4.1 Acredito que

We begin by studying the mood selection patterns of acredito and the related semantic aspects. As previously mentioned, the function of acredito que is to modalize the speaker’s utterance, emphasizing the subjective status of the content presented. The results in Table 17 show that the selection of the subjunctive mood with acredito que is very frequent in Portuguese, and, what is more, remains high even if the belief predicate is not in the scope of the negation operator (figuring as não ‘not’, nem, or nem sequer). This clearly distinguishes Portuguese from Spanish, where the indicative turns out to be the general mood in affirmative contexts selected by the equivalent expression creo que (see Quer, 1998, pp. 23–25, 44–46, 83–84; Delbecque & Lamiroy 1999, p. 2008f.).

If we sum up the occurrences of these mood forms for both Portuguese varieties, we can appreciate the overall distribution between the indicative and the subjunctive in Portuguese. In the overall corpus, 45.4% of the forms selected by acredito que are present subjunctives (16,011/35,240), whereas 54.6% (19,229/35,240) of the forms are present indicatives. The presence of the subjunctive is strong across the board, though a little bit less pronounced in Brazilian Portuguese (with a ratio of 56.8% IND to 43.2% SUBJ). In European Portuguese, the subjunctive is actually predominant after acredito que (47% IND: 53% SUBJ). However, negation has to be taken into account as a strong conditioning factor when it comes to mood selection. If we exclude the occurrences in the scope of negation, the presence of the subjunctive decreases to 37.8% (vs. 62.2% indicative) in BP and to 41% (vs. 59% indicative) in EP. Still, the subjunctive remains a strong option in both varieties of Portuguese.

Contexts of negation bias the selection of mood in favour of the subjunctive. Correspondingly, in both Brazilian and European Portuguese, the matrix predicate acredito selects the subjunctive in 84.1% of cases in BP and in 92.8% of cases in EP where it is in the scope of negation.

The analysis of the mood selection patterns of acredito que suggests a clear meaning difference. The subjunctive marks an utterance as a subjective or personal stance on its belief content, as in example (8), in which the Brazilian speaker confesses his personal conviction concerning the efficiency of a monarchic regime in Brazil. Similarly, the Portuguese speaker of (9) expresses his personal opinion on the virtues of cocaine.

In contrast, the indicative is used when speakers take or claim the role of an expert or authority on a certain matter. This is the case in examples (10) and (11), in which an Egyptologist and a trainer assess a state of affairs that falls into their respective domains of expertise.

Generally speaking, the indicative is associated with propositional content supported by strong evidence provided by the speaker, who is, or claims to be, an expert on the matter under discussion.

4.2 Acredita que

Table 18 summarizes the mood distribution of acredita in BP and EP.

Comparing Table 18 with Table 17 for acredito reveals significant differences in mood selection patterns between acredito que and acredita que. While the indicative and the subjunctive are distributed relatively evenly with acredito que (in affirmative contexts and, to a slightly lesser extent, in negated contexts), the indicative is overwhelming preferred with acredita que in both varieties. In addition, as the figures in Table 18 show, the selection of the subjunctive turns out to be marginal in contexts without negation in EP. In this variety, the subjunctive appears in 9.1% of affirmative contexts, whereas the form occurs at nearly twice this proportion in BP (18.1%).

Although negation clearly inclines the balance in favour of the subjunctive, its impact is weaker than in the case of acredito que. As the figures show, the subjunctive is selected in only 2/3 of the contexts under negation in BP (69% Subj: 31% Ind), while its proportion rises to 71.6% (Ind: 28.4%) in EP.

Overall, the conclusion can be drawn that mood contrasts play a minor role in the use of acredita que compared to acredito que, especially in EP.

The relatively low number of subjunctives in subordinate clauses selected by acredita que reveals the very specialized, not to say residual, function of the subjunctive in third person doxastic contexts. In the European variety, the indicative is so prevalent that it appears to be the default choice for acredita que in affirmative contexts.

Semantically, the residual cases of the subjunctive mood indicate that the form does not signal a subjective attitude on the part of the belief holder, as is the case for acredito que. On the contrary, a typical example such as (12) demonstrates that representing another person’s belief generally triggers the indicative, regardless of the speaker’s own perspective.

A plausible hypothesis is that the subjunctive can (but need not) be used if the speaker clearly wants to distance himself from another person’s belief or conviction. Take example (13), in which the speaker seems to resort to the subjunctive in order to highlight that the belief content he reports borders on the absurd:

In order to test this hypothesis, a larger sample of examples based on frequent indicative and subjunctive forms (são, estão, seja, sejam, tem, tenha) in the complement clause was analyzed from a qualitative perspective. A distinction was made between different types of beliefs or positions adopted a) by individuals (personal opinion), b) by groups (collective opinion/folk opinion), and c) by experts (expert judgements). The results show that mood alternations between the indicative and the subjunctive occur in both varieties under scrutiny, regardless of the type of belief or conviction expressed in the subordinate clause. This is illustrated by the following examples:

The first category corresponds to a personal/subjective opinion of an individual which can be expressed using either the subjunctive, as in example (14) and (15), or the indicative, as in examples (16) and (17).

Given examples like (15), which contains an absurd belief (“eleven is equal to twelve”), it is tempting to interpret the subjunctive as a marker of the speaker’s distance from the belief. However, this interpretation is challenged by the indicative examples (16) and (17) from the two varieties under examination. In these examples, questionable beliefs (such as those of the troubled minds in (16)) or contestable beliefs (as seen in (17), where the speaker clearly contradicts the person’s view) are reported. It is particularly revealing that the speaker in example (18) does not use the subjunctive to distance himself from the absolutely scandalous position reported of a highly renowned scientist in the field of genetics. This points to the fact that the subjunctive is not (or is no longer) regarded as a device for signaling and highlighting the speaker’s rejection of, or at least his disengagement from, the content of a belief.

The results for the second category, i.e., collective opinions or folk beliefs, point in the same direction. We find cases of the indicative in EP as well as in BP, as shown in examples (19) to (21).

At the same time, however, we also encounter examples of the use of the subjunctive in both varieties, as shown by examples (22) and (23):

Interestingly, we find the same context – the summary of a survey on a specific topic – with varying moods. The indicative is selected in the survey on the role of entrepreneurs reported in (20), while the results of the survey on snoring are presented in the subjunctive mood in example (22). Controversial opinions such as the provocative and blatant elitism in (19) are marked by the indicative mood. The same holds for the critical and polemic stance of the speaker in (21), who clearly speaks out against the alleged view of the gay community. Finally, the belief of the police – as a collective – is expressed in a subjunctive sentence, although the belief holders are quite sure about their assumptions (“muito provável”, ‘highly probable’).

The third category – expert beliefs and opinions – presents a particularly interesting picture.

Most of the belief complements are subjunctive clauses in both varieties. Although a long list of examples could be presented, only one example per variety is cited here, for the sake of brevity, in order to illustrate mood selection in this particular context.

In examples (24) and (25), the convictions of experts on their specialized topics are presented with the subjunctive mood.

We also spotted examples with the indicative mood, like (26); these, however, are much lower in frequency:

To sum up, mood distinctions do not systematically mark degrees of distancing between the speaker and belief content of the utterance. As illustrated by our examples, even scandalous opinions or opinions contested by the speaker are frequently expressed using the indicative mood. On the other hand, it is striking that expert opinions are more frequently expressed using the subjunctive than the indicative. Furthermore, we often find mood variation in very similar contexts, as, for instance, the survey examples have illustrated.

These results suggest that the subjunctive combined with acredita que cannot be interpreted as a general marker of distancing by the speaker. Additionally, we can conclude that different mood selection patterns in very similar contexts highlight the significant role of inter-speaker variability in mood selection. Finally, the stronger correlation (which still requires quantification) between the subjunctive and reported beliefs and convictions (such as those of experts) may point to a quotative interpretation of the subjunctive in these contexts. In line with this hypothesis, the subjunctive would highlight the reported character of a belief that is attributable to a specific – prominent and/or authoritative – source. However, this hypothesis still needs to be tested empirically in a more detailed future study. The same applies to an alternative hypothesis (suggested by a reviewer) that the subjunctive is associated with social prestige and, therefore, is preferred in more formal speech, such as that produced by scientists.

In the next subsection, we will discuss the mood patterns of the alternative expressions creio que and crê que, highlighting their specific peculiarities in contrast to acredito que and acredita que.

4.3. Creio que

Strikingly, the mood selection profiles of creio que and crê que are very different in comparison to the competing forms of acreditar. This general observation applies particularly to the European Portuguese variety, as will become clear in the following two subsections. Table 19 gives the results for creio que. We can observe, from the data in Table 19, that the selection of the indicative strongly prevails in both varieties, the total ratio amounting to 1:2 in favour of the indicative. This tendency is even more pronounced when taking into account the role of negation as an important factor in triggering the subjunctive mood. As indicated in Table 19, the proportion of the subjunctive decreases from nearly 35% to about 25% in BP, and from 30% to a residual 9% in European Portuguese, when considering only affirmative contexts. These data suggest that selection of the subjunctive after creio que is strongly associated with the negation operator (which manifests as não, nem or nem sequer), especially in European Portuguese. This role of the negation operator becomes all the more evident if we appreciate the fact that, in EP, the subjunctive is selected in more than 98% of the cases in which the matrix predicate is in the scope of negation. In other words, negation correlates strongly with the subjunctive mood.

Conversely, the indicative is selected in EP in more than 90% of affirmative contexts. This selection pattern of creio que in the European variety indicates a strong convergence between the mood systems of Spanish and Portuguese in the doxastic/epistemic domain, as we have already pointed out in the previous section. Nevertheless, approximately 9% of occurrences select the subjunctive in affirmative contexts – an option that is completely unavailable in European Spanish. Brazilian Portuguese, on the other hand, continues to exhibit a substantial proportion of subjunctives – still 1 in 4 cases – in affirmative belief contexts.

The clear indicative bias in EP manifests itself in examples such as (27).

Although the speaker expresses a highly personal conviction about his emotions, he employs the indicative mood. In the Brazilian variety, the selection of the subjunctive is more likely in this context, as illustrated by example (28).

By using the subjunctive, the speaker marks his contribution to the blog as a very personal stance. The subtle differences concerning the mood selection in both varieties could be further explored in a more detailed study that combines corpus analysis results with speakers’ acceptability ratings. In the occasional cases in which the speakers – optionally – choose the subjunctive, they tend to relativize the strength of a statement or to minimize its impact. Consider example (29), in which the speaker signals that his explanation should not be taken too seriously or even literally – a message highlighted by his mood selection and, in addition, by his meta-comment “rs” (for risa ‘laughter’):

Finally, we recall that the indicative bias of creio results from its use in typical, quasi-formulaic Christian belief contexts and, specifically, in formulaic confessions of belief. This tradition of credere goes back to the Latin translation of the New Testament, as, for instance, Hieronymus’ translation, the famous Vulgate, shows (see Becker, 2014, p. 187f.). This bias of the indicative, entrenched by a tradition of use in contexts with creio que, is mirrored in the following example, which is just one of many examples in Davies’ Web/Dialect corpus.

4.4. Crê que

The selection patterns of crê que, summarized in Table 20, are also quite specific in both varieties under scrutiny. To begin with, the expression crê que is rare in the corpus, with a frequency of only 577 occurrences. The mood ratio between the two varieties suggests that the subjunctive has even less weight with the third person than with the first person singular. Its marginal role in affirmative contexts, especially in European Portuguese, can be grasped by its minimal frequency in absolute terms, with only 16 (!) occurrences. In relative terms, we can state that the present subjunctive appears in 24.3% of cases in BP and in 17% of the present tense forms in EP. This overall tendency again becomes more pronounced if we exclude contexts of negation. So, in affirmative contexts, the proportion of the subjunctive drops to 17.2% in BP and to a mere 11.5% in EP. Interestingly, negation has a strong impact in favour of the selection of the subjunctive in both varieties, increasing its use to over 70% in each. However, these results demonstrate that the indicative still maintains a relatively strong presence in contexts of negation, and suggest that there is no deterministic relationship between the subjunctive mood and the negation operator.

The marginal status of the subjunctive in affirmative contexts, especially in European Portuguese, is echoed by its very limited range of use. This raises the question of whether the subjunctive in those very rare contexts expresses a particular meaning. Does the speaker want to mark his maximum distance or dissociation from a belief content he is reporting? Note that this kind of possible residual function – an expressive marking of strong dissent from another individual’s belief state – was characteristic of the evolution of the doxastic subsystem in other Romance languages, especially in French. The French subjunctive was also progressively ousted from affirmative doxastic/epistemic contexts in the 16^th and early 17^th centuries. In this period, the subjunctive was purely optional and served to underline the speaker’s strong rebuttal of the belief content under consideration (see Becker, 2014, pp. 263–265). It seems that, at least in French, the optional and particularly expressive use of the subjunctive in affirmative belief contexts represents the last stage of an evolution that announces the complete disappearance of the form in these modal contexts.

In order to cast some light on this hypothesis, we analysed residual subjunctive uses in affirmative contexts determined by crê que. Note that, especially in EP, the subjunctive in affirmative contexts is absolutely marginal, with only 16 occurrences (see Table 20), whereas the BP data yield a slightly higher frequency of subjunctive occurrences.

Interestingly, the contexts of usage of the subjunctive in both varieties do not differ from the indicative ones. Complement clauses that express a hypothesis can be found with both subjunctive and indicative verbal forms. The modal verb poder occurs with both moods and expresses epistemic modality, as in (31) and (32).

Similarly, collective convictions or folk beliefs are expressed using both moods in both EP and BP (although with differences in their frequencies, as mentioned above). The following examples, (33) and (34), illustrate the mood alternation with respect to this type of belief content in EP:

Examples (35), (36), and (37) illustrate the same mood alternation in the BP variety:

The Brazilian examples illustrating mood selection after crê que show – in analogy with the selection patterns for acredita que – that the personal stance of the speaker does not motivate the mood alternation. In the indicative example in (35), the speaker distances himself from the false beliefs of the superstitious character that he is denouncing. In the subjunctive example in (36), on the other hand, the speaker marks explicitly (with “e com razão”) that he sides with the collective conviction (“muita gente”). In example (37), however, the speaker simply reports the results of an opinion poll.

It is striking that reported expert opinions and convictions generally appear with the indicative mood in both varieties. We quote indicative examples that summarize the standpoints of philosophers like Habermas (the European Portuguese example in (38)), but also of an expert in the field of sociology (the Brazilian example in (39)):

A very rare example of the subjunctive in a complement clause expressing an expert opinion, or rather a hypothesis, is found in the BP example in (40).

Even rarer are examples of expert opinions marked by the subjunctive in EP, such as the one in (41).

What conclusions can we draw from mood alternations in affirmative contexts determined by crê que? As has been illustrated, neither the speaker’s disagreement with nor distancing from a point of view, nor the subjective character of the propositional content of the complement clause, is a sufficient trigger for the subjunctive mood. As the quantitative data show, there is a strong bias towards the general selection of the indicative mood after crê que. This preference for the indicative may be strongest for expert opinions or convictions. The rare subjunctive occurrences refer to tentative hypotheses, though both moods appear with the epistemic modal poder – pode, as in example (40), and possa, as in example (41). Based on the quantitative results, we may conclude that there is an overwhelming trend that suggests the generalization of the indicative in the contexts determined by crê que. Thus, this doxastic expression foreshadows the convergent trends in the development of the mood systems of European Portuguese and European Spanish within the domain of doxastic modality.

4.5 A regression analysis

In this subsection, we aim to disentangle the factors that determine mood selection and test the statistical significance of their predictive power for mood choice in BP and EP. Given the focus of this paper, we are especially interested in highlighting subtle differences in mood selection patterns between the two major varieties of Portuguese. Therefore, the crucial question is whether clear-cut differences regarding mood selection patterns between EP and BP can be identified and whether these differentiating tendencies possess statistical significance.

For our quantitative analysis, we considered three variables, which were already highlighted in previous Section 4.4. These variables are:

1. Variety: Brazilian Portuguese vs. European Portuguese

2. Negation: Negated vs. non-negated and

3. Person: first person vs. third person.

In order to tease apart the three factors examined in this study and test the statistical significance of their predictive power for mood choice, we ran two logistic regression models, one for acreditar and one for crêr. We used the glm function in R (R Core Team, 2024) to compute models that attempt to predict mood choice using these three independent variables and their two- and three-way interactions.

Given the focus of our article, we are particularly concerned with the impact of the variable “variety”. The model on acreditar is based on the corpus data summarized in Table 21.

For ease of interpretation, we report the coefficient estimates of our models as odds ratios. Table 22 displays a summary of the acreditar model.

We can interpret the odd ratios against the backdrop of the model’s reference level – which corresponds to Brazilian Portuguese (BP), first person (acredito que), without negation (non-negated). The model summary can be interpreted as follows:

• - The intercept has the value of 0.61, meaning that the odds of the subjunctive being selected in BP for acredito in contexts without negation is 0.61 times “higher” than of the indicative being selected under the same circumstances. In other words, the choice of the subjunctive is clearly less probable than of the indicative, but it is far from rare.

• - Interestingly, the odds of the subjunctive being selected under the same circumstances in EP (the matrix form being acredito, without negation) is 1.23 times higher (with an odds ratio of (0.61*1.23=) 0,75) than in BP. This indicates that EP has a stronger subjunctive bias in non-negated contexts of acredito than BP.

• - Negation is a very strong predictor for the selection of the subjunctive in both varieties. In the case of non acredito in BP, the odds of the complement clause being in the subjunctive are 8.68 times higher than without negation.

  In European Portuguese, the odds of the complement being in the subjunctive in contexts of non acredito is 12.92 (0.61*1.23*8.68*2.01=12.92) times higher than of the complement being in the indicative.

  Therefore, we can conclude that if negation is already a strong predictor for the selection of the subjunctive per se, it is a particularly strong predictor for the selection of the subjunctive in EP.

• - The third person has a much stronger bias for selecting the indicative than the first person. The odds of the subjunctive being selected with the third person and under the non-negated condition are only 0.22 (0.61*0.36=0.22) in BP, and even less in EP, at just 0.1 (0.61*1.23*0.36*0.37=0.1).

  In the negated condition, the odds of the subjunctive being selected rise to 2.21 (0.61*8.68*0.36*1.16) in BP and to 1.47 (0.61*1.23*8.68*0.36*2.01*0.37*1.25) in EP.

We were also interested in comparing probabilities of specific combinations of variables. For this purpose, we used the emmeans function (Lenth 2020). The results are presented in Table 23.

Table 23 shows that across both varieties (BP and EP), negation significantly increases the probability of the subjunctive being selected. For instance, in BP, the probability rises from 37.8% to 84.1% for the first person, and from 18.1% to 69% for the third person. In EP, the probabilities rise from 42.7% to 92.8% for the first person and from 9.1% to 71.6% for the third person.

The table of probabilities confirms that the third person (acredita) generally shows a much lower probability of selecting the subjunctive compared to the first person, especially in the non-negated condition. If the probability of the subjunctive being selected with non-negated acredita is already quite low in BP (at 18.1%), it turns out to be marginal in EP (at 9.1%).

The EP variety shows higher probabilities of selecting the subjunctive in both the negated and non-negated conditions than does BP. This difference is particularly evident in the negated condition for the first person (92.8% for EP vs. 84.1% for BP).

In the second part of this subsection, we focus on tracing the quantitative profile of crer with regard to mood selection. The model predicting mood choice with crêr is based on the following data, which summarizes the different occurrences of acreditar in the corpora under study.

For ease of interpretation, we again report the coefficient estimates of our models as odds ratios. Table 25 summarizes the results for the crer model.

• - The intercept has the value of 0.33, which means that the odds of the subjunctive being selected by creio in BP are 0.33 times that of the indicative being selected under the same circumstances. However, in EP the odds of the subjunctive being selected under the same circumstances are much lower: 0.1 (i.e., 0.33*0.3). In other words, it is quite rare for the subjunctive to be selected in this specific context.

• - Negation again has a strong impact on the odds of the subjunctive being selected in both BP and EP. However, the effect is much more pronounced in EP (odds for creio: 0.33*0.3*27.99*20.94=58) than in BP (0.33*27.99=9.24).

  The odds drop considerably in the third person condition (crê) in BP (odds for crê: 0.33*27.99*0.63*0.41=2.97) and even slightly more in EP (0.33*0.3*27.99*0.63*20.94*2.09*0.41*0.08=2.5).

We were also concerned with comparing the probabilities of specific combinations of variables. The results are presented in Table 26.

Table 26 shows that negation significantly increases the probability of selecting the subjunctive in both BP and EP, regardless of person. For instance, in BP the probability of selecting the subjunctive with negation in the first person jumps from 24.7% in non-negated contexts to 90.2% in negated ones. Similarly, in EP the probability of the subjunctive being selected with negation increases from 8.99% to as high as 98.3%.

Apart from that, the first person condition consistently shows a higher probability of selecting the subjunctive than the third person condition, especially when negation comes into play.

BP generally shows a higher probability of selecting the subjunctive compared to EP. In BP, the probability of the subjunctive being selected with creio is relatively high (24.7%); it remains reasonably high even under the third person condition in non-negated contexts (17.2%). By contrast, in EP the propensity of the subjunctive to be selected under the non-negation condition is quite low with the third person (11.5%), and even lower with the first person (8.99%).

Finally, we plotted the model’s estimates with 95% confidence intervals using various plotting functions. The plots (see Figure 1) visualize the peculiarities and patterns of mood selection with respect to the two belief predicates acreditar and crer in BP and EP. The y-axis represents the probabilities from 0.0 (=0%) to 1.0 (=100%). The plots convey an impression of the differences concerning mood selection patterns between the two varieties EP and BP, summarizing the impact of the three factors (person, negation, and variety) considered in this paper.

Figure 1

Predicated mood selection patterns for acreditar and crêr in BP and EP.

The plots highlight clear-cut differences between the mood selection patterns of acreditar and crer in EP and BP. The following aspects can be emphasized in our summary:

• - Non-negated acredito is associated much more strongly with the subjunctive than non-negated creio in both varieties, with a slightly stronger bias for the subjunctive in EP. In EP, non-negated creio rarely occurs in combination with the subjunctive mood.

• - Non-negated acredita is associated more strongly with the subjunctive in BP than in EP. As concerns crê in both varieties, the subjunctive mood is not very frequent, but it is not marginal, either.

• - Negation has a strong impact on acreditar and crêr. Interestingly, the probability of the subjunctive being selected after acredito and creio under the negated condition is very high in both varieties, but is consistently higher in EP. On the other hand, there are no significant differences between BP and EP as regards the third person under the negated condition.

To summarize, we can highlight the following three major results:

• 1.) Crer vs. acreditar: The selection of the subjunctive remains significantly stronger with acreditar than with crer under the non-negated condition in both EP and BP. Under these same conditions, the impact of the subjunctive is stronger in BP than in EP, with the exception of acredito, which exhibits a slightly higher subjunctive bias in EP. Notably, creio is particularly infrequent with the subjunctive mood in EP.

• 2.) Negation: Negation has a very strong impact on the selection of the subjunctive in both varieties. However, the impact of negation with acredito and creio is consistently stronger in EP than in BP for both speaker-oriented forms.

• 3.) First vs. third person: The subjunctive bias is generally weaker in third person contexts than in first person contexts. This tendency is even more pronounced in EP than in BP in the non-negated condition. However, in the case of acredita and crê in the negation condition, no significant differences between the two varieties are discernible.

4.6 Summary of the analysis

In conclusion, it can be observed that the subjunctive, when combined with acredito que, serves a distinct function within the context of modalization, particularly in marking and highlighting the subjective nature of the speaker’s stance. This role is somewhat more pronounced in EP than in BP. In contrast, the function of marking the speaker’s subjectivity in contexts involving creio que + subjunctive is significantly weaker in both varieties, playing only a marginal role in EP. This trend is even more evident in the case of the third person forms of the two belief predicates: although still relatively frequent with acredita que in BP, the subjunctive tends to be almost marginal in EP. Negation has a very strong impact on the selection of the subjunctive in both EP and BP. However, the subjunctive bias of negated acredito que and creio que is consistently stronger in EP than in BP for both speaker-oriented forms. This indicates that the subjunctive functions primarily as a marker of disbelief on the part of the speaker, with this effect being more pronounced in EP than in BP.

The in-depth qualitative analysis of the corpus data reveals that the subjunctive, when following acredita que and crê que, no longer functions as a device for marking the belief holder’s (maximal) distance with respect to the belief content (if indeed it ever did). On the contrary, our analysis reveals that mood alternations occur in similar contexts and appear to be independent of the degree of subjectivity of the belief holder. This significant new insight can be illustrated by a scale of belief types, ranging from personal convictions to expert opinions, with intermediate stages including collective and folk beliefs. As a result, we were unable to identify general cues or principles that consistently motivate and explain the selection of one mood over the other.

Future research should aim to determine whether some speakers alternate freely between the two moods (free variation) or whether those who do alternate still select the moods in a systematic manner, according to clearly identifiable cues and principles. If the latter proves to be the case, it would suggest the co-existence of competing grammars within the speech community of contemporary Portuguese (for an explanation of linguistic change in terms of competing grammars, see Kroch, 1994; for the application of this approach to the evolution of Brazilian Portuguese, see Galves & Kroch, 2016). This potential co-existence of competing grammars should be systematically contextualized and described using sociolinguistic variables such as age, social status, education, and register.

5.1 The inflected infinitive complement clauses selected by acreditar que and crer que in BP and EP

As already commented on above, inflected infinitive constructions are very rare. The detailed analysis of all the examples displayed in the corpus casts some interesting light on the correlation between the matrix form and the person agreement on the bare infinitive (Table 21). Conspicuously, acredito combines almost exclusively with inflected infinitive forms of the third person plural (e.g., serem) in the European variety (55/59 occurrences, corresponding to 93.2%). Other inflected infinitive forms turned out to be marginal: we encountered only one or two occurrences of the second person singular form (seres), the first person plural form (sermos), and the second person plural form (serdes). In the Brazilian variety, this trend is a bit less pronounced. The third person plural forms (serem) clearly prevail (with 166/186 = 89.2% of occurrences). However, first person plurals do yield 10.8% (i.e., 20/186) of occurrences.

As regards acredita, the attested combinations present an even more “monolithic” picture (see Table 22). In BP, acredita combines with an infinitive marked for the third person plural (mostly serem) in 100% (!) of the cases. In EP, the share amounts to 98.5% (64 out of 65 occurrences), with a single occurrence (1/65 = 1.5%) attested with the second person plural (serdes).

As shown in Table 23, the form creio is also overwhelmingly associated with the third person plural (serem) – in 78.8% of the cases (corresponding to 67 out of 85 occurrences) in BP and in 69.7% of the instances (i.e., 46 out of 66 inflected infinitive forms) in EP.

Similarly to acredita, crê also only selects personal infinitives in the third person plural (serem). In other words, speakers typically refer to the belief content of a distant, i.e., “third party”, belief holder who, in turn, refers to individuals and properties not present in the situation of utterance. In comparison to acredita, however, crê only marginally selects an inflected infinitive clause, as can be seen from the figures in Table 24.

To round off the picture, a further important aspect must be highlighted. It appears that inflected infinitives are “recruited” from a very limited set of verbs. As Table 21 shows, serem is the dominant form, amounting to 122 out of 166 occurrences in BP and 25 out of 65 occurrences in EP. The correlation is somewhat weaker with acredita, where serem is found in 80 out of 129 instances in BP and 24 out of 64 occurrences in EP. The same holds for creio and crê: serem occurs in 37 out of 67 instances selected by creio in BP, and in 30 out of 46 instances in EP. With crê, we find serem in 18 out of 21 occurrences in BP and in 2 out of 9 occurrences in EP. Note that the second most frequent inflected form is terem, followed at a considerable distance by estarem (see Tables 21, 22, 23).

This clear correlation between the belief predicates under scrutiny and a small number of inflected infinitive forms, especially terem and serem, may point to the routinization of usage patterns with highly frequent verbs. This tendency appears to be more accentuated in the Brazilian than in the Portuguese variety. The question of whether this is a sign of the more conservative character of BP, which preserves these well-entrenched structures due to the high frequency of these verbs (see Rosemeyer, 2014), or whether we are dealing with a loss of productivity in BP, which ensues from the more limited set of infinitival candidates, should be dealt with in more depth in a separate study.

Apart from that, clear selection restrictions also come to the fore: the set of inflected infinitive forms attested in the corpus falls exclusively into the domain of stative verbs. This list of verbs (with one or two attestations in the corpus queries) also includes inflected forms of deontic verbs, such as dever(em), and of evaluative predicates like merecer(em) (‘to deserve’) and abundar(em) (‘to be plentiful’, ‘to be abundant’). Inflected forms like fazerem (‘to do’) and virem (‘to come’), which are not stative per se, occur as support verbs in collocations and periphrastic constructions, such as fazer sentido (‘make sense’), fazer sucesso (‘to be successful’), and vir a ser (‘to be going to’). The results of our corpus queries confirm, therefore, Ambar’s (1994) theoretical considerations addressed in Section 2.1. It would be interesting to discuss the consequences of the stativity requirement for a semantic analysis of the infinitival structures selected by belief predicates. From a semantic point of view, it seems that these infinitival structures suspend the time axis associated with the temporal structure of an event.

Finally, a very curious and unexpected result of our queries is the fact that some speakers combine both construction types under scrutiny, i.e., the finite and the inflected non-finite construction. More precisely, the speakers introduce the complementizer que, followed by an inflected infinitive form (“infinitivo flexivo”). This phenomenon is very rare and is restricted to the first person (acredito que + inflected infinitive with 9 occurrences in the corpus, and creio que + inflected infinitive with 6 occurrences). However, it is systematic and therefore cannot be shrugged off as a casual grammatical or orthographic error. This is also shown by the fact that a Google query of the type acredito que termos yields an impressive number of occurrences. The examples in (42) and (43) – taken from each of the Portuguese varieties under scrutiny – illustrate this curious phenomenon.

For the time being, we cannot come up with a sound explanation. However, the phenomenon may give an insight into the way the inflected infinitives are stored in the memory.

5.2 The subject position of inflected infinitives selected by acreditar and crer in BP and EP

In this subsection, we return to the question of the position of the subject in inflected complement clauses, with a particular focus on the availability of preverbal subjects. As highlighted in Section 2.1, the generative literature on inflected infinitive clauses is especially concerned with the subject position in these constructions, which differs in accordance with the verb class involved. In the case of doxastic predicates, the generative research claims that preverbal subjects are excluded in principle for structural reasons (see Section 2.1). However, in very rare cases (not specified by the generative literature) they do occur and, according to Mensching, these occurrences are more acceptable in Brazilian Portuguese than in European Portuguese (see Mensching, 2000, p. 28).

Generally speaking, our data confirm this observation. The number of examples with preverbal subjects in belief constructions with inflected complement clauses is minimal. However, these examples do exist and they attest to a marginal option in the grammatical system of the Portuguese language. Contrary to the literature, we were unable to verify that the pre-position of the subject is marginally more acceptable in BP than in EP. On the contrary, Davies’ Web/Dialect corpus contained examples of a nominal preverbal subject only in the European variety of Portuguese. In addition, another interesting aspect comes to light: the pre-position of personal pronouns seems to be, in general, more acceptable than that of a preverbal nominal phrase. However, our findings also allow us to refine other observations, further differentiating between the two varieties under scrutiny.

To begin with, the typical position of the subject in the infinitive clause is postverbal in both European and Brazilian Portuguese. See the standard examples (44) and (45) for both varieties under scrutiny.

As regards the availability of preverbal subjects in inflected infinitive clauses, a more differentiated picture emerges in the Davies’ Web/Dialect corpus. As the data suggest, preverbal nominal phrases (NP) are marginally accepted in EP but seem to be excluded in BP. At least, we did not find a single instance of this pattern in BP but there were some isolated examples in EP, such as (46).

More importantly, the preverbal position seems (at least more) acceptable with certain grammatical items in both varieties under consideration. The following categories are attested in both EP and BP, as the illustrative examples in (47) to (52) show.

These examples show that different kinds of pronouns, with their anaphoric potential and minor “syntactic weight” (in contrast to full NPs), are attested, if infrequently, in the preverbal position in inflected infinitive clauses.

A further frequently recurring structure comes to light in the Web/Dialect subcorpus, namely inflected infinitive complements of doxastic/epistemic verbs nested in a relative clause. Examples (53) and (54) are representative of these structures in the two varieties.

In these constructions, creio and acredito do not constitute simple parenthetical segments. If they were parentheticals, the verb following the belief predicate would be a finite form. Take the following example (55).

In the construction in (55’), the nominal phrase, i.e., the direct object of the main clause, is left-dislocated from the following base structure (55’).

In a next step, the construction is left-dislocated (55’’) and is finally integrated into a relative clause (55’’’).

Note that nested structures seem to be particularly amenable to being complement clauses of belief predicates that constitute copula sentences.

We close this subsection with a final interesting observation. There is a specific postverbal pronominal position of the matrix clause that functions as a focus position. This focus position allows the speaker to contrast or to emphasize the subjective or personal character of his/her stance on or conviction about a certain state of affairs (see Posio, 2012, 2014, p. 15), This (contrastive/emphatic) focus position can be occupied by the first person pronoun eu, as illustrated in (56).

This postverbal focus position in combination with doxastic predicates is not limited to infinitive clauses. It is also available in finite complement clause constructions, as examples (57) and (58), with initial and finite verb positions, illustrate.

To sum up, we can conclude that there are clear differences in the very restricted availability of preverbal NPs between European and Brazilian Portuguese. As the examples show, EP allows them marginally; for BP, however, we did not find any occurrences in the corpus. There is still a need for acceptability tests to determine whether preverbal NPs are completely excluded in BP, i.e., systematically banned in that variety. There is, however, a strong convergence between the two varieties concerning the marginal possibility of placing anaphoric pronominal material in front of the inflected infinitive. These first results need to be complemented by further experiments testing speakers’ acceptance of preverbal pronominal and nominal subjects in the two varieties of Portuguese considered in this paper. Apart from that, the (restricted) availability of anaphoric pronominal material in preverbal position must be accounted for in an appropriate manner by integrating these findings into a formal syntactic model.