1.1 Prosodic prominence and language processing

A number of studies have investigated the interplay of prosodic prominence on language processing, especially on word recall (Fraundorf et al. 2010; Savino et al. 2020; Kember et al. 2021; Koch & Spalek 2021). Fraundorf et al. (2010), for instance, compared the effects of the H* accent to the L+H* accent on word recall in speakers of American English. Following Pierrehumbert & Hirschberg (1990), the accent type H* is considered to introduce new information and to not show contrastive focus while the accent type L+H* is considered to not introduce new information and to show contrastive focus. The procedure was split into two parts: First, the participants had to listen to all auditorily presented experimental sentences (24 critical and 24 filler sentences), that were embedded into small discourses (one context sentence, one experimental sentence). Afterwards, participants were asked in a forced-choice recognition memory task whether a word appearing on the screen had been presented in the experimental sentences or not. The results showed that target words bearing the pitch accent L+H* were recalled better than target words accented with an H* accent. The authors attributed the improved word recall of the accent type L+H* over H* to the fact that the encoding of the word bearing an L+H* accent is facilitated in discourse.

A similar effect is reported by Kember et al. (2021). The authors examined speakers of Australian English or of Korean in a recall task where prominence was manipulated syntactically and prosodically. Prosodic prominence was manipulated in terms of a high pitch accent. The results showed that target words which were prosodically prominent were recalled better than target words that were syntactically prominent. Kember et al. (2021) argued that prosodic prominence facilitates lexical processing which leads to a better recall.

While the aforementioned studies measured the influence of prosodic prominence on off-line language processing, eye-tracking studies have investigated the influence of prosodic prominence on on-line language processing (see e.g., Weber et al. 2006; Chen et al. 2007; Ito & Speer 2008; Braun & Biezma 2019).¹ Ito & Speer (2008), for example, investigated anticipatory effects of intonation in an instructed visual search task conducted with English-speaking participants. Participants had to decorate a holiday tree with ornaments that were presented on a vision board. Different ornaments were hung up in different cells. Each cell consisted of four ornaments of the same type but differing in colour. The participants’ task was to decorate a holiday tree by executing the instructions that were auditorily presented. While doing that, their eye movement was monitored with a head mounted eye-tracker. Ito & Speer (2008) manipulated the position of the contrastive L+H* accent that was either on the prenominal adjective or on the noun in instruction sentences like ‘Hang the green drum. Now hang the BLUE drum./Now hang the blue DRUM.’ (capitalization indicating the location of the accent; Ito & Speer 2008: 545). The authors found more early fixations and a steeper initial rise, i.e., more frequent fixations to the target cell, in the condition where the prenominal adjective was manipulated with an L+H* accent. The authors argued that the contrastive accent on the adjective facilitated fixation to the target cell because participants expected the same type of ornament than in the preceding sentence.

Similar findings were reported by Weber et al. (2006). The authors presented participants with four pictures (e.g., purple scissors (first referent), red scissors (target; contrastive referent), red vase (non-contrastive referent), and green clock (distractor)). Participants were instructed to click on the object that was specified in the sentence they heard. Participants first heard an instruction sentence such as ‘Click on the purple scissors.’, with scissors carrying an H* accent. Then, participants heard the instruction sentence ‘Click on the red scissors.’ with either red or scissors carrying an L+H* accent. Weber et al. (2006) found that participants looked faster to the target referent when the preceding adjective was manipulated with an L+H* target compared to when the adjective was unaccented. The authors argued that their participants “exploited accents on preceding adjectives rapidly enough to anticipate target referents even before the referent noun was mentioned” (Weber et al. 2006: 386).

While these studies looked at prosodic prominence and its function in marking contrast, there is only sparse psycholinguistic evidence on the issue of how prosodic prominence influences on-line language processing devoid of an explicit contrasting context. Cutler & Swinney (1987) conducted a word monitoring task with English-speaking children and adults. Participants were asked to listen to pre-recorded sentences and to push a response key as soon as they recognized a target word, that was presented auditorily before the onset of the sentence. The primary sentence accent was either on the target word or appeared elsewhere in the sentences (e.g., The family was already at the summer cabin; bold indicating the accented target word; Cutler & Swinney 1987: 151). Cutler & Swinney (1987) found that their ten adult participants reacted faster when the target word carried the primary sentence accent compared to when the accent appeared elsewhere in the sentence. These findings suggest that prosodic prominence affects language processing by facilitating word identification, even if the accent is devoid of its function, i.e., does not explicitly mark contrast.² The study replicated findings by Cutler & Foss (1977), who manipulated the stress of target phonemes in carrier words. Cutler & Foss (1977) asked participants to listen for previously defined target phonemes in carrier words in sentences and to push a button as soon as they heard the target phoneme. The authors found that participants recognized the target phonemes faster when they appeared in words which bore the main accent of the sentence compared to when they occurred in words that did not carry the main accent.

Further evidence that prosodic prominence might facilitate word recognition and processing comes from a study by Cole & Jakimik (1980). Cole & Jakimik (1980) examined whether stress influenced the detection of mispronunciations and asked participants to click on a button as soon as they heard a word that was mispronounced. In the mispronounced words, syllable-initial /p/ and /k/ were replaced by /b/ and /g/, respectively. The mispronounced words were either stressed, i.e., prosodically prominent, or unstressed, i.e., not prosodically prominent. Cole & Jakimik (1980) found that mispronunciations were detected significantly more often when the mispronounced word was stressed (82%) compared to when it was unstressed (47%). The authors argued that “acoustic features [were] more prominent in stressed than unstressed syllables” (Cole & Jakimik 1980: 968), making the voicing in stressed words more salient and resulting in a higher detection rate of mispronunciations in stressed compared to unstressed words.

To summarize, off-line studies have quite consistently shown facilitatory effects of prosodic prominence on word recall across studies and languages (Fraundorf et al. 2010; Kember et al. 2021; Koch & Spalek 2021; see also Savino et al. 2020 for a recall task of digits). Additionally, eye-tracking studies have revealed that prosodic prominence facilitates referent identification when the prominent entity indicates a contrast (Weber et al. 2006; Chen et al. 2007; Ito & Speer 2008; Braun & Biezma 2019). However, there is only little psycholinguistic research on the effect of prosodic prominence on on-line language processing when prosodic prominence is used devoid of context and function, for instance without signalling an explicitly mentioned contrast (Cutler & Foss 1977; Cutler & Swinney 1987; see also Cole & Jakimik 1980). Moreover, while previous research focused on the effects of one or two specific accent types, the issue whether the graded scale of prosodic prominence established by Baumann & Röhr (2015) leads to graded effects on on-line language processing has not been investigated in psycholinguistic research so far. The goal of the present study was to close these gaps by investigating the influence of prosodic prominence on language processing in German.

1.2 Aims and predictions

The aim of our study was to follow-up on the research by Cutler & Swinney (1987) in investigating whether and how prosodic prominence influences on-line language processing in German when it is employed devoid of its function in a given context. In addition, we sought to examine whether the graded scale of perceptual prominence established by Baumann & Röhr (2015) would also lead to graded effects of prosodic prominence on on-line processing. To this end, we conducted a word monitoring task that was similar to that of Cutler & Swinney (1987), where participants were asked to push a response key as soon as they recognized the previously defined target word in a sentence that was auditorily presented to them. In contrast to Cutler & Swinney (1987), we did not only test the difference between accented and deaccented words but investigated the processing of three different accent types (L+H*, L*+H, L*) and of deaccentuation (ø), each varying in perceptual prominence ratings, in prenuclear sentence position. By testing these four types, we aimed to cover the extreme ends of the prominence scale suggested by Baumann & Röhr (2015): while accents L+H* and L*+H were rated as most prominent in the study by Baumann & Röhr (2015), the accent L* and deaccentuation (ø) were rated as least prominent.

According to Posner (1980: 4) attention orienting is “the aligning of attention with a source of sensory input”. Prosodic prominence is assumed to centre the attention of a listener to the prosodically prominent entity during language processing (e.g., Himmelmann & Primus 2015). The listener’s attention is involuntarily drawn by an external stimulus, i.e., the prominence-lending prosodic cues, towards the prosodically prominent word. Thus, the orienting of attention is not under control of the listener but is an involuntary reaction to an external stimulus, the prominence-lending prosodic cues. Following James (1890: 403 f.), attention is defined as “taking possession by the mind, in clear and vivid form, of one out of what seem several simultaneously possible objects or trains of thought”, which implies a “withdrawal from some things in order to deal effectively with others”. This implies that while processing resources are drawn to the entity attention is oriented towards, information that is not in focus is neglected.

How might prosodic prominence influence word monitoring? While numerous psycholinguistic models have been put forward to explain auditory word recognition (e.g., Marslen-Wilson & Welsh 1978; McClelland & Elman 1986; Norris 1994; Cutler & Clifton 2001), most of them commonly assume the following three stages: initial contact, selection and integration (see e.g., Dahan & Magnuson 2006). During the initial contact stage, words that share the same onset get activated. As the auditory input unfolds, more and more activated words in this initial cohort get excluded until one candidate is selected. During the integration process, the selected lexical item is then integrated into the discourse in consideration of the semantic and syntactic structure of the input. We suggest that prosodic prominence might facilitate the selection process, resulting in a faster identification of the prosodically prominent word. Prosodically prominent words are hyperarticulated (e.g., de Jong 1995), which might make the distinction between similarly sounding words easier. This might be due, for instance, to a smaller initial cohort for the selection process or to a quicker weeding-out of cohort candidates. As a consequence, prosodically prominent words might be more accessible for the listener and get accessed faster. This might also hold when a prosodically prominent accent type is used devoid of its pragmatic function, for example when a contrast inducing L+H* accent is used without explicitly signalling contrast. Given the assumption that prosodically prominent entities draw attention, a prosodically prominent target word should be identified quicker in a word monitoring task compared to a target word that is presented without a prosodically prominent accent. Moreover, given the graded scale of prosodic prominence proposed by Baumann & Röhr (2015), we might expect a graded effect of prosodic prominence on word monitoring times. Specifically, prosodically prominent accent types, here the accent types L+H* and L*+H, should be more effective in drawing attention to the target word, compared to target words presented with the less prominent accent type L* or for target words that are deaccented (ø). Hence, we expected to find the following scale for the identification times of the target word in our study: ø > L* > L*+H > L+H*. These expectations led to the following research question:

RQ 1: Do differently prominent accent types (i.e., L+H*, L*+H, L*) and deaccentuation (ø) on a target word influence identification times in a word monitoring task differently?

We further assumed that the centring of attention to the prosodically prominent entity and the processing of prosodic prominence binds processing resources. In turn, these processing resources are no longer available for parsing the elements immediately following the prominent entity. Accent types that are high in prosodic prominence should bind more processing resources than accent types that are prosodically less prominent. In a word monitoring task, this would lead to the assumption that target words following a word with a prosodically prominent (L+H*, L*+H) accent type should be identified slower compared to target words that follow a word with a less prosodically prominent accent type (L*) or that follow a deaccented word (ø). This leads to the second research question that we aimed to answer:

RQ 2: Do differently prominent accent types (i.e., L+H*, L*+H, L*) and deaccentuation (ø) on a word directly preceding the target word influence identification times in a word monitoring task?

In order to answer this research question, the word directly preceding the word identified as target word in the word monitoring task was either manipulated with the accent types L+H*, L*+H, L* or was deaccented (ø). Following our line of argumentation, we assumed that the previously proposed scale of word identification times should be reversed: ø < L* < L+H* < L*+H.

Given the predictions sketched out for our first and second research question, we also postulated that a comparison of those experimental conditions in which the target word or the word directly preceding the target word are manipulated with the same highly prominent prosodic accent types (L+H* or L*+H) should lead to different recognition times of the target word. Assuming that prosodic prominence draws attention, a prominent accent on the target word should lead to a facilitation of the recognition of the target word. At the same time, we expect that the recognition of the target word should be hindered when the word before the target word is manipulated with that exact accent, as the processing of prosodic prominence is assumed to bind processing resources. We expect this difference in word identification times to be strongest for the two highly prominent accent types (L+H* and L*+H), leading to an interaction between accent type and position of the accent type in the sentence, i.e., on the target word or on the preceding word. Our third research question, therefore, was:

RQ 3: Do prosodically prominent accent types (i.e., L+H* and L*+H) influence the identification times in a word monitoring task differently when they are on the target word compared to when they are on the word directly preceding the target word?

Summarizing, to investigate the influence of different levels of prosodic prominence on identification times in a word monitoring task, either the target word or the word directly preceding the target word were presented in four different accent conditions (L+H*-, L*+H- or L*-accent or deaccentuation, ø).

3.1 Accent manipulation on the target word (RQ 1)

Figure 4 shows the distribution of the reaction times including the median for the conditions L+H*, L*+H, L* and ø on the target word. For the condition L+H* on the target word, the mean reaction time was 553 ms (SD = 145). In the condition L*+H on the target word, the participants had a mean reaction time of 544 ms (SD = 159). In the condition L* on the target word, the mean reaction time was 563 ms (SD = 161) and in the condition ø on the target word the mean reaction time averaged at 545 ms (SD = 139). We predicted that the mean reaction times for the accent types L+H*, L*+H and L* on the target word would be faster than for ø on the target word respectively. Furthermore, we expected that the accent type L+H* would lead to the fastest word identification times with an increase of the mean reaction time from L*+H and L* to deaccentuation (ø) and with deaccentuation leading to the slowest word identification times. However, these predictions were not born out. In contrast to our expectations, the accent type L*+H led to the fastest reaction times (544 ms), whereas the accent type L* led to the longest reaction times (563 ms). Also, mean reaction times in the conditions L+H* on the target word and L* on target were slower than the condition ø on the target word. Only the condition L*+H showed a numerically marginally faster reaction time than the baseline. For a plot of the mean reaction times, see Figure 4.

We ran a linear mixed-effects model with random intercepts for participant and random intercepts for items. The output of this model is given in Table 2 and shows non-significant effects for all experimental conditions. Post hoc comparisons (Tukey corrected) showed no significant differences between experimental conditions (see Table 3). Thus, we found neither a graded effect of prosodic prominence nor an effect of prosodic prominence per se.

Figure 4

Violin plot of the mean reaction time in ms per experimental condition. ø serves as the baseline condition and is the condition where both the target word and the word preceding the target word are deaccented.

3.2 Accent manipulation on the word preceding the target word (RQ 2)

Figure 5 shows the distribution of the reaction times including the median for the four conditions L+H*, L*+H, L*, or deaccentuation on the word preceding the target word (i.e., the sentential subject). In the condition L+H* on the word preceding the target word, the mean reaction time was 522 ms (SD = 143). In the condition L*+H on the word preceding the target word the mean reaction time averaged at 515 ms (SD = 141). Participants had a mean reaction time of 549 ms (SD = 167) in the condition L* on the word preceding the target word and a mean reaction time of 545 ms (SD = 139) in the condition ø on the word preceding the target word. We predicted that the mean target identification times for the accent types L+H*, L*+H and L* on the word preceding the target word would be slower than for a deaccented word preceding the target word respectively, since a prominent accent on the word preceding the target word should bind processing resources. These processing resources would then be unavailable for identifying the target word. Furthermore, we anticipated that reaction times would increase as follows: ø < L* < L*+H < L+H*, i.e., we expected that deaccentuation, ø, would lead to the fastest word identification times with an increase of the mean reaction time from L* and L*+H to L+H*, with the accent type L+H* leading to the slowest word identification times. However, these predictions were not born out. In contrast to our expectations, the accent type L*+H led to the fastest reaction times (515 ms), whereas the accent type L* led to the longest reaction times (549 ms). Also, mean word identification times in the condition with the prominent L+H* or L*+H accents on the word preceding the target word were faster than mean word identification times in the baseline condition (ø). For a plot of the mean reaction times, see Figure 5.

Figure 5

Violin plot of the mean reaction time in ms per experimental condition where the word preceding the target word was accentually manipulated. ø serves as the baseline condition and is the condition where both the target word and the word preceding the target word are deaccented.

We ran a linear mixed-effects model with random intercepts for participant and random intercepts for items. The output of this model is presented in Table 4 and shows non-significant effects for all experimental conditions. Post hoc comparisons (Tukey corrected) showed no significant differences between experimental conditions (see Table 5). Thus, we found neither a graded effect of prosodic prominence nor an effect of prosodic prominence per se.

3.3 Interaction of accent and position (RQ 3)

Figure 6 shows the distribution of the reaction times including the median for the conditions L+H* on the target word or on the word preceding the target word as well as for the conditions L*+H on the target word or on the word preceding the target word.

We expected faster target word identification times when the prominent accent types (L+H* and L*+H, respectively) were on the target word compared to when they were on the word preceding the target word. However, when the accent L+H* was on the word preceding the target word, the mean reaction time was 31 ms faster compared to when it was on the target word itself. In the condition L*+H, the mean reaction time was 28 ms faster when on the word preceding the target word compared to when on the target word.

Figure 6

Mean reaction time in ms per experimental condition where either the target word or the word preceding the target word were manipulated with an L+H* or L*+H accent respectively.

We ran a linear mixed-effects model with random intercepts for participant and random intercepts for items. The output of this model is given in Table 6 and shows non-significant effects for all experimental conditions. Post hoc comparisons (Bonferroni corrected) showed no significant differences between experimental conditions (see Table 7).

3.4 Post-hoc experiment

The mean rating of how highlighted, thus prominent, the presented target words sounded was 1.75 (SD = .91) in the condition ø and 4.25 in the condition L+H* (SD = .97). We ran a mixed effects ordinal model (Christensen 2023) in R (R Core Team 2021). The model revealed a significant effect for experimental group (ß = –4.8, p < .0001), which confirms that the target words manipulated with an L+H* accent type were rated as more highlighted than the target words that were deaccented. Figure 7 shows the distribution of the ratings for the material in these two experimental conditions.

The mean rating of how highlighted, thus prominent, the word preceding the target words sounded in the experimental material was 1.75 (SD = .88) in the condition ø, and 2.3 (SD = 1.38) in the condition with an L+H* accent on the word preceding the target word. The statistical analysis confirmed a significant difference for the rating of the prosodic prominence of the preceding word in these two experimental conditions (ß = –4.4, p < .0001), suggesting that the preceding words manipulated with an L+H* accent type were rated as more highlighted than the preceding words that were deaccented.

Figure 7

Distribution of the rating of the perceptual prominence in the condition ø on the target word and L+H* on the target word. The scale 1 to 5 indicates how highlighted the target word sounded, 1 being not highlighted at all, 5 being highly highlighted.

4.1 Accent manipulation on the target word (RQ 1)

There was no significant effect of accent type on target word recognition times. We expected prosodic prominence to draw attention to the target word, thereby facilitating word identification. More specifically, we predicted faster recognition times if the target word was manipulated with a prominent accent type (L*+H or L+H*) compared to when the target word was manipulated with a less prominent accent type (L*) or was deaccented (ø). By contrast, the mean reaction time in the baseline condition (deaccentuation, ø) was numerically faster than the mean reaction times in the conditions L+H* and L*. These results do not support the predicted scale of identification times (L+H* < L*+H < L* < ø) but rather imply that a prosodically prominent accent on the target word does not facilitate identification of these target words.

We controlled for a number of possible confounding variables that might have influenced word-identification times, such as the number of phonemes of the target words, the duration in ms of the target word and the word frequency of the target word. We furthermore had the written sentences rated before we distributed them amongst the experimental groups. Additionally, the results of a post-hoc study, where we asked the participants to rate how highlighted the target words sounded in the conditions ø and L+H*, confirmed that participants indeed perceived the difference in prosodic prominence between these two experimental conditions. The condition L+H* was rated as significantly more highlighted, thus prominent, than the condition ø. In the light of the finding that differences in prosodic prominence in the experimental material were clearly perceived, we conclude that the lack of an effect of prosodic prominence on language processing in our experiment is sound and reflects that differences in the prosodic prominence of a target word do not affect its identification times in a word monitoring experiment.

This result contrasts with findings by Cutler & Swinney (1987), who report a significant effect of prosody on word identification times in a word monitoring experiment. The ten tested adult participants reacted about 36 ms faster when the primary sentence accent was on the target word compared to when the primary sentence accent occurred elsewhere in the sentence. Note, however, that Cutler & Swinney (1987) presented the same sentences once with and once without the prosodic manipulation on the target word. It is hence possible that due to the relatively small number of filler sentences (six) and experimental sentences (16), participants remembered the sentences and the position of the target word within the sentence, affecting their word identification times. In addition, it might have been the case that the participants expected the target word to be in the particular position in the sentence where it occurred because of the syntactic and prosodic structure of the experimental sentences.

Another difference between our experiment and the one by Cutler & Swinney (1987) concerns the placement of the accent in the experimental sentences. Cutler & Swinney (1987) manipulated the accent in the nuclear position of the sentence (their wording was primary sentence accent), while in our study the accentual manipulation was in the prenuclear position of the sentence. Could this difference account for the observation that we found no effect of prosodic prominence on word monitoring times contrary to the findings of Cutler & Swinney (1987)? There is an on-going discussion in the literature with regard to the status of prenuclear accents. Some researchers argue that prenuclear accents are purely ornamental without serving a function (see e.g., Büring 2007; Calhoun 2010; Jagdfeld & Baumann 2011; Kapatsinski et al. 2017; Baumann et al. 2021). For instance, Jagdfeld & Baumann (2011) found that when participants were asked to judge if a word sounded accented or not, they were less sensitive to accents in prenuclear position compared to accents in nuclear position. Similar results were reported by Kapatsinski et al. (2017) who found that adults weighed pitch peaks at the end of an utterance, equivalent to nuclear accents, as more important than in the beginning of the sentence. Against this background, one could assume that participants in our study ignored the accentual manipulations of the target words in our experiment since this manipulation occurred in the prenuclear position, thus accounting for the lack of an effect of prosodic prominence on language processing in our data. Note, however, that the data of our post-hoc study indicate that participants indeed perceived the difference in prosodic marking in the experimental material although it occurred in prenuclear position. This provides evidence against the assumption that the lack of an effect of prosodic prominence on word monitoring in our study can be attributed to the position of the accent manipulation in our experimental material. Moreover, the assumption that prenuclear accents are purely ornamental and not functional is controversial. A number of studies have argued that prenuclear accents are not purely ornamental but that they do serve a function, such as marking contrast (e.g., Féry & Krüger 2008; Braun & Biezma 2019). For instance, Braun & Biezma (2019) employed a visual-world paradigm where they auditorily presented participants with a declarative sentence (e.g., The swimmer wanted to put on flappers) while showing them four pictures on the screen, the target (flappers), an unrelated filler, a related but non-contrastive filler (sports) and a contrast alternative to the subject (diver). The prenuclear subject was either manipulated with an L*+H or with an L+H*. Braun & Biezma (2019) found that a prenuclear L*+H accent on the subject led to higher fixation rates on the contrast alternative, i.e., more looks to the diver when the subject swimmer was accented. This led the authors to conclude that the prenuclear L*+H accent serves to activate discourse alternatives. Similarly, Féry & Krüger (2008) found that the tone of the prenuclear accent changes, depending on the information status of the word in nuclear position, i.e., whether the entity in the nuclear position is given or new. These observations suggest that prenuclear accents are not purely ornamental. Given these findings and given that in our study the participants did perceive the prenuclear accent manipulation, we suggest that the lack of an effect of prosodic prominence on word monitoring cannot be attributed to the participants disregarding the accent manipulations in the prenuclear position. However, future research should specifically address the role of nuclear and prenuclear accents on word monitoring tasks, for instance by employing the word monitoring paradigm with sentences that differ in the position of the accentual manipulation (i.e., nuclear vs. prenuclear position) but are kept equal otherwise.

4.2 Accent manipulation on the word preceding the target word (RQ 2)

With respect to our second research question, we assumed that a prosodically prominent accent on the word preceding the target adverbial (i.e., on the subject) would draw attention, thereby binding processing resources to this word, leading to longer word identification times of the successive target word in comparison to a condition where the preceding word was deaccented. Specifically, we expected that the participants would react fastest when the word preceding the target word was deaccented and that identification times of the target word would increase when the word preceding the target word was presented with increased prosodic prominence (L*, L*+H, L+H* respectively). However, there was no significant effect of accent type on word identification times when the word preceding the target word was prosodically manipulated. Counter to our expectations, the mean reaction times in the conditions L+H* and L*+H on the preceding word led to numerically faster recognition times than when the preceding word was deaccented. Moreover, pairwise comparison showed neither differences between the three accent types and the baseline condition (deaccentuation, ø), nor between the different accent types (L+H* vs. L*+H, L+H* vs. L*, and L*+H vs. L*). Thus, our results do not support the initially postulated hypothesis that prosodically prominent accents bind processing resources, leading to slower identification times of the target word that directly follows the prosodically prominent word. Furthermore, the suggested identification time scale (L+H* > L*+H > L* > ø) was not confirmed. In contrast, these results support the findings described in section 4.1., that prosodic prominence does not influence on-line sentence processing in a word monitoring task.

Although the accent on the preceding word (here the subject) is not in the nuclear but the prenuclear position, it might still be considered more natural to have the sentential subject accented compared to having an adverbial accented. Thus, the processing of the target word might have been slowed down in those experimental conditions where the target word (i.e., the adverbial) was prosodically manipulated (conditions 1a-d), due to a general ‘unnaturalness’ of prosodically marking an adverbial. This could account for the observation why we did not find an effect of prosodic prominence in these experimental conditions. Note, however, that we nevertheless would have expected an effect of prosodic prominence in those experimental conditions where the subject was prosodically manipulated (conditions 2a–d) and where prosodic manipulations might be considered more ‘natural’. Contrary to this, however, our prosodic manipulations in these latter conditions did not influence word recognition times, although they were well perceived by the participants as indicated by our post-hoc study. Thus, we conclude that the position of the accent (on the subject vs. on the target adverbial) is not a decisive factor in explaining our results. Rather, the results suggest that prosodic prominence had no influence on word monitoring times in our study.

4.3 Interaction of accent and position (RQ 3)

Further evidence against the assumption that prosodic prominence draws and binds processing resources to the prosodically prominent entity, thus affecting on-line language processing, comes from the observation that we found no interaction between the type of accent (prominent L+H*/L*+H) and its position on the target word or on the word preceding it. We expected that the prominence of accent types would influence word identification times differently, depending on whether the accent was manipulated on the target word or on the preceding word. Specifically, we assumed prosodically prominent accent types on the target word to have a facilitatory effect on target word identification, while we expected prosodically prominent accent types on the word preceding the target word to bind processing resources and to slow down word identification of the following target word. However, this assumption was not supported by the statistical analyses. Neither of the two pairwise comparisons (L*+H on target word vs. on preceding word, and L+H* on target word vs. on preceding word) was significant.

4.4 General discussion

Accents are not produced in isolation but it is argued that for instance the contrast-signalling accent type L+H* might evoke a contrastive context and activate contrast alternatives even when the context and alternatives are not mentioned explicitly (e.g., Rooth 1985; Watson et al. 2008; Braun & Tagliapietra 2010). For instance, Watson et al. (2008) compared the looks evoked by the accent types L+H* (signalling a contrast) and H* (signalling a new referent) in an eye-tracking study, employing the visual-world paradigm. The authors created contrast pairs by first introducing two referents (“Click on A and B”), then they made one referent more salient (“Move B to the right of C”) and finally they either re-mentioned one referent or introduced a new referent (“Now, move A/D below E”). The referents A and D received either the accent L+H* or H*. Watson et al. (2008) found that the accent type L+H* evoked more looks to the contrast alternative (i.e., A) whereas the accent type H* equally triggered looks to new (i.e., D) and to contrast alternatives (i.e., A). Adding on this, Braun & Tagliapietra (2010) investigated the effect of contrastive intonation in a cross-modal priming study. They presented participants with a recorded sentence (e.g., “In Florida he photographed a flamingo”, ibid.: 1041) that either included a contrastive accent on the sentence-final word or not. After each sentence, a word appeared on the screen and participants had to indicate whether it was a real word or not. The word on the screen was either a contrast alterative to the last-mentioned word of the sentence (e.g., “pelican”), non-contrastive (“pink”) or a control word (“celebrity”). The results revealed that when the sentence-final entity (i.e., “flamingo”) received a contrastive accent, participants reacted faster to the contrast alternative (i.e., “pelican”). The authors argued that contextual alternatives became activated and more salient, although the sentences were uttered in isolation and the contrast alternatives were not mentioned explicitly.

These studies suggest that the contrastive accent L+H* might activate contrast alternatives even when they are not explicitly mentioned in the context. These observations could imply that in our experimental conditions (1a and b), the presentation of rising accents (L+H* and L*+H) might have evoked the activation of unmentioned contrast alternatives. The activation of unmentioned contrast alternatives to the target adverbial might lead to a slowdown in language processing because participants would have to process not just the actual word but also the co-activated contrast alternatives. This slowing down of processing might have cancelled out the potentially facilitating effects of prosodic prominence on target recognition, leading to a zero-sum situation and accounting for the lack of an effect of prosodic prominence in our word monitoring task. Evidence against this assumption, however, comes from experimental conditions (2a and b) where the word preceding the target adverbial was presented with rising accents (L+H* and L*+H). In these two conditions we would, hence, expect an addition of two hindering effects on language processing: First, the contrast-evoking rising accents on the word preceding the target word might co-activate unmentioned contrast alternatives, which would slow down processing. In addition, processing the prosodic prominence of the word preceding the target word should also slow down recognition times of the target word, as resources are bound to processing the prosodic prominence of the preceding word and are, hence, lacking for parsing the target word immediately following the prominent word. Consequently, these two experimental conditions should not only lead to the slowest recognition times in our word monitoring task but should also result in significantly slower recognition times compared to the baseline condition (ø) where neither co-activation of unmentioned contrast alternatives nor the processing of prosodic prominence would hinder the recognition of the target word. Note, however, that the data display no evidence for a significant difference in word recognition times of these two experimental conditions against the baseline condition. Rather, recognition times in both experimental conditions (2a and b) were numerically shorter than recognition times in the baseline condition. Likewise, the comparison between experimental conditions with rising accents on the target word and conditions with rising accents on the word preceding the target word (RQ 3) indicated no differences in word recognition times. Such differences, however, would have been expected under the assumption that rising accents lead to the activation of unmentioned contrast alternatives. This should have resulted in an accumulation of processing costs in experimental conditions where the word preceding the target word was prosodically manipulated. We, therefore, conclude that the possible activation of lexical competitors was not decisive for the lack of an effect of prosodic prominence on word recognition times in our study.

While we found no evidence that a prominent accent affects word identification times in an on-line word monitoring task, studies using the off-line measure of word recall have quite consistently reported an effect of prosodic prominence on the recall of accented words both in sentences (e.g., Fraundorf et al. 2010; Kember et al. 2021) and serial lists (e.g., Savino et al. 2020), indicating that prosodic prominence serves to anchor words in working memory more deeply. Furthermore, eye-tracking studies, for example, have repeatedly shown facilitatory effects of prosodic prominence in visual search tasks (e.g., Weber et al. 2006; Chen et al. 2007; Ito & Speer 2008; Braun & Biezma 2019), i.e., participants found target referents earlier when they were presented with a prosodically prominent accent. However, in these studies prosodic prominence served its function of indicating contrast to an element given in a preceding context sentence or in a visual display. This suggests that it was not prosodic prominence alone that facilitated the search for the target referent but it was the prediction generated from the interplay of the accent type used and the contrast relationship between the referents. However, in this present study, we sought to investigate whether prosodic prominence per se affects on-line language processing, i.e., when it is presented without an explicitly contrast-inducing context. In contrast to findings reported by Cutler & Swinney (1987) who also conducted a word monitoring task to investigate the influence of prosodic prominence on on-line language processing, we found no evidence that word identification times were affected by prosodic prominence. We assumed that prosodic prominence leads to hyperarticulation and speculated that this might result in a faster lexical selection of the prosodically prominent word and, hence, in faster word recognition times in our word monitoring task. However, our results provide no indication for such a facilitatory effect of prosodic prominence on word recognition times. This might suggest that the process of selecting the correct lexical item is a stage during language processing that is not influenced by prosodic cues. More research using different on-line language processing measures is needed to determine whether this suggestion holds true and to gain a more detailed view on the influence of prosodic prominence on on-line language processing.