The boundary-crossing constraint revisited: movement verbs across varieties of Spanish

2.1 (Un)learning the boundary-crossing constraint

In this section previous studies about the acquisition and translation of boundary-crossing events are reviewed, since a large body of literature available on the BCC took a cross-linguistic perspective. According to the Thinking for Speaking (TFS) hypothesis proposed by Slobin (1996a), the Talmian language type will influence what speakers pay attention to while speaking, since they are limited by the available linguistic forms of their language. Slobin proposed that the TFS patterns will be transferred from the first language (L1) when a second language (L2) is being learned. A large amount of research has examined whether, or to what extent, learners will be able to acquire the L2 TFS patterns (e.g., Cadierno 2010; Cadierno and Ruiz 2006; Hasko 2010; Hijazo-Gascón 2018; Lewandowski and Özçalışkan 2021; Stam 2010). When one zooms into the literature addressing boundary-crossing events, the two main areas on which previous studies have focused are: (i) finding out whether L2 learners are influenced by their L1, that is, whether they transfer patterns from their L1 (Alonso 2013; Muñoz and Cadierno 2019) and, in some cases, testing whether there is an advantage for learners whose L1 is typologically similar to the L2 (Cadierno 2010; Cifuentes-Férez and Molés-Casés 2020); (ii) testing if incidental learning and unlearning of the boundary-crossing constraint from the input occurs (Alghamdi et al. 2019; Treffers-Daller and Calude 2015). Some of these studies also test if learners are closer to native norms when their proficiency level is higher.

Cadierno (2010) analysed how Spanish, Russian, and German learners of Danish as an L2 with an intermediate level described Motion Events with boundary crossing in their L2. Cadierno asked whether the L1 would influence the patterns of the participants’ L2, since Russian, German, and Danish are part of the same group (satellite-framed languages), whereas Spanish is a verb-framed language. The results showed that those learners whose L1 matched the L2 in being satellite-framed (i.e., Russian and German) did better than those whose L1 (Spanish) is typologically different from the L2, Danish.

Muñoz and Cadierno (2019) examined bidirectional transfer between Spanish and English in both boundary-crossing and ‘non-crossing’ situations. An innovation in comparison with previous studies was that the stimuli they used for eliciting motion-event descriptions were realistic and dynamic rather than static. They found that English learners of L2 Spanish were influenced by the L1 patterns in their L2 production of Motion Events both in boundary-crossing and non-crossing situations, regardless of their proficiency level. They found influences of L2 in the learner’s L1, at least partially (Muñoz and Cadierno 2019: 61).

Regarding the translation of descriptions of boundary-crossing events, both Alonso (2013) and Cifuentes-Férez and Molés-Casés (2020) found similar results. On the one hand, Alonso (2013) examined translation from Spanish to English by Spanish students of translation with a C1 level in their L2 English and found that Manner tended to be mapped onto a separate constituent and not the main verb, indicating transfer from the L1 to the L2. On the other hand, Cifuentes-Férez and Molés-Casés (2020) examined the translation from English to their L1 by Spanish or German translation students and found that the German translators did better than the Spanish translators since the loss of Manner information was less frequent and the crossings were expressed more often. Thus, even advanced students were influenced by their motion-event L1 patterns in the L2. Regarding the Spanish translations in the latter study, the Manner was more often lost in vertical and slow boundary crossings, while in quick crossings the Manner information was more often preserved.

Treffers-Daller and Calude (2015) tested whether the frequency of different movement verbs in (L1 English) learners’ (L2 French) input determines the frequency of the movement verbs in the learners’ own L2 output. Their results indicate that, in general, the input frequency helps because the most frequent movement verbs in the input are also the most frequent in the learners’ output, and the more advanced the learners were, the closer they got to L1 speakers’ use of specific movement verbs. However, one exception to the frequency benefit found in this study was the boundary-crossing constraint, since most of the English learners of L2 French did not respect the restriction in the oral elicitation task, and even the more advanced learners were not found to be better than the less advanced ones. Thus, the results suggested that the lexical frequency of motion verbs in the input is not enough for acquiring the BCC.

Alghamdi et al. (2019) focused their study exclusively on boundary-crossing situations comparing English speakers, who preferred a manner verb in describing boundary crossings in their L1, with Arabic speakers, who preferred other constructions in their L1, such as path verbs or not mentioning the crossing. The L1 Arabic learners of L2 English, whose L1 is a verb-framed language and whose L2 is a satellite-framed language (i.e., they had to unlearn the constraint), were somehow in between, being more similar to native English speakers as their level of English increased. When measuring incidental learning, the researchers discovered that L2 motion verb frequency, as measured using the British National Corpus, did not predict the use of manner verbs in boundary-crossing situations by these learners. Even the most advanced learners acquiring English in immersion settings could not completely learn the L2 patterns from the input, and this, together with the probable lack of negative feedback from teachers which is “normally not given as the learners produce grammatically correct structures albeit different from the preferences of the target language” (Alghamdi et al. 2019: 99), highlights the need to teach the target forms explicitly. Laws et al. (2021) showed that a treatment based on the principles of the Input Processing methodology (Lee and VanPatten 2003; VanPatten and Cadierno 1993) helped learners of L2 English with different verb-framed L1s to unlearn the boundary-crossing constraint in English.

From the studies reviewed so far, it can be concluded that learners are clearly influenced by their L1 when expressing Motion Events with boundary crossing in their L2, especially if the L1 and L2 are from typologically distinct groups, and that the role of both the language proficiency and the lexical frequency of motion verbs is limited, because even for L2 users with a high proficiency level, and even for high-frequency target forms, the BCC still remains a challenge.

2.2 The boundary-crossing constraint revisited

In the previous section we reviewed research showing that the native language patterns of encoding boundary-crossing events exert a strong influence on production in a second language. Despite the clear role of the BCC cross-linguistically, a number of studies stand in opposition to the full validity of the BCC in verb-framed languages. The first researchers to notice the violation of this restriction were Naigles et al. (1998), when comparing how Spanish and English speakers described scenes with boundary crossings: “For the horizontal boundary events, this was the case: 100 % of the responses with a, de or para (n = 40) occurred with path verbs […]. However, for the vertical boundary events, 97 % of the responses with a, de or para occurred with manner verbs” (Naigles et al. 1998: 542). They concluded that the reason might be that in verb-framed languages, manner verbs are allowed in vertical boundaries and/or in involuntary motion, such as in scenes where a Figure jumps into a pool, in which the motion event is initiated intentionally but the boundary crossing itself is not intentional any more. Martínez Vázquez (2001: 43) documents other examples in Spanish where a manner verb is used in boundary crossings in the CREA corpus; see example (4):

Özçalışkan (2015) compared how native speakers of English (a satellite-framed language) and of Turkish (a verb-framed language) described scenes with boundary crossings in their L1. The author concludes that the constraint is accurate as a test of whether a language belongs to one type or the other, since the English speakers produced more manner verbs (except in a scene where the Figure unintentionally falls into a net) and fewer path verbs in boundary crossings than the Turkish speakers did. However, in some contexts the Turkish speakers did use manner verbs in boundary-crossing contexts, namely quick or instantaneous crossings, such as dive into a lake or leap over a hurdle, though the Turks never used manner verbs “in the description of activity-type events that are temporally extended (e.g., crawling, walking, running)” (Özçalışkan 2015: 496). Another factor that might play a role, according to this author, is the character of the boundary, since the crossings of fluid boundaries (e.g., between air and liquid, such as jump into a pool) were more frequently described with a manner verb and more manner verbs were used to express crossing over an unenclosed bidimensional boundary (such as leap over a hurdle or flip over a beam) in comparison with crossing out of or into a space.

The categorisation of verbs as either manner or path verbs remains a challenge. Zlatev and Yangklang (2004: 167) wrote about the verb fall, which was considered a path verb by Slobin (1996b) since it mainly indicates movement downwards, but it can be argued that it also indicates manner since the movement is through the air (or another medium) and it indicates the Figure’s loss of control. That is the reason why Zlatev and Yangklang proposed an intermediate category called “Manner + Path verbs” (2004: 167). Lewandowski and Mateu (2020) proposed a semantic continuum of verbs: path verbs (e.g., exit) > directional manner verbs (e.g., walk) > pure manner verbs (e.g., float) > non-motional manner verbs (e.g., whistle). The difference between directional and pure manner verbs is that directional manner verbs imply a change of location, while pure manner verbs do not necessarily do so. For testing the acceptability of the use of either directional or pure manner verbs followed by the preposition a ‘to’ by Spanish native speakers, Lewandowski and Mateu (2020) conducted an acceptability judgement task with Spanish native speakers (n = 20) from different countries. These Spanish speakers rated as acceptable sentences with a directional manner verb followed by the preposition a, such as Juan corrió al baño ‘John ran [in]to the bathroom’, while they rated as unacceptable sentences with a pure manner followed by the preposition a, such as Juan bailó a la habitación ‘John danced [in]to the room’. According to these authors, while the preposition a indicates that the Figure is “in contact with the exterior boundary of the [G]round” (Lewandowski and Mateu 2020: 12), the preposition en ‘in’ indicates that the Figure is inside the Ground or on its surface (en el cajón ‘in the drawer’ or en la mesa ‘on the table’). This study did not have enough participants for it to be possible to check whether there were any differences between varieties of Spanish in the use of the preposition a with directional or pure manner verbs. There are studies on the alternation of Spanish prepositions with path verbs, such as Ibarretxe-Antuñano’s study (2003) on entrar en/a ‘enter in/to’ which proposed semantic differences based on Talmy’s force dynamics (Talmy 1988) and on Langacker’s profiling (1987, 1991, 2000 cited in Ibarretxe-Antuñano 2003), such as deictic differences (Ibarretxe-Antuñano 2003: 48). Other authors argued that the choice between en/a (‘in/to’) with verbs that denote movement into a place (e.g., entrar ‘enter’) is a matter of diatopic variation, the preposition a being preferred in Latin American varieties (for a review, see Eslava Heredia 2012).

Beavers et al. (2010: 362–365) emphasise the role of context and, by means of a review of several studies, they enumerate factors that might influence the interpretation (either locative or directional) of the prepositions in different languages: the proximity of the Figure to the Ground, the type of Ground (an enclosed space with clear boundaries or open space), punctual actions as opposed to processes, different types of manner verbs (pure vs. directional), Manner prominence, etc.

The goal of the present study is to test whether speakers of Spanish (both Peninsular and Latin American) use manner verbs to describe boundary-crossing motion events (production) and whether they interpret manner verbs as descriptions of boundary crossings (comprehension). In addition, on the basis of the previous studies reviewed above, we test which specific factors may modulate the acceptability of manner verbs for boundary-crossing meanings. Specifically, this study addresses the following research questions and subquestions:

1. Do speakers of Spanish use manner verbs as main verbs when describing dynamic scenes with boundary-crossing situations? Is the probability of using a manner verb to describe boundary-crossing events higher for movement towards the inside of an enclosed space as opposed to exiting a space?

2. Do speakers of Spanish rate (on a five-point Likert scale) as suitable sentences with main manner verbs describing the same dynamic scenes with boundary-crossing situations? Are there differences between the ratings of different verbal constructions (see Section 3.3) or the usage of the prepositions en ‘in’, a ‘to’ and de ‘out of’?

3. Do speakers of Spanish interpret as instances of boundary-crossing sentences with main manner verbs? Is the boundary-crossing interpretation with directional manner verbs (run, walk) more likely than with pure manner verbs (float, roll)? Is the boundary-crossing interpretation with small Grounds (la ardilla trepó en el pequeño hueco ‘the squirrel climbed upwards in(to) the small hole’) more likely than with larger Grounds (… en la casa del árbol ‘… in(to) the treehouse’)? Is the boundary-crossing interpretation more likely with vertical than with horizontal movements?

For each of these three questions and their respective subquestions, we also asked whether there are differences between different varieties of Spanish.

3.1 Participants

A total of 180 native Spanish speakers took part in the study. An additional 15 participants were recruited and completed the experiment, but they were excluded because either they failed the attention checks (n = 11) described in the next section, 3.2. Procedure, or Prolific categorised their performance as ‘timed-out’, i.e., they took longer than 67 min (n = 4). Eighty-eight participants from Spain, 66 from Mexico, 20 from Chile, and six from other Latin American countries completed an anonymous online survey made with Google Forms. The participants were recruited using the online research platform Prolific (https://www.prolific.co/) during the month of September 2022. They were rewarded financially for completing the survey. The study was distributed among the pool of registered participants on Prolific (the Standard sample option) and the inclusion criteria were as follows: nationality (all Spanish-speaking countries), first language (Spanish), fluent languages (Spanish), raised as monolingual, language-related disorders (none), earliest language in life (Spanish), and approval rate on previous studies on Prolific (minimum approval rate 70 out of 100). The study was run in two rounds to make sure that the number of participants was balanced between speakers from Spain and from Latin America. The participants could use a mobile, tablet, or desktop for completing the survey, and they took 19.69 min on average to complete it. Out of the total 180 participants, 62 identified as female and 118 as male. The participants’ mean age was 29.6 years, SD 9.27 (for the participants from Spain it was 31.3, SD 10.3, range 19–66, and for those from Latin America it was 28.0, SD 7.88, range 19–68).

3.2 Procedure

After consenting to participate in the study, recording their Prolific ID, and responding to a question checking the fulfilment of the main requirement for participation (namely Is Spanish your native language?), the participants completed a survey in Google Forms comprising three tasks in a fixed order: Task 1 was an open-ended description of videos depicting boundary-crossing scenes, Task 2 was a multiple-choice task with sentences containing main manner verbs in which the participants were asked to choose between interpretations of these sentences either involving a boundary crossing or not, and in Task 3 the participants rated (on a five-point Likert scale) the suitability of different sentences for describing the same videos that had been presented in Task 1. For the data collection of Task 1 and Task 3 a set of ten animated videos was made. A printable version of the survey with the three tasks, as well as the set of animated videos, is available at https://osf.io/9tquh/.

In Task 1, the participants were instructed to watch a video and describe what had happened in a short sentence. An example was provided with a similar video not depicting a boundary-crossing event. Figure 1 shows screenshots of the ten target animated videos, lasting between 4 and 10 s, that were made (using the Procreate software) specifically for the present study to investigate Motion Events with boundary crossings. Five videos depicted motion into a place and five out of a place for each of the following actions: running, crawling, flying, jumping, and falling. The first four scenes, as ordered in Figure 1, show horizontal boundary crossings, scenes numbers 1 and 2 capturing quick crossings, 3 and 4 slow crossings. The rest of the scenes show vertical boundary crossings: 5 and 6 depict intentional slow crossings, 7 and 8 intentional (at least at the beginning of the movement in 7) quick crossings, and 9 and 10 depict unintentional and quick or instantaneous crossings. The order of the videos presented in Task 1 was random, but fixed for all participants, and it was as follows: scenes 1, 7, 10, 4, 5, 2, 9, 3, 6 and 8.

Figure 1:

Screenshots of the short videos in the critical moment of the boundary crossings.

In Task 2, 26 short sentences containing main manner verbs, plus four fillers used as attention checks (Curran and Hauser 2019), were presented (after brief instructions for completing the task) to the participants, who were asked to choose one of the four following options according to their own interpretation of each sentence: a) a paraphrase of the boundary-crossing interpretation of the sentence, b) a paraphrase of its non-boundary-crossing interpretation, c) both meanings are possible, and d) I don’t understand the meaning of the sentence. The attention checks were for screening out careless participants: those who failed two or more attention checks through the experiment (there was an additional one in Task 3) lost their entitlement to the monetary compensation in Prolific and all of their responses were excluded from the study. Example (5) gives four of the sentences used, variants of each other, with potential vertical boundary crossings into a clearly defined space and with directional manner verbs (Lewandowski and Mateu 2020). Out of them (5a) and (5b) present a small Ground, whereas (5c) and (5d) present a large Ground; (5a) and (5c) include the preposition a ‘to,’ whereas (5b) and (5d) include the preposition en ‘in.’ Appendix 1 lists all of the 26 target sentences used in Task 2. It can be seen that the other sentences included, as main verbs, pure manner verbs (Lewandowski and Mateu 2020), with a potential horizontal crossing either into or out of a space, the latter with the preposition de ‘out of’.

The response options in Task 2 varied according to the sentence to be interpreted; e.g., for the sentence (5a) the options were: la ardilla entró al hueco desde fuera ‘the squirrel entered into the hole from the outside’, la ardilla se acercó al hueco pero no entró ‘the squirrel approached the hole but did not enter it’, ambos significados son posibles ‘both meanings are possible’, and no entiendo el significado de la frase ‘I do not understand the meaning of the sentence’.

In Task 3, after brief initial instructions, the participants first had to watch the same videos as those presented in Task 1, but in a different order, which was again fixed for all participants, namely scenes 6, 9, 1, 8, 3, 5, 2, 7, 4 and 10. After rewatching each video, they had to read and rate for suitability three or four short sentences (three sentences for videos depicting movement out of a place and four for movement into a place), each offered as a potential description of what had happened in the video. A five-point Likert scale was used, with 1 corresponding to very unsuitable description and 5 to very suitable description. When the Figure crossed a boundary and entered a place, two of the sentences offered used the typical pattern of a verb-framed language (Talmy 1985, 1991, 2000), i.e., a main Path verb and a separate element, out of which one used the preposition a ‘to’ and the other one the preposition en ‘in’. The other two sentences used the typical pattern of a satellite-framed language, i.e., a manner verb followed either by the preposition a ‘to’ or en ‘in’. See example (6) for the proposed descriptions of the video of a Figure running into a house:

For videos of a Figure crossing a boundary and leaving a space, sentences following the prototypical pattern of a verb-framed language (salió gateando de la casa ‘she/he exited the house crawling’) and sentences following the prototypical satellite-framed pattern (gateó de la casa, ‘she/he crawled out of/away from the house’) were offered for rating. There was also one attention-check item in Task 3, unrelated to the video with which it was presented.

The study was first piloted with Spanish-speaking friends via privately sharing a link to the online survey. After the study had been piloted with 27 participants, each of whom was debriefed, some attention checks were changed, and the option of I do not understand the meaning of the sentence was added to Task 2, along with other minor edits. The data from the pilot was entirely excluded, since some of the participants might have known the purpose of the study right from the beginning.

3.3 Coding

This section explains the labels used for coding the verbal constructions used by the participants in the open-ended description task (Task 1). The labels can also be seen in Table 1 of the results section below, where there are additional examples taken from the data.

1. Manner verb: in this category sentences describing the boundary crossing with main manner verbs, such as arrastrarse ‘to crawl’ or saltar ‘to jump’, were included.

2. Manner + Path verb: sentences with main Manner + Path verbs as defined by Zlatev and Yangklang (2004: 167), such as caer(se) ‘to fall’ or huir ‘to flee/run away’, are included in this category.

3. Path verb: sentences where a path main verb was used, such as David entra en casa ‘David enters the house,’ including sentences with gerunds, such as Persona entrando en casa ‘Person entering the house’, because the gerund clearly indicates the crossing.

4. Path verb… Manner element: here sentences with the prototypical verb-framed construction pattern were included, i.e., a path main verb and a separate manner element, such as Juan entró corriendo a la casa ‘Juan entered the house running’.

5. Other: participants’ sentences that did not clearly express the boundary crossing were included in this category, i.e., sentences with until-markers (Beavers et al. 2010) such as hasta ‘until’ or sentences with prepositions that do not entail a boundary crossing, such as hacia ‘towards’; see Table 1 for an example of this; sentences with gerunds as main manner verbs where the participants are clearly focussing on the process and not the crossing itself, such as corriendo a casa ‘running to the house’; sentences in the present tense that do not mention the Ground, such as David gatea ‘David is crawling’ or in the past but without mentioning a clearly-defined enclosed Ground, such as El ángel voló al cielo ‘the angel flew to the sky’; sentences where both the Path and the Manner are expressed in two main verbs that are not part of the same sentence, i.e., coordinated or subordinated sentences, such as David corrió y entró a la casa ‘David ran and entered the house’ or Adrián cae por el precipicio al salir de su casa ‘Adrián fell from the cliff when exiting his house’; sentences without any verb, such as dramática caída ‘dramatic fall’; and sentences that did not describe what happened in the video, such as David es Jesucristo ‘David is Jesus’ or did not describe any movement, such as día de verano ‘summer day’. In general, constructions that did not clearly belong to any of the other categories used were included in this category.

For Task 2, in which participants interpreted the meanings of sentences, only sentences with the construction from the category Manner verb were presented. Some of them incorporated pure manner verbs and others directional manner verbs, as defined by Lewandowski and Mateu (2020).

4.1 Description production task

First, the open-ended descriptions of the videos depicting the boundary-crossing scenes produced by the participants in Task 1 were analysed. The responses that were elicited were coded for the verbal construction used (see Section 3.3), choosing between the options listed in Table 1.

Figure 2 shows the percentage of each construction split by the participant’s variety of Spanish. It can be seen that, irrespective of variety, approximately a quarter of the descriptions used either a Manner verb (e.g., correr or gatear) or a Manner + Path verb (as defined by Zlatev and Yangklang 2004; e.g., caer or huir) and less than a half of them used a Path verb with or without a separate Manner element to describe the boundary-crossing events shown in the videos.

Figure 2:

Percentages of verbal constructions of the different types elicited in Task 1 split by the participants’ variety of Spanish.

The probability of the participants using a manner verb (i.e., the first two constructions in Table 1) as opposed to them not using one (i.e., the remaining three types of constructions in Table 1) in their descriptions was modelled by a mixed-effects logistic regression model with Variety of Spanish (Spain, Mexico, Chile, Other, sum coded for four levels, Other coded as −1, −1, −1), Crossing direction (sum-coded: entering into 1, exiting out of −1) and their interaction as the fixed effects, and Participant and Item as the random effects with varying intercepts. The model R formula was as follows:

Table 2 gives the model’s coefficient estimates for the fixed effects and Figure 3 plots the values fitted by the model to this data. Regarding the random effects, the standard deviation estimated for by-participant varying intercepts was 1.090 logits and for by-item varying intercepts (n = 10) it was 1.617 logits. Since three statistical models in total (see below) were fitted for data elicited from the same participants, the alpha level was Bonferroni-corrected to 0.05/3 ≈ 0.0166. Crossing direction alone was not found to have a reliable overall effect on the probability of manner verb usage. However, a reliable effect was found for the participants from Chile (interaction term logit estimate: 0.46, SE = 0.18, z = 2.61, p = 0.009), and for the remaining varieties the raw counts of manner verbs used to describe entering into events were higher than those used to describe exiting out of events (cf. Figure 3).

Figure 3:

Values fitted by the logistic regression model of the probability of manner verbs in the participants’ video descriptions split by Variety and Crossing direction. Error bars indicate 95 % confidence intervals.

The descriptions data was inspected for which specific manner verbs were used by the participants to describe the videos with boundary-crossing events. Figure 4 lists them all and it plots the raw counts of their occurrence, marking the construction type used. Table 3 then zooms into the six most frequent verbs (two of which were Manner + Path verbs), giving an example video description for each.

Figure 4:

The count of occurrence of each manner verb used in the participants’ video descriptions and the construction type used.

4.2 Description rating task

This section reports the participants’ ratings of the suitability of the video descriptions presented in Task 3. These results are presented before those of Task 2 because the same videos were used as in Task 1. A mixed-effects ordinal logistic regression model was fitted to this data with the ratings (on a five-point Likert scale) as the response variable and verbal Construction type (Manner verb, Manner + Path verb, or Path verb … Manner element), Preposition (a, de, or en), and the participant’s continent of origin (henceforth Origin; levels: Spain vs. Latin America), as well as their interactions as the fixed effects. Participant and Item were included as the random effects with varying intercepts. The two-level predictor Origin was used, rather than the four-level one Variety, in order to reduce the number of parameters estimated by the model. The model R formula was as follows:

Table 4 gives the model’s coefficient estimates for the fixed effects and Figure 5 plots the fitted values. As regards the random effects, the standard deviation estimated for by-participant varying intercepts was 0.8327 logits and for by-item varying intercepts (n = 36) it was 1.0364 logits. The analyses revealed a high probability of the best ratings of the suitability of a Manner verb in combination with the preposition a for describing the boundary-crossing scenes shown in the videos (specifically, the probability of response 4 or 5 was 0.90, computed as 1 – plogis (−2.1929), i.e., by applying the logistic function to the third intercept in Table 4 and subtracting from 1), about as high as it was for the Path verb … Manner element construction type (see Figure 5, upper leftmost and rightmost facets, respectively). Using a Manner verb in combination with the preposition de resulted in lower suitability ratings (logit difference: −2.50, SE = 0.75, z = −3.35, p = 0.0008) with the response probability showing a mild peak at response 4 (Figure 5, middle left), whereas using a Manner verb in combination with the preposition en resulted in clearly lower suitability ratings (logit difference: −5.01, SE = 0.75, z = −6.68, p < 0.0001) with the response probability culminating at the worst grade (Figure 5, lower left). The ratings of the suitability of the Manner + Path verb constructions to describe boundary-crossing events generally rose moderately towards the suitable end of the Likert scale, irrespective of the preposition used (Figure 5, central facets). Only two terms in the model estimate output out of those that included Origin were significant. Both involved the triple interaction between Construction, Preposition, and Origin, showing that the Latin American participants gave reliably lower ratings to the suitability of a Path verb … Manner element construction in combination with the prepositions de (logit difference: −1.05, SE = 0.30, z = −3.50, p = 0.0005) and en (logit difference: −1.65, SE = 0.31, z = −5.34, p < 0.0001) to describe boundary-crossing events than did the peninsular Spanish participants (see Figure 5, middle and lower right facets).

Figure 5:

Values fitted by the mixed-effects ordinal logistic regression model for the participants’ video description suitability ratings (1 = very unsuitable, 5 = very suitable) split by verbal Construction, Preposition, and continent of Origin. Error bars indicate 95 % confidence intervals.

4.3 Sentence comprehension task

Finally, the participants’ choices of the meanings of the sentences presented in Task 2 were analysed. Figure 6 shows the percentages of the interpretations chosen by participants with different varieties of Spanish. Recall that each sentence used a manner verb and offered three interpretations: one involving a boundary crossing, another one not, and one stating that both meanings were possible. The fourth option provided was always that none of the meanings matched the sentence, i.e., I do not understand the meaning of the sentence. Figure 6 shows that for each variety more than half of the responses chose interpretations with a boundary-crossing meaning (i.e., either the boundary-crossing interpretations offered or the both meanings possible option) and that the distribution of the interpretation options was similar across varieties (except for the minority of six participants in the Other category who had a somewhat higher percentage of the both meanings selections and a lower percentage of the no crossing selections).

Figure 6:

Percentages of the interpretations chosen by the participants for the sentences presented in Task 2 split by their variety of Spanish.

The probability of the participants choosing a boundary-crossing interpretation (i.e., either the boundary-crossing option or the both meanings possible option) of the sentences in Task 2 (always with a manner verb) was modelled by a mixed-effects logistic regression model with Ground size (large vs. small), Verb type (directional vs. pure), movement Direction (horizontal, vertical upward, or vertical downward), Preposition (a, de, or en), and the participant’s continent of Origin (Spain vs. Latin America) as the fixed effects, as well as Participant and Item as the random effects with varying intercepts. The model R formula was as follows:

Table 5 gives the model’s coefficient estimates for the fixed effects. As for the random effects, the standard deviation estimated for by-participant varying intercepts was 0.8558 logits and for by-item varying intercepts (n = 26) it was 1.0432 logits. None of the predictors was found to have a reliable effect on the probability of a boundary-crossing interpretation, with one exception, namely the preposition used. Figure 7 plots the predicted values for the different prepositions used. The probabilities for the preposition a (0.89, the Intercept in Table 5) and the preposition de (0.91) did not differ reliably, but the probability of a boundary-crossing interpretation for the preposition en (0.35) was much lower than that of a (logit difference: −2.74, SE = 0.47, z = −5.86, p = <0.0001).

Figure 7:

Model-predicted values of the probability of boundary-crossing interpretations chosen by participants in Task 2 for sentences containing different prepositions. The error bars indicate 95 % confidence intervals.