Prominence scales and variation in differential object marking: experimental evidence from Ibero-Romance

2.2.1 Portuguese: deities and animate nouns

As already briefly mentioned in the introduction, a-marking of DOs is a marginal phenomenon in Portuguese. Stressed pronouns are obligatorily marked,^[5] some quantifiers like todos “all” also receive the marker (Dubert and Galves 2016), but nominal DOs are rarely marked. Furthermore, Portuguese is one of the primary languages discussed by Caro Reina (2020) in order to distinguish different types of names on the EAH. The literature on Portuguese reports that a-marking of DOs is basically found when reference to God is made, whereas other names or nouns denoting human or animate reference do not receive it. Therefore, Caro Reina argues on the basis of this language that deity names are a separate category on the EAH. According to him, deity names are the highest category of names in terms of prominence since referents of other names are context dependent. Deity names, in turn, refer unequivocally and irrespective of context, since they are treated as unique entities. While Caro Reina (2020) also includes names of saints and devils in this category, his only example for this language is the deity name Deus “God”. This is quite typical in the literature – only a very limited number of such names receiving a-marking have been explicitly reported, such as Deus and Cristo (Pires 2017), usually appearing as DOs of a particular set of verbs such as adorar “worship” and amar “love.” To the best of our knowledge, it has never been verified empirically whether the generalization to a whole category of deity names as proposed by Caro Reina (2020) actually can be confirmed for Portuguese. In principle, the reported data would also be compatible with a construction-based analysis where more or less fixed verb-noun combinations might historically have been related to some prominence scale, but which in modern Portuguese simply are fossilized uses in highly specific contexts. If the explanation in terms of the EAH holds true for Portuguese, not only should Deus and Cristo be a-marked but also other deities, including those of the ancient world.

Beyond deity names, Cyrino (2017) claims that DOM is in fact optional with certain animate nouns, such as homem. Her example is (6), an intuitional report.

Beyond the example in (6), she provides another one from a journal where homem is used as a generic term. Cyrino also discusses cases where more complex syntactic configurations can lead to a-marking, such as dislocation or coordination. As far as canonical SVO-sentences are concerned, she only differs from Caro Reina (2020) in claiming that at least in BP, a-marking is not restricted to deity names on the EAH but may also occur with animate nouns. This assertion stands in sharp contrast to Döhla (2014: 279), who, based on extensive but unsuccessful attempts of eliciting a-marking on human and animate nouns in Brazil as well as Portugal, concludes that “DOM does not exist anymore in modern spoken Portuguese.” Thus, at a closer look, we have empirically unvalidated claims as well as contradictory statements about DOM in this language. The literature agrees that a-marking occurs with Deus and Cristo, but it is unclear whether this generalizes to the whole category of deity names. Furthermore, while some authors claim that at least in certain varieties a-marking of human nouns is optional, others deny that it even exists. This calls for further empirical studies. Given that, overall, in Portuguese neither a-marking nor deity names other than Deus and Cristo are frequent in everyday language use, it would be difficult to gather enough data even from large-scale corpora. In contrast, we can test these very specific claims in a straightforward manner in an AJT. If the claims in Caro Reina (2020) are correct, they predict that a-marking will be more acceptable on deity names than on proper names in a balanced sample. Conversely, the absence of a-marking would lead to decreased acceptability for deity names in contrast to proper names. If human nouns were optionally a-marked in BP, they would receive higher acceptability ratings than inanimate nouns when marked, whereas no contrast is expected without marking. Furthermore, there could be a contrast between BP and EP, since Cyrino (2017) explicitly limits her observations to the former, however, without making clear claims regarding the latter. In Section 4.4, we explain in further detail how we tested these predictions.

2.2.2 Catalan: animacy, optionality, and nonhuman animates

Detailed studies of DOM in Catalan have up until very recently been sparse. Historically, DOM has often been considered a mere contact phenomenon caused by the proximity and influence of Spanish and many grammarians urged writers to avoid the structure whenever possible, since it was not considered proper Catalan (Badia Margarit 1962; Fabra 1933, among others). Nonetheless, there were a few early defenders of the autonomous origins of Catalan DOM (cf. Müller 1971; Meier 1945), and further confirmation can be found in more recent studies such as Irimia and Pineda (2019) and Pineda (2021). In accordance with this traditionally restrictive view of DOM, Catalan normative grammars only accepted a-marking in very few contexts (cf. Fabra 1933), such as strong personal pronouns, certain quantifiers (e.g., tots “all,” tothom “everybody”), and contexts of reciprocity (l’un a l’altre “each other”) as well as to avoid ambiguity. In the newest revision of the Catalan normative grammar (IEC 2016), we still find quite a narrow field of permitted DOM uses. The permitted contexts mentioned in Fabra (1933) are supplemented with contexts where a change in word order (i.e., topicalizations, dislocation to the right) favors DOM. The authors also mention regional variation, pointing out that in Majorcan even some inanimates can receive a-marking in a dislocated position. Nevertheless, they insist that in Catalan, as a general rule, DOs do not appear with a preposition, supplying example (7) (IEC 2016: 730).

Regional grammars and studies on DOM show a slightly different picture, highlighting the still considerable gap between standard Catalan and spoken Catalan throughout the territory (cf. also Sancho Cremades 2008). For instance, Escandell-Vidal (2009) finds that even Central Catalan, the dialect closest to standard Catalan, shows a certain divergence regarding the contexts where DOM appears. She cites examples of a-marking with proper names, human definite noun phrases, and occasionally even with nonhuman animate DOs. Majorcan and Valencian show a considerably larger extension of DOM along both animacy and definiteness scales. For Valencian, examples can be found that show strong parallels to Spanish DOM, e.g., a-marking of indefinite DOs (8) and marking of collectives of humans as in (9) (AVL 2006: 303).

For animals, it seems that both presence and absence of a-marking are possible in Valencian, cf. (10a) versus (10b) (RACV 2016). Escandell-Vidal (2009) also mentions a certain degree of optionality for DOM with animals in Catalan in general, without going into further details. Overall, animal DOs have not received a lot of attention either from a theoretical or from an empirical perspective.

Based on the cited literature, the AJT in Zeugin (2021) tested regional differences between Catalan varieties in the a-marking of human animates, animals, and inanimate DOs. Since no robust differences were found for regional contrasts, we do not further discuss regional variation here and rather focus on the robust gradual difference found between the three nominal categories across all varieties.

Catalan is interesting in this regard because a-marking is not obligatory for any of these noun classes. Hence, in contrast to what is expected for Spanish, all nonmarked DOs should be perfectly acceptable. Since there is no detailed discussion of a-marking of animal-denoting nouns for Catalan in the literature, we have to refer to the literature on Spanish. Weissenrieder (1991) relates the possibility of a-marking to factors which, according to Egetenmeyer (2019), ultimately can be subsumed under the notion of saliency, such as degree of domestication, size, etc. All of these, however, do not reach the saliency of human nouns. From this, two important observations can be derived. First, not all animal nouns are expected to have the same acceptability for a-marking, and hence in an experiment, the types of animal nouns used should be controlled for. Second, for marked DOs, the prediction for acceptability ratings is a decrease from human to animals to inanimate nouns, following the animacy scale. Section 4.2.2 explains in more detail how these predictions were tested in the experiment.

2.2.3 Spanish: optionality, regional variation, and different types of inanimates

DOM in Spanish has been discussed from a wide variety of perspectives. Fábregas (2013) provides a detailed overview of the main findings in the literature, Pensado (1985) and Laca (2006) present the main diachronic evolution and tendencies. Most research has focused on the so-called nominal factors, i.e., how different types of nouns and their uses interact with a-marking. The definiteness/referentiality and animacy scales have been identified as highly relevant in reflecting both the evolution of DOM as well as the main contrasts in the current system. Nonetheless, the variation is highly complex, and many open questions remain. As will be shown in Section 3, first experimental results have corroborated these findings in a broad sense. The (in)definiteness and (in)animacy contrasts appear quite consistently in such studies. However, although most studies test different categories from the prominence scales, they tend to focus on binary contrasts. The question as to how the scale itself might be reflected in gradient acceptability judgments is hardly ever raised. A notable exception is Bautista-Maldonado and Montrul (2019: 36), where at least one truly scalar effect is reported: acceptability ratings increase gradually along the definiteness scale for unmarked DOs with animate referents. They find the lowest ratings with unmarked nouns and report a gradual increase while moving down the scale passing proper names and definite NPs until, for unmarked indefinite NPs, they receive intermediate ratings. This is also a first indication that acceptability judgments can be used to map the variation of DOM along larger parts of prominence scales, not only for binary contrasts. Such evidence is highly relevant since it would allow for testing of the stronger predictions of the EAH in contrast to binary scales. However, this is an isolated finding, and more research is needed here. Previous studies have not only paid less attention to this question, the sentences used for the different categories of the scale also often involved highly differing linguistic material, a choice that might explain the absence of gradual effects. In order to test the idea of gradually increasing or decreasing acceptability along prominence scales, ideally only the nominal category of the DO should be changed across experimental conditions, and at least three categories of the scale should be tested in order to go beyond binary contrasts. Thus, in a first experiment, similarly to the Catalan one, we tested the idea of graduality in judgments by including an intermediary category between human nouns on one side and inanimate concrete nouns on the other. This category, labeled as “variable” in the experiment, included animals, and inanimate nouns of different types, which according to a native speaker consultant could receive a-marking. Based on the findings of this experiment, a follow-up study was conducted, testing more systematically those nominal categories that were prominently represented in the “variable” group, namely collectives of humans, institutions, and other abstract nouns. Whereas the predictions for the first experiment are quite straightforward and parallel to those of the Catalan experiment, the second needs some further discussion.

In his corpus-based study, Egetenmeyer (2019: 444–486) discusses the semantic features of these three nominal classes. Roughly summarized, Egetenmeyer (2019: 472) implicitly argues that collectives of humans have human denotations, just like nouns with human referents. This generalization is surprising given his otherwise very fine-grained semantic analysis. Arguably, a team is not a human entity, it only consists of such: hence, its denotation should not be collapsed with that of human-referring nouns. Institutions, in turn, are abstract nouns in his account because they imply some feature of internal organization that is more prominent than the individuals that belong to them. From this perspective, he assumes that collective nouns should be marked more often than institutions. This, however, is not borne out by his corpus data, where institutions receive a-marking in 75 % of occurrences and collectives of humans only in 35 %. Therefore, he suggests that other factors within his account of semantic complexity have to be responsible for this finding. However, it might well be the case that the results are not robust enough for drawing strong conclusions given the small sample of occurrences (68 collective and 45 institution nouns), which is acknowledged by the author himself. Thus, more robust and balanced empirical data might confirm the intuition of Egetenmeyer, which is also suggested by the preliminary findings in our third experiment. This is not the place to derive a precise semantics for nouns denoting collectives and institutions. Still, the analysis has to be explicit enough for deriving the predictions of the second experiment. To this end, we propose using the “independent existence” criterion of proto-agents (Dowty 1991; García García 2014). We assume that the meaning of a given (human) collective noun could be paraphrased simply by “consists of human individuals that share some characteristic.” Thus, there is no independent existence of a collective without its individuals; hence, these are highly prominent. In institution-referring nouns, this is not necessarily the case. Institutions involve statutes that define them or some other sort of regulation, typically places or buildings that belong to them, and further internal structures and functions that go considerably beyond group membership of human individuals. This implies that, borderline cases notwithstanding, independent existence of the institution itself is assumed on several levels and, consequently, human individuals that belong to a given institution are less prominent in its conceptual structure than in collective nouns. Thus, we propose that both collectives of humans and institutions denote abstract entities, but they can be distinguished on the prominence scale as given in (11), also predicting decreasing acceptability for a-marking. Since a-marking is not obligatory for any of these noun classes, absence of it should result in more or less acceptable judgments.

Experiments 1 and 2 were conducted in parallel in four regions of the Spanish-speaking world because variational literature revealed that especially in River Plate Spanish and Mexican Spanish, a-marking is more flexible than in Spain (cf. Section 3 for details and references). Thus, we tested these three regions separately and included another important hub of this polycentric language, namely Peru, since little is known regarding possible particularities of DOM in its varieties.

4.1.1 Participants

The general profiles of participants from Argentina, Mexico, Peru, and Spain can be found in Table 2 above. Online recruitment of participants was more difficult in the Americas than in Spain. Nevertheless, at least five participants judged each of the experimental lists. On average, participants were around 10 years older than those of Exp. 3, and less homogeneous with regard to age. Also, in this first experiment, we did not exclude participants with a background in language-related studies.^[8]

4.1.2 Stimulus materials

The stimulus materials were constructed such that the DO slot could receive a human noun, a concrete object (inanimate) noun, and a third category with nouns for which a-marking was optional according to native speakers. Table 3 gives an example of one item in all six conditions.

As can be seen in Table 3, the items were canonical SVO sentences followed by a prepositional phrase, which helped to provide some context. Sentences were also controlled for length. The 30 verbs were chosen such that they fitted with the three noun classes under investigation, but further control was very difficult to achieve. All verbs were conjugated in the Compound Perfect (Pretérito Perfecto Compuesto, PPC), the ending of which has a vowel quality distinct from the following DOM marker, thereby making it salient in contrast to verb forms ending with -a.^[9] Subject and DOs were always definite singular NPs. Half of the DO nouns were masculine, the others feminine. Grammatical gender was maintained for the three nouns of each sentential item (see Table 3). Whereas the noun classes of human and concrete objects are straightforward, the “optional” category (C2/C5) requires further explanation. We asked native speakers for nouns that could be used with and without a-marking in definite NPs. From the gathered nouns, we chose 30 that fell into four larger categories – nouns referring to animals, collectives of humans, institutions, and inanimate nouns referring to humanoid forms. There were also a few further isolated abstract and concrete inanimate nouns.

4.1.3 Results

For the visualization of the results, we use a combination of aggregated mean judgments with confidence intervals and so-called violin plots (the curved shapes around the means). On the y-axis, we plot the z-transformed mean judgment scores. This transformation standardizes the judgments of individual participants (which each might have interpreted the scale somewhat differently), while maintaining the distances between the judgments, thus making them more comparable.^[10] The numbers on the axis represent standard deviations from the overall mean. Thus, the higher a point on the y-axis, the higher the acceptability of the respective category was judged. The 95 % confidence intervals (CI) around the means can be interpreted as follows: if the experiment were repeated infinite times, in 95 % of them, the mean is expected to remain within this interval. Thus, if the intervals do not overlap across conditions, this indicates a statistically robust difference. Violin plots, in turn, show the dispersion of individual judgments in the dataset. For each condition, the higher the numbers of judgments for a specific value, the “bulkier” the plot gets. The results visualized in Figure 1 show that, at first glance, the predicted gradient acceptability between noun classes is largely borne out, including the directionality of the gradual decrease and increase. The results show a kind of “mirror image” for marking and nonmarking, suggesting that for the human and inanimate categories, the presence and absence of marking is not optional. The “variable” category, in turn, ranges between the two other ones and does not show strong contrasts between the marked and the unmarked conditions. However, for this category, the violin plots also indicate that judgments are more heterogeneous than in the other two, especially in the marked conditions in Argentina, Peru, and Spain. This might reflect either the more heterogeneous composition of this group of stimuli or the fact that speakers did not behave homogeneously, but rather showed clustered divergent judgments. Note, furthermore, that for Peru, there is no apparent difference between the unmarked “variable” and inanimate classes. An interpretation of this finding is given in the discussion in Section 5 by comparing the results with those of the other experiments. Finally, while the “ungrammatical” conditions (marked inanimate, unmarked human) are rated equally low in Argentina and Peru, they show an inverse behavior in Mexico and Spain. For Mexican speakers, marked inanimates are more acceptable than unmarked humans, while for Spanish speakers, marked inanimates are less acceptable than unmarked humans. This finding will also be revisited in the discussion.

Figure 1:

Confidence intervals of mean judgments and violin plots across the six conditions of experiment 1.

For the statistical test of the effects, a linear mixed effects model with R (R Core Team 2020) was fitted on the judgments, using the package lme4 (Bates et al. 2015). The fixed effects were noun class and DOM. Participants and items were included as random effects, the judgment being the dependent variable. P-values for the independent variables were obtained by comparing the model with fixed and random effects to a model without the respective independent variable in a likelihood ratio test.^[11] For the interaction, an interaction term was added to the full model and compared to the version without it. Conditional and marginal r² values were calculated for the interaction model to estimate the amount of variance in the data accounted for by the fixed and random effects using the r.squaredGLMM() command of the MuMIn package (Bartoń 2023). Table 4 summarizes the results for Exp. 1 in all regions. The first three columns with numbers provide the chi-square value (with the difference in degrees of freedom in brackets), and the associated p-value for the likelihood ratio test of the respective variable (model comparison with and without the variable). Following standard procedure, we consider a p-value below 0.05 as an indicator for a statistically robust effect for the given factor. The last column indicates how much of the variance is accounted for by the fixed effects (marginal r²) and by both fixed and random effects (conditional r², with possible values ranging from 0 to 1).

In order to interpret these findings, it is important to remember that the results for marked versus unmarked noun classes are roughly mirror-images of each other. As a consequence, the means across categories level each other out and, except for Spain, no main effect for DOM is found in the statistical analysis. The same is true for the factor noun class in Mexico. Thus, the robust effect found for the interaction has to be interpreted as a crossover effect. There are clear effects for DOM and noun class, as seen in Figure 1, but mathematically, both factors individually are neutralized across conditions, hence, the high p-values in some cells of the DOM and noun class column of Table 4.

4.2.1 Participants

As can be seen in Table 2, a total of 240 participants took part in the Catalan experiment, the largest group being from Central Catalonia (90) and the smallest from North-Western Catalonia (36). Thus, for each region and each experimental list, at least 6 participants gave their judgments on the sentences. Participants were recruited via Facebook, personal contacts, and through cultural associations. There were no limitations concerning the age of participants, as can be seen by the wide range and considerably larger SD of Exp. 2 compared to Exp. 3–5. All participants were native speakers of one of the four major Catalan dialects (participants reporting that they spoke several varieties were eliminated since their judgments could not be clearly categorized). As in Exp. 1, participants with a background in language studies were not excluded in this experiment.^[12]

4.2.2 Stimulus materials

The two manipulated factors were presence/absence of DOM (via the marker a) and the three animacy levels (human, animal, inanimate). Table 5 presents one of the test items in all six resulting conditions.

As can be seen in Table 5, the sentences were of canonical SVO word order. All items presented the same basic structure consisting of a definite singular subject NP, a transitive verb conjugated in Periphrastic Perfect (Perfet Perifràstic, PfP), a DO in the form of a singular definite NP containing the possessive adjective, and finally, a prepositional phrase in order to add some context. The 30 verbs were chosen based on their compatibility with all three noun classes. The DOs of each category were kept as homogeneous as possible: human DOs were terms denoting family members, animals were common ones (i.e., pets or farm animals). For the inanimate DOs, only terms denoting material objects were employed, thereby avoiding abstract concepts, which can be attributed a certain degree of animacy due to metonymy. The possessive meva/meu “my” was included in the DO to avoid possible influences from contact languages such as Sardinian, where DOM is blocked in definite DOs that only contain a definite article.^[13] The co-occurrence of definite article and possessive illustrated in the example item is inherent to Catalan in general. Each subject, object, and verb was used only in one of the 30 items, and the prepositional phrases were not repeated either. The sentences in general were of a similar length and within each item the three DOs were kept as parallel in form as possible. Hence, the DOs used in one item show the same grammatical gender and number of syllables. Furthermore, half of the DOs were feminine, half masculine, and the gender distribution between subjects and DOs was kept equal (i.e., out of 30 items, 7 had both masculine subject and DO, 7 both feminine subject and DO, and 8 feminine subject and masculine DO, and vice versa).

The language used for the items was the standard version of Catalan, allowing use of the same experimental set in all four regions in order to achieve fully parallel and comparable results.

4.2.3 Results

As expected (Section 2.2.2), the aggregated results for Catalan show that unmarked DOs are always highly acceptable, but interestingly, marked human nouns receive a similar degree of acceptability (compare the left and right side of the violin plots in Figure 2). Consequently, it seems that DOM with human definite NPs is, firstly, well accepted in all Catalan varieties, and, secondly considered an optional choice. For the animal and inanimate categories, the predicted gradual decrease in acceptability in the marked conditions (the left side of Figure 2) also becomes visible. Hence, for these two noun classes, there is a clear preference for the unmarked DO. Note the difference in overall pattern compared to Exp. 1, where marked and unmarked categories create a mirror-image pattern. For Catalan, instead of a mirror image, we see a reflection of the animacy scale for marked NPs and a nearly uniform acceptability level for unmarked NPs.

Figure 2:

Confidence intervals of mean judgments and violin plots across the six conditions of experiment 2.

The statistical analysis was conducted as described for Exp. 1. The likelihood ratio tests confirmed a significant effect for both independent variables DOM (χ² (1) = 785.05, p < 0.0001) and animacy (χ² (2) = 368.32, p < 0.0001), as well as for the interaction between them (χ² (2) = 282.24, p < 0.0001). As mentioned, the differences between the Catalan varieties were marginal, and, consequently, there was no significant main effect for variety. Nonetheless, it seems that there is some variation concerning DOM, as seen in the significant interaction term for DOM and variety (χ² (3) = 61.074, p < 0.0001). Model fit was assessed by calculating marginal r² (0.14) and conditional r² (0.39) for the model with interactions.

4.3.1 Participants

The participants were university students (excluding students of subjects where linguistic analysis plays a major role, such as philology, literature, and linguistics) and had to have lived in the respective cities for several years. They were recruited in the capitals of the countries in which Exp. 1 had been conducted, except for Argentina. Instead, we recruited students from Montevideo, a city located in front of Buenos Aires on the opposite side of the Río de la Plata, which is equally representative of River Plate Spanish.

4.3.2 Stimulus materials

The stimulus materials were constructed such that the DO slot could receive three kinds of inanimate nouns: collectives (of humans), institutions, and abstract nouns. Table 6 gives an example of one item in all six conditions.

As in Exp. 1, the items were canonical SVO sentences followed by a prepositional phrase, which helped to situate them in some context. Furthermore, sentences were controlled for length. The 30 verbs were chosen such that they were compatible with the three noun classes under investigation; further control appeared to be very difficult to achieve. All verbs were conjugated in the Simple Past (Pretérito Perfecto Simple, PPS).^[14] Subject and DOs were always definite singular NPs. The goal of equal grammatical gender distribution for DOs could not be met entirely given the strong restrictions on DO semantics. Thus, whereas the distribution is even for collectives and institutions, 60 % of the abstract nouns are masculine.

4.3.3 Results

The results of Exp. 3 show that the predictions of gradual differences are borne out for the marked conditions in all languages in the expected direction: as can be seen in Figure 3, collectives received the highest ratings, followed by institutions. Other abstract nouns received the lowest ratings. In the unmarked conditions, the picture is somewhat different. Here, only in Madrid, the expected three-way distinction was found, although the contrasts are not as strong as for the marked conditions. Thus, the overall pattern is not as close to a mirror image as in Exp. 1. For the other three cities, there appears to be no difference between unmarked collectives and institution nouns. In Mexico City, there is even a slight tendency toward a higher acceptability of unmarked collectives. Furthermore, for the marked conditions, the contrasts appear to be strongest in Madrid. Whereas for Montevideo, Lima, and Mexico City, the distance between the highest and lowest z-scores is 0.8, for Madrid, it is 1.4 (confidence intervals having a range of 0.2 or less).

Figure 3:

Confidence intervals of mean judgments and violin plots across the six conditions of experiment 3.

The statistical analysis was conducted as described for Exp. 1. Table 7 summarizes the results for Exp. 3 in all cities.

4.4.1 Participants

The participants were university students (excluding students of subjects where linguistic analysis plays a major role, such as philology, literature, and linguistics). They were recruited in Lisbon and Rio de Janeiro. Besides the other criteria, they had all lived in the respective cities for several years.

4.4.2 Stimulus materials

The stimulus materials were constructed such that the DO slot could receive three kinds of nouns: deity names, proper names, and concrete inanimate nouns. Table 8 gives an example of one item in all six conditions.

As in all experiments, the sentences were canonical SVO sentences that were controlled for length. The 30 verbs were chosen such that they could be combined with the three noun classes under investigation and they were restricted to verbs of adoration, contempt, or description. All verbs were conjugated in the Simple Past (Pretérito Perfeito Simples, PPS). The sentences were framed in a religious context and deity names were taken from different religions and mythologies. Deus “God” was not used because, according to Caro Reina (2020), it is a general noun denoting a deity, not a name. Subject NPs were always definite plurals. Given that the a-marker and the feminine definite article are single vowels of the same quality and fuse obligatorily, feminine gender was avoided in this experiment. Only one item has a feminine plural noun as subject due to the described scenario, all others are masculine. The materials were first elaborated and tested for BP. In order to conduct the experiment in Portugal, the lexical material had to be adapted. Brazilian Candomblé deities, such as Oxum, are not well known in Portugal, and they were, therefore, replaced with further mythological deities in these sentences. Subject nouns were changed to fit these deities where necessary.^[15] Note also that in the Brazilian experiment, only the control conditions with the inanimate DO feature a definite article, since article use with proper names is not mandatory in BP. For the EP experiment, definite articles were inserted for all conditions, making the design even more consistent.

4.4.3 Results

Although the results of Exp. 4 and 5 are not as directly comparable as the regional variants of the other experiments, they are presented here together for convenience. It should be kept in mind, however, that the stimulus materials do not overlap completely. Even so, the two parallel experiments should still provide an adequate level of comparability, while at the same time avoiding external effects due to culturally incompatible vocabulary, which would have resulted in skewed data. The inclusion of the definite article in the Lisbon experiment seems to be negligible since, if anything, it would have produced a global effect, which is not visible in the data. As can be seen in Figure 4, both in BP and EP, unmarked DOs are always more acceptable than marked ones, even for the divinity category. However, at least in Lisbon, participants showed more acceptability for marked divinities than marked proper names, and unmarked divinities were less acceptable than unmarked proper names. In Rio de Janeiro, no robust contrast can be found within the unmarked conditions, but there does seem to be one in the marked DOs, since concrete nouns are deemed less acceptable than the other two conditions (cf. the red values for Rio de Janeiro in Figure 4). Thus, for both languages, the contrasts within marked or unmarked conditions are at most binary.

Figure 4:

Confidence intervals of mean judgments and violin plots across the six conditions of experiments 4 and 5.

The statistical analysis was conducted as described for Exp. 1. Table 9 summarizes the results for Exp. 4 (Rio de Janeiro) and 5 (Lisbon).

Summing up, Section 4 showed that the predictions based on the EAH were partly confirmed in the experiments, but not across the board.