Braun et al. (2019), but unlike in their production experiment, participants in the present study were not provided with complete sentences on the screen, but were asked to combine sentence fragments to full grammatical utterances. They were allowed to add material not given as sentence fragments as well as to choose the syntactic make-up of their utterances. Of the prosodic parameters tested in Braun et al. (2019) and related work on the prosody of RQs, only intonation (edge tones and nuclear accents) and speaking rate were included in the present study, because the quality of the sound files elicited in the online setting did often not allow for an analysis of voice quality. We address the following research questions:

1. I. Do speakers prefer a specific syntactic structure (e.g., wh, polar) when producing RQs in German?

2. II. Do speakers use specific lexical or morpho-syntactic means to highlight the rhetorical meaning of the question?

3. III. Does the use of lexical and/or morpho-syntactic cues to rhetorical meaning go along with the use of prosodic cues, or does the use of prosody become less important or even redundant? If the latter, are all prosodic parameters affected in the same way?

2. The experiment

In this section, we report on the production experiment designed to address the research questions given above.

2.1. Methodology

The material consisted of 44 context situations triggering either a rhetorical or an information-seeking reading of the questions (Table 1; cf. Braun et al. 2019). Contexts for ISQs were created such that the answers to the questions were unknown by the speaker (i.e., they were not obvious from the context) and that the speaker’s intention was to elicit information. Contexts for RQs entailed the answer, i.e., the answer was obvious to all discourse participants and thus known by the speaker (i.e., the participant of the experiment). Contexts were the same for polar and wh-questions.

Table 1 Example contexts and sentence fragments presented on screen.

Target questions in Braun et al. (2019) had the following form: (i) polar questions: verb DiP(denn) subject(anyone) object, e.g., Mag denn jemand Limonen? ‘Does DiP anybody like limes?’; (ii) wh-questions: wh-word(wer ‘who’) verb DiP(denn) object, e.g., Wer mag denn Limonen ‘Who likes DiP limes?’. The current experiment set out to test which syntactic structures participants would use if entire sentences were not available and if participants were free in their syntactic choice, thus sentence fragments were used, which participants had to combine to form complete grammatical questions (see Table 1). ⁶ The sentence fragments for half of the material (corresponding to polar questions in Braun et al. 2019) were a verbal infinitive (e.g., mögen ‘like’) and the object noun (e.g., Garnelen ‘shrimps’). For the other half (corresponding to wh-questions in Braun et al. 2019), the wh-word wer ‘who’ was also given. Participants were free to add more material to their utterances and while wh-words were given in half of the stimuli, participants were generally free to choose their preferred syntactic structure for their utterances. Fragments were thus used to be maximally explorative with respect to the use of syntactic structure and DiPs. At the same time, we used the same materials as in Braun et al. (2019), including the verb along with the object noun and the wh-word for half of the materials for the two studies (Braun et al. 2019 and the current one) to be maximally comparable.

Participants were told that the goal of the study was to understand the structure of German RQs in comparison to ISQs, but they were not primed with respect to syntactic structure (polar, wh, other). A short explanation and examples of RQs (The speaker does not expect an answer because she believes it to be obvious) and ISQs (The speaker expects an answer because she does not know it) were provided before the experiment started to ensure that participants knew the difference between the two. The study was web-based and conducted using the platform SoSci Survey (https://www.soscisurvey.de; Leiner 2019). During the experiment, participants were presented with 44 test items: 22 verb/noun and 22 wh-word/verb/noun. Participants were asked to use the sentence fragments displayed on the screen but were allowed to add as much linguistic material as they needed to formulate a question – RQ or ISQ – as they thought was appropriate given the context. There were four experimental lists, each of which contained 44 items. The items were pseudo-randomized such that there were at least four other items between two items of the same pair (i.e., between ISQ and RQ containing the same object noun). The lists were balanced such that each context was used for stimuli with and without wh-word with equal frequency. They were randomly assigned to participants by the SoSci Survey platform. Participants used their own equipment. Their productions were recorded in .wave format using the built-in microphone of their laptops/headsets.

2.2. Data treatment and analyses

The data set contained 748 utterances produced by 17 speakers. A total of 93 utterances had to be excluded due to technical issues. The remaining 655 utterances were analysed using Praat (Boersma & Weenink 2018). All utterances were orthographically transcribed on a tier of the respective TextGrid by student assistants who were native speakers of German. On another tier, the transcribers broadly classified the sentence type (e.g., wh, polar) and annotated the occurrence of DiPs. There were 471 wh-questions, 171 polar questions, 12 V2-structures, and one alternative question. The syntactic structure was later analysed in a more fine-grained way by the second and third authors. Each utterance was phonologically annotated by the second author following GToBI. Speaking rate was operationalized as the assumed number of syllables per second (disregarding phonological processes, e.g., word-final schwa elision), with the actual duration of the sentence as the frame of reference, to which the number of syllables for each sentence was set in relation.

For the statistical analysis of the occurrence of DiPs (DiP or no DiP), we ran a general linear mixed effect regression model with illocution type (RQ and ISQ) and syntax (sentence type) as fixed factors, and participants and items as crossed random effects. The variable syntax consisted of only two levels (wh, polar), i.e., only those items entered the statistical analysis. As shown in the subsequent section, some structural variation was observed within polar and wh-questions (e.g., extended wh-phrase, embedded clauses), however, without changing the syntactic type. The binary partition into wh- and polar questions without further divisions kept the statistical models manageable. The twelve V2 utterances and one alternative question were excluded for the following reasons, which may result in Type2 errors (Columb & Atkinson 2016): (i) their infrequency in the data set, (ii) the fact that they were exclusively realized within one illocution type (all twelve V2 sentences were rhetorical), and (iii) the rare occurrence of DiPs (N = 2). Models with and without random slopes were compared using the anova() function in R-studio. If the model fit improved, random slopes were added and kept in the model (Bates et al. 2015; Matuschek et al. 2017). In backward selection, non-significant (p > 0.05) interactions were removed from the models if this did not deteriorate the fit of the models.

For the statistical analysis of the effect of DiPs on speaking rate, we ran a linear mixed effect regression model with illocution type (RQ vs. ISQ), the occurrence of DiPs (DiP vs. no DiP) and syntax (wh vs. polar) as fixed factors and participants and items as crossed-random factors.

All p-values were calculated using the Satterthwaite approximation in the R package lmerTest (Kuznetsova, Brockoff & Christensen 2017) and were adjusted by means of the Benjamini–Hochberg correction (Benjamini & Hochberg 1995).

2.3. Results

2.3.1. Question match and sentence structure

The contexts and sentence fragments presented to participants suggested the production of polar and wh-questions. However, participants were free to use other syntactic structures too, and they did (see Figure 3). Figure 3 is divided by stimuli without wh-word and stimuli with wh-word. Participants generally preferred wh-questions (N = 471, 72% of utterances) over polar questions (N = 171, 26% of utterances). This tendency was observed for RQs in particular, which were predominantly realized as wh-interrogatives. As Figure 3 shows, if no wh-word was given, ISQs were in most cases (N = 178, 80%), realized as polar questions, but RQs were mostly realized as wh-questions (N = 130, 80%). In contrast, when a wh-word was given, utterances were almost exclusively produced as wh-questions. Within RQs, 165 (out of 168, 98%) of targets were realized as wh-questions. Within ISQs, 155 (96%) of the 162 targets were produced with wh-syntax. Twelve utterances, all RQs with no wh-fragment, were produced with V2-declarative syntax (2%); one ISQ with no wh-fragment was produced as an alternative question; these are summarized as ‘other’ in Figure 3.

Figure 3 Produced syntactic structure; left panel: stimuli presented without wh-fragment, right panel: stimuli presented with wh-fragment; white: ISQs, grey: RQs.

Figure 4 shows lexical and morpho-syntactic details within the four produced syntactic types. Explanations and examples for each pattern listed on the y-axis are given in the Appendix; the most frequent ones are addressed here. Within produced polar questions, the preferred syntactic structure for an ISQ was a simple V1-interrogative without any additional lexical material or further syntactic complexity (N = 58, 43%; ‘polar’ in Figure 4; see (1)). The second most frequent realization of polar ISQs (N = 41, 31%) had an extended subject pronoun (jemand von euch/einer von euch ‘anybody/one of you’; ‘polar complex’ in Figure 4). The most frequent realization of polar RQs (N = 10, 28%) was a V1-interrogative with the intended RQ as an embedded clause (‘polar, sub cl’ in Figure 4). The second most frequent polar RQ realization was a V1-interrogative containing a DiP (‘polar DiP’ in Figure 4; N = 8, 22%) and a simple polar question (‘polar’, N = 8, 22%), i.e., the same that was the most frequent one for polar ISQs.

Figure 4 Details of syntactic structures across illocution types (ISQs white, RQ grey) in (from left to right) produced polar questions, wh-questions, V2 constructions, alternative question; syntactic patterns are explained in the Appendix.

Within wh-questions, participants preferred simple wh-structures without any additional lexical material when producing an ISQ (‘wh’ in Figure 4; N = 77, 42%; see (2)). The second most frequent wh-ISQ structure had an extended wh-phrase (‘wh, complex’ in Figure 4; N = 68, 36%, e.g., Wer von euch ‘Who of you’). By contrast, the most common wh-RQ realization contained a DiP (‘wh, DiP’ in Figure 4; N = 202, 71%). Within V2 constructions, all rhetorical, the most frequent structure was a simple V2 sentence with a finite main verb (‘V2’ in Figure 4; N = 9, 69%). The alternative question did not contain a DiP.

2.3.2. Discourse particles (DiPs)

The most frequent non-prosodic means to signal RQs were DiPs. In twelve cases, participants also used initial discourse markers, such as Hallo (‘Hello’; Hallo, wer isst denn schon Rosenkohl? ‘Hello, who eats DiP DiP brussels sprouts?’) or interjections such as Ieh (‘Ugh’; Ieh, wer mag schon Lebertran? ‘Ugh, who likes DiP cod-liver oil?’). Nine of the twelve were RQs. The three ISQs were produced in one case with an initial Entschuldigung ‘excuse me’ and in two cases with So ‘so’. Comparative structures or structures with the combination of subjunctive and negation as described in Section 1 (see (3), (4)) did not occur.

Overall, 358 utterances contained a DiP and 296 did not. The statistical analysis did not yield an interaction between illocution type (RQ vs. ISQ) and syntactic realization (wh vs. polar) (p = 0.6), i.e., their joint effect is not greater or smaller than possible main effects. For example, it is not the case that wh-RQs show stronger effects than polar RQs (or vice versa). Two main effects were found: (i) RQs contained DiPs significantly more often than ISQs (RQ: N = 260 vs. ISQ: N = 36; ß = 4.80, SE = 0.66, z = 7.25, p < 0.001), and (ii) wh-questions showed a significantly higher number of DiPs than polar questions (wh: N = 280 vs. polar: N = 14; ß = 2.84, SE = 0.41, z = 7.0, p < 0.001). Of all wh-RQs, 247 (87%) contained a DiP and 36 (13%) did not. A total of 33 (18%) wh-ISQs contained a DiP and 155 (82%) did not. Of all polar RQs, 11 (31%) were produced with a DiP and 25 (69%) without, and only three polar ISQs (2%) had a DiP while 132 (98%) did not. Two (17%) of the twelve V2 sentences used as RQs contained a DiP.

Figure 5 shows the distribution of DiPs for wh- and polar ISQs and RQs. The most common DiP in wh-RQs was denn (N = 98, 35%). Other common DiPs in wh-RQs were denn in combination with another DiP (‘denn x’ in Figure 5); e.g., denn schon, denn auch (N = 66, 23%) and schon on its own (N = 66, 23%). The most common DiP in polar RQs was etwa (N = 9, 25%). The most common DiP in wh-ISQs was denn (N = 27, 14%). Denn was also the only DiP occurring in polar ISQs (N = 3, 2%). Two V2 sentences contained doch.

Figure 5 Distribution of DiPs in produced polar questions (left panel), V2 sentences (center panel), and wh-questions (right panel); white: ISQ, grey: RQs; x represents any second DiP in a sequence.

Since DiPs turned out to be the most (and, in our data, only) frequent non-prosodic marker of rhetorical interpretation, we investigate the relation between their occurrence and prosodic cues.

2.3.3. Prosodic parameters

In this section, we report the results for each prosodic parameter in turn. For edge tones and nuclear accents results are reported by DiPs.

2.3.3.1. Speaking rate

For the variable speaking rate, there was no significant interaction between the predictor variables illocution type (ISQ vs. RQ), syntax (wh vs. polar) and occurrence of DiP (DiP vs. no DiP) (p = 0.93). However, there were main effects for illocution type and syntax. The average speaking rate at which RQs were produced (4.7 syllables per second) was significantly slower than that of ISQs (5.2 syllables per second) (ß = −0.47, SE = 0.07, t = −6.36, p < 0.001). Regarding syntax, the average speaking rate at which wh-questions were produced (4.9 syllables per second) was significantly slower than that of polar questions (5.3 syllables per second) (ß = −0.27, SE = 0.09, t = −2.89, p < 0.05). The presence of DiPs in any of the syntax/illocution types increased the speaking rate, except for wh-ISQs. On average, questions without DiP had a slower speaking rate (4.8 syllables per second) than questions with DiP (5.2 syllables per second). V2-utterances without DiP had a slower average speaking rate (4.4 syllables per second) than those with DiP (5.1 syllables per second). However, the effect of DiP was not significant (p > 0.05). Figure 6 shows the average speaking rate for polar and wh-RQs and ISQs with and without DiP.

Figure 6 Average speaking rate in produced polar and wh-RQs (gray) and ISQs (white) with and without DiP; left two panels: polar questions; right two panels: wh-questions.

2.3.3.2. DiPs and edge tones

Polar ISQs were predominantly produced with the high rise H−^H% (N = 97, 89%), polar RQs with the mid-high plateau H−% (N = 17, 47%), wh-ISQs with H−^H% (N = 95, 51%), and wh-RQs with low terminus L−% (N = 268, 95%). See Figure 7 for edge tones in relation to DiPs. We present the results for each DiP separately.

Figure 7 Distribution of DiPs and edge tones (!H−%: red, H−%: olive green, H−^H%: green, L−%: blue, L−H%: pink) in RQs and ISQs. Top: polar questions; bottom: wh-questions, right: ISQs; left: RQs.

none (no DiP): Polar questions typically did not contain a DiP (N = 157, 92%). Polar ISQs without DiP were mostly realized with a high rise H−^H% (N = 95, 72%) followed by low-rise L−H% (N = 20, 15%), fall L−% (N = 12, 9%), and plateau H−% (N = 5, 5%). Polar RQs without DiP most often ended in H−% (N = 10, 40%), followed by L−% (N = 5, 20%) and L−H% (N = 5, 20%). Three polar RQs without DiP ended in H−^H% (12%) and two in a down-stepped plateau !H−% (8%). Of all wh-questions, 191 (41%) did not contain a DiP. Wh-ISQs without DiP mostly terminated in H−^H% (N = 93, 60%), followed by L−% (N = 39, 25%), L−H% (N = 15, 9%), H−% (N = 6, 4%), and !H−% (N = 3, 2%). Within wh-RQs, 29 utterances without DiP were realized with L−% (81%) and seven (19%) with H−^H%.

denn: Of the three polar ISQs which contained denn, two terminated in H−^H% (67%) and one in L−H% (33%). There were no polar RQs with denn. Wh-ISQs with denn were mostly realized with L−H% (N = 14, 52%), followed by L−% (N = 11, 40%), H−^H% (N = 1, 4%), and H−% (N = 1, 4%). Wh-RQs with denn were mainly realized with L−% (N = 98, 98%). Two wh-RQs with denn terminated in H−^H% (2%).

denn x (denn + additional DiP): This combination of DiPs only occurred in wh-RQs (N = 66, e.g., Wer tanzt denn bitte Lambada? Lit.: Who dances DiP DiP lambada ‘Oh please, who dances lambada?’). All occurrences terminated in a low edge tone (L−%).

etwa: This DiP only occurred in polar RQs (N = 9; e.g., Magst du etwa Kamille? Lit.: Like you DiP chamomile ‘Do you really like chamomile?’). Five of them (56%) terminated in H−% (see Figure 1), three in H−^H% (33%), and one in L−H% (11%).

schon: The DiP schon only occurred in wh-RQs (N = 66, e.g., Wer mag schon Sellerie? Lit.: Who likes DiP celery ‘Who likes celery?’). Of these 66, 63 (95%) terminated in L−% (see Figure 2), two (3%) in H−% and one (2%) in L−H%.

other: Several other DiPs occurred only sporadically, such as, for instance, überhaupt (lit.: actually, at all), jetzt (lit. now), and wohl (lit. possibly, perhaps). They were subsumed under the category ‘other’. Within polar questions, two polar RQs contained ‘other’ DiPs, and both terminated in H−%. Within wh-questions, five wh-ISQs contained other DiPs; four of them (80%) terminated in L−% and one (20%) in H−^H%. Other DiPs were most frequent in wh-RQs, where they mostly co-occurred with an L−% edge tone (N = 14, 82%). Two wh-RQs terminated in H−^H% (12%), and one in H−% (6%).

2.3.3.3. DiPs and nuclear accents

The most common nuclear accents in polar ISQs were L* (N = 94, 71%) and L+H* (N = 20, 18%), in polar RQs it was (L+H)* (N = 19, 53%). Wh-ISQs also were mostly realized with L* (N = 94, 50%) and L+H* (N = 33, 16%). Wh-RQs were mostly realized with (L+H)* (N = 166, 59%; see Figure 2) and L+H* (N = 44, 16%). See Figure 8 for nuclear accents in relation to DiPs.

Figure 8 Distribution of nuclear accents ((L+H)*: red, H*: orange, H+!H: olive green, H+L*:green, L*: turquoise, L*+H: blue, L+!H*: purple, L+H*: pink) in RQs and ISQs, top: polar questions; bottom: wh-questions, right: ISQs; left: RQs.

none: Polar ISQs without DiP often had a L* nuclear accent (N = 94, 71%), followed by L+H* (N = 20, 15%), L*+H (N = 9, 7%), L+!H* (N = 3, 6%), H* (N = 3, 6%) and (L+H)* (N = 3, 6%). Polar RQs without DiP were most often produced with (L+H)* (N = 13, 52%), followed by L* (N = 5, 20%), L*+H (N = 4, 16%), L+!H* (N = 2, 8%) and H* (N = 1, 4%). Within wh-questions, wh-ISQs without DiP were mainly realized with L* (N = 92, 59%), followed by L+H* (N = 25, 15%), H+!H* (N = 15, 10%), H* (N = 6, 4%), (L+H)* (N = 4, 3%), L*+H (N = 4, 3%) and L+!H* (N = 4, 3%). Wh-RQs without DiP most often contained (L+H)* nuclear accents (N = 21, 58%), followed by L* (N = 6, 17%), L+!H* (N = 3, 8%), and L+H* (N = 2, 6%), H* (N = 1, 3%), H+!H* (N = 1, 3%), H+L* (N = 1, 3%), and L*+H (N = 1, 3%).

denn: Within polar questions, denn only occurred in three polar ISQs; two of them had a L* nuclear accent and one a down-stepped rising accent L+!H*. Within wh-questions, 27 wh-ISQs and 98 wh-RQs contained denn. The most frequent nuclear accent in wh-ISQs with denn was L+!H* (N = 8, 37%), followed by L+H* (N = 8, 30%), (L+H)* (N = 4, 14%), H* (N = 3, 11%), H+L* (N = 1, 4%), and L*+H (N = 1, 4%). Wh-RQs with denn were mostly realized with (L+H)* (N = 58, 59%), followed by L+H* (N = 18, 18%), L+!H* (N = 11, 11%), H* (N = 5, 5%), L*+H (N = 2, 2%), L* (N = 2, 2%), and H+!H* (N = 2, 2%).

denn x (denn + additional DiP): The combination of denn with another DiP did not occur in polar questions and not in wh-ISQs, but it did in 66 wh-RQs. Of these, 40 (61%) were realized with (L+H)*, followed by L+H* (N = 11, 17%), H* (N = 5, 8%). There were four instances of L*+H and four of L+!H* (6% each) and single cases of H+!H* and H+L*.

etwa: The DiP etwa only occurred in polar RQs (N = 9). Five of them (56%) were realized with the (L+H)* nuclear accent (see Figure 1), followed by L* (N = 3, 33%) and L+!H* (N = 1, 11%).

schon: The DiP schon only occurred in wh-RQs (N = 65). The respective utterances were mostly realized with (L+H)* (N = 40, 62%; see Figure 2), followed by L+H* (N = 10, 15%), L+!H* (N = 6, 9%), L*+H (N = 4, 6%), H* (N = 3, 5%), and H+L* (N = 2, 3%).

other: There were two polar RQs with DiPs that were allotted to ‘other’, occurring with (L+H)* and L*+H, respectively. Regarding wh-questions, the five ISQs in this category were H+!H* (N = 2), L* (N = 2), and H+L* (N = 1). RQs realized with other DiPs mostly had an (L+H)* accent (N = 7, 41%), followed by L+H* (N = 3, 18%), L* (N = 3, 18%), L+!H* (N = 2, 12%), H+L* (N =, 1 6%), and H+!H* (N = 1, 6%).

3. Discussion

The current study set out to investigate the relation between lexical and morpho-syntactic means to signal rhetorical meaning, on the one hand, and prosodic means, on the other. The current section discusses the research questions given at the end of Section 1 in turn against the background of the experimental results.

Research Question I is concerned with whether or not speakers prefer a specific syntactic construction when producing RQs. The answer is that they do. Our results show that when speakers are free to choose the morpho-syntactic form and lexical material (here, only infinitives, object nouns, and wh-words were presented as sentence fragments instead of giving a full utterance, lexical additions and any syntactic structure were allowed), they prefer wh-questions to polar questions (as well as to any other syntactic structure) for RQs. This preference can possibly be accounted for by the presence of the wh-word as a segment fragment, but this explanation would only account for one half of the stimuli. The assumption of a general bias toward wh-questions triggered by the presence of the wh-fragment would not account for the fact that when a wh-word was not among sentence fragments in the stimuli, only RQs are to a high extent produced as wh-questions, but not ISQs. Indeed, 80% of RQ targets without wh-fragment were also produced as wh-questions, in contrast to 20% of ISQ-targets without wh-fragment. We do therefore not assume a general bias of the occurrence of the wh-word, but instead conclude that when speakers have the choice (and in the experimental setting only little lexical material was given), they prefer wh-interrogative syntax for RQs but polar interrogative syntax for ISQs. Interestingly, twelve RQs without wh-fragment in the stimulus were produced not as polar or wh-questions, but with V2 declarative surface syntax, a syntactic form that did not occur in any of the other conditions. These twelve cases deserve mention because of the pragmatic closeness of RQs and declaratives (or: assertions). The use of a V2 structure may thus be due to the illocution of an RQ, which is essentially assertive (e.g., Meibauer 1986; Han 2002). The ultimate goal of an RQ is that its propositional content becomes part of the common ground of all discourse participants (Caponigro & Sprouse 2007; Biezma & Rawlins 2017). In that regard, RQs have a high resemblance to (V2) assertions (see Farkas & Bruce 2010 for assertions). Hence, it is conceivable that in some cases speakers opted for a V2 construction, which is commonly used for assertions (e.g., Glück 2010), when intending rhetorical meaning, instead of an interrogative form. The fact that V2 syntax was never observed for RQ stimuli with wh-fragments is perhaps a consequence of the presence of the wh-word: participants did not ignore these wh-words but formed wh-questions almost throughout. The fact that there were only twelve instances of V2 may well be a task effect, since instructions only introduced RQs and ISQs (i.e., only interrogative structures). Against this background, we cannot exclude that some of the V2 sentences were intended as V2 interrogatives rather than declaratives, thus not assertions but non-neutral questions of some kind (Gunlogson 2003; Krifka 2011). However, while we have evidence for the affinity of RQs to assertions, we have to leave the issue of potential V2 interrogatives with rhetorical meaning to future research. In any case, the occurrence of V2 syntax is a unique feature of (polar) RQs in our data, marking a possible syntactic difference between rhetorical and information-seeking meaning. Whether V2 (declarative or interrogative) would be more frequent in another experimental setting and how common it is in spontaneous speech is another interesting question for future research.

Syntactic differences between ISQs and RQs were also observed within question types. Within polar questions, most ISQs were simple questions, or simple questions with an extended pronominal subject (together 75% of all polar ISQs), while RQs were simple in this sense only in 22% of the cases and made use of embedded clauses as well as DiPs. This takes us to Research Question II, which asked whether speakers use specific lexical or morpho-syntactic means to signal rhetorical meaning. Again, the answer is clearly yes, especially as far as DiPs are concerned. DiPs occurred significantly more frequently in RQs than in ISQs. Given that the provided sentence fragments were the same for ISQs and RQs, we take this result as strong evidence that DiPs, particularly denn + another DiP, etwa and schon, provide an important lexical cue to rhetorical meaning. The most frequent DiP, denn, occurred in both ISQs and RQs and in both polar and wh-questions, although it did not occur in polar RQs. We therefore consider denn neutral regarding illocution type, which fits in with prior research on DiPs in RQs stating that denn does not favour one or the other reading of a question (Thurmair 1989; Viesel & Freitag 2019) and is equally possible in both interrogative sentence types (Theiler 2021). However, we found that DiPs etwa, denn + x and schon occurred in RQs, only, and we therefore consider them markers of rhetorical meaning. This is in line with the literature on use of DiPs to signal rhetorical meaning, remember Bayer & Obenauer’s (2011) example of schon given in Section 1 (see also Zaefferer 1984; Meibauer 1986).

Other lexical or morpho-syntactic means, such as the use of the subjunctive in combination with negation, comparative structures, or NPIs, which have been argued to occur with RQs (see Section 1), were non-existent in the current data set. This may be an effect of the experimental environment or their general infrequency and markedness in contrast to DiPs. In the experimental environment, participants added some lexical material, but they did not complicate things by using non-canonical sentence structures or negate sentences. Different methodology, including corpus analyses, will shed more light on this issue in future research. In any case, since DiPs were frequently added especially to RQs, even though they were nowhere given as sentence fragments, we conclude that they are the main lexical means of choice for speakers to mark a question as rhetorical.

Research Question III was concerned with whether or not the use of lexical and/or morpho-syntactic cues to rhetorical meaning has an effect on the prosodic realization of RQs. Previous research on the prosody of German RQs has shown that in comparison to ISQs, RQs show certain distinct prosodic features: slower speaking rates, frequent occurrence of the rising-falling pitch accent (L+H)*, an almost obligatory fall to L−% in wh-RQs, and frequent occurrence of the H−% mid-high plateau edge tone in polar RQs (Braun et al. 2019; Braun et al. 2020). The current data fully supports these observations. First, RQs were generally produced with slower speaking rates than ISQs. Second, we predominantly find high-rising edge tones (H−^H%) with polar ISQs, but H−% with polar RQs and L−% in wh-RQs, all irrespectively of the presence of a DiP. And third, we found the typical rhetorical (L+H)* nuclear accent in RQs.

So how do lexical and morpho-syntactic means interact with prosody? Given our data, we focus on the interaction of DiPs with prosodic cues. First, regarding speaking rate, RQs were slower than ISQs, and wh-questions were slower than polar questions. There is a tendency for the presence of DiPs to increase the speaking rate. However, this tendency did not reach significance and, more importantly, there was no interaction with any of the other factors, including illocution type. In other words, the effect holds for RQs and ISQs alike. It is therefore not possible to conclude, for example, that the presence of a DiP would reduce the necessity of a slower speaking rate in RQs because a DiP already signals rhetorical meaning. Moreover, the tendency applies to all DiPs, including denn, which is neutral with respect to illocution type. Relating to Research Question III, we therefore conclude that speaking rate may be affected by the presence of a DiP, but not in any illocution type specific way.

The second prosodic parameter we looked at was edge tones (Figure 7). Here we compare RQs with ISQs, as well as different DiPs within RQs, because given that DiP denn is neutral in terms of illocution type, we cannot generalize across all DiPs. Comparing polar ISQs with polar RQs, they do not share a DiP; we observe denn in ISQs, but etwa and ‘other’ DiPs in RQs. Within polar RQs, none (i.e., no DiP) is more frequent than the two occurring DiPs etwa and ‘other’ DiPs. Polar RQs without DiPs show the greatest variety with respect to edge tones, while the two polar RQs containing ‘other’ DiPs both ended in the plateau H−% typical of polar RQs, and the percentage of H−% is also higher for polar RQs with etwa than with no DiP. While numbers are low, if anything, the presence of a DiP in polar RQs reinforced the use of the RQ-typical utterance-final plateau.

Wh-RQs mostly ended in L−%, as opposed to more variation in wh-ISQs (in line with Braun et al. 2019; Braun et al. 2020). This pattern is the same across DiPs. Within wh-RQs, we do not observe meaningful differences between utterances containing neutral denn, on the one hand, and utterances containing DiPs that only occur in RQs, i.e., denn x and schon, on the other. Again, the predominant edge tone is L−%, suggesting no effect of the DiP such that a DiP signalling rhetorical meaning would allow more variation in terms of edge tones. Consequently, like with polar RQs, we find simultaneous rhetorical marking by both DiP and edge tone.

The third prosodic parameter was the use of nuclear accents in relation to the use of DiPs (Figure 8), again comparing RQs and ISQs, but also DiPs within RQs, for the same reasons as for edge tones. Like for edge tones, no comparison is possible between polar ISQs and polar RQs, because they do not share specific DiPs. Nuclear accents in utterances without DiPs (‘none’ in Figure 8) are in line with earlier results for the prosody of German RQs (Braun et al. 2019; Braun et al. 2020), in particular, we observe the occurrence of the (L+H)* nuclear accent in more than 50% of polar RQs. Since neutral denn does not occur in our set of polar RQs, no comparison is possible within polar RQs between denn and other DiPs. What we do observe is the occurrence of (L+H)* to approximately the same extent as in RQs without DiP, thus we conclude that the presence of a DiP signalling rhetorical meaning (etwa in particular) does not prevent speakers from also using the nuclear accent typical of RQs, again suggesting independent contributions of lexicon and prosody.

Comparing wh-ISQs with wh-RQs, again we clearly find more use of the nuclear accent typical of RQs across question types, (L+H)*, irrespective of the presence or absence of a DiP, and also, within wh-RQs, irrespective of which DiP was used. So, as for polar questions, we conclude that the use of DiP does not alter the prosodic pattern.

Taken together, and in reply to the second part of Research Question III, we conclude that all types of prosodic cues to RQs under scrutiny here (speaking rate, edge tones, nuclear accents) are used regardless of the presence of DiPs already signalling rhetorical meaning. This may be interpreted as an additive effect, which may suggest reinforcement of the rhetorical message: the more markers, from more than one area of the grammar, the clearer the rhetorical meaning. Whether this is the case or not will best be tested in a perception experiment including all versus some or none of these parameters. Alternatively, lexicon and prosody (as well as syntax) may all be considered independent of one another. Prosody may simply not be ‘interested’ in whether an RQ is already marked as such by a lexical marker, or it may not have access to that kind of information because the different areas of the grammar proceed in parallel.

Future research will shed more light on these open questions. In any case, we can safely conclude that there is no such thing as ‘reduction of prosody’ or ‘lack of prosodic marking’ of an RQ due to the presence of cues to rhetorical meaning from other areas of the grammar.

4. Conclusion

Participants preferred wh-questions over polar questions when producing RQs and used certain DiPs to lexically signal rhetorical meaning. Prosodically, RQs were found to show typical features already identified in previous research. There was no interaction between prosodic and lexical features of RQs, meaning that the different means did not replace one other or make each another superfluous. Instead, speakers combined DiPs with prosodic features to mark rhetorical meaning. Whether or not RQs that combine both strategies are perceived as stronger or as less ambiguous in meaning than RQs with only lexical or only prosodic marking will have to be investigated in future perception experiments.