While there are various reasons why the NA can occur on a non-final element in English (e.g. informational focus, contrastive focus, presence of an utterance-final indefinite pronoun; see Ladd 2008), this study investigates NA placement in broad focus intransitives.

Semantic/syntactic factors, such as verb type, have often been held responsible for at least part of the prosodic variability in English intransitives (e.g. Zubizarreta & Vergnaud 2006; Zubizarreta & Nava 2011; Irwin 2012). Some researchers have taken a syntactic approach in which NA placement is affected by the status of intransitive verbs as unaccusative versus unergative (e.g. Irwin 2012). Unaccusative verbs are analyzed as having an internal argument whereas unergative verbs have an external one (Levin & Hovav & Keyser 1995). Unaccusative verbs are non-agentive, such as verbs that denote appearance (e.g. appear, arrive) or change of state (e.g. melt, freeze). Unergative verbs, on the other hand, denote more agentive actions, such as walk or yell. While the specifics of the relationship between verb type, syntax and prosody may vary from model to model, it is generally assumed that NA assignment in English is closely intertwined with the syntax. Researchers have claimed that, due to the argument structure, sentences with unaccusative verbs trigger NA placement on the subject, whereas unergative verbs are said either to trigger NA placement on the verb (e.g. Kahnemuyipour 2009) or to exhibit variability, with the NA on the verb or the subject.

Others, however, have rejected the claim that NA placement can be based on verb type, even partially. For example, Bolinger (1954) proposed that NA assignment rests on the semantic or informational importance of words in an utterance, with words judged by the speaker to carry the most importance being more likely to receive the NA. Hirsch & Wagner (2011) found that once information structure or pragmatic factors, such as topicality, are controlled for, syntactic verb type does not matter for NA placement with intransitives, only discourse and pragmatic factors do. For example, Hirsch & Wagner examined verbs of appearance and disappearance, which are both unaccusative. They found that verbs of disappearance, where the subject could be construed as topical (something must already be introduced into the discourse in order to disappear), were more likely to elicit the NA on the verb, whereas verbs of appearance occurred more frequently with an accented subject.

Zubizarreta & Nava (2011) make a similar proposal that pragmatic factors influence NA placement in intransitives, but also tie pragmatic factors to syntactic verb type. In this approach, the NA is assigned to the subject in thetic (eventive) statements but to the verb in categorical statements (also known as topic-comment structures). Thetic statements are statements that introduce an event as a whole, whereas topic-comment statements contain two elements: a topic already established in discourse (the subject) and a comment about the topic. Zubizarreta & Nava argue that statements containing unaccusative verbs, due to the inherent lexical semantics of unaccusatives, are construed as thetic, and therefore are more likely to have an accented subject. Unergative verbs, on the other hand, can be construed as occurring in either thetic or categorical statements, depending on the context, which then leads to more variable NA placement on either the verb or subject. (5) provides an example of a thetic statement with an unaccusative verb and (6) provides an example of a categorical statement with an unergative verb.

The distinction between thetic and categorical statements with an unergative verb is said to be driven by pragmatic factors, such as noteworthiness, predictability, or expectedness. If a verb is less predictable (denoting an unexpected event), it is more likely to be accented, whereas if it is more predictable, there is a higher chance the subject will be accented. For example, the verbs in both (7) and (8) are unergative. However, in (7), the subject is more likely to be accented since birds are expected to chirp, whereas the verb is more likely to be accented in (8), as a bird burping is very unexpected, and therefore a noteworthy event:

Zubizarreta & Nava provide a limited amount of experimental evidence to support the above claim. They conducted an experiment examining the prosodic patterns of both native English speakers and L1 Spanish L2 English learners. The participants were asked to read simple intransitive sentences aloud; the sentences were contextualized with a preceding question. NA placement was determined through perceptual annotation. In this study, native English speakers placed the NA on the subject 97% of the time for unaccusative sentences, but only 42% of the time for unergatives; the remaining 58% of unergative utterances were produced with the NA on the verb. Zubizarreta & Nava used these findings to support their claims, noting that there is little variability in unaccusatives and high variability in unergatives. However, they did not explicitly test whether the factors they proposed to be responsible for NA placement variability in unergatives, such as predictability or expectedness, were in fact behind the variation found in unergatives.

NA placement in Spanish generally occurs at the end of an utterance, regardless of any information structure factors (Zubizarreta 1998; Hualde 2005; Zubizarreta & Nava 2011). Spanish has been categorized as a non-plastic language (Vallduví 1991), indicating that syntactic rather than prosodic manipulations are used to mark words in narrow focus. Zubizarreta (1998) has provided what is perhaps the most prominent model in the field, which accounts for the relative prosodic inflexibility of Spanish, while predicting prosodically-motivated syntactic movement to mark narrow focus. Non-final NA placement in non-plastic languages like Spanish is possible for marking contrastive focus as is seen in (10), which Zubizarreta’s model accounts for. In other cases, such as informational focus, focus is expressed through word order inversions, with the focused element occurring at the end of a sentence, as is shown in (9). This stands in direct contrast to a language like English, which has rigid word order, but greater flexibility in NA placement.

The lack of prosodic plasticity in Spanish, however, cannot be entirely taken for granted. Recent experimental studies have indicated that Spanish may have more prosodic flexibility than previously thought. With regard to narrow focus, written production tasks have found that native Spanish speakers do not always put the focused element at the end of a sentence, instead preferring to mark narrow focus in-situ (Hertel 2003; Hoot & Leal 2020). Oral production studies have corroborated these findings, showing that speakers are able to prosodically augment an element in narrow focus even when it occurs non-finally. Such findings have been found across various dialects, including Mexican Spanish (Kim 2016), Venezuelan Spanish (Calhoun et al. 2018), Castilian Spanish (Vanrell Bosch & Soriano 2013) and Argentinean Spanish (Gabriel 2010).

While much of the experimental research on narrow focus in Spanish has contradicted previous theoretical approaches about lack of prosodic plasticity in Spanish, less experimental work has been conducted to examine NA placement in broad focus contexts. One exception is Calhoun et al. (2018), who conducted an oral production task to test the effects of verb type (unaccusative versus unergative) and information status on word order and NA placement in Venezuelan Spanish intransitives. This study found that SV order occurred more frequently in Spanish intransitives than predicted. However, regardless of word order, the NA largely occurred on the final word. Landblom (2020) found similar results. This study examined Spanish word order and prosody by examining how both verb type and subject (in)definiteness affect word order inversions in a word order preference task and prosody in broad focus intransitives in a production task. While both factors affected the preference of SV versus VS word order, L1 speakers of Spanish were found to most commonly produce sentence final NA placement regardless of word order. These studies suggest a preference for utterance-final NA placement in Spanish broad focus intransitives. However, these studies are limited in scope and there is still more we could learn about whether prosodic flexibility in Spanish broad focus intransitives is affected by dialect or by pragmatic factors.

Given that Romance and Germanic languages vary predictably with regard to NA placement, they have often been contrasted in studies of L2 acquisition of plastic versus non-plastic prosody. Studies have examined whether native speakers of a non-plastic language can acquire the prosodic flexibility of a plastic language and vice versa (e.g. Krahmer & Swerts 2001; Rasier & Hiligsmann 2009; van Maastricht et al. 2016). Most studies have focused exclusively on how NA placement in narrow focus constructions is acquired. For example, Rasier & Hiligsmann (2009), a bi-directional production study, found that French speakers tended to transfer L1 accentual patterns into L2 Dutch, and as a result, were less likely to deaccent given/old information in their Dutch productions than did native Dutch speakers. L1 Dutch/L2 French speakers, on the other hand, were more likely to imitate native French patterns, exhibiting less evidence of transfer. Based on these findings, the researchers propose that acquiring a plastic prosodic system is more difficult than acquiring invariant prosody in a non-plastic system. They used the Markedness Differential Hypothesis (Eckman 1977) to explain their findings, arguing that NA placement rules in French are only structurally driven, whereas those in Dutch are driven by both structural and pragmatic rules, making it the more marked of the two, and hence more difficult to acquire.

Van Maastricht et al. (2016) also conducted a bi-directional experiment examining L2 acquisition of narrow, contrastive focus in L1 Dutch/L2 Spanish and L1 Spanish/L2 Dutch speakers. The results of this study provided evidence of transfer of L1 prosodic patterns in both directions. Additionally, speakers at higher levels of proficiency (for both speaker groups) had relative maximum pitch values (specifically the maximum pitch of the stressed syllable of the first word minus that of the second word) closer to those of the native speakers, indicating that transfer of the L1 patterns was decreasing with proficiency, and the learners were converging on the target L2 prosodic patterns.

While most L2 studies of prosody have examined narrow focus, English and Spanish also differ regarding the prosodic structure of broad focus intransitives, as discussed above. To examine whether speakers can acquire a non-final NA when it does not exist in the L1, and to learn more about the trajectory of acquiring flexible NA placement in an L2, it is important to examine broad focus as well as narrow focus contexts. Zubizarreta & Nava (2011), described in Section 2.1, is one of the few studies to include broad-focus intransitive stimuli in their investigation of NA placement in the L2 English of Spanish speakers. They suggest that L2 learners only acquire non-final NA placement at high levels of proficiency. In unaccusative sentences, the high proficiency learners produced an accented subject at a rate of 36% (compared to 97% for the native English speakers). In comparison, the intermediate L2 learner group accented the subject only 4% of the time. For unergatives, the rate of NA placement on the subject was 42% for native speakers, 39% for high-proficiency learners, but only 16% for intermediate learners.

Zubizarreta & Nava’s findings illustrate that it is difficult for Spanish speakers to acquire variable English prosody in broad focus intransitives. However, further investigation is required. As discussed in section 2.1, according to Zubizarreta & Nava (among others), NA placement in intransitives is driven by two factors: verb type and pragmatic factors, such as expectedness. Therefore, L2 English learners need to learn two distinct properties. First, they need to relate verb type to NA placement: on Zubizarreta & Nava’s proposal, unaccusatives occur in thetic statements, which have non-final NA placement on the subject, while unergatives can occur either in thetic statements or in categorical statements, and hence either the subject or the verb may be accented. Second, L2 English learners need to learn that pragmatic factors such as expectedness/predictability affect whether an unergative verb occurs in a thetic or a topic-comment statement, and hence whether the NA is assigned to the subject or the verb. The exact contributions of verb type versus pragmatic factors to NA placement have not yet been fully teased apart in studies with either native speakers or L2 learners of English. Our study aims to do just that, by independently manipulating verb type and (un)expectedness. Additionally, by recruiting speakers of differing proficiency levels, we hope to start examining the learning trajectory in more detail, by exploring whether verb types or pragmatic factors that affect English NA placement are acquired earlier.

Our study is a partial replication and extension of Zubizarreta & Nava (2011). The goal of our study is two-fold. First, we examine the effects of both verb type and expectedness on NA placement in English broad-focus intransitives. As discussed above, Zubizarreta & Nava proposed that unaccusative verbs favor NA placement on the subject due to their inherent lexical and syntactic properties; in contrast, unergative verbs exhibit more variability as they can occur in either thetic or categorical statements, depending on the discourse. However, while Zubizarreta & Nava demonstrated that unergatives show more variability than unaccusatives, they did not explicitly manipulate their stimuli for expectedness, only for verb type. In this study, we manipulate the two factors separately, asking the following research question:

The second goal of this study is to examine the L2 acquisition of intransitives more closely, by testing L1 Spanish L2 English learners and comparing their performance to that of native English speakers. Following the prior literature discussed above, we assume that English exhibits more flexibility in NA placement in broad-focus intransitives than Spanish, which nearly always places the NA sentence-finally. Therefore, L1 Spanish L2 English learners need to acquire the fact that English uses variable NA placement to distinguish between thetic and categorical statements; as discussed above, it is an open question as to whether the relevant factors behind variable NA placement are verb type, pragmatic expectedness, or both. Regarding L2 acquisition, we therefore ask the following research question:

As discussed in section 2.2, while there is more prosodic flexibility in NA placement Spanish than had been originally thought, this flexibility has been noted primarily in narrow-focus utterances. The studies that have looked at broad-focus intransitive utterances find that NA placement is usually utterance-final, regardless of word order. This is in sharp contrast with English, where NA placement in broad-focus SV intransitives is affected by both verb type and pragmatic factors. Following the prior studies on broad-focus intransitives in Spanish, we assume that transfer from Spanish should lead L1 Spanish L2 English learners to prefer NA placement on the verb in intransitive SV utterances.

Examining RQ2 with learners of different proficiency levels can provide insight into the trajectory of L2 acquisition of NA placement with broad focus intransitives, informing us as to whether it is easier to acquire non-final NA placement based on structural factors such as verb type, or on pragmatic factors like expectedness.

The norming task1 had a 2x2 design in which verb type (unaccusative versus unergative) was crossed with predicted expectedness (expected versus unexpected); 32 token sets were built. Verbs were drawn from Levin et al. (1995), who listed and categorized a large number of intransitive verbs. After the verbs were chosen, target sentences were created, each with a contextual question designed to elicit broad focus and to create a situation in which half of the target sentences would be expected as answers to the question, and the other half unexpected. We attempted to keep the target question the same for all conditions in each token set. However, due to the differences in meaning between unergative and unaccusative verbs, and the difficulty in setting up expected and unexpected contexts for each verb type, some token sets had two different contextual questions, one for both unaccusative verbs and one for both unergative verbs. Most token sets (n = 21) had only one question for all four verbs, as in Table 1, while the remaining 11 had two questions, as in Table 2: in the latter case, we tried to keep the contextual questions as similar as possible.

The norming task was administered to 42 participants via Amazon Mechanical Turk. Two participants were excluded due to being native in a language other than English; three participants who reported being bilingual in English and another language – one speaker each for French, Spanish and Mandarin – were kept in the analysis as they met the criteria of being a native speaker in English and reported English being spoken in the home while growing up. Following an example, participants saw question-answer pairs and were asked to rate how expected the response was as an answer to the question provided, on a scale from 1 (very unexpected) to 5 (very expected). The norming task consisted of one experimental list with the full 128 items (32 token sets X 4 tokens per set).2

To determine which of the token sets would be used for the production experiment, the ratings were averaged across participants. Then, for each token set, a difference score was calculated between the predicted expected and unexpected tokens, for both unaccusative and unergative verbs. The 12 token sets with the highest overall expectedness difference scores for both verb types were used in the oral production task. Table 3 shows the average ratings by condition for the 12 token sets selected.

To gauge whether the unergative and unaccusative stimuli differed in perceived expectedness, a post-hoc analysis was conducted, in the form of a linear mixed effects regression on expectedness ratings of the 12 selected token sets which was run using the lmer function from the lme4 package in R (Bates et al. 2015). Verb type (2 levels) and expectedness (2 levels) were introduced as fixed effects, and an interaction between these terms was also introduced. Both token and participant were included as random effects. Table 4 shows the output. Participants gave significantly lower ratings to unexpected than to expected verbs, but verb type did not have an effect, and did not interact with expectedness.

The 12 token sets selected for the experiment corresponded to 48 distinct sentences (12 per condition). A post-hoc frequency analysis was conducted on the verbs from these sentences to ensure any differences between conditions in the main experiment could not be attributed to word frequency. We examined the frequency of occurrence of each verb in the Corpus of Contemporary American English (COCA). Table 5 shows the average number of occurrences by condition. An ANOVA was conducted on the frequency scores, with verb type and expectedness as the main effects. The ANOVA results yielded a significant effect of expectedness (F(1,44) = 7.48, p < .05). Neither Verb Type (F(1,44) = 2.59, p = .11) nor the interaction between the two factors (F(1,44) = 1.39, p = .24) turned out to be significant, despite unaccusatives having a much greater frequency than the unergatives.

These results indicate that frequency is confounded with expectedness, as the expected verbs are much more frequent than the unexpected ones; thus, any effects of expectedness in the oral production task can just as easily be attributed to frequency. Crucially, however, the relationship between expectedness and frequency holds for both unergative and unaccusative verbs.

Prosodic units, such as nuclear accent, have generally been challenging to measure. Acoustic cues marking NA placement can vary, even within the same language. For example, the NA may be marked with an obvious pitch excursion at times, whereas other times the accented word is made prominent through other acoustic means, such as duration or intensity (e.g. Ladd 2008). There is also variation across languages, which has important implications for any study on the L2 acquisition of prosody. Because the NA can be difficult to measure, many studies have relied on perceptual annotation to establish NA placement (e.g. Zubizarreta & Nava 2011). However, this is subject to its own shortcomings, as this method relies on trained annotators to complete the task, which requires time and resources. Even though annotators may be trained, the nature of the task allows for the introduction of human error or bias. Because of this, it can be helpful to use a combination of acoustic measurements in determining NA placement, which can be collected automatically and are not reliant on the availability of hired annotators. This approach has been taken in numerous studies (e.g. Irwin 2011; Kim 2016). It is also possible to use acoustic measures in conjunction with perceptual annotation, to provide more robust, converging evidence of NA placement (e.g. Calhoun et al. 2018; Landblom 2020).

Many studies examining prominence at the phrasal level have investigated the acoustic correlates of narrow focus (e.g. Cooper & Eady & Mueller 1985; Xu & Xu 2005; Burdin et al. 2015). However, there are studies which have included broad focus constructions as well (e.g. Nguyen et al. 2008; Morrill 2011). There is no one unified way of measuring prominence, and methodologies vary. One approach which has been employed in various studies is to examine the relative difference in various acoustic measures of the stressed syllables between two adjacent words (e.g. Kim 2016; van Maastricht et al. 2016). In this approach, a variety of acoustic measurements are taken from the stressed syllables of both words, and then the measurements from word 2 are subtracted from that of word 1 for a difference measurement. The assumption behind this approach is that prominent words, those assigned the NA, will be acoustically augmented (e.g. higher intensity, higher maximum pitch, or pitch range), while unaccented words will be more acoustically compressed. Studies that have used this method with L1 and/or L2 speakers have included Nguyen et al. (2008), Hirsch & Wagner (2011) and Irwin (2011) for English, as well as Kim (2016) and van Maastricht et al. (2016) for Spanish. The present study adopts this methodology and uses the relative values of acoustic measurements between the subject and the verb to determine NA placement.

NA placement was analyzed through both quantitative acoustic measurements and perceptual annotation.6

As discussed in Section 4.2.1, the analysis includes the PC coordinates as opposed to each acoustic measurement, so for space reasons we include only those in this section. Figure 1 shows the overall values of each observation on the new PC1 coordinates and Figure 2 shows those for PC2. Based on these boxplots, it appears that for the L1 speaker group, the distributions for the expected versus unexpected tokens are separating for unaccusative verbs, but less so for unergative verbs. The PC values for the L2 speaker group, on the other hand, appear very similar across verb types and expectedness conditions with an exception for PC2 values, which show some separation between unaccusative verbs.

The lmer function from the lme4 package in R (Bates et al. 2015) was used for the regression models. Two different linear mixed effects models were run on both PC coordinates. The fixed effects included in both models were verb type (unaccusative versus unergative), expectedness (expected versus unexpected) and sex (female versus male). The reason for including sex in the models is that the PCs include pitch measurements. Model1 included speaker group (L1 versus L2). Model2 was built to examine the effect of proficiency for the L2 speaker group; this model included only the data from the L2 speakers, with proficiency test scores as a covariate. For all models, interactions between the fixed effects were evaluated as well. Speaker (n = 43) and Token (n = 48) were included as random effects. The final models were selected using the drop1 function in R. In this approach, all fixed effects and interactions are included in the original model. The drop1 function then evaluates the model and indicates whether a fixed effect or interaction can be dropped because it does not explain the variation in the data and does not contribute to the model. Interactions and fixed effects not contributing to the model are dropped one by one until there are no longer any predictors that can be dropped. Thus, everything included in the final model contributed significantly to the outcomes. P-values were computed from the final model using the lmerTest package in R (Kuznetsova et al. 2015). Follow-up pairwise comparisons were calculated using emmeans in R (Lenth et al. 2018) to explore significant interactions.

For PC1, Model1, the final regression model included verb type, expectedness, speaker group and two two-way interactions, between verb type and speaker group, and between expectedness and speaker group. Verb type was significant as a main effect. Both interactions were also significant (Table 8).

Further pairwise comparisons were conducted to better explore the two-way interactions. When expectedness was held constant, unaccusative and unergative verbs differed significantly for L1 speakers (t = 2.78, p < .05) but not for L2 speakers (t = 1.06, p = .29). Holding verb type constant, there is no significant effect of expectedness for either group (L1: t = 1.09, p = .28; L2: t = –.55, p = .58).

For PC2, Model1, the final regression model included verb type, expectedness and speaker group as main effects, as well as an interaction between verb type and expectedness, all of which were significant (Table 9).

Pairwise comparisons indicate that there is a significant difference between the two verb types in the expected condition (t = 3.9, p < .05), but not the unexpected condition (t = .74, p = .46). Sentences with unaccusative verbs have higher PC2 values than all other conditions.

Thus, as both PC1 and PC2 indicate, both verb type and expectedness have an effect on L1 speaker productions. PC1 indicates a significant effect of verb type, with higher values for unaccusatives than unergatives. PC2 indicates an interaction between verb type and expectedness, with the unaccusative, expected condition having higher values overall – a trend that is seen in the boxplots for both PC1 and PC2.

In comparison, the L2 speaker group, as a whole, appears not to be consistently distinguishing among conditions. It does appear, according to the results for PC1, that values for unaccusatives are higher than for unergatives, but this does not reach significance as it does for the L1 speaker group. Moreover, there is also a significant effect of speaker group, indicating that L1 speakers may have higher values than L2 speakers, which could signal a higher rate of sentence-final NA placement for the L2 speaker group. Such invariable NA placement would be consistent with L1 transfer.

Model2 for PC1, the regression model including only the L2 speakers and their proficiency scores, indicates that only higher proficiency speakers are beginning to produce the NA on the subject. The final regression model evaluating proficiency effects included only the main effect of proficiency score, which was significant, as is seen in Table 10.

For PC2, Model2 included the main effects of expectedness and proficiency, both of which were significant (Table 11).

Both the PC1 and PC2 results suggest that as L2 speakers’ proficiency scores increase, so do the PC values, which could indicate that higher proficiency speakers are more likely to start producing a non-final NA on the subject.

After the annotations were completed, two binomial mixed effects regression models were run to test the effects of verb type, expectedness and speaker group on identified NA placement. Model3 tested the effect of speaker group (L1 versus L2) as well as the other relevant factors, while Model4 included the L2 data only and tested the effect of proficiency. The dependent measure was the likelihood of a non-final NA. Item and speaker were included as random effects in both models barring any problems with convergence. Each model was selected using a forward stepwise regression selection method in which first a minimal model (including only random effects) is evaluated. Predictors are then added until the Akaike Information Criterion (AIC) number becomes larger instead of smaller (Akaike 1974). Models were run using the glmer function from the lme4 package in R (Bates et al. 2015).

Overall, the perceptual annotation results support the findings reported in Section 4.4 for the acoustic results. Figure 3 shows the mean percentage of NA placement on the subject as identified by the annotators. There is a preference for an accented subject in sentences with expected, unaccusative verbs for L1 speakers only. A final, accented verb is preferred across all other conditions in the L1 group and for the expected verb in the L2 group. For the unexpected verbs, there is about a 50/50 split for the NA identified on the subject versus on the verb in the L2 group.

The final binomial regression model for Model3 included verb type, expectedness and speaker group as fixed effects as well as an interaction between speaker group and expectedness. All main effects and the interaction were found to be significant (Table 12).

Pairwise comparisons using emmeans showed that there was a significant difference between expected and unexpected tokens for L1 speakers (t = 6.29, p < .05), but not for L2 speakers. Additionally, L1 speakers differed from L2 speakers significantly only in the expected conditions (t = 3.03, p < .05), and not in the unexpected conditions. Overall, L1 speakers are more likely to have the NA identified on the subject when the verb is expected versus when it is unexpected, whereas the L2 speaker group is more likely to have the NA identified on the verb regardless of expectedness.

Figure 4 shows the correlation of the proportion of tokens with an identified NA on the subject with the proficiency scores of the L2 participants.

Model4, on the effect of proficiency score for the L2 speakers only, included verb type and proficiency score as main effects, both of which were significant. Expectedness was not significant, nor was the interaction between the two main effects (Table 13). Only speaker was included as a random effect due to issues of convergence.

These results indicate that the higher in proficiency a speaker is, the more likely they are to place the NA on the subject. Sentences with unaccusative verbs have higher levels of non-final NA placement than those with unergative verbs. Visually, there is evidence that expectedness makes a slight difference within the unaccusative constructions, with expected verbs more likely to exhibit non-final NA placement than unexpected verbs. However, this trend is not strong enough to reach significance.

Given the results, RQ1, repeated below, can now be addressed.

Findings indicate that both verb type and expectedness play a role in NA assignment for L1 English speakers. The acoustic analysis provides evidence of an effect of both verb type and expectedness, with unaccusatives and expected conditions having higher values than other conditions. The perceptual annotation analysis supports these results, as it found that NA placement on the subject was most likely for L1 speakers for the expected, unaccusative condition. More variability is seen for unergatives, which is especially apparent from the perceptual annotation results.

These results partially replicate those from Zubizarreta & Nava (2011), who found that L1 speakers of English were more likely to produce the NA on the subject for unaccusatives, and that NA placement in unergatives was highly variable. However, there are two main differences. First, we found an effect of expectedness in unaccusative verbs, which was not predicted by the theoretical model proposed in Zubizarreta & Nava. Secondly, it was predicted that expectedness would affect NA placement in unergatives, but this was not found to be the case.

It is possible that the results found here are in part a reflection of frequency effects (see section 3.1). Frequency has been found to correlate with accentuation in that more frequently occurring words are less likely to be assigned prominence, as frequency correlates with predictability, which in turn contributes to the prominence structure of an utterance (e.g. Watson & Arnold & Tanenhaus 2008; Bell & Brenier & Gregory & Girand & Jurafsky 2009; Cole & Hualde & Smith & Eage & Mahrt & de Souza 2019). However, frequency cannot be the only explanatory factor: if it were, we would expect to see a greater degree of non-final NA placement in the expected, unergative verbs as well, as this condition had the second most frequently occurring verbs.

It is surprising that utterances with unergative verbs were produced overwhelmingly with final NA placement. One question that could be investigated further is whether this could be an effect of the read-aloud task. It would be good to compare the results of a more spontaneous speaking task to the findings in this study to see if sentences with unergative verbs still show a preference for utterance-final NA placement. Additionally, it may be the case that when speakers have only limited context, in this case, a single preceding question, it may be difficult to construe a statement as thetic. Schmerling’s (1976) examples show that thetic sentences are often uttered when there is ample context – more than what may be provided in one question. Additionally, there were no checks to see if participants were actually listening to or reading the contextual question. It may be helpful in a future study to find a way to ensure that participants are engaged in the dialogue. For unaccusative expected verbs, their inherent properties may lead to their being interpreted as thetic, in which case, context would matter less. This would lend support to Zubizarreta & Nava’s proposal of inherent lexical properties of unaccusative verbs leading to a thetic construal which, in turn, helps with the assignment of NA placement on the subject.

A limiting factor of this study, as well as of Zubizarreta & Nava (2011), is the fact that the differences between unaccusative verbs (cf. Sorace 2000) are not taken to account. Zubizarreta & Nava used a small set of unaccusative verbs, including many verbs of appearance. In contrast, a sub-experiment on unaccusatives from Hirsch & Wagner (2011) manipulated the type of unaccusative, using verbs of disappearance and of appearance only. The assumption was that verbs of disappearance lead to inherently more topical sentences than verbs of appearance, as the subject must already be implied in the surrounding discourse in order to disappear. This factor was not controlled for in our study, and coincidentally, there were four verbs of disappearance (six tokens) in the unexpected unaccusative condition (die(x2), expire, perish, vanish(x2)). Furthermore, many verbs used in the expected unaccusative condition were verbs of appearance (appear(x3), arrive(x3), come(x2)). Zubizarreta & Nava (2011) proposed that unaccusative verbs are construed as thetic due to their lexical semantics (p. 654). This seems to be only partially true, as a difference is seen between the two unaccusative verb types in NA placement.

Our findings are consistent with Hirsch & Wagner’s claim that topicality can be an important factor for NA placement in unaccusatives. While Hirsch & Wagner further claim that verb type plays no role in NA assignment, there may still be room for this factor, as the inherent function of such verbs as appear, arrive or come in a broad focus context is to introduce a referent into the discourse. If the referent has already been introduced, then it becomes old information and is prosodically treated as such (i.e. either reduced to a pronoun or prosodically reduced). It may be that the verb semantics interacts with topicality, thus driving NA placement. It would be fruitful to control for verb type in future studies. If there were more verbs of disappearance in the expected conditions, and more verbs of appearance in the unexpected conditions, would the patterns of NA placement change? Additionally, our findings suggest that topicality does play a role in NA placement assignment, which can extend across verb types, and should not be ignored in future studies.

We now address RQ2, repeated below:

Overall, it was found that L2 speakers as a group were less likely than the L1 group to make a distinction among conditions. The perceptual annotation results indicate a higher proportion of utterance-final NA placement. This is similar to prior findings for Spanish as an L1, suggesting a role for L1-transfer, especially considering the finding that the pattern of final-NA placement is more likely at lower levels of proficiency (when transfer is likely more pervasive).

For the second part of the question, there is evidence that L2 speakers are acquiring non-final NA placement as proficiency increases, as was shown in both the acoustic and the perceptual analyses. The perceptual analysis found an effect only of verb type, whereas the acoustic analysis found evidence of an effect of expectedness only. Taken together, these findings suggest that L2 speakers are learning patterns on non-final NA placement on the subject. It is still unclear what aspects of NA placement – pragmatic or syntactic – learners are acquiring. The acoustic analysis indicates an effect of expectedness only. However, since expectedness correlates with frequency (see section 3.1), this could also be a frequency effect, with learners struggling to pronounce words that are less frequent, and therefore quite possibly less familiar to them. The perceptual annotation analysis indicates that speakers may be acquiring non-final NA placement based on verb type, which would indicate that speakers are acquiring structural factors before pragmatic ones. However, since this was only found in the perceptual annotation analysis, it would be important to confirm this with more acoustic data. It may be that the relatively small sample size (23 L2 participants) made it difficult to see any trends more clearly. It would be beneficial to collect data from a larger group of L2 speakers who vary in proficiency, in order to determine whether NA placement based on verb type or pragmatics is acquired first.

The fact that there are no clear patterns of non-final NA placement replicates Zubizarreta & Nava’s (2011) findings that acquiring this prosodic pattern in English is difficult for speakers of Spanish. While there is an effect of proficiency, with higher proficiency speakers becoming more target-like, the effects found for the L1 speaker group, such as higher rates of non-final NA placement in sentences with expected, unaccusative verbs, are harder to see for the L2 speaker group, even for speakers of higher proficiency levels.

It would also be fruitful to conduct a study where participants record sentences in both their L1 Spanish and their L2 English. As discussed in Section 2.2, this study assumes utterance-final NA placement for broad-focus unaccusatives in native Spanish, which is an assumption based on the results of prior studies (Calhoun et al. 2018; Landblom 2020). However, neither of these studies examined the effects of pragmatic factors such as expectedness in native Spanish. Collecting both Spanish and English data from the same set of participants would strengthen our conclusions about the role of L1 transfer.