4.2. Materials and Design

The target sentences in the experiment were German transitive verb-first structures. They consisted of a monosyllabic finite auxiliary verb, a monosyllabic subject d-pronoun, a disyllabic gradable adjective or quantifier modifying the object, the disyllabic object, and a trisyllabic non-finite lexical verb with stress on the second syllable:

The target sentences were embedded in a dialogue between two speakers (see (3)) so that the context and the punctuation disambiguated the speech act, and the context disambiguated the three IS conditions: New, contrastive or given information. In the new condition, speaker 1 introduces a hypernym of the object of the target utterance (i.e., in (3) domestic animals as a hypernym of llamas). Speaker 2 mentions the object (only) in the target sentence. In the given condition, speaker 1 introduces the object occurring in the target sentence by using the same expression as in the target utterance (i.e., there is across-speaker repetition of lexical material). In the contrastive condition, speaker 1 introduces a focus alternative to the object of the target sentence (i.e., guinea pigs in (3)). Speaker 2 mentions this alternative, brushes it aside in their first sentence(s), and then utters the target sentence.²

Concerning the new condition, we note that elements for which a hypernym has been mentioned in the context have been considered to be accessible rather than new (Baumann & Riester, 2013). However, as has also been observed (Baumann et al., 2015), the hypernym-hyponym (superset-set) relation is sensitive to the distinction between referent vs. lexical item. In a dialogue like A: Do you like animals? B: I like all dogs., dogs cannot be deaccented because we cannot identify dog referents on the basis of an animal referent superset. Thus, referentially, ‘dogs’ are not accessible; dogs is only lexically a hyponym of animals. In comparison, when a hyponym denotes a referent that has been introduced otherwise, there can be deaccentuation. Consider the following sequence from Baumann & Grice (2006): Ole was a talented sportsman. He was well-known in his region. The local press praised the tennis player. Observe that tennis player, which is a hyponym of sportsman, can be deaccented. However, also observe that the referent of tennis player has been introduced by the name Ole, which really is the trigger for the deaccentuation (Baumann & Grice, 2006). In our materials, the referents of the object noun are not identifiable on the basis of the superset (or otherwise). This is especially true because the superset always is very large. For this reason, we are speaking of new information focus. Future research must show if there are prosodic differences between completely ‘unprepared’ information (which is an option we did not choose, to ensure coherence), and information for which there is a referential superset in the context.

The experimental materials consisted of nine items in 2 × 3 conditions each, for a total of 54 recordings per participant.

The experiment also contained nine filler items in three pseudo-conditions each.³ These fillers were included as distractors but we also used them as controls when some of the results for the experimental items turned out to be unexpected. Therefore, we describe the fillers here in detail and discuss some of the prosodic characteristics of the filler sentences produced by the participants in Section 6, where we discuss the experimental results. The fillers were string-identical in the two recording sessions but represented a different speech act in each session. In session 1 (testing polar exclamatives), they were negative declarative questions, see (4). In session 2 (testing polar questions), they were negative declarative sentences used as rejections, see (5). Thus, across experimental items and fillers, each session contained a balanced number of questions and non-questions.

The entire lexical content of the fillers was given. They could thus serve as a comparison for the accentuation of given material in two more non-assertive speech acts.

All fillers began with Der hat keine… ‘He has no…’, followed by the direct object and the non-finite lexical verb. The metrical structure and segmental make-up were less controlled than in the experimental items.

4.3 Participants

The experiment had 27 participants (mean age: 21.8, range: 18–28). All participants were female native speakers of German who were recruited from the student population of the University of Cologne, and were either paid or received course credit.⁴ The study was run in the XLinC Lab Cologne under the laboratory ethics approval #2016-09E2-200213 by the German Linguistic Society. Participants gave written consent.

4.4 Procedure

The recordings were made in two sessions to reduce repetitiveness and the duration of the experimental session(s). The sessions were separated by at least six days. In session 1, the 27 exclamatives were recorded, in session 2 the 27 questions were recorded.

The participants took the role of speaker 2 in the dialogs illustrated in Section 4.2. The dialogs were presented via headphones and through accompanying text on a computer monitor. Each dialog was preceded by a caption (e.g., A discussion at the institute for biology at a university), with a photograph of this setting. Then the dialog between the two speakers, who were represented by silhouettes, began. The part of speaker 1 was presented in a speech bubble and also played through headphones. After a button press, the participants read the part of speaker 2 in another speech bubble. Another button press started the recording. Participants could repeat their utterances until they were satisfied with them. Another button press stopped the recording and advanced to the next dialog. The dialogs were presented in four pseudo-random orders. Every participant saw and recorded every dialog in every condition. We used Presentation (Neurobehavioral Systems. https://neurobs.com/) for the presentation of the stimuli and the recordings, which were made at a sampling rate of 44.1 kHz and a bit depth of 16 bits, using an omnidirectional DPA 4060 microphone.

4.5 Dependent variables

We investigated the following variables on the syllable level: Mean F0, maximum F0, minimum F0, F0 excursion (F0_max – F0_min), duration (log-transformed), and intensity. F0 was measured in semitones relative to 1 Hz. Across variables involving F0, we only compared accented syllables of the same accent types.

On the utterance level, we investigated speech rate, pitch excursion, mean pitch, and intensity (mean and range). We operationalized speech rate as the number of (underlying) syllables per second (i.e., nine syllables divided by the total duration in seconds).

As categorical variables, we investigated accent placement, accent type, prominence levels of accents, and boundary tones. Many syllables were accented either so often or so rarely that there were issues with the model fit. We will note these issues below as they come up. The accent types that we investigated were the GToBI accents (Grice et al., 2005). The prominence levels that we investigated are those of the DIMA scheme (Kügler et al., 2015, 2019): Level 1 for weak prominences (post-focal prominences, rhythmic accents and phrase accents), level 2 for strong prominences (nuclear accents), and level 3 for extra-strong prominences (emphatic realizations of nuclear accents).

4.6 Data preparation

Recordings containing slips of the tongue or deviations from the lexical materials were excluded from the analysis (30 exclamatives and 11 questions out of a total of 1458 recordings). One participant’s recordings of the questions were discarded because of a technical error. This left 699 exclamatives and 691 questions for analysis.

Large outliers in any of the dependent variables were corrected manually if they represented measurement or annotation errors, or were left in the data if they represented real outliers.

The recordings were annotated in Praat (Boersma & Weenink, 2021) by two trained research assistants. We used a modified version of the DIMA scheme (Kügler et al., 2015, 2019),⁵ using the following tiers: Syllable boundaries, GToBI tones, prominence levels, final boundary tone. We did not use a phrase boundary tier, since the target utterances were expected to constitute single intonational phrases, which indeed they did. After an initial round of annotation, the research assistants checked each other’s annotations and reached a consensus for conflicts in the annotations.

Pitch was sampled every 10 ms. We manually inspected and corrected the pitch tracks produced by Praat, in case there were octave jump errors or spurious voicing. The summarizing statistics were calculated directly in R (R Core Team, 2021), using R package rPraat (Boril & Skarnitzl, 2016).

There were two practical challenges regarding syllabification. First, some items contained sequences of liquids and/or nasals that resulted in the frequent reduction of the number of realized syllables (e.g., /ŋən/ → [ŋː]). When this happened, we annotated the unrealized syllable such that it only covered the final period of the nasal. The resulting bimodal distribution of syllable durations allowed us to treat unrealized syllables (and syllables preceding them) differently. Second, potentially ambisyllabic consonants (e.g., /m/ in /hʊmɐ/), were treated as the onset of the subsequent syllable: /hʊ.mɐ/, to ensure annotation consensus.

5.1. Utterance-level measures

Figure 1 shows the means of the utterance-level measures.

Figure 1

Utterance-level acoustic measures. Means are marked by diamonds.

Mean pitch. There was an effect of Speech Act in both models: Mean pitch was higher in questions than in exclamatives (New vs. Contrast model_SA+OS: b = 1.18, SE = 0.1, t = 11.5, p < 0.001; New vs. Given model_SA: b = 1.3, SE = 0.1, t = 11, p < 0.001). There was also an effect for the New vs. Contrastive comparison (b = –0.23, SE = 0.04, t = –5.7, p < 0.001), which was involved in an interaction with Speech Act (b = –0.07, SE = 0.03, t = -2.2, p < 0.05): Mean pitch was lower in utterances with Contrastive than with New objects, and the effect was more pronounced in questions than in exclamatives.

Pitch range. There was an effect of Speech Act in both models: Pitch range was higher in questions than in exclamatives (New vs. Contrast model_SA+OS: b = 1.7, SE = 0.18, t = 9.6, p < 0.001; New vs. Given model_SA: b = 2.3, SE = 0.22, t = 10.8, p < 0.001). Additionally, there was an effect of Contrastive focus (b = 0.54, SE = 0.1, t = 5.4, p < 0.001), which was involved in an interaction (b = –0.57, SE = 0.07, t = –8.4, p < 0.001): Contrastive focus led to a higher pitch range, but only in exclamatives.

Mean intensity. Mean intensity did not show significant differences for either Object Status or Speech Act, only tendencies. Descriptively, mean intensity tended to be lower for questions (New vs. Contrast model_SA*OS: b = –1.9, SE = 0.9, t = –2, p < 0.2; New vs. Given model_SA+OS: b = –1.8, SE = 0.9, t = –1.9, p < 0.2) and for utterances containing contrastive objects (b = –0.14, SE = 0.08, t = –1.9, p < 0.1).

Intensity range. Intensity range did not exhibit significant differences for Object Status. As for Speech Act, intensity range was lower in questions than in exclamatives in both models (New vs. Contrast model_SA+OS: b = –1.3, SE = 0.2, t = –8.7, p < 0.001; New vs. Given model_SA+OS: b = –1, SE = 0.2, t = –5.6, p < 0.001).

Syllables per second. There was an effect of Speech Act in both models: Speaking rate was higher in questions than in exclamatives (New vs. Contrast model_SA+OS: b = 0.36, SE = 0.03, t = 11.3, p < 0.001; New vs. Given model_SA: b = 0.36, SE = 0.03, t = 11.2, p < 0.001). There were no significant differences for Object Status and no significant interactions.

We assume that most of the significant differences on the utterance level can be traced to differences on the syllable level or to other local effects. For instance, if most questions ended in a final rise/high boundary tone, pitch range might be higher in questions than in exclamatives just because of the boundary tone. Similarly, pitch range in exclamatives containing a contrastively focused object might well be higher than in exclamatives containing a new object because the nuclear accent on the object is more prominent than in the other two conditions (i.e., without pitch range on all syllables being higher). We return to this issue after the investigation of accent placement in Section 5.3, since accent placement determines which acoustic comparisons can be made.

5.2. Final boundary tone

In the following, we will speak of high boundary tones and/or final rises if the boundary tone is H-^H% or L-H%, and of low boundary tones and/or final falls if the boundary tone is L-%.

The difference between the speech acts in terms of boundary tones was very stark: Questions overwhelmingly ended with a high boundary tone (i.e., a final rise), and exclamatives overwhelmingly ended with a low boundary tone (i.e., a final fall). Exceptions were few: 10 out of 699 exclamatives were rising, and nine out of 691 questions were falling.

5.3. Accentuation patterns

Figures 2 and 3 show the accentuation of stressed syllables in the words of the target sentences, split by GToBI accent types and annotated prominence levels. More prominent accents are represented by darker colors. Note that these data are pooled across all utterances (i.e., the figures do not show which accents co-occurred within the same utterance). We return to the latter issue further below (for accent types see Section 5.3.1; for perceived prominence levels see Section 5.3.2).

Figure 2

Distribution of accented syllables in the exclamatives.

Figure 3

Distribution of accented syllables in the questions.

The most salient patterns in Figures 2 and 3 are as follows: (i) In both speech acts, the most frequent accent carrier was the object, (ii) the d-pronoun very often carried an accent in exclamatives, but virtually never in questions, (iii) the clause-final lexical verb quite frequently carried an accent in questions, but almost never in exclamatives, (iv) with very few exceptions, L* accents occurred only in questions. Furthermore, the auxiliary appears to be able to carry an accent only in exclamatives, while the adjective/quantifier⁶ may carry an accent in both speech acts, although it did so more often in questions than in exclamatives.

The influence of IS and speech act on accent placement can be summarized as follows: (v) In both speech acts, the IS status of the object has a comparatively small impact on the accentuation of the object itself, and a comparatively larger impact on the accentuation of one other element: The d-pronoun in exclamatives, and the adjective in questions, (vi) contrastive focus leads to high accentuation rates on the object and to low accentuation rates on the respective non-object element, (vii) in questions with a given object, the object is accented slightly less often than in questions with a new object, whereas the accentuation rate of the adjective is slightly higher in questions with a given, than a new object.

5.3.3. Contours: Combinations of accent types and combinations of prominence levels

We next turn to the issue of which accented syllables tended to co-occur in one utterance. Figure 4 shows the most common accent type combinations (i.e., contours) by experimental condition. Figure 5 shows examples for the two most common contours in each speech act.

Figure 4

Percentage of contours within conditions per speech act. The question contours are ordered top-to-bottom by occurrence rate; the exclamative contours are roughly ranked by (presumed) cumulative prominence: Double L+H* accents > combinations of one L+H* accent and one H* accent > double H* accents > single H* accents. DSub = subject d-pronoun; Other = contours occurring in fewer than 5%, pooled.

Figure 5

Example contours for the two most common contour types in each speech act. Top: Exclamatives. Bottom: Questions. Each utterance was produced by a different participant. Gloss/Translation of the example: Has he wild bulls ridden; ‚Has he ridden wild bulls’.

Exclamatives exhibited a greater variety of contour types than questions, and they also had more contours with multiple prominences. In questions, the most common contour type in each condition was a single L* accent on the object. In both speech acts, Contrastive focus led to an increase of single accents on the object (i.e., to a reduction of accentuation elsewhere in the utterance). For exclamatives, this means that the accent on the d-pronoun, which in the New and Given conditions regularly occurs as a second accent, is present less frequently in the Contrastive condition. The single accent was the more prominent L+H* accent.

Figure 6 shows the co-occurrence of accented syllables by prominence levels. The top facets illustrate the competition for prosodic prominence between the object and the d-pronoun in exclamatives that we saw above from another perspective: When the object was New or Given, the additional accent on the d-pronoun was more often more prominent than the accent on the object (level 2 vs. 1). This suggests that it was the nuclear accent in almost half of the utterances. In exclamatives with Contrastive objects, the object had a higher prominence than the accent on the d-pronoun (when it was not the only accent anyway).

Figure 6

Percentage of prominence level combinations within conditions per speech act. See caption of Figure 5 for abbreviations.

5.4. Syllable-level measures

We now turn to the continuous measures on the syllable level. For the pitch-related variables maximum pitch, minimum pitch, and pitch range, we divided the data into subsets according to the respective syllable’s starred tone: We pooled syllables carrying H* and L+H* accents, and syllables carrying L* and L*+H accents.⁷ For the two variables that are not related to pitch – intensity and duration – we pooled all accented syllables regardless of GToBI accent. The theoretical motivation for these subsets is that our hypotheses regarding the impact of the IS status of the object on its prosodic characteristics only differentiate between high/rising and low/falling accents, instead of between specific GToBI types within these broader groups. An additional, practical motivation is that the choice of GToBI accent and the IS status of the object were correlated, as described in Section 5.3. For example, H* accents were quite rare on contrastively focused objects in exclamatives, while L+H* accents were rare in the other two conditions. Such imbalance produces issues with model fits.

We investigated maximum pitch only for syllables with high starred tones and minimum pitch only for syllables with low starred tones. We investigated pitch range for both types of starred tones. High starred tones occurred both in exclamatives (on the object and on the d-pronoun) and in questions (on the object). Low starred tones occurred only in questions (on the object and on the lexical verb).

In the following, we will discuss the results by syllable and speech act, starting with the object in both speech acts.

6.1. Contrast: Prenuclear effects, nuclear effects, and phonological categories

Contrastive focus largely was marked as expected, although there were also some important unexpected results. As hypothesized (H.2), contrastive focus led to increased prosodic prominence in both speech acts. We found more frequent accentuation of the focused object in both speech acts, and a significant increase in the number of L+H* accents in exclamatives. We also found phonetic differences indicating increased prominence: Contrastive objects in questions were characterized by lower pitch, while in exclamatives, they were associated with higher pitch, larger pitch excursions, and longer durations.

The most important unexpected finding was that in utterances with a contrastive object, an element that in non-contrastive utterances was prosodically prominent – the d-pronoun in exclamatives and the adjective in questions – was no longer as prominent. There were much fewer accents on these two elements when the object was contrastive. Thus, contrast was not only marked by prominence increase but also by prominence decrease. Another unexpected finding was the lacking duration effect for contrast marking in questions. This is curious. Possibly the high speaking rate of questions made the durations of individual syllables fairly rigid.

Overall, our results show two important things.

1. Contrast marking is characterized on the one hand by categorical prosodic patterns (accent placement, accent type). Some of these are used probabilistically, similar to what has been reported for categorical patterns for the choice of accent type in givenness and focus marking in assertions (Röhr & Baumann, 2010; Mücke & Grice, 2014; Baumann et al., 2015; Grice et al., 2017; Roessig & Mücke, 2019; Roessig, 2021). On the other hand, contrast marking is done with gradient patterns, like the pitch-related measures. Both types of patterns differ depending on the speech act, which itself also is marked by both types of patterns.

2. Contrast marking occurs both locally – by a prominence increase on the contrastive element –, and non-locally – by a prominence decrease on another, prenuclear element.

Both (i) and (ii) raise important issues for the prosody-meaning interface. Regarding (i), the observation that both phonological and phonetic patterns are used to mark contrast and speech act, and that they do so inter-dependently suggests that the often assumed division of prosody-meaning mapping by meaning type is not supported by our data. We did not find that syntactic and (discourse-)semantic information, being part of the grammar, maps onto phonological categories which via the phonology-phonetics interface are implemented in the phonetics, whereas pragmatic or paralinguistic information has a direct interface with the phonetics.⁸

We think that a view like that taken by Grice et al. (2017, p. 105), who describe phonology and phonetics as “two sides of the same coin, best understood as a single system”, is more likely to adequately capture the phonetics-phonology-meaning relation. Consider our findings regarding the relationship between L+H* as a phonological category and contrast: (a) In exclamatives, contrast increased the number of L+H* accents on objects, but L+H* accents were also found on new or even given objects; (b) many L+H* accents in exclamatives were found on the d-pronoun, which was never contrastive and always given. Thus, the relationship between L+H* and contrast is probabilistic. Contrast is neither necessary nor sufficient to license the occurrence of L+H*. Considering the relationship of L+H* and speech act, the same probabilistic characteristic emerges. L+H* is not reliably used as the exclamative accent to mark the speech act, but it occurs often.

The acoustic properties of L+H* and H* accents also lie along a continuum (Grice et al., 2017; Roessig, 2021). It seems that an accent with a high target is perceived as L+H* if the target is high and/or late enough. What counts as “enough” most likely is different between individuals, both for speakers and for hearers: Some speakers choose between L+H* vs. H* to mark an information-structural difference, whereas others mark the same difference by gradient differences within an accent type (Grice et al., 2017).

In sum, we think that at least in the case of L+H* and H* in German, it is inappropriate to speak of phonological categories (i.e., tonemes). There is no categorical distinction between the distribution of L+H* and H* accents as meaning-distinguishing. The labels H* and L+H* are probably best viewed as shorthand descriptors of particular F0 contours, but these represent idealized end points of a continuum. Real productions of either accent type will probably always lie somewhere in between these endpoints (cp. Cangemi & Grice, 2016).⁹

Turning to the issue arising from observation (ii), we may ask whether the F0 contours just mentioned can be viewed as clearly meaning-distinguishing in the context of the current study. The answer is that this is probably not the case: Apart from prominence-increasing local marking, there often (also) is non-local prominence-decreasing marking. Thus, it does not always seem to be sufficient to have a prominence increase on the contrastive element.

The combination of prominence increase and prominence decrease is of course familiar from focus marking, where there is increased prominence on a focused element and reduced prominence in the post-focal region, viz. post-focal compression, which in assertions is realized as lower pitch in many languages. The lower pitch has been analyzed as deaccentuation (Ladd, 1980; for German, e.g., Féry, 1993) but also as a result of a reduced pitch register (Kügler & Féry, 2017; for American English, see Xu & Xu, 2005, who explicitly argue against post-focal deaccentuation). In our experiments, the reduction of prominence occurs in the prenuclear region. Baumann et al. (2006) also observe that there may be deaccentuation and reduced gradient prominence in the prenuclear region in sentences with corrective focus. Niebuhr et al. (2010) observe prenuclar deaccentuation in double-checking declarative questions, where the nuclear accent also marks narrow focus. Kügler and Féry (2017) suggest that in the prenuclear region there is no deaccentuation, but at most slight compression. Yet, similarly to Baumann et al. (2006) and Niebuhr et al. (2010), we did observe a significant drop in accent frequency if the object was contrastive, contradicting Kügler and Féry’s conjecture.

The reduction of accentuation of the d-pronoun requires closer scrutiny. After all, it is a speech-act-marking device in exclamatives without a contrastive object, and therefore might be considered to carry a high functional load. Importantly, the constructional prosodic default suggested by Repp and Seeliger (2020) only requires that there be a highly prominent accent in an exclamative, without specifying the position of that accent. So deaccenting the d-pronoun is compatible with the constructional default as long as there is a highly prominent accent elsewhere, for instance on the contrastive object.

Having established that contrast in both speech acts is marked by a combination of prominence increase and decrease, and that this is compatible with earlier observations in the literature, let us ask what this finding tells us about the meaning-prosody mapping. We might describe our finding in terms of what we may call prominence balance. This balance must be positive for the meaning category contrast, which means that contrast increases the prominence for the contrastively focused word, while at the same time potentially reducing the prominence of other potentially accented words.¹⁰ The prominence balance is negative for the meaning category givenness (see next section): An element that is given comes with reduced prominence whereas other elements in the clause might come with a prominence increase.

Prominence balance is fed by many different ingredients (pitch, duration, gradient phonetic measures ‘feeding’ into accent type etc.). An important question is what ‘kind of animal’ prominence balance is. One option is to assume a phonetic-phonological optimality-theoretic account as has been proposed for segmental phonology by Flemming (2001), with constraints over phonetic detail and with constraints making reference to the maximization of phonological contrasts/distinctiveness. Let us assume that we have only few phonological contrasts in intonation (cp. Gussenhoven, 1984). One such contrast could be between prominent vs. non-prominent elements as signaling contrastive focus vs. non-focus. The reduction of prominence on the d-pronoun would maximize the distinctiveness of this d-pronoun from the highly prominent contrastive object, thus ensuring a positive prominence balance for the contrastive object. These ideas obviously need careful discussion in future research because prominence is of course gradient, and information-structural meaning categories have been associated with different degrees of prominence.

An optimality-theoretic account would also have the advantage of being able to capture the non-deterministic mapping from a meaning category to prominence balance more generally: The mapping seems to be more like a ‘wish-list,’ for two reasons. The first is that the requirements of different meaning categories might conflict with each other. We found that contrast marking can ‘overwrite’ speech-act specific default accent positions in exclamatives. This is possible because of the flexibility of the ‘wish-list’ for exclamative speech act marking (the constructional prosodic default), which does not prescribe a position for the ‘exclamative accent’. The second reason is that there are other lexical elements in the clause with their own requirements concerning their prominence balance, which might also conflict, for instance if there is a double focus. In the next section we will see another instance of conflicting requirements stemming from discourse-semantic restrictions.

An alternative way to view prominence balance is in terms of a gestalt strategy. Different gestalts would then be associated with different information-structural categories. This idea requires careful thought about the compatibility of gestalts with the auto-segmental theory we are assuming here. All these considerations raise many important theoretical questions and must be addressed in greater detail in future research.

6.2. Givenness: Speech-act specific findings and a semantic-pragmatic account

Our hypothesis regarding the marking of givenness (H1) was borne out for exclamatives but not for questions: In neither speech act was givenness clearly marked prosodically, neither on the categorical scale of accentuation, nor on the continuous scales of pitch and duration. For questions, we hypothesized given objects to be accented less often than new objects, to be shorter, and to have higher pitch. What we found was that givenness leads only to a slight, statistically non-significant reduction in the number of accents and to no significant differences in the continuous measures. A lack of givenness marking was expected for the exclamatives, but not for the questions. We do not think that this outcome is a consequence of the hypernym-hyponym relation of the new objects with an element in the context, which we argued in Section 4.2 to be insufficient to make new objects accessible from a discourse-semantic perspective. Furthermore, in our data new objects are regularly accented. This is expected. It is the regular accentuation of given objects that is surprising.

To investigate this surprising outcome, we examined the accentuation patterns in the fillers. Recall that these were negative rejections and negative declarative questions like Der hat keine Blumen mitgebracht!/? ‘He didn’t bring flowers !/?’. In both speech acts, all referring expressions including the objects were lexically given. The only non-given words were the auxiliary and the negative determiner keine ‘no’.¹¹ If given objects are also accented very often in the fillers, our experimental paradigm might be fundamentally flawed. Note that neither declarative questions nor rejections require their lexical contents to be all given: Both can theoretically also contain contrastive focus. All-given declarative questions and rejections thus represent the same ‘special’ case that all-given exclamatives and polar questions do.

Figure 7 shows the number of accented syllables in the negative rejections and the negative declarative questions, split by accent type and prominence level. We see that there were differences between the speech acts regarding object accentuation. The negative declarative questions pattern with the positive polar questions from the experimental items: The object, although given, was accented in the vast majority of utterances (92%).¹² These accents virtually always represented the nuclear accent, being at least of prominence level 2. In the rejections, in comparison, objects were accented in only 19% of the utterances, and roughly half of those accents are of prominence level 1, (i.e., post-nuclear accents). Given objects in rejections were thus mostly deaccented. We can therefore exclude the possibility that the participants simply always placed ‘default’ accents on objects, regardless of IS: We do find categorical deaccentuation in rejections.

Figure 7

Distribution of accented syllables in the fillers.

Our account of the differences in givenness marking across the four speech acts rests on the assumption that prosodic constituents must be headed (Selkirk, 1984). The default prosodic head of the intonation phrase in the target sentences (in an all-new context) is the stressed syllable of the object noun: It carries the nuclear accent. If IS requires that the object noun be deaccented due to its givenness, another syllable in the intonation phrase must become the head of the phrase and thus carry the nuclear accent. Importantly, the nuclear accent cannot just go anywhere: An accent shift can and usually will have semantic and pragmatic effects. In the following, we use the term accent shift instead of deaccentuation in order to emphasize that the nuclear accent has to move elsewhere if it ‘cannot stay’ in its default position. Crucially, we suggest that if accent shift is semantically or pragmatically illicit, the nuclear accent must be realized even on ‘fully’ given elements (i.e., on given elements that are not (contrastively) focused). We propose that the dialogue contexts in the experiment did not allow an accent shift in the polar questions and in the declarative questions for specific semantic-pragmatic reasons. In the rejections, however, accent shift was possible, and we found that speakers considered it close to mandatory.

To develop our account, let us first consider the element(s) that the accent could be shifted to in the negative declarative questions and in the negative rejections in the fillers. We saw that the object’s main competitor for prosodic prominence is a different word in the two speech acts. In the declarative questions, the negative determiner keine ‘no’ is accented just a little less frequently than the object. In the rejections, by far the most prominent syllable is the auxiliary, which is accented in about 90% of the utterances and virtually always carries the nuclear accent. Now, accenting the negative determiner in the declarative questions seems intuitive if we assume that what is marked is focus on the negation (i.e., on the polarity of the sentence). The polarity (i.e., truth) is at issue in the discourse. Accenting the finite auxiliary in the rejections serves the same purpose: We propose that it marks verum focus (i.e., also focus on the polarity of the utterance (e.g., Höhle, 1992; Romero & Han, 2004; Gutzmann, 2012; Lohnstein, 2012, 2016)). Since neither the negative determiner in the declarative questions nor the auxiliary in the rejections represent given information, a shifted accent on these elements also seems legitimate from the point of view of givenness. An interesting question is why the two filler speech acts show different accentuation patterns: Both regarding the choice of negative determiner vs. auxiliary, and regarding the accentuation of the given object. Why do rejections behave in a unique fashion and allow deaccentuation of the given object?

Before we answer these questions, let us turn to the polar questions in the experimental items. There is only one candidate for an accent shift away from the given object: The auxiliary. All other words were also lexically given, and a nuclear accent on any one of them with attendant deaccentuation of the object would have indicated narrow focus on that word, which would have been incoherent in the context. Yet, why was the auxiliary not accented? After all, positive polar questions ask about the polarity just as negative declarative questions do, and if there is no negative marker which could be accented to signal focus on the polarity, verum focus in principle should be licensed.

Our explanation for the lack of accent shift in the polar questions and the declarative questions builds on the precise pragmatic effects that verum focus has in questions. In the literature, different effects have been proposed, depending on the question type and the surrounding context (e.g., Höhle, 1992; Creswell, 2000; Romero & Han, 2004; Biezma, 2009; Gutzmann & Castroviejo Miró, 2011; Gutzmann, 2012; Lohnstein, 2012, 2016). Relevant for our purposes are the following two effects. First, verum can indicate that the speaker had or has an epistemic bias against the prejacent of the question (Romero & Han, 2004). For instance, Did Peter come? can convey that the speaker of the question had assumed that Peter did not come. Second, Verum can give rise to what Biezma (2009) called the cornering effect.¹³ Roughly, the cornering effect is the ‘flavor’ of impatience that a question signals when it is used to finally settle an issue, either after a previous refusal by the addressee to give a straight answer, or after multiple commitments by several interlocutors to both p and ¬p (also Lohnstein, 2012, 2016). We argue that neither the polar questions nor the declarative questions in the experiment occurred in contexts that would license one of these effects of verum focus. We can see this when we examine the contextual setup in more detail.

The polar questions in the experiment were preceded by turn-initial interjections like Ach echt? ‘Oh, really?’, which indicate a fairly high willingness of the speaker to accept the contextually implied information as true. Such interjections are incompatible with the intention to corner the addressee, which would require conflicting information about the prejacent of the question or a lack of cooperation by the addressee. As for the function of verum to indicate epistemic bias, this was blocked by the context following the question (e.g., ‘I would find that exciting’), which signals that the speaker had no previous expectations regarding the questioned proposition. Therefore, verum focus was not licit in the polar questions. This, in turn, made a full accent shift away from given objects impossible.

Declarative questions do not seem to be able to give rise to cornering effects in general, which might have to do with the fact that these questions always are biased: There must not be evidence against the questioned proposition in the context or even evidence for it (e.g., Gunlogson, 2003; Šafařová, 2005; Sudo, 2013; Trinh, 2014; Seeliger & Repp, 2018; Seeliger, 2019). We leave this issue for future research. For the current purposes we note that an accent shift to the auxiliary in declarative questions can only give rise to the inference that the speaker was previously biased with respect to the underlying proposition and is now resolving an epistemic conflict between the contextual evidence and the previous assumption. To illustrate, in (6) Belinda, who asks the question, assumed that Carl had never been to France because Carl himself told her so. Ann’s assertion in the preceding context creates an epistemic conflict, which in Belinda’s question is marked by verum focus.

In our experiments, the speaker of the declarative questions always received new information and was asking a double-check question – as was also indicated by turn-initial interjections like Moment mal ‘Wait a moment’. Nothing in the context supported the necessary background assumption that the speaker was previously biased about the proposition that is questioned in the declarative question. The follow-up questions (e.g., ‘Doesn’t he work in a plant nursery?’ after ‘He didn’t bring flowers?’) arguably are too weak to license verum: They only indicate that the speaker has made a weak, defeasible inference (e.g., if someone works in a plant nursery there is a good chance that they bring flowers). Something much stronger would be needed to place the nuclear accent on the auxiliary (e.g., ‘I thought he brought half the plant nursery!’). In other words, a previous weak assumption does not seem to be enough to license verum focus in declarative questions, previous strong conviction is required.

What about the other potential target for the shifted accent: The negative determiner keine ‘no’? Polarity focus can be expressed by negation, and we saw that participants often accented keine. Yet the accent on keine tended to be pre-nuclear, instead of attracting the nuclear accent. It turns out that an accent on keine can also indicate a different focus – depending on the prosodic characteristics of the remainder of the clause. Intuitively, an accent on keine only expresses polarity focus if the object is accented as well. If the object is deaccented, there is a narrow focus reading where the focus alternatives are quantifiers: {none, a few, several, many, …}. In (7), where the stretch of speech after the prominent syllable of the determiner is compressed (marked by italic font), the speaker is double-checking whether the subject referent really brought no flowers at all – as opposed to just a few, which appears to be the speaker’s original epistemic bias with such a contour. Crucially, this reading is not salient in the experimental contexts.

As for why the accentuation of the object, even though it is given, appears to be necessary to mark polarity focus, we tentatively suggest that polarity – or rather negation – is a propositional operator and for focus on this operator to project to its scope position, the phrase containing the negative determiner keine needs to be focus-marked. Just accenting the determiner apparently cannot serve this purpose. This requires more scrutiny in future research on the syntax-prosody-semantics interface of negative sentences.

Finally turning to rejections, the nuclear accent can shift to the auxiliary, indicating verum focus, as we saw in the data. The requirement imposed by verum that the speaker and/or the context be biased with respect to the rejected proposition is fulfilled trivially in rejections: One can only reject a proposition if one is biased against that proposition. A single accent on the object would induce a contrastive reading (i.e., the speaker would reject only that the proposition is true for the object referent). This contrastive reading was not supported by the experimental context (no focus alternatives to the object denotation). Double accentuation – a nuclear accent on the auxiliary and a post-nuclear accent on the object – is apparently dispreferred in rejections. As the exponent of the polarity focus is not the same as in the negative declarative questions (auxiliary vs. negative determiner), we think that this finding is not at odds with our proposal for the accentuation pattern in the negative declarative questions.

Summarizing our discussion of givenness marking in the questions, we assume that deaccentuation of given material is possible in questions, be they polar or declarative, but the nuclear accent has to move to another element for structural prosodic reasons. If there is no element available that could host the accent without triggering semantic-pragmatic effects that are unlicensed in the context, given material needs to be accented. This account combines well with the idea developed in Section 6.2, that there is no deterministic mapping from a meaning category to prosody. The marking of givenness by a negative prominence balance in polar questions conflicts with a structural prosodic requirement (headedness) and with other meaning-prosody mappings (verum focus), which according to our findings ‘win over’ givenness marking.

Exclamatives are different from the two question types and rejections. A nuclear accent on the auxiliary with full deaccentuation of the object in principle appears to be possible, and does not seem to give rise to verum-related readings or to contrast. This accentuation choice was thus compatible with the experimental contexts. We take the fact that this contour is only very rarely attested in the data¹⁴ as evidence for a genuine reduction of givenness marking in the exclamatives. This view is further supported by the finding already discussed above, viz. that the d-pronoun, which is given and not contrastive is a typical attractor of the exclamative accent. Intuitively, it is unclear whether a nuclear accent on the d-pronoun with full deaccentuation of the object leads to a contrastive focus reading of the d-pronoun. This needs to be tested in future research.