Glossa Psycholinguistics

2.1 Default mechanisms

Speakers exhibit several preferences when ordering the words in an utterance, such as syntactic preferences. For instance, they preferably place the constituent that serves as the subject of the sentence before the object (SO-order), mostly in sentence-initial position. This is the canonical word order in German (see, e.g., Jacobs, 1988; Lenerz, 1977; Reis, 1986; von Stechow & Sternefeld, 1988; among others). The seminal corpus study by Hoberg (1981) shows for the middle field (i.e., the region following the finite verb, as in Heute trifft sie ihn, lit., ‘Today meets she him’), that SO-sentences make 96% of all sentences in their corpus and are, thus, strongly preferred over OS-sentences. With respect to sentence-initial position, subjects are placed in this position in 50% of the cases (e.g., Sie trifft ihn heute), and objects, in 7% of the cases (e.g., Ihn trifft sie heute) (Bohnacker & Rosén, 2008).

This preference is also observable from the perspective of language comprehension, as shown by Bader and colleagues for German (e.g., Bader et al., 2000). In locally ambiguous sentences, such as (1), from Bader et al. (2000), speakers initially interpret the first noun phrase (NP) Fritz as the subject. When they process the second NP die Oma (‘the grandma_NOM’) and the dative verb geholfen (‘helped’), the case marking on the definite article leads to a reanalysis, evidenced by longer reading times on, or following, the disambiguating region. However, Weber and colleagues (2006), using a visual-world eye-tracking paradigm, showed that the SO preference can be modulated by intonation.

Bader and Häussler (2010) suggest three types of reasons for why speakers deviate from the preferred SO-order: discourse-related reasons (topics are likely to be placed in sentence-initial position, see Reinhart, 1981), semantic reasons (agents are likely to be animate, to be realized as the subject, and to be placed in first position – the “agent-first-principle”; see Clark, 1965; Jackendoff, 1972), and phonological reasons, such as constituent weight (short NPs precede long NPs; for the phenomenon called Heavy-NP Shift, see Arnold et al., 2000; Wasow, 2002). In a corpus study, Bader and Häussler (2010) investigated these three factors and found that lexical-semantic and discourse-related constraints both play a significant role. According to Bader and Häussler, the main discourse-related constraint is the requirement that topics occur in sentence-initial position. In most cases, subjecthood and topichood go hand in hand, since subjects are prototypical topics (Reinhart, 1981), and objects are likely to be the focus of the utterance.

If, however, topichood and subjecthood do not coincide, and if the topic is not the subject of the sentence, speakers deviate from their preference for placing the subject in initial position, and, instead, place the topic in initial position, even when it is the object of the sentence. In (2), for instance, Peter is the topical subject and is placed in sentence-initial position. Pasta is the object; it is the AC of the focus particle auch and, as such, has the information structural status of the focus of the utterance (the AC of auch is indicated by square brackets, the particle itself is in boldface).

An OS-structure occurs when the object is the topic, as in (3), where not the AC, but the focus particle itself is the focus.

In (3), Pasta is the AC of the stressed, following auch, and, as such, it is defocused. Keeping the object Pasta in the middle field, following the finite verb, and the subject Peter in the initial position does render the sentence strongly marked, or even ungrammatical (see (4)).

Based on the factors presented above, we assume that the default option for speakers would be to integrate the particle auch in its unstressed, preceding version, as in (2). However, speakers have a choice, and both pairs of utterances, (2) and (3), are possible. Which version is the preferred one is an empirical question that has not been systematically answered so far. The goal of the present study is to investigate experimentally whether – beyond the factors presented above – speakers are influenced by structural properties of the context, known as structural priming, when constructing an utterance that contains the focus particle auch.

2.2 Structural priming

Structural priming (see Bock, 1986; Branigan, 2007; Pickering & Ferreira, 2008; Ziegler et al., 2019; Ziegler & Snedeker, 2019) can be defined as the tendency to repeat a syntactic form in successive utterances (Bock, 1986). A classic example of structural priming is the use of a structure containing a prepositional object versus a double object structure. The seminal study by Bock (1986) showed that prepositional primes (e.g., The boy is handing a valentine to a girl) increased the incidence of prepositional utterances by 23%, relative to their frequency following double object primes (The boy is handing the girl a valentine), and double object primes increased the incidence of double object utterances by 22%, relative to their frequency following prepositional primes. The alternations involved variation in word order, and the phenomenon, thus, does not depend on superficial relationships between successive sentences, but on more abstract similarities. As stressed by Branigan (2007), the results show that priming alters the relative likelihood of producing one structure or another. It does not uniquely determine which structure has to be produced (see also Mahowald et al., 2016, for a meta-analysis of 73 peer-reviewed journal articles on syntactic priming).

Besides prepositional objects versus double object structures, priming has been observed for several other structures, including passives (Bock, 1986), transitives (Bernolet et al., 2016), relative clause attachment (Scheepers, 2003), temporarily ambiguous sentences with a main clause interpretation or a reduced relative clause interpretation (Pickering & Traxler, 2006), PP-attachment ambiguity with a VP-attachment or an NP-attachment (Branigan et al., 2005), production/omission of optional complementizers (Ferreira, 2003), and the form of complex NPs (Cleland & Pickering, 2003). It is known that priming occurs between expressions that involve the same syntactic categories, irrespective of whether they are content words or function words (e.g., Bock, 1989). There are several studies that investigated structural priming with respect to function words, such as complementizers (Ferreira, 2003), verb-particle placement (Gries, 1999; Konopka & Bock, 2009), and prepositions (e.g., Bock, 1989; Bock & Loebell, 1990). However, we are not aware of any study that reports on structural priming with respect to focus particles.

Priming appears to be sensitive to surface word order. Hartsuiker and Westenberg (2000), for instance, show that the position of PPs and the relative order of auxiliary and main verbs can be primed, whereas Pickering et al. (2002) show that expressions with the same constituent categories, but different surface word orders, do not prime each other. This suggests, as pointed out in the meta-analysis by Branigan (2007), that the relevant representations are specified for linear order.

An open question, when it comes to structural priming, is whether speakers are primed only by linear order, that is, by the surface structure of a sentence, or also by the underlying structure (known as the deep structure in transformational grammar, e.g., Chomsky, 1965). Bock et al. (1992) did not find that the deep structure of a prime sentence influenced the production of a target sentence with respect to passives. In a passive sentence such as Five people were carried by the boat, the subject five people is the underlying object of the corresponding active sentence (The boat carried five people), and the NP the boat, which is part of the optional by-phrase, is the underlying subject of the corresponding active sentence. In passive versus active sentences, thus, the surface structure and the underlying structure differ. Bock et al. (1992) used an experimental procedure that involved the production of a priming sentence followed by the description of an event (there is a tendency for the forms of priming sentences to be repeated in picture descriptions; see Bock, 1989). Half of the primes were actives, and half were passives; half had animate subject arguments with inanimate object arguments, and half had inanimate subject arguments with animate object arguments. The results show that inanimate subjects in prime sentences tended to elicit inanimate subjects in active sentences (e.g., The alarm clock wakes the boy), irrespective of whether those prime sentences were active (e.g., The boat carried five people) or passive (e.g., The boat was carried by five people). However, animate subjects in passive primes (e.g., Five people were carried by the boat) did not elicit animate objects in active sentences (e.g., The alarm clock wakes the boy). Thus, participants tended to repeatedly map entities with particular animacy features to the same surface position, indicating that there is a direct mapping of semantic representation onto surface structure representations, not onto deep structure representations. In the present study, we deploy structural priming in order to investigate to what extent speakers deviate from default mechanisms when constructing an utterance in discourse. While Experiments 1 and 2 target the linear order of constituents, Experiment 3 investigates whether the underlying syntactic structure of a context sentence influences the construction of a target utterance.

Priming has also been found to occur in dialogue (see Branigan et al., 2000; Cleland & Pickering, 2003; Haywood et al., 2005). The results of Branigan et al. (2000) have even shown larger priming effects than in Bock’s (1986) original study. Therefore, we expect to find priming effects in our dialogue study, as well. The next subsection is devoted to introducing the German focus particle auch in more detail, and to presenting the study that the present study is based on.

2.3 The German focus particle auch

What is special with respect to the German additive particle auch, compared to English also, is that German distinguishes between two variants of auch, namely, a stressed, following variant and an unstressed, preceding variant. From an information structural perspective, the AC of stressed, following auch is claimed to have the status of a contrastive topic (see Dimroth, 2004; Krifka, 1999; Sæbø, 2004; Sudhoff, 2010) or an aboutness topic (Dimroth, 2004; Sudhoff, 2010), whereas the AC of unstressed auch is the focus of the utterance. Stressed, following auch is often related to the AC being the subject of the utterance, and unstressed, preceding auch, to the AC being the object (e.g., Höhle et al., 2009). However, several of these factors are intermingled, and, as mentioned above, speakers might rely on default mechanisms when choosing between the two variants of auch: If the AC of auch is the subject, it most likely has the information structural status of a topic (Reinhart, 1981), which preferably is the AC of stressed, following auch.

Reimer and Dimroth (2022) conducted a sentence fragment arrangement task (see Gauza, 2018) in order to investigate to what extent speakers’ choice of preceding versus following auch indeed depends on the syntactic function of the AC (whether it functions as the subject or the object). As part of a short dialogue, speakers read a context sentence (see (5–6)). The context sentences contained a claim about something (e.g., about Peter and Mary wanting to eat fruits) and a subsequent claim about one of the two (e.g., Peter only having eaten peaches). The speakers had to reply to that latter claim by formulating a contrastive statement. The contrast was between the lexical items nur (‘only’) and auch. For instance, in the context sentence, it was claimed that Peter was the only person who ate peaches, and the target utterance claims that this is not true, because Mary ate peaches, too. At the same time, the focus particle nur in the context sentence, which either associated with the subject (5) or the object (6), indicated which expression should function as the AC of auch in the target utterance. Both the a) and b) sentences are felicitous continuations to the context.

In order to construct the target utterance (the a) and b) sentences in (5) and (6)), participants saw the words in randomized order on a screen and had to adjust them (by dragging and dropping) in such a way that they built a meaningful sentence and a coherent reply to the context sentence. After constructing the target sentence, speakers were asked to read it out loud. Reimer and Dimroth measured whether speakers integrated auch so that it preceded or followed the respective AC. Note that the audio files indicated that a preceding auch was always unstressed, and a following auch was always stressed.

The data of the experiment indicate that speakers have a strong preference for following auch when the AC is the subject (see (5b)) and for preceding auch when the AC is the object, (see (6a)), indicating that speakers rely strongly on default mechanisms. However, while the preference for preceding auch in the object condition was 93% (close to ceiling), the preference for following auch in the subject condition was only 69%. That is, there was a reduced preference for following auch in the subject condition, which Reimer and Dimroth (2022) attribute to the possible influence of the context sentence: As can be seen in (5), the context sentence contained the German focus particle nur (‘only’). Compared to auch, there is only one version of nur in German, which always precedes its AC, even in the subject condition. Thus, from a surface structural perspective, nur is similar to preceding auch, but different to following auch. Reimer and Dimroth (2022) claimed that speakers might have been primed by the context sentence, which led them to insert preceding auch (see (5a)) in a condition in which speakers actually have a preference to insert following auch (see (5b)). Note that there was no such clash between default mechanisms and structural priming in the object condition. Both the preference for preceding auch when the AC is the object and the SVO order of the context sentence lead to the insertion of preceding auch.

These data provide a first indication that speakers’ choices can be affected by surface structural properties of a context sentence (the presence of another focus particle that precedes its AC), which lead them to deviate from their default mechanisms. However, there is one further possible explanation. It is claimed in the literature that unstressed, preceding auch and stressed, following auch are related to different questions under discussion (QUD; see Roberts, 2012; also, Klein & von Stutterheim, 1987, von Stutterheim & Klein, 1989, for earlier proposals). While an utterance with unstressed, preceding auch answers an underlying wh-question (Who ate pears? → Mary ate pears, and also Peter ate pears), an utterance with stressed, following auch answers an underlying polar question (Mary ate pears. What about Peter, did he ate pears? → Yes, Peter ate pears, too) (see Dimroth, 2004). As a consequence, nur is the alternative to unstressed auch (not only Mary, but also Peter), while negation is the alternative to stressed auch, as semantically, both modify polarity (No, Peter did not eat any pears/Yes, Peter ate pears, too). Thus, it is possible that participants used more instances of unstressed, preceding auch because they were primed by the lexical item nur. In order to examine these instances of priming more closely, we conducted three experiments using a sentence fragment arrangement task. We utilized not only the focus particle nur (Peter hat nur [die Äpfel] gegessen ‘Peter only ate the apples’), but also negation (Peter hat [die Birnen] nicht gegessen/Peter hat keine [Birnen] gegessen ‘Peter did not eat any pears’; see below). The context sentences suggest that we are dealing with focus-associating uses of negation, as negation has scope over the direct object, not over the entire VP. Hence, we take all these particles – auch, nur, and negation – to be focus-sensitive operators.

In Experiment 1, we tested whether the strong preference for unstressed, preceding auch in the object condition found by Reimer and Dimroth (2022) can be mitigated by (a) surface structural properties of the context, and (b) lexical properties of the context. We investigate the influence of the surface structure of the context by using two lexical items that precede their associated constituents – the focus particle nur ‘only’ and the negative pronoun kein(e) ‘no’, and one item that follows its associated constituent – the negation particle nicht ‘not’.

At the same time, nur is a semantic alternative to unstressed, preceding auch, and kein(e) and nicht (both referred to as negation in the remainder of this article) are alternatives to stressed, following auch. In Experiment 2, we focus on kein(e) and nicht and manipulate the intonational pattern (hat contour versus two pitch accents) in order to disentangle the influence of the surface structure from the influence of intonation. Experiment 3 seeks to distinguish between the influence of the surface structure of the context sentence and the influence of its underlying structure.

In all three experiments, we operationalize the extent to which participants are primed by the context as the proportion of times that they deviate from the established preference for preceding auch in the object condition and, instead, position auch following its AC. In all three experiments, the prime and target sentences are part of a short dialogue, while the seminal investigation by Bock (1986) studies sentences in isolation. If sentence production in isolation enhances structural effects, we expect to find priming effects of smaller effect sizes than in the study by Bock (1986).

3.1 Materials and methods

3.1.3 Procedure

The questionnaire was created with SoSci Survey (Leiner, 2014) and made available to participants on www.soscisurvey.com. The experiment started with a short introduction. Participants were instructed to silently read the context sentences (speaker A’s and speaker B’s utterances) and to assemble speaker A’s second utterance by using a set of words which were presented with capital letters in small boxes in randomized order. In order to avoid a linearization effect, the word boxes were arranged not in linear order, but in zigzag fashion (see Figure 1). Above the single words there were five empty boxes of equal size arranged in linear order. Participants had to click on the word boxes given or drag and drop them into the empty boxes. It was possible for participants to remove the word boxes they had placed in the empty boxes and reposition them. There were four practice items, similar to the filler sentences.

Figure 1: Presentation of an experimental item of the sentence fragment arrangement task.

3.2 Results

The data of two non-native participants were discarded. Thirteen answers (3.4%) had to be excluded, due to an ordering of the fragments that led to an ungrammatical sentence or to an utterance that was not coherent with the specific context, for instance, because the AC of the particle was the subject instead of the object. The remaining sentences were classified as preceding ([X+ AC]: Ich glaube, dass Peter auch [die Birnen] gegessen hat/Ich glaube, dass auch [die Birnen] Peter gegessen hat) or as following ([AC + X]: Ich glaube, dass Peter [die Birnen] auch gegessen hat). Overall, speakers had a general preference for auch preceding its AC (grand average: 81.4%), and following auch was chosen in only 18.6% of the cases.

The two positions were analysed using a mixed-effects logistic regression model (glmer) in R (R Core Team, 2017; package lme4, Bates et al., 2015) with the optimizer “bobyqa” to avoid convergence issues. The fixed-effects factor was the element in the context (3 levels: nur/keine/nicht), and participants and items were added as crossed random effects (random intercepts; the inclusion of random slopes for the within-group factor context led to non-convergence). The model was refitted after removing 11 data points with residuals larger than 2.5 sd (3% of the data points). The results showed a significant effect of condition (χ² = 17.9, df = 2, p < 0.001). Tukey-corrected pairwise comparisons showed significant differences between nicht and the other conditions (β_nur = 1.5, SE = 0.5, z = 3.4, p < 0.005, β_kein = 1.8, SE = 0.5, z = 3.8, p < 0.001), but there was no difference between nur and keine (β = 0.2, SE = 0.5, z = 0.5, p = 0.8); see Figure 2 and Table 1. In other words, relative to the nur- and keine-conditions (on average, 14.5% choice of auch following its AC), this proportion increases to 27.1% in the nicht-condition, and a priming effect of 12.6% in the nicht-condition.

Figure 2: Results of Experiment 1 in % for the three conditions (nur/kein(e)/nicht); error bars refer to standard error of the mean.

3.3 Discussion

Based on the literature and on Reimer and Dimroth’s (2022) study, we expected participants to show a preference for auch preceding its AC when the AC is the object. This was borne out by the results of Experiment 1. Our participants showed an overall preference for auch preceding its AC (81.4%). This preference was not as high as in Reimer and Dimroth’s (2022) study, where the preference was 92.8%. We attribute this decreased preference to the fact that our experimental context sentences included not only preceding nur, but also preceding kein(e) and, crucially, following nicht, as described in the following.

The main focus of the study was whether speakers are influenced by the surface structure of the context, when its structure (e.g., [AC + X]) deviates from the structure they actually prefer for the target sentence (e.g., [X + AC]), or whether they are influenced by lexical properties of the context. Our results clearly point to an effect of structural priming: the results for the kein(e)-condition pattern with the results for the nur-condition – both share the structure that the particle precedes its AC. Compared to these two conditions, participants used more instances of following auch when the context contained the lexical item nicht, which also follows the object AC. Thus, when the context contains information that is in conflict with default preferences regarding the structure of the sentence, speakers are structurally primed by the context and deviate from their actual preference (integrating preceding auch) by 12.6%. The differences in outcomes between kein(e) and nicht indicate that the type of lexical alternative (in both cases, negation) did not influence the choice between unstressed, preceding auch and stressed, following auch.

To summarize, speakers strongly rely on default mechanisms when constructing their utterances. However, they also pay attention to the surface structure of the context sentence when choosing between the two variants of the particle auch. There were more instances of stressed, following auch when the context contained the negation particle nicht (which also follows its AC), compared to when the context contained nur and kein(e), which both precede their ACs. While we excluded effects of lexical priming in Experiment 1, Experiment 2 is designed to pursue another possible factor that might have led to more uses of following auch in the condition where the context contained nicht, namely, intonation.

The possibility cannot be excluded that participants silently read the three conditions with different intonation contours (see Brown, 1958; Fodor, 1998; Rayner et al., 2012; on the activation of stress and intonation patterns during silent reading). Compared to nur and kein(e), nicht is expected to be produced with a nuclear accent, which is generally perceived as more prominent than the expected prenuclear accents in nur and kein(e) (in which the following noun receives the nuclear accent). Post-hoc productions of the KEINE-sequences and NICHT-sequences in (7B’) and (7B’’) in the given context (7A) by four speakers each (in a between-subjects design, so that each speaker produced only one of the negations) tested this assumption. The results showed that speakers typically produced a rising prenuclear accent on kein(e) and the nuclear accent on the noun (Birnen ‘pears’) was downstepped (!H*) or an early-fall (H+L*). Nicht was typically produced with a rising (L+H*) or high nuclear accent (H*) and the preceding noun had either a rising accent or was unaccented. In order to control for unwanted confounds due to silent reading, Experiment 2 presented the contexts auditorily in two different intonation contours to test for (a) potential intonational priming (main effect of intonation) and (b) intonational influence on syntactic priming (interaction between intonation and particle in the context).

4.1 Materials and methods

4.1.2 Materials

In order to achieve a similar number of items per condition, we extended the set of 20 dialogues of Experiment 1 by twelve further dialogues, leading to 32 dialogues. Compared to Experiment 1, we included only the conditions keine [X + AC] and nicht [AC + X], which differ in their structure, and manipulated the pitch accent of the AC and the negation in the second context sentence (speaker B’s utterance; see (9) and (10)). The complete context sentences were recorded in the two conditions in a sound-attenuated cabin in the PhonLab at the University of Konstanz by two trained phoneticians (male and female). The duration of the recorded audio file was manipulated so that the two intonation conditions of each item were the same, using the PSOLA manipulation in the PRAAT software package (Boersma & Weenink, 2009). While one intonational realization contained a hat pattern (a rising L*+H on the first word, signaled by “/”, and an early fall, H+L*, on the second word, signaled by “\”; see (9B) and (10B)), the other formed a double-peak pattern (H* on the first word, H* on the second; see (9B’) and (10B’)); see Figure 3. All audio-files can be found in the OSF archive associated with this article. The manipulation of the lexical element (which is related to a specific structure either preceding or following the AC) and the intonation led to four conditions.

1. (9)
1. A:

1. Peter hat Äpfel gegessen.
2. ‘Peter ate apples’

1.  
1. B:

1. Ich wette, Peter hat /KEIne BIRnen\ gegessen.
2.                                            L*+H H+L*

1.  
1. B‘:

1. Ich wette, Peter hat KEIne BIRnen gegessen.
2.                                           L+H*    L+H*
3. ‘I bet Peter didn’t eat any pears.’

1. (10)
1. A:

1. Peter hat die Äpfel gegessen.
2. ‘Peter ate apples.’

1.  
1. B:

1. Ich wette, Peter hat die /BIRnen NICHT\ gegessen.
2.                                                   L*+H  H+L*

1.  
1. B‘:

1. Ich wette, Peter hat die BIRnen NICHT gegessen.
2.                                                  L+H*    L+H*
3. ‘I bet Peter didn’t eat any pears.’

As in Experiment 1, the target sentence (speaker A’s second utterance) was presented in fragments that had to be arranged by the participants.

The 32 dialogue quartets were allocated to four lists by means of a Latin Square design. Each participant received each item in only one of the four conditions. Sixteen filler sentences were added to each list; half of them had a hat pattern, the other half had a double-peak. This time, the filler sentences contained the particle auch. Compared to the experimental sentences, the AC of auch in the filler sentences was always the subject. Subject ACs are often related to following auch (see Reimer & Dimroth, 2022). The fillers could, thus, counterbalance the preference for preceding auch that we expected as the default in the target items, where the AC was the object.

Figure 3: Intonation contours of Experiment 2; hat pattern and double-peak for the kein(e)-condition; hat pattern and double-peak for the nicht-condition.

4.2 Results

The data of six participants had to be excluded. Two participants did not have German as their native language (L1); one participant created sentences in which the AC of the particle was the subject, instead of the object, in 97% of the cases; two participants created verb-second sentences in 30%–40% of the cases; and one participant showed mixed errors in over 30% of the cases.

Of all remaining answers, 88 answers (6.3%) had to be excluded, due to an ordering of the fragments that led to an ungrammatical sentence, or to an utterance that was not coherent with respect to the specific context, for instance, because the AC of the particle was the subject instead of the object.² The resulting 1320 answers were classified in the same way as in Experiment 1. Speakers had a general preference for auch preceding its AC (grand average: 70.1%) and following auch was chosen in 29.9% of the cases.

The initial data analysis was similar to Experiment 1. The fixed-effects factors were the element in the context (keine/nicht) and the intonation in the context (hat pattern/double-peak); participants and items were added as random effects (random intercepts, as the inclusion of random slopes led to convergence issues). The final model was refitted after removing 8 data points with residuals larger than 2.5 sd (0.6% of the data points). The interaction term was not significant at α = 0.05, and was, therefore, removed (β = 0.5, SE = 0.3, z = 1.8, p = 0.08); the main effect of intonation was not significant either (β = 0.1, SE = 0.15, z = 1.0, p = 0.3). The final model showed a main effect of condition only (β = 1.4, SE = 0.16, z = 8.7, p < 0.001).

As in Experiment 1, the presence of nicht led to more uses of auch following its AC (38.3%) than the presence of keine (19.9%). In other words, relative to the keine-condition, we observe a structural priming effect of 18.4% in the nicht-condition (see Figure 4 and Table 2). This pattern was not influenced by the intonation of the precontext.

Figure 4: Results of Experiment 2 in % for the two conditions (kein(e)/nicht) and the two intonations (hat pattern/double-peak); error bars refer to standard error of the mean.

4.3 Discussion

In Experiment 2, we tested whether participants’ responses are influenced by intonation, in addition to structural priming. If intonation had an influence, we would expect more uses of following auch in the hat pattern-condition, irrespective of the structure of the sentence.

As in Experiment 1, participants had a general preference for integrating auch in such a way that it preceded its AC. Crucially, we only found a main effect of element in the context, similar to the pattern reported in Experiment 1. When the context contained the negation nicht, speakers used more instances of stressed auch that followed its AC than when the context contained the negation kein(e). Again, participants were primed by the structure of the context. With a priming effect of 18.4%, participants integrated auch more often in such a way that it followed its AC – against their actual preferences, but in accordance with the ordering of the negation in the context sentence.

We did not find a main effect of intonation, indicating that the hat-pattern intonation in the context did not lead to more uses of stressed, following auch in the target sentence. This allows us to refute an intonational priming account. It is possible that the close succession of the two accents (in most cases, there were only one or two syllables between the accents) reduced the perceptual salience of the hat pattern (which maintains a continued high pitch between two accents). Note that, numerically, the number of instances of stressed, following auch was, on average, higher in Experiment 2 than in Experiment 1, in which there was no auditory presentation (nicht: 38.3% vs. 27.1%; kein(e): 20.5% vs. 14.4%). It is conceivable, hence, that the accents on both the particle and the noun led to a higher proportion of stressed, following auch per se.

In any case, the results did not show a main effect of intonation, suggesting that the prior perception of an intonational contour that is more typical for stressed auch, following its AC, did not influence the positioning of auch in the target sentence. There is, hence, no reason to assume that the results of Experiment 1 were driven by different kinds of silent prosody, depending on the element. Importantly, we replicated the main effect of element in the context reported in Experiment 1 (nicht led to more uses of following auch), which strengthens the structural priming interpretation.

It is still an open question, however, whether it is indeed the surface structure that primes the participants, or whether it is actually the underlying structure. In Experiments 1 and 2, kein(e) and nicht differed in their surface structure, but at the same time, they differed in their underlying structure. We therefore conducted a third experiment that further investigated this question.

5.1 Materials and methods

5.2 Results

The audio files were transliterated by a research assistant. The data of two participants had to be excluded. One participant did not have German as L1, and one participant recorded all answers with an infelicitous question/uncertainty intonation. The sentences were classified as in Experiments 1 and 2 regarding the position of the particle auch.

The model fitting was the same as in Experiment 1. The sole fixed-effects factor was the element in the context (keine/nicht). Participants and items were added as random intercepts. Models with random slopes did not converge. Data points with residuals larger than 2.5 sd from the regression line were removed (n = 23, 3.4%) and the model was refitted. As in Experiments 1 and 2, the results show that speakers have a general preference for auch preceding its AC (82.3%). Compared to Experiments 1 and 2, however, the difference between keine and nicht was not significant (β = 0.01, SE = 0.55, t = 0.02, p = 0.9), although there was a numerical trend for the presence of nicht to lead to more uses of auch following its AC (19.1%) than the presence of keine (16.4%) (see Figure 5 and Table 3). To statistically corroborate the difference in outcome between the underlying order and the surface word order (i.e., between Experiments 1 and 3), we created a dataset consisting of the data from Experiment 1 without the condition nur and the data from Experiment 3. The mixed-effects regression model contained the fixed effects experiment (1/3) and element in the context (keine/nicht). The model contained random intercepts for participants and items and was refitted after removing data points with residuals larger than 2.5 sd (n = 23, 3.4% of the data points). The results showed a significant interaction of experiment and element in the context (β = 1.32, SE = 0.63, t = 2.1, p < 0.05). While there is a difference between the conditions keine and nicht in Experiment 1, there is no difference between these conditions in Experiment 3.

Figure 5: Results of Experiment 3 in % for the two conditions (kein(e)/nicht); error bars refer to standard error of the mean.

5.3 Discussion

Experiment 3 tested whether the structural priming effect of Experiments 1 and 2 is triggered by the surface order or influenced by the underlying structure. To that end, we used the same negations as in Experiments 1 and 2 – kein(e) (underlyingly [X + AC]), and nicht (underlyingly [AC + X]) – but changed the surface structure and presented the sentences in such a way that both negations followed their respective AC. Thus, kein(e) and nicht differed in their underlying structure ([keine + AC]/[AC + nicht]), but were identical on the surface ([AC + X]). The results showed a general preference for unstressed auch, indicating that speakers strongly rely on default mechanisms. This overall preference is also in line with the results of Experiments 1 and 2, indicating that the change in modality (participants did not type in their utterances, but recorded them) did not change the general result. Compared to Experiments 1 and 2, however, there was no significant difference between kein(e) and nicht with respect to the choice of preceding auch – there was merely a tendency for preceding auch to be used more often if the context contains kein(e), compared to nicht. We hypothesized that if the underlying structure of a context sentence influenced speakers in their construction of a target sentence, the difference between kein(e) and nicht would be the same across experiments. However, this was not borne out. We therefore conclude that it is the surface structure of the context sentence that influenced speakers’ sentence planning. This finding is also in line with Bock’s (1989) findings.

Note, however, that we modified the introductory sentence in Experiment 3, in order to enable participants to be more flexible in their construction of the utterances. In contrast to Experiments 1 and 2, it introduced a main clause (Ich glaube, …). Due to this modification, the target sentence had a prefield (i.e., a preverbal slot), while the target sentences in Experiment 1 and 2 did not have a prefield. It is an open question whether this modification had an effect on the results of Experiment 3. The prefield of a German main clause is a prominent position for topics, which, in turn, are likely to be the AC of stressed, following auch. Following this reasoning, we would have expected participants to integrate auch in such a way that it follows its AC, independent of the element in the context. This, in turn, should have led to more uses of stressed, following auch compared to the other experiments. This was not the case, however, as only 17.7% of the uses of auch were instances of stressed, following auch, compared to 27.5% in the corresponding conditions in Experiment 1. A further difference, however, was that we used indefinite object NPs in the target sentences, while we used definite object NPs in Experiment 1 and 2. If speakers decide to place the subject pronoun in the prefield, they might prefer preceding auch if the object NP is indefinite, as following auch seems to be less acceptable with an indefinite object NP (e.g., #Ich glaube, er hat Äpfel auch gegessen vs. Ich glaube, er hat die Äpfel auch gegessen). This might have led to the decrease in production of stressed, following auch in Experiment 3.

We are aware of the fact that the structure used in Experiment 3 is highly unnatural and that modified sentence structures are often accompanied by specific intonational patterns. An anonymous reviewer pointed out that stranded kein(e) is heavily focused and as intonationally prominent as the stressed particle nicht, which, in turn, might have led to the result observed in Experiment 3. We conducted a further experiment, similar to Experiment 3, in which participants read the target sentence and typed in their answer. If the results of Experiment 3 were indeed intonationally driven, and if reading and typing eliminate or, at least, de-emphasize intonational effects, we would have expected to find a difference between kein(e) and nicht in this further experiment. However, as in Experiment 3, we did not find a difference between kein(e) and nicht, which leads us to conclude that it is not the intonation that influenced the speakers, but the surface structure. Of course, more controlled experiments are needed to study this specific sentence structure further.