1.1 Prosodic prominence

One of the most essential aspects of spoken communication is an appropriate interpretation of an utterance’s prosody. That is, we only understand all of the intended meaning of an utterance in a discourse context if we are able to process and judge a speaker’s accent placement and intonation correctly. A central function of prosody is highlighting, i.e. making elements prominent in relation to neighbouring elements. Acoustically, the cues fundamental frequency (F0) movement, increased duration and intensity as well as spectral emphasis and vowel quality serve to lend prominence to a syllable or word (at least in Germanic languages and other intonation languages; see e.g. Fry 1955; 1958; Sluijter & van Heuven 1996; Kochanski et al. 2005). The illustrated German examples show the difference between a (prominence-lending) pitch accent on Kamilla (Figure 1) and lack of accent on the same word (Figure 2). Accentuation is cued here primarily by a rising local pitch movement on the lexically stressed syllable -mil- (vs. lack of movement in the unaccented version). In the example in Figure 2, the final accent has been “shifted” to the second syllable of gewunken (‘waved’), which now carries the most decisive pitch movement (and is longer and louder than in the first utterance).

Figure 1

Oscillogram and superimposed F0 contour for the utterance Wir haben Kamilla gewunken (‘we waved to Kamilla’) with a nuclear pitch accent on Kamilla. The accented syllable -MIL- is printed in capital letters.

Figure 2

Oscillogram and superimposed F0 contour for the utterance Wir haben Kamilla gewunken (‘we waved to Kamilla’) with a nuclear pitch accent on gewunken (‘waved’) and lack of accent on Kamilla. The accented syllable -WUN- is printed in capital letters.

The final pitch accent in an intonation unit is commonly referred to as the nucleus or nuclear accent, which has a special status in most phonological theories (most explicitly stated in the British School; see e.g. Crystal 1969). Structurally, the nucleus is most prominent, since it is the only obligatory element in its unit, often carrying the most distinct tonal movement. More formally oriented approaches define the nucleus as the head of an (intermediate) intonation phrase, assigning it a central role in the prosodic hierarchy (e.g. Beckman & Edwards 1990; 1994). Semantic-pragmatically, the nuclear accent marks the most important element, in the sense that its position determines the interpretation of an utterance’s information structure. Applied to the utterance in Figures 1 and 2, e.g., this means that the first utterance could be interpreted as broad focus, since its prosody makes it an appropriate answer to a question like “What happened?”, whereas the accent placement in the second utterance suggests a narrow focus context, in which Kamilla is treated as background information, i.e. as derivable from the previous context. (The relation between prosody and information structure will be discussed in more detail in section 1.2.)

Nuclear accents can thus be considered primary in terms of phonological strength, while prosodically prominent words or syllables before and after the nucleus – pre- and postnuclear elements, respectively – may be regarded as secondary in nature. This prominence relation (which is syntagmatic in nature) is illustrated in the metrical tree in Figure 3.

Figure 3

Metrical prominence relation (s = stronger, w = weaker) between prosodically defined elements (adopted from Liberman 1975).

Nevertheless, there is an important difference between the two types of secondary prominences: While prenuclear prominences can surface as fully-fledged pitch accents (usually being indicated by a combination of longer duration, higher intensity and local pitch movement), postnuclear prominences cannot, since the nuclear accent is by definition the last pitch accent in the phrase. Postnuclear prominences are not marked by pitch movement but mostly by increased duration and intensity. We will refer to this type of secondary prominences as phrase accents as proposed by Grice, Ladd & Arvaniti (2000). They define a phrase accent as an edge tone that is additionally associated with a lexically stressed syllable, lending a certain degree of prosodic prominence to the constituent in question. Several examples will be given in the course of the paper.

There is a growing body of behavioural studies investigating the perception of prosodic prominence. It has been shown, in particular for Germanic languages, that even untrained listeners are able to judge prominence consistently. One of the elicitation methods currently used is the so-called Rapid Prosody Transcription (RPT) method developed by Cole and colleagues (see overview in Cole & Shattuck-Hufnagel 2016) in which naive listeners have to indicate all words on a transcript which they feel to stand out in a given utterance.¹ The method records patterns of inter-transcriber agreement since it calculates the prominence score (p-score) for each word and can at the same time capture inter-transcriber differences in the prosodic annotation. Furthermore, although the prominence judgments are binary in nature, the resulting p-scores are (quasi-)continuous-valued since they can be translated into percentages, which not only indicate the probability of a word to be marked as prominent but also – however only indirectly – suggest different degrees of perceived prominence. In fact, the “graded prosodic labels […] can be used to test the contribution of individual acoustic cues or other non-acoustic predictors to the perception of prominence” (Cole & Shattuck-Hufnagel 2016: 10), which has been done in several previous studies. For spontaneous American English, Cole and colleagues (Cole, Mo & Baek 2010; Cole, Mo & Hasegawa-Johnson 2010; Mahrt et al. 2012) showed that duration and overall intensity (RMS) were particularly important for prominence perception. For German, Baumann & Winter (2018) found that pitch movement in the vicinity of the stressed syllable was the most important factor. Figure 4 indicates that the prominence scores for fully-fledged pitch accents (in prenuclear and nuclear position) were higher than the scores for postnuclear prominences. As can be seen in Figure 4 (on the left), phrase accents (referred to as “postnuclear accents” in the figure) were identified only as slightly more prominent than “no accents” by most listeners (mean: 7.4% vs. 2.1%), whereas prenuclear accents had a much higher prominence score (mean: 36.5%).

Figure 4

Percentages of average prominence marks for different accent positions (left) and accent types (right) in a Rapid Prosody Transcription task on German (Baumann & Winter 2018: 31).

The same study revealed a systematic ranking of pitch accent types affecting prominence perception (irrespective of their position in the utterance; see Figure 4 on the right). The classification was based on another metalinguistic prominence rating task (see Baumann & Röhr 2015) accounting for the factors direction of pitch movement (on)to the accented syllable (rising > falling), degree of pitch excursion (steep > shallow) and height of the accentual tone (high > mid/downstepped > low).

However, it has been shown that it is not only acoustic highlighting that leads to the perception of an element’s prominence but also expectations derived from the listener’s knowledge about the linguistic structure of a language. Cole, Mo & Hasegawa-Johnson (2010) found that word frequency and textual givenness influence a listener’s judgment of prominence as well (both standing in an inverse relation with perceived prominence). New information is typically processed faster when it is accented and given information shows processing advantages when being deaccented (e.g. Bock & Mazzella 1983; Terken & Nooteboom 1987; Birch & Clifton 1995). These effects are further modulated by information structural notions like focus, contrast and the presence of licit accents elsewhere in a clause (e.g. Nooteboom & Kruyt 1987; Sedivy et al. 1999; see Cutler et al. 1997; Birch & Clifton 2002 for a comprehensive overview). The interplay of acoustic cues and expectations has further been observed during the processing of intonational boundaries and in segmentation (e.g. Brown et al. 2011; Buxó-Lugo & Watson 2016). In addition, the more attentional resources are required in processing a signal that stands out or is otherwise unexpected (e.g. by encountering strong acoustic/prosodic cues and by the cognitive inaccessibility – i.e. newness – of lexical items), the higher is the probability that a word or syllable will be perceived as prominent (for attention allocation see Corbetta & Shulman 2002; Ranganath & Rainer 2003).

Actually, the expectation evoked by the discourse context may modulate a listener’s speech perception so that the information conveyed by the acoustic signal is reinterpreted. This has been shown by Bishop (2012), who conducted a prominence rating study on American English SVO constructions where acoustically identical target sentences with a potentially ambiguous focus structure (nuclear accent on the object in sentences such as …because I bought a MOtorcycle) were presented with context questions that induced varying focus expectations. Although previous (production) studies have shown that the prosodic differences between broad and narrow focus are often only subtle, and at the same time subject to speaker-specific variation (see Snedeker & Trueswell 2003), it was still revealing that the context-induced expectations led to systematic differences in prominence perception: the object in a narrow focus structure was judged as more prominent (and the verb as less prominent) than the object in a broad focus sentence (where the verb was judged as more prominent than under narrow focus). The results were consistent across individual subjects, although they can be assumed to differ, among other things, in their pragmatic skills (Bishop 2017). Bishop (2012) concludes that listeners are actually aware of the subtle differences in prosody-meaning mapping that speakers employ but that these differences usually do not surface in laboratory settings but only in more natural contexts utilizing a purposeful communicative goal (as e.g. in Breen et al. 2010).

In the current research, we are particularly interested in the real-time processing of secondary prominences and how they relate to two information structural notions, focus and newness. To this end, we turn to cases of Second Occurrence Focus, which will be described in more detail in the next section.

1.2 The relation between prosodic prominence and information structure

Following the logic of the allegedly universal effort code (Gussenhoven 2004) – i.e. the more important an item is for a speaker, the more articulatory effort s/he will spend to produce it – we may assume a more or less linear relation between linguistic importance and prosodic prominence. In fact, there is some evidence, at least for Germanic languages, that the information structurally relevant concepts of focus and newness are marked by greater prosodic prominence whereas given elements in the background are produced in a prosodically less prominent manner. To be more concrete, in English, German and Dutch, an item that is at the same time discourse-new and (narrowly) focused is often produced with a high or rising pitch accent on that item (as movies in (1B)),² while a discourse-given item is deaccented (indicated by “∅”), since it is predictable from the context, either textually or inferentially (as movies in (2B) which is lexically given). Deaccentuation implies lack of pitch movement, generally accompanied by reduced duration and intensity (see the description of Figs. 1 and 2 above).

More interesting than information-structurally and prosodically clear-cut examples like these are hybrid cases whose (prosodic) encoding and (cognitive) decoding is far less well understood. A case in point is the so-called Second Occurrence Focus (SOF), which is defined, in its classic form, as a contextually given expression that is at the same time morpho-syntactically focused by virtue of a focus-sensitive particle (Büring 2013; Baumann 2016).³ In the famous example (3) the second mention of vegetables is both focused (due to only) and textually given, while the first mentions of both vegetables and Paul are focused and new (and are thus called First Occurrence Focus, FOF).

The assumption of standard “association with focus” theories (e.g. Jackendoff 1972) is that a focus particle like only is associated with a syntactic constituent which contains a semantic focus. This semantic focus in turn contains at least one element which is marked by prosodic prominence. Acoustic and articulatory production studies have shown that FOF elements are generally marked by fully-fledged pitch accents (indicated in (3) by capital letters on the accented syllables), whereas SOF elements are marked by phrase accents, i.e. postnuclear prominences expressed by increased duration and intensity – in comparison with Background elements – but not by tonal movement (indicated in (3) by small capitals) (see Rooth 1996; Bartels 2004; Beaver et al. 2007 for American English; Féry & Ishihara 2009 and Baumann et al. 2010 for German).

As mentioned in section 1.1 above, phrase accents can be regarded as secondary prominences, here reflecting the combination of boosting (FOCUS) and inhibiting (GIVENNESS) information structural factors (see Féry & Ishihara 2009). Following up on this line of argumentation, FOF elements can be claimed to be marked by two boosting factors (FOCUS and NEWNESS), leading to a primary prominence, and Background elements (as e.g. Mary in example (3)) to be marked by two inhibiting factors (NON-FOCUS and GIVENNESS), resulting in complete lack of prominence. Equivalent weighting procedures with factors that have a “positive/amplifying” or a “negative/weakening” influence on an element’s surface prominence have been proposed by Selkirk (2008) and Beaver & Velleman (2011). What all these approaches have in common is the assumed relevance of three distinct levels of prosodic prominence that mirror three distinct levels of information structural weight or importance. (4) translates this relation into a system of binary features.

(4)

In the largest empirical study on SOF to date, Beaver et al. (2007) investigated not only the production but also the perception of SOF elements, in comparison with Non-Focus items (here: background elements that are not in the scope of a focus particle, see examples (5) and (6), which are adopted from Beaver et al. (2007); ⁴ FOF elements were not tested in the perception part of the study). Since it has been claimed that SOF is an “inaudible” focus (see Krifka 2004), it was important to test whether the acoustic and articulatory correlates found for SOF actually are perceptible.

As stimuli for the perception test the authors selected 40 minimal sentence pairs like (5C) and (6C) from the production data taken from the same speaker. Subjects had to judge in which of the two isolated sentences a target word (here: Sid) was more prominent than a competitor (here: court). The experiment revealed that SOF targets were judged as more prominent than Non-Focus elements in 63% of the cases (all 14 subjects performed above chance). This result can be taken as support for the assumption that the secondary prominence of SOF does not only manifest itself in acoustic features (increased duration and relative energy⁵) but also has perceptual relevance.

There has been some debate about whether only the “classic” SOF cases with their verbatim repetitions of first occurrence expressions are marked by secondary prominences or also elements that are accessible from the context but which are expressed by lexically new material (or at least not by identical copies of the first occurrence expressions). Krifka (2004: 203) called this specific group of cases “quasi second occurrence expressions”, and we will borrow this term for the present study. In fact, we will refer to elements which display a combination of boosting and inhibiting factors but which are not morpho-syntactically marked as focus as Quasi-SOF items.

An example from Büring (2007) for this type of information structural hybrid is given in (7). Here, the butcher constitutes an epithet as described by Clark (1977), namely as a bridging inference that adds some derogatory information about an antecedent.

That is, the butcher is identical with the previously mentioned Dr. Cremer (and thus referentially given or coreferential) but at the same time consists of a discourse-new expression (i.e. it is lexically new). This distinction between a referential and a lexical level of givenness⁶ leads to a more fine-grained differentiation of the information structure categories discussed so far, refining the strictly binary model in (4) above.

This is illustrated in (8). Note that for the purposes of the present study, “±focus” refers to the presence or absence of a focus particle, which will be restricted here to only. In the framework of alternative semantics (based on Rooth 1992), only is an (exclusive) expression whose interpretation is sensitive to focus (see also the stimuli description in 2.2, Beck 2016, as well as Beaver & Clark 2008 for an elaborate definition of only and its semantic effects). That is, only associates with the target words as the semantic foci in FOF and SOF structures, which makes them “important”, but they differ in whether they are at the same time contextually new (more “important”, FOF) or given (“less important”, SOF). Nevertheless, SOF is ranked above Quasi-SOF because the former is focused by the exclusive expression only, which may be considered stronger than standing in “free focus” (i.e. only contextually focused but not associated with a focus-sensitive operator, see Büring 2013) like Quasi-SOF expressions. Furthermore, Quasi-SOF items are only lexically but not referentially new, indicated by the combined ± feature for newness, which is considered weaker than the + feature for focus in SOF.⁷ Finally, Background elements are neither contextually nor morpho-syntactically focused and additionally given information, which clearly makes them least “important”.

(8)

Actually, the degree of prosodic prominence (or rather lack of it) of Quasi-SOF items like the butcher in (7) is disputed and probably subject to a large amount of variation. In German and English, such expressions will either be deaccented (as proposed by Büring 2007) or carry some kind of secondary prominence (as suggested e.g. by Rooth 1996 and Krifka 2004). However, these proposals are based on individual intuitions rather than an extended data analysis. In any case, though, a fully-fledged pitch accent on the butcher is clearly prohibited, since it would rule out the intended coreference reading. For our experiment, we chose to produce Quasi-SOF items as deaccented, following Büring’s (2007) original intuition.

While previous research on SOF has employed behavioral measures, we investigate the real-time correlates of SOF and Quasi-SOF using electrophysiological measures during language comprehension. This will allow us to assess the contribution of prosody, focus and newness to the processing of SOF elements.

1.3 Neurocognitive processing of prosody and information structure

Event-related brain potentials (ERPs) allow for a fine-grained characterization of the time-course of the underlying processes (see Bornkessel-Schlesewsky & Schumacher 2016). They represent synaptic changes that are time-locked to a cognitive event and that are recorded from electrodes placed on the listeners’ scalp.

From a neurocognitive perspective, prominence-lending cues such as pitch movement or increased duration are computed as the sensory input unfolds, and expectations are incrementally built up for upcoming entities, including their prosodic realization (see Hickok & Poeppel 2015; Bornkessel-Schlesewsky & Schumacher 2016 for overviews). Input that mismatches these expectations results in a prediction error (see Friston 2010; Bornkessel-Schlesewsky & Schlesewsky 2019), which has, for instance, been shown to yield a pronounced N400 – a negative potential peaking around 400 ms after the onset of a critical entity. The N400 has been observed during semantic processing: the less expected a word, the more pronounced the amplitude of the N400 (e.g. Kutas & Federmeier 2011; inter alia); it has also been observed during referential processing: the less accessible a discourse entity, the more pronounced the amplitude of the N400 (e.g. Burkhardt 2006; Schumacher & Hung 2012). In prosody, inappropriate and thus unexpected accent types have also been shown to elicit an N400 (e.g. Heim & Alter 2006; Toepel et al. 2007; Baumann & Schumacher 2012).

Previous neurophysiological studies on the processing of information structure and prosody further reported on another relevant ERP marker, known as expectancy negativity (EN). This potential has been observed in anterior regions and as early as 200 ms after the onset of a target entity. It is commonly interpreted as being triggered by the expectation of a focused, and accented, constituent in a given discourse context. The contextual licensors in earlier studies consisted either of a focus-eliciting question (e.g. Hruska & Alter 2004; Toepel et al. 2009) or a focus particle assigning focus to its right-adjacent constituent (e.g. Heim & Alter 2006; 2007). These findings indicate that different cues (focus particles or context questions) can lead to similar processing responses to accented constituents.

Prominence-lending cues further serve as attention orienting signals. The mechanism of attention orienting requires listeners to update their mental model (e.g. when expressing a topic shift or introducing new information), which has been claimed to engender a late positivity (henceforth “LP”), i.e. a positive-going potential with a peak latency around 600 ms after the critical entity (see Burkhardt 2006; Schumacher & Hung 2012; Brouwer & Hoeks 2013; Wang & Schumacher 2013 – see also Bornkessel et al. 2003 for earlier onset latencies). This LP has further been shown to occur with prosodic cues: It was found for prosodically marked new information and focused constituents (e.g. Hruska & Alter 2004; Toepel et al. 2007) as well as with deaccentuation (Baumann & Schumacher 2012). Thus, the results for prosodic cues are mixed. Unexpected accents, too, may cause a LP (in addition to an N400), which may be interpreted as an instance of mental model repair, since unexpected accents can conflict with information structural cues and this conflict must be resolved (e.g. Magne et al. 2005; Toepel et al. 2007; Schumacher & Baumann 2010; Dimitrova et al. 2012; Brouwer & Hoeks 2013).

Baumann & Schumacher (2012) have shown for German that prosody and information structure are processed by evoking similar combinations of N400 and LP; that is, less expected cues engendered a more enhanced N400 and attention-drawing cues showed a pronounced LP. They crossed information structure (discourse-new vs. discourse-given) and accent type (L+H* vs. deaccentuation), as illustrated in (9)–(12), and observed effects of information structure in both the N400 and the LP time windows (new > given) as well as a biphasic N400-LP response to prosody (deaccentuation > L+H* accent). In this study, crucially, the mismatches – i.e. missing accent on the noun (here: Winzer, ‘winegrower’) in (10) and superfluous accent on the noun in (11) – did not pattern together relative to the matching conditions. Rather, prosodic cues (deaccentuation > pitch accent) and information structural cues (new > given) grouped together. This was observed both at the noun position (representing referential givenness) and at the position of the adjective in the target sentence (constituting lexical givenness; see Baumann & Riester 2012; 2013 for different types of givenness).

Interestingly, the biphasic patterns for prosody and information structure mapped onto different topographical distributions on the scalp: Prosodic cues led to ERP effects with an anterior maximum while information structural cues evoked ERP profiles that were most pronounced over posterior regions. This led the authors to suggest that prosodic and information structural cues are subject to the same general mechanisms (expectation-based processing and mental model updating) but that they may be processed by discrete underlying networks that surface on the scalp in an anterior-posterior divide. This latter claim is partly supported by previous studies on information structure in the written modality that show effects with a posterior maximum (e.g. Burkhardt 2006; Schumacher & Hung 2012) while research on ambiguity resolution associates discourse complexity with effects over anterior electrode sites (e.g. Kaan & Swaab 2003; van Berkum et al. 2007). Prosodic effects like the expectancy negativity have surfaced over anterior regions, but findings from the N400 or LP have been mixed.

Previous ERP research on prosody has usually employed a mismatch paradigm and contrasted the processing of appropriate and inappropriate prosodic realizations as a function of context (e.g. Hruska & Alter 2004; Magne et al. 2005; Heim & Alter 2006; 2007; Toepel et al. 2007; 2009; Dimitrova et al. 2012; Baumann & Schumacher 2012). In the following, we want to investigate the processing of prosodic prominence in a more direct way. Usually, two clearly different prominence values occur in matching vs. mismatching contexts (e.g. expressed by a steeply rising “contrastive” accent on the one hand and complete lack of accent on the other, see Toepel et al. 2007; Baumann & Schumacher 2012). A first step to investigate different degrees of accentuation was Schumacher & Baumann’s (2010) investigation of two accent types (H*, H+L*) plus deaccentuation, representing three degrees of prominence. H* stands for a high pitch on an accented syllable and H+L* for a falling accent with high pitch on the pre-accentual syllable and low pitch on the accented syllable. They investigated the comprehension of inferential relations in German (e.g. Sabine repairs an old shoe. In doing so, she cuts the sole.) and varied the accent types on the sentence-final inferentially linked definite expression. Previous research has identified an H+L* accent to be the most appropriate accent for this whole-part relation (Baumann & Grice 2006). And indeed, deviations from the expected prosodic realization were reflected in N400 effects, which were further modulated by the severity of the deviation (H+L*<H*<deaccentuation) and followed by a LP. Nevertheless, the aim of that study was once more to examine the contextual appropriateness of the target items and not the processing of their prosodic prominence.

In the present study, we are not concerned with mismatches but exclusively with contextually appropriate prosody, expressing three different degrees of prominence. In particular, we investigate the processing of primary and secondary prominences by comparing First Occurrence Focus, Second Occurrence Focus and Background in the response to short dialogue sequences. As outlined in (4) and (8) above, these conditions differ in terms of their information structural importance (±focus, ±new) and prosodic prominence (±pitch, ±dur). In addition, we will include a condition reflecting Quasi Second Occurrence Focus, which differs from SOF by being referentially given yet lexically new.

1.4 Research questions

The studies discussed so far provide evidence that secondary prosodic prominences, e.g. in the form of (postnuclear) phrase accents, may serve as markers of linguistically meaningful distinctions. Here, we presented Second Occurrence Focus (SOF) as a case in point. Investigations on SOF have shown that an intermediate level of prominence has a phonetic basis, and that it can also be perceived as different from higher as well as lower levels of prominence. Furthermore, different levels of prosodic prominence have been shown to be perceived differently in a wide range of behavioural studies, both metalinguistic and related to a linguistic task.

The present study is looking for evidence for the assumption that secondary prominence (here: phrase accents) is also neurocognitively processed in a different way than primary prominence (here: pitch accents) on the one hand and lack of prominence (here: deaccentuation) on the other. SOF is an ideal testbed for this research question, since it additionally provides clearly defined information structural categories such as focus, newness and (textual) givenness whose role in cognitive processing has been tested before but not in a sufficiently controlled way.

2.1 Hypotheses

Our hypotheses are concerned with the processing of prominence and with the potential difference between the processing of information structural and prosodic cues. Critically, the present study does not intend to investigate mismatches between the two levels of linguistic description but measures contextually licensed prosody. In other words, we claim that all prosodic realizations used in the test materials are appropriate in their given contexts – a claim which is based on the results of previous production and perception studies discussed above.

The processing of prominence affects both expectation-based mechanisms and mental model updating. Cues given by the context are used to generate expectations for upcoming entities and incoming prosodic or information structural cues modulate the amplitude of the N400. Prominence-lending cues further have the capacity to initiate updates of the mental representation, which is reflected in a more enhanced LP.

The current research seeks to tease apart effects of prosody (primary prominence, secondary prominence, no prominence) and information structure (newness and focus) by looking at ERP effects associated with expectation-based processing and mental model updating that have been shown in previous research. Accordingly, we formulate separate hypotheses for prosody (H1) and information structure (H2) below.

Hypothesis 1: As to the prosodic contrasts, we are primarily interested in the status of phrase accents as secondary prominences, i.e. we predict that they represent an intermediate level of prominence between pitch accents and deaccentuation.

The three levels of prosodic prominence tested may trigger stepwise differences in ERPs. Based on the findings of Baumann & Schumacher (2012) we expect decreasing N400 and LP amplitudes from deaccentuation (∅) through phrase accents to fully-fledged pitch accents. The ranking in (13) uses the phonetic parameters PITCH MOVEMENT (±pitch) and DURATION (±dur) somewhat simplified as binary features. In combination, they reflect the three levels NO PROMINENCE, SECONDARY PROMINENCE and PRIMARY PROMINENCE.

Translated into the information structural conditions which carry the three degrees of prominence, H1 can be stated as:

Building on the observation that prosodic and information structural mechanisms show distinct topographical distributions (see Baumann & Schumacher 2012), the prosodic effects are predicted to surface over anterior brain regions.

Hypothesis 2: Regarding the information structural contrasts, we investigate whether information structural effects are driven by ±newness or ±focus or a combination of them. If ±newness is the driving force, then SOF and BG should pattern together relative to FOF. If the distinction ±focus is primarily used, then SOF and FOF on the one hand should differ from BG on the other hand. If both newness and focus contribute to the underlying processes, a graded distribution is predicted.

In this latter case, we would expect increasing N400 and LP amplitudes from BG through SOF to FOF. The ranking in (15) uses the binary feature FOCUS (±focus) and the simplified feature NEWNESS (±new). The combined features indicate three levels of importance ranging from low for BG to high for FOF.

Following Baumann & Schumacher (2012), the information structural differences should have a maximum over posterior regions.

H1 and H2 work in opposite directions according to the results found in previous studies. If the two levels of investigation, i.e. prosodic prominence and information structural importance, operate independently and assuming different topographical distributions, we should observe the patterns predicted in (13)–(15). If the two levels interact with each other and are subserved by the same neural networks, we might observe that one cue outranks the other (reflected in ERP signatures in line with either H1 or H2). Alternatively, prosodic and information structural cues may interact with each other in an unweighted manner; since we are testing contextually licensed utterances, FOF and BG continuations are more predictable than secondary prominences, and the latter might thus evoke more processing effort with respect to expectation-based mechanisms and mental model updating.

Hypothesis 3: As to the exploratory investigation of the processing of Quasi-SOFs, we have the following predictions: With respect to prosody, Quasi-SOF items will be processed in a similar fashion as BG items, since both are deaccented. With respect to information structure, the Quasi-SOF condition can shed further light on whether newness relies on referential or lexical information: Quasi-SOF elements are lexically new (but referentially given) and should be processed similarly as SOF elements because they show attenuating (=coreference) and boosting (=lexical newness) effects just like SOF items (which are given but focused). Hence Quasi-SOF should pattern with SOF but lexical differences may evoke a moderate N400 amplitude, however less pronounced than (referentially and lexically new) FOF.

2.3 Results

Figure 10 depicts the grand-average ERPs time-locked to the onset of the critical word (e.g. Bier resp. Zeug in the examples above). It illustrates different ERP effects for anterior vs. posterior electrode sites. Over anterior regions, BG (red solid line) and SOF (black dotted line) show a biphasic pattern compared to FOF (blue dashed line), reflected in a more pronounced negative deflection between 250–400 ms that is followed by a positive-going wave between 750–950 ms. Crucially, BG and SOF do not differ from each other. Note also that there appears to be a very early effect for SOF; however, since this difference already emerges before the onset of the critical word, we refrain from interpreting the early negativity for SOF. We suspect that the pre-stimulus differences arise from length variation between the target sentences of the different conditions. The Quasi-SOF condition differs from SOF and BG: BG (red solid line) and SOF (black dotted line) evoke more pronounced negativities between 250 and 400 ms relative to Quasi-SOF (grey solid line). Subsequently, Quasi-SOF reveals a reduced positivity in the 750–950 ms time-window relative to BG and SOF.

Figure 10

Grand-average ERPs at selected electrodes for the contrast BG (red solid line) vs. SOF (black dotted line) vs. FOF (blue dashed line) vs. Quasi-SOF (grey solid line). Negativity is plotted upwards. Time course on horizontal axis spans from 200 ms before until 1400 ms after the onset of the critical word. An 8 Hz low pass filter was applied for visual presentation.

Over posterior electrode sites, the FOF condition (blue dashed) shows a pronounced negative amplitude roughly around 400–650 ms after the onset of the critical word relative to the other two conditions. Regarding the Quasi-SOF condition, electrodes over posterior sites reveal an enhanced negativity between 400 and 650 ms in contrast to SOF and BG, which is less pronounced than the FOF amplitude. In the window from 750–950 ms the FOF condition shows the least positive-going trend relative to the other conditions.

Statistical analyses for the time-window between 250 and 400 ms revealed a CONDITION × ROI interaction [F(3,60) = 21.83, p < .001]. Resolution of this interaction by ROI registered significant effects of CONDITION over anterior regions [F(3,60) = 13.78, p < .001] but not over posterior regions [F(3,60) = 1.27, p > .29]. Pairwise comparison in the anterior ROI yielded a significant effect of CONDITION for the contrast BG vs. FOF, SOF vs. FOF, BG vs. Quasi-SOF and SOF vs. Quasi-SOF but no effect between BG and SOF and between Quasi-SOF and FOF (see Table 2 for statistical details).

Analyses for the 400-650 ms window registered a main effect of CONDITION [F(3,60) = 9.80, p < .001] and an interaction of CONDITION BY ROI [F(3,60) = 17.45, p < .001]. Resolution of the interaction by ROI showed an effect of CONDITION over posterior sites [F(3,60) = 25.37, p < .001] and no effect over anterior sites [F(3,60) = 1.09, p >.35]. Pairwise comparisons in the posterior ROI revealed differences between all contrasts except for BG vs. SOF.

In the 750–950 ms time-window, there was a main effect of CONDITION [F(3,60) = 7.41, p < .001] and a CONDITION × ROI interaction [F(3,60) = 8.77, p < .001]. Resolving the interaction by ROI registered CONDITION effects over anterior [F(3,60) = 10.06, p < .001] and posterior regions [F(3,60) = 4.25, p < .01]. Planned comparisons in this window yielded an anterior difference between BG and FOF, and SOF and Quasi-SOF, a posterior difference between FOF and Quasi-SOF as well as differences in the anterior and posterior ROIs for FOF vs. SOF.