2.1 Defining focus

We examine the notion of focus, defined as the part of the sentence that is not presupposed but provides new information, evokes alternatives, or closes an open variable in the discourse context (Rooth 1992; Vallduví 1992; Krifka 2007; López 2009). Researchers have distinguished between types of focus, based either on the pragmatic inference conveyed or on the size of the focus domain. Specifically, previous work differentiates information (or presentational) focus from contrastive (or identificational) focus; the former simply conveys new information while the latter encodes additional pragmatic interpretations such as correction, contrast, or exhaustivity (É. Kiss 1998). Regarding domain, the term wide (or broad) focus is commonly used to denote instances where the focus extends to the entire sentence, whereas focus on a single constituent constitutes narrow focus. Here, we investigate narrow information focus, illustrated in (4), where the focus of the sentence—marked with the subscript F—is determined by the discourse context. For brevity, we refer to this phenomenon simply as focus hereafter.

Crosslinguistically, focused constituents have been argued to carry prominence (Jackendoff 1972; Truckenbrodt 1995), but languages differ in how they achieve such prominence. Intonation languages, such as English and German, mainly use prosody, shifting the nuclear pitch accent to match the focus location (Selkirk 1986; 1995; Büring 2001). French, specifically colloquial French, is known to use different syntactic constructions and most notably clefting (Lambrecht 1994; Destruel 2013), while languages like Japanese, Korean, and Gùrùntùm use morphological markers (Hartmann & Zimmermann 2009). Finally, languages like Italian or Hungarian use syntax to alter the word order so that the focal constituent occupies a structurally prominent position, often either by moving the focus to the left periphery or by scrambling other constituents to leave the focus rightmost (Rizzi 1997; Zubizarreta 1998; Szendrői 2017). This last strategy is what has been argued to happen in Spanish.

4.1 Participants

We tested 42 monolingual native speakers of Mexican (n=20) and Chilean (n=22) Spanish in Puebla, Mexico and Santiago, Chile, where they resided. We tested Mexican speakers because this variety is well documented in the experimental literature on information structure (see Hoot 2016; Leal, Destruel & Hoot 2018, among others). Conversely, Chilean Spanish was selected because few studies have focused on it regarding information structure. We also considered the fact that Chilean Spanish is not in significant contact with Mexican Spanish or English and is not a Caribbean dialect, because Caribbean Spanish has different syntactic restrictions on subject position (Camacho 2006; Ortiz López 2009). Participants were compensated for participation.

4.2 Procedure and materials

Participants watched short clips from a silent film—Charlie Chaplin’s The Circus (1928)—in front of a computer. At the end of each clip, participants had ten seconds to answer a wh-question that appeared on the screen. The software, SuperLab 5.0, did not allow for video replay. To discourage single-constituent answers, the instructions asked participants to answer using complete sentences and to provide copious details. Answers were recorded with the software Audacity.

We manipulated the grammatical function of the focus element: wh-questions targeted either the direct object or the prepositional phrase, as illustrated in (7) and (8). Object focus was compared to PP focus because PPs appear finally both when focused and in the canonical word order.

Experimental items included a subject, a transitive verb, an object, and a prepositional phrase. Each condition included five lexicalizations, matching the actions in the clips. The task also included 65 fillers. We distributed the ten experimental items and the fillers across three lists (25 items each), so each participant saw three or four experimental items and 21–22 fillers. Because our prompts were based on the story depicted in the videos, it was not possible to control for the frequency, complexity, or prosodic weight of the constituents. The full instrument is available in Supplementary File 2.

4.3 Results

We transcribed responses orthographically with the aid of a research assistant and then, to ensure accuracy, randomly checked 10% of the transcriptions. When finding discrepancies, we reviewed the entire transcript of a given participant. We coded the responses following conventions adopted in our previous work (Leal, Destruel & Hoot 2018). Among in-situ realizations, we distinguished between instances in which the object was focalized in its canonical position (In-Situ, i.e. (S)V[O]_FPP) and instances where the object’s surface position was sentence final because the PP was elided (Elision, i.e. (S)V[O]_F). Within ex-situ realizations, we found three movement strategies: (i) sentence initial (Movement:Fronted, e.g., [O]_FV(S) and [PP]_F(S)VO), (ii) sentence final (Movement:Final, e.g. (S)VPP[O]_F), and (iii) movement to positions other than sentence-initial or sentence-final (Movement:Other). Finally, we used Other and Misunderstood/No response for answers that did not fit into the previous categories or when participants misunderstood or failed to answer. Misunderstood/No Response codes were excluded from analysis. Similarly, because Other strategies were mainly single-constituent answers, mostly uninformative for our analysis, we do not consider them hereafter. Finally, all authors reviewed every case to ensure consistent application of the codes. (Supplementary File 1 details our coding system.)

Figure 1 illustrates the proportion of responses in the object and PP focus conditions for the two groups.

Figure 1

Distribution of strategies for object focus marking and PP focus marking by group.

Overall, speakers largely prefer in-situ marking, that is, canonical SVOPP order, for object focus (42% and 60% for Mexican and Chilean speakers, respectively). This percentage is even higher for PP focus (in which case both groups average 70% and 72%). This result holds regardless of the object’s phonological weight.

Because marking focus via movement was uncommon, we collapsed all three movement types under the “ex-situ” label in Figure 1; nonetheless, some movement types were more common than others. Objects that were not in canonical position most frequently appeared in final position (18% for Mexican speakers, 14% for Chilean speakers)—precisely the strategy predicted under Zubizarreta’s (1998) analysis (see example 9). Fronting was very rare (see example 11); no examples were attested in the Chilean group, and only 6% in the Mexican group.

When examining PP focus movement strategies, we find dialect-related differences; only Mexican speakers use Movement: Other, as in (10) (2%), and only Chilean speakers use fronting, as in (11) (9.3%).

A less-common strategy (which does not apply to PP focus) was finding objects in rightmost position where post-focal material was elided, as in (12). Thus, speakers could leave the focus element in rightmost position without syntactic movement. Finally, clefts were never used to mark PP focus and only infrequently for object focus.

To determine statistical significance, we ran a generalized mixed logistic regression, using the glmer function in R (R Core Team 2014), which predicted focus-marking realization from the two fixed-effect predictors, treatment-coded prior to analysis: Group (Chilean: 1, Mexican: –1) and Focus Constituent (Object focus: 1 vs. PP focus: –1). For ease of analysis, we collapsed our categorical dependent variable (i.e., the open-ended responses given) into a binary outcome. We categorized responses depending on whether the sentence’s focus was realized in its canonical, in-situ position or not. Our models included random by-item intercepts and random by-participant intercepts and slopes for the two fixed-effect predictors, as well as their interaction. We report the estimate (β), standard error (SE), t-scores, and p-values, taking a p-value of 0.05 as the threshold for statistical significance.

Our analysis revealed no main effect of Group (β = 8.73, SE = 1.34, t = 0.65, p = 0.51) or Constituent (β = –2.75, SE = 1.33, t = –0.205, p = 0.83), and no interaction between the two (β = –6.14, SE = 1.34, t = –0.63, p = 0.52), suggesting that both native groups realize object and PP focus in similar ways: in-situ marking is favored overall.

Nonetheless, it is important to note that the order in roughly one in six responses to object focus questions across both groups was (S)VPP[O]_F, with the object in final position, whereas (S)V[PP]_FO never occurred under PP focus (as expected). Because there were no (S)VPPO instances in PP focus, we could not carry out a statistical analysis of the difference, but these results nevertheless show a relationship between final position and focus for direct objects.

5.1 Participants

We tested monolingual native speakers of Yucatecan Spanish in Merida, Mexico (n=42).⁵ Participants reported their language history, dominance, ideology, and preferences via the Bilingual Language Profile (BLP; Birdsong, Gertken & Amengual 2012). Their demographic and language history characteristics are presented in Table 1.

We excluded speakers who spoke a different variety (e.g., anyone who grew up outside the state of Yucatan), had not learned Spanish in early childhood, reported significant contact with a language other than Spanish prior to age 14, or reported language-related or cognitive impairments. Fifteen people, not included in the count above, were excluded for these reasons. Participants were naïve and uninformed about the object of the study and were compensated.

5.2 Procedure and materials

Participants completed the forced-choice task and self-paced reading task in the same session.⁷ The forced-choice task lasted around 10–15 minutes and was delivered online via the software Qualtrics, but it was conducted in person (on a computer and, on a few occasions, on a mobile device).⁸ Participants completed the BLP during the same session.

The task included written instructions in Spanish, along with two practice items after which there was no feedback. Participants completed 64 forced-choice items, including 16 critical items—eight in each condition—and 48 fillers, randomized by participant. Our fillers tested other types of focus as well as unrelated structures. Each item was followed by three choices, also randomized by participant.

During each trial, participants read a question creating the context for object focus or focus on the adjunct prepositional phrase (PP). Then, participants chose the most natural sentence to answer the question from three word-order options. Questions were preceded by drawings depicting the action in the target sentences (see Figure 2).

Figure 2

Sample forced-choice item.⁹

The forced-choice task had a 2 × 3 design. The word-order options were (canonical) VOPP, VPPO, and Fronting (PPVO/OVPP). This last category was included because this study is part of a larger project investigating cross-linguistic influence in bilingual speakers. In Yucatec Maya, the most common way to realize focus is via fronting the focal constituent to sentence-initial position (Verhoeven & Skopeteas 2015), which is why the fronted options were included. While focus fronting is also possible in Spanish, it signals contrast rather than information focus (López 2009). Examples are found in Table 2.

To instantiate these structures, we created sixteen lexicalizations. Participants saw each lexicalization only once, either in the object-focus condition or in the PP-focus condition. Because lexical frequency can affect acceptability judgments, the words used were among the 5,000 most common Spanish words (Davies 2006). Additionally, because phonological weight affects pre- and post-posing of arguments (Heidinger 2013; 2015), direct objects and PPs were comprised of the same number of syllables. The full instrument is available in Supplementary File 3.

5.3 Results

The results of the forced-choice task are shown in Figure 3.

Figure 3

Forced-choice task results.

Figure 3 shows that VOPP is clearly the preferred word order for both focus contexts. To determine whether the proportions were different in the two focus contexts, we conducted three binary logistic regressions using the GENLINMIXED procedure in SPSS. For each regression, Focus Type (Object or PP focus) was the only fixed effect. To account for repeated measures, we included a random intercept by subject.¹⁰ We report the estimate (β), standard error (SE), t-scores, and p-values, taking a p-value of 0.05 as the threshold for statistical significance.

The first test compared the probability of VOPP word order across the two conditions (object/PP focus). The effect of Focus Type on the probability of VOPP was not significant (β = –0.34, SE = 0.20, t = –1.69, p = 0.09), so VOPP was preferred under both object and PP focus questions.

The second test compared the probability of VPPO word order across the two focus conditions. This test revealed a significant effect of Focus Type on the probability of VPPO (β = –1.02, SE = 0.29, t = –3.48, p = .001). Pairwise comparisons revealed that VPPO was substantially more probable under object focus than under PP focus.

Finally, the third test compared the probability of Fronting across the two conditions, revealing a significant effect of Focus Type on the probability of Fronting (β = 1.50, SE = 0.29, t = 5.13, p < .001). Pairwise comparisons revealed that Fronting, unlike VPPO, was substantially more probable under PP focus than under object focus.

6.1 Participants

The same participants who took part in the forced-choice task completed the self-paced reading task. See section 5.1.

6.2 Procedure and materials

The SPR task was completed in the program Linger (Rohde 2003). At the outset, participants read on-screen instructions and completed five practice items. The 96 SPR items were randomized by participant and presented in three blocks of 32 each. Participants could take breaks between blocks and between tasks. Overall, the task lasted approximately 45 minutes.

Each item was preceded by a non-moving context establishing given and new information and ending in a question targeting either the direct object or the PP. The stimulus then appeared on the screen, masked by dashes, and participants advanced through the sentence by pressing the spacebar. The task used a non-cumulative moving window paradigm, such that each press of the spacebar revealed one segment of the sentence, allowing participants to set their own pace for reading (Just, Carpenter & Woolley 1982). The program measured the time in milliseconds between button-presses. Participants could not go back and reread.

Each stimulus was followed by a true/false statement to check comprehension; participants responded by pressing J if false and F if true. Half the statements were true, half false. Half targeted the context and half targeted the experimental sentence, but none targeted the constituent in focus. Wrong answers were marked as incorrect so that readers would slow down. One participant whose answers were less than 80% accurate on the comprehension questions was excluded from the study.

The self-paced reading task had a 2 × 2 design manipulating two independent variables: Focus Type (object focus/PP focus) and Word Order (VOPP/VPPO). Table 3 illustrates sample items in the four conditions; critical regions are underlined.

To reduce variation due to external factors, we controlled for the following: All direct objects were indefinite, masculine, inanimate nouns (e.g., un mosaico ‘a mosaic’). PPs were either temporal adjuncts (por la noche ‘at night’) or locative adjuncts (en el bosque ‘in the forest’), including a preposition and its complement definite DP. In a given sentence, the direct object and the PP had the same number of syllables (either four or five), to control for phonological weight. Verbs, always three syllables long, were core transitive verbs appearing in the third-person singular of the preterit. The combined critical region (verb, object, and PP, underlined in Table 3) was controlled for overall length (between 26 and 32 characters).

We controlled for definiteness and phonological weight because they interact with information structure. We also controlled for plausibility, so we were not able to control for word frequency. To address potential problems with infrequent lexical items, however, we conducted a pilot study in which native Spanish speakers (n=9) were asked for explicit feedback on plausibility and comprehensibility. They were asked in particular for feedback on lexical choice. Their comments resulted in small changes to introductory contexts; critical items were all judged comprehensible and plausible.

We embedded the items in a carrier phrase so that the critical region would be neither the first nor the last part of the sentence.¹¹ The portion of the sentence after the critical region was always a new clause, so that a constituent appearing at the end of the critical region was also at the end of its clause, i.e., in the predicted focus-prominent position.

We created 32 lexicalizations and distributed them across eight lists, so that each participant saw eight items in each cell of the design, for a total of 32 experimental items. Additionally, each participant saw 64 filler items. The full instrument is available in Supplementary File 4.

6.3 Results

Before conducting the analysis, we log-transformed the reading times (RTs) and examined them to exclude outliers who had RTs more than 2 standard deviations from the overall mean, resulting in two exclusions. To analyze the data, we then trimmed the raw reading times conservatively at a minimum of 100 milliseconds and a maximum of 10,000 milliseconds. Finally, we created length-adjusted RTs following Fine et al.’s (2013) length-adjustment procedure and then combined the RTs of the three regions of interest (verb, direct object, and PP). The residual RTs from the adjustment procedure indicate how much faster or slower participants read the items compared to the expected time for a segment of that character length. The lower the RT, the faster those items were read. Figure 4 presents the length-adjusted RTs by condition and word order.

Figure 4

Length-adjusted reading times by condition.

We analyzed the length-adjusted RTs by fitting a linear mixed model following Eddington’s (2015) model-fitting procedure. The model included two fixed factors: Focus (object or PP) and Order (VOPP or VPPO), as well as their interaction. The model accounted for repeated measures by including a random intercept by subject; it also included a random intercept by item.

Type III tests of fixed effects revealed a main effect for Order (F(1,1237) = 4.69, p = .031), but no main effect for Focus (F(1,1237) = 0.02, p = .886) and no interaction (F(1,1237) = 0.86, p = .355). Pairwise comparisons confirmed that VOPP was read faster than VPPO overall (p = .031), irrespective of focus context.

7.1 Summary of results

Overall, the results of our three tasks converge. These results show that in-situ realizations (i.e., canonical orders) are the most commonly produced, most highly rated, and most rapidly processed for non-final object focus. Beyond this overwhelming preference, however, we find that speakers do produce and accept object-final word orders more often under object focus than otherwise, which fits in with the dominant understanding in the syntactic literature. The results from our three experiments are most compatible with a view of information structure in which canonical word orders can realize focus on any constituent, while non-canonical word orders are restricted to specific contexts (e.g., Höhle 1982; Hopp 2009; Weskott et al. 2011). Specifically, while canonical SVOPP word order is felicitous in any context, SVPPO is mostly found in contexts in which the constituent in final position is in focus. This contrasts with approaches that propose a strict one-to-one relationship between focus-marking and grammatical function, or those in which movement is triggered by pragmatic features, whether directly (e.g., Rizzi 1997) or indirectly, via an association with stress (e.g., Zubizarreta 1998; Büring & Gutiérrez-Bravo 2001; Samek-Lodovici 2001).

7.2 The role of task characteristics

Across the three tasks, we find limited evidence that task characteristics influenced the observed results; in fact, there is remarkable similarity across the findings. This is in spite of the fact that the tasks did not use the same items and that the production task was done with a different group of speakers. In what follows, we examine the results of each task one by one and compare them to previous findings in an effort to determine what effects could be attributed to the task design.

7.3 Asymmetry by argument type

We have noted that languages differ regarding whether they mark focus on subjects and non-subjects using the same strategies (Zimmermann 2016). The fact that we do not observe effects of the context on reading times for object focus in the SPR task, whereas our previous results indicate that context does matter for processing subject focus, leaves open the possibility of such an asymmetry in Spanish, a possibility endorsed by Skopeteas and Fanselow (2010). Recall from section 2.3 that the existence and nature of this asymmetry is not uncontroversial; some evidence shows that Spanish displays the typological asymmetry that Skopeteas and Fanselow discuss, while at the same time the mainstream syntactic accounts propose the same focus-marking mechanism across constituents. What do our results contribute to this discussion? We can compare our object focus results to previous results on subject focus, both our own and in the literature, in search of the relevant asymmetry. In Table 4, we summarize this comparison. Our production and judgment tasks both find that object focus is preferentially realized in situ, although we found some evidence for movement to final position as a focus-marking strategy. Directly comparing with previous production results on subject focus, we see that speakers are even more likely to produce focused subjects in situ, while producing comparatively more movement instances under object focus.

Comparing our present results to previous literature (Table 4) is complex because a multitude of design-related factors could play a significant role (e.g., whether or not the sentence includes a direct object, whether or not canonical orders are included, etc.). However, we do find evidence for similar strategies (movement and in-situ marking) for both subjects and objects. Overall, then, this comparison reveals no clear asymmetry by argument position. Interestingly, however, the lack of asymmetry is not the one predicted by the syntactic accounts (that any argument under focus must be equally marked with movement to final position). Instead, the Spanish speakers in our sample pattern more with what happens in English, where no special word order is required, regardless of constituent.¹²

At first blush, the results from the SPR seem to offer the clearest case of an asymmetry. Using the same methodology with the same population reported here, our previous research examining the processing of subject focus showed evidence that the proper context resulted in felicitous orders being processed faster than non-felicitous orders—that is, it showed contextual licensing—in alignment with previous processing research. In the present experiment, on the other hand, we see no such contextual advantage. This could be taken to indicate a subject/non-subject asymmetry in the spirit of Skopeteas and Fanselow, whereby subjects move to final position under focus, but objects do not. However, a more plausible explanation is that task characteristics are involved: our previous experiment on subject focus did not contrast canonical and non-canonical orders but rather compared the processing of two non-canonical orders (VOS vs. VSO). Thus, the comparison between tasks is not a minimal pair: the question of how canonical orders (SVO) are processed against the non-canonical option that is presented as felicitous in the syntactic literature (VOS) is an open question that future research should address.

Overall, as summarized in Table 4, we do not find strong evidence of an asymmetry of marking between subjects and objects. However, because the tasks in the previous literature differ in multiple dimensions (population, constructions elicited in the prompts, etc.), more research paying crucial attention to oft-overlooked methodological choices is required to determine whether this lack of asymmetry generalizes.

7.4 Limitations of the present work

Two limitations of our work involve prosody, which is an important part of focus marking. Although the specific prosodic realizations of focus in Spanish are under debate (Kim & Avelino 2003; Martín Butragueño 2005; Kim 2016), psycholinguistic investigations of focus have consistently shown that speakers not only encode information structure via prosody, but that they also utilize prosodic cues to determine the structure of utterances and to predict what lies ahead in terms of the structuring of information (see Kaiser 2016 for a review).

First, in all three tasks, the prompts were presented in writing, not aurally. For the judgment and SPR tasks, the focus was therefore not marked prosodically, and we could not control what prosody participants may have been mentally assigning to the sentences as they read them. Clearly this is a limitation which arose as a trade-off for making the tasks more feasible for participants to complete in a reasonable time. Such decisions are common in all research, but we must nonetheless acknowledge that the lack of prosodic information in the sentences being judged and processed is a limitation.

Second, although participants responded aloud in the production task, we did not design it to be suitable for prosodic analysis: while it is possible that speakers used intonation to mark focus, the task is not amenable for this type of analysis. For example, the descriptions of the characters in the video clips contained many voiceless consonants, making examining pitch tracks or using other usual measures of pitch accent difficult. Impressionistically, though, we did not find any of the typical patterns used to mark focus in the production task. That is, in our estimation, participants did not produce a clearly audible stress on the in-situ object when they produced SVOPP word order in response to object focus questions. Instead, most sentences carried a neutral nuclear accent at the end, as is the typical pattern for Spanish. Although this finding is merely descriptive, it should be noted that previous studies do not always find that participants stress in-situ focus (e.g., Kim & Avelino 2003; Ocampo 2003; Labastía 2006; Kim 2016; Vanrell & Fernández-Soriano 2018). This lack of correspondence between focus and stress has not, to our knowledge, been a central consideration in the theoretical literature on focus. Such a finding can potentially be of importance for the theoretical treatment of information focus: If the lack of correspondence between focus and nuclear stress holds for other investigations, this would constitute a violation of the focus prominence rule—one of the most basic assumptions on the focus-prosody interface.¹³

This leads to obvious questions regarding the viability of syntactic models such as Zubizarreta’s (1998, 2016), which depart from the assumption that “[t]he focused constituent must contain the rhythmically most prominent word” (Zubizarreta 2016: 166). This is the case because data by Gabriel (2010) and many others, including our present and past findings, show that Spanish does not appear to be restricted to this particular realization of nuclear stress. Zubizarreta (2016) has noted that, for dialects such as Argentinean Spanish, A-deaccenting and NS-shift, although “a marked option in Spanish” (Zubizarreta 2016: 183), might be preferable to changes in the word order or to V-initial orders.

The problem is that this preference does not seem to be limited to Argentinean Spanish and, moreover, it also appears to be a poor candidate to be explained via dialectal variation. At the moment, we have evidence that Catalonian Spanish, Mexican Spanish, and Chilean Spanish (as measured by different tasks) behave in very similar ways to Argentinean Spanish—speakers of these varieties prefer what Zubizarreta calls the “marked” option in Spanish and avoid what she considers to be the most felicitous one (p-movement). Where does that leave us then? In our view, it is probably time to question whether A-deaccenting and NS-shift are really marked options in Spanish.