Since the 1990s, a number of studies on the history of Brazilian and European Portuguese wh-interrogatives have evinced syntactic change regarding (a) the expression and placement of the subject constituent, (b) the usage frequency of clefting strategies and (c) the preferred position of the wh constituent (Duarte 1992; Lopes-Rossi 1993; Lopes Rossi 1996; Kato & Mioto 2005; Kato & Ribeiro 2005; 2009; Fontes 2012a; b; Pinheiro & Marins 2012; Kato 2014; De Paula 2015; 2016; 2017). Regarding (a), several studies demonstrate that over time, null subjects (2a) and postposed subjects (2b) came to be replaced with preposed subjects (2c). For instance, Pinheiro & Marins (2012: 172) find that in a corpus of theater texts dated between the 19th and 20th century, SV word order rose from a relative frequency of 7 to 76 percent, to the detriment of the two other patterns of subject expression.

At the same time, there is a marked increase in the use of clefted wh-interrogatives (see 1b) in their data, from 0 to 74 percent. There thus appears to be a relationship between the changes in subject use and clefting. As already observed by Duarte (1992), early tokens of CleftWh are much more likely to display SV word order than non-clefted wh-interrogatives, which is why she hypothesized that CleftWh served as a catalyst for the spread of SV from declarative sentences to wh-interrogatives.

This study focuses on the change in (c), i.e. the rise in the usage frequency of InSituWh in Portuguese, documented in several of the cited studies. Lopes Rossi (1996: 68) finds for BP that the relative usage frequency of InSituWh increased from 0 percent in the first half of the 19th century to over 28 percent in the second half of the 20th century. For EP, she only documents an increase from 0 to 2.9 percent in the same period. It has to be noted, however, that Lopes Rossi only analyzed one theater play per 50-year period and dialect, which very much weakens the representativeness of her results. De Paula (2016), analyzing a much bigger sample of theater texts, confirms Lopes Rossi’s results for BP and EP. The author documents an increase of the relative usage frequency of InSituWh in n = 7 BP theater texts from 0 percent in 1845 to 15.9 percent in 1992. In her much larger sample of EP theater texts (n = 40), the relative usage frequency of InSituWh oscillates between 1 and 3 percent (2016: 73). In De Paula (2017), the author also analyzes the diachronic distribution of wh-interrogatives in sociolinguistic interviews in EP and BP. She finds an increase in the relative usage frequency of InSituWh in the interviews in BP conducted in the 1970s and the 2010s, from 4 to 10 percent. For EP, she actually documents a decrease in the relative usage frequency of InSituWh, from 17 percent in the interviews conducted in the 1980s to 5 percent in the interviews conducted in the 2010s. In summary, these results suggest that the use of BP InSituWh has become more frequent in theater texts and possibly even in spoken language.

Although there is thus solid evidence for the existence of this change, to my knowledge no study has yet offered an explanation for it. Likewise, the previous studies do not investigate the possibility of changes in the usage contexts of InSituWh. However, as we shall see in the next section, the possibility of such changes is the key to explaining the changes in the usage frequency of InSituWh.

Descriptions of the semantics of wh-interrogatives distinguish between the proposition of the interrogative (henceforth P) and the referent of the interrogative pronoun or adverb (henceforth X). Following Hamblin (1973), the meaning of a wh-interrogative like Where did John go? can be described as the sum of all possible answers (e.g., ‘John went to London’, ‘John went to Barcelona’ etc.). The use of a wh-interrogative presupposes the validity of P. Thus, while Where did John go? can be uttered in a context in which the proposition has been mentioned in or pragmatically inferred from the previous context, the questioning speech act can also be felicitous in out-of-the blue contexts. In such contexts, the speaker can exploit the fact that the hearer will accommodate the presupposition, i.e. integrate the assumption that John left into her belief system on the basis of the inference that the speaker of the interrogatives appears to believe so.

In Romance languages such as French (Coveney 1990; Obenauer 1994; Chang 1997; Mathieu 1999; Cheng & Rooryck 2000; Adli 2006; Myers 2007; Boucher 2010; Cheng 2013), Spanish (Dumitrescu 1992; 1993; Escandell-Vidal 1999; 2002; Dumitrescu 2008; Rosemeyer 2018a) and Portuguese (Pires & Taylor 2007; Oushiro 2010; 2011a; b; De Paula 2016: 105), InSituWh is most frequently used in contexts in which the proposition has been mentioned in the previous context and can consequently be described as activated. Consider, for instance, example (3), taken from the C-ORAL BRASIL, a corpus of informal spoken conversation between Brazilians from Minas Gerais (Raso & Mello 2012).1 FLA did not understand EMM’s utterance in line 1, which is why she uses the InSituWh interrogative aparece o quê? in line 3 in order to ask FLA to repeat the utterance. Such echo question readings exist in all of the mentioned Romance languages.

However, there are great differences between the Romance languages with respect to whether or not the use of an InSituWh interrogative requires the proposition to be activated. In particular, BP InSituWh can be used with non-activated propositions, as demonstrated by examples such as (4). While tidying up the room, LAO and MBA find Julio’s trousers and subsequently discuss how much he loves these trousers and how difficult it was for Ana – supposedly his partner – to convince him to let go of them (lines 1–10). In line 11, MBA discovers another item belonging to Julio, a sandal, and asks LAO how many days this sandal has been lying on the floor. The possibility of such out-of-the blue uses is also documented for French (Adli 2006: 184; Boucher 2010: 109; Larrivée to appear), but is mostly unacceptable in Spanish (Rosemeyer 2018a).

Note that whereas in (3), the interrogative aparece o quê? is used with an ascending intonation (indicated by ? in the transcript), in (4) the interrogative receives a falling intonation (indicated by; in line 12). This observation conforms to the generalization established in Kato (2013) for BP that echoic InSituWh usually receives a rising intonation, whereas non-echoic InSituWh is typically pronounced with a falling intonation.

Crucially for our purposes, the parameter of the activation of P is not only related to intonation but also to the degree of to which the interrogative even needs to be answered. This idea can be couched in terms of the notion of conditional relevance (Schegloff 1968). Wh-interrogatives that express information questions occur as the first part of an adjacency pair, making the second part (i.e. the answer) highly relevant. Both (3) and (4) can be described as such contexts, since in each example the speaker expects the hearer to answer the question. However, sometimes wh-interrogatives are not used to ask for information. Consider, for instance, example (5), in which the speaker uses the InSituWh construction as a rhetorical question. VER clearly assumes that everyone knows the answer to his question, which is why he does not even wait for the interlocutors to provide this answer; rather, he provides it himself. The answer consequently has a very low degree of conditional relevance and VER could have easily left the answer out.

The three examples in (3–5) thus suggest an intricate relationship between the parameters of activation of P, conditional relevance and intonation in the use of BP InSituWh. Whereas echoic uses of InSituWh typically receive an ascending intonational contour, non-echoic uses receive a falling contour. The non-echoic uses can be either information questions or rhetorical questions, and may have propositions that are discourse-new.

These considerations are relevant to the analysis of the changes in the use of InSituWh in BP because some authors contend that the parameter of activation of P played an important role in the history of French InSituWh. For instance, Larrivée (to appear) analyzes the use of InSituWh in a comparative corpus of French sociolinguistic semi-directed interviews conducted between 1969 and 1974, and in 2014. Following Dryer (1996), the author annotates the data according to the parameter of activation of P, distinguishing between propositions that are explicitly activated and those that are discourse-new. He documents an increase of the relative frequency of InSituWh in contexts in which the proposition is discourse-new, from 46 percent (n = 6/13) in the 20th century data to 86 percent (n = 60/70) in the 21st century data. However, as acknowledged by Larrivée himself, follow-up studies analyzing a greater quantity of data are necessary in order to confirm this result.

In line with Larrivée’s (to appear) analysis of the diachrony of French InSituWh, this paper explores the question of whether the increase in the usage frequency of BP InSituWh was accompanied by a change in the activation of P. In particular, it tests the hypothesis that InSituWh became less strongly bound to contexts in which P is activated.

Figure 1 illustrates the changes in the usage frequency of InSituWh in the corpus of BP theater texts. Each point in the graph represents the log-transformed normalized usage frequency of InSituWh for one year (in turn representing one or more plays published that year), whereas the line represents the result of a local polynomial regression analysis indicating the trend. Figure 1 demonstrates that InSituWh interrogatives have become more frequent in BP theater plays, especially after the beginning of the 20th century, confirming the results from previous historical studies (see Section 2).

Table 3 gives the non-normalized usage frequencies of InSituWh per 25-year time period, again demonstrating that InSituWh is very infrequent before the second half of the 20th century.

The increase in the usage frequency of InSituWh goes hand in hand with distributional changes. First, as hypothesized in Section 2, there are changes with respect to whether or not InSituWh can be used with discourse-new propositions. Figure 2 visualizes this change, masking the scarce (n = 5) number of tokens of InSituWh before 1925. Whereas in the first half of the 19th century, a majority of InSituWh tokens occur in discourse-old contexts (i.e. contexts in which the proposition has either been mentioned or is logically dependent on a previous proposition), after 1950 InSituWh gradually advanced into discourse-new contexts. In the 21st century data, almost 75 percent of the InSituWh tokens occur in discourse-new contexts.

In line with the description of the typology of discourse functions of InSituWh from Section 4, one might interpret this result as a change from anaphorical to cataphorical discourse functions. If this were the case, the change could be interpreted as an intrusion of the use of InSituWh into information question functions previously reserved to ExSituWh interrogatives.

Crucially however, it was shown in Section 4 that rhetorical InSituWh tokens can also appear in discourse-new contexts. As it turns out, the increase of InSituWh in discourse-new contexts is largely due to a change from non-rhetorical to rhetorical discourse functions. Figure 3 visualizes this change, collapsing the three groups of discourse functions (ANA = anaphorical, CATA = cataphorical and RHET = rhetorical) and the parameter of whether or not the proposition is activated (NewP vs. OldP).

Figure 3 clearly illustrates that the change from usage contexts with activated propositions to usage contexts with non-activated propositions is moderated by discourse function. Most importantly, there does not appear to be a strong increase in the relative usage frequency of non-rhetorical cataphorical discourse functions in contexts in which the proposition is discourse-new (orange color). The relative usage frequency of these contexts only increases from 25 percent in 1925–1949 to about 36 percent in the 21st century. Note however that due to the fact that most of the InSituWh tokens are found in texts dated between 1974 and 2016, the increase in the frequency of these contexts between the two last time periods might still be significant.

We document a stronger and more consistent increase in the use of rhetorical discourse functions, especially with discourse-new propositions (green color), from about 13 percent in 1925–1949 to about 36 percent in the 21st century. This increase coincides with a decrease in the usage frequency of the use of InSituWh in non-rhetorical cataphorical contexts with a non-activated proposition, i.e. with the discourse function of Elaboration (blue color). These results do not suggest that the historical trend of BP InSituWh towards a higher usage frequency in contexts in which the interrogative proposition is discourse-new illustrated in Figure 2 is symptomatic of a functional change of InSituWh towards information question use. Rather than intruding into information question contexts, the construction becomes more frequent with rhetorical question readings.

The next steps in the analysis aim at confirming the results from the last section using inferential statistical methodology. In a first step, a logistic regression model was calculated in order to validate the coding of discourse functions by evincing correlations between the interpretation of the data and more objective contextual criteria.

Based on the results from the last section, a binary dependent variable RHET_NEWP was constructed that distinguished between rhetorical uses of InSituWh in contexts with non-activated propositions (RHETNEWP = ‘True’) and all other discourse functions of InSituWh (RHETNEWP = ‘False’). The regression model thus measured the probability for an InSituWh token to be used in contexts classified as rhetorical with non-activated propositions. Table 4 summarizes the predictor variables used in the regression model. These predictor variables were hypothesized to have an influence on the likelihood for an in-situ wh-interrogative to realize a rhetorical reading on the basis of the discussion of the properties of rhetorical questions in section 4 and the descriptive findings in section 5.1, with the exception of the variable LOGORALITY.

The predictor LOGORALITY was included in order to test the intuition that rhetorical uses of wh-interrogatives are more likely to occur in formal than in informal registers. Although the corpus used for the analysis can be characterized as homogeneous in terms of genre, the theater plays still represent orality to a greater or lesser degree, depending on the degree to which the authors heed the norms of written BP. Consequently, it was hypothesized that in plays that represent spoken language more accurately, the use of rhetorical questions is less likely. The predictor LOGORALITY establishes a measurement of the degree to which the plays in the Brazilian Portuguese corpus represent orality, by using Biber & Finegan’s (2004 [1987]: 68) dimension of “involvement” of the oral/literate dimensions of variation.3 Five linguistic variables, listed in Table 5, were extracted from the corpus of BP theater texts. These variables can be said to represent orality due to the fact that their use is contingent on temporal, spatial or discourse deixis (present progressive, demonstrative neuter pronouns, time and place adverbs and discourse markers) or because they represent intellectual states prone to expression in orality (the type of verbs that Biber & Finegan call “private verbs”). Both realizations typical for EP (e.g., estar + infinitive progressives) and BP (e.g., estar + gerund progressives) were included in order to capture all variants in all temporal periods. As proposed in Biber and Finegan’s study, the frequencies of the five variables were aggregated for each text. Afterwards, they were normalized and log-transformed in order to ensure comparability of the numerical variable across the different texts. The higher the value of a text for the resulting variable LOGORALITY, the more likely it is to represent spoken language more accurately. As summarized in Rosemeyer (to appear a), one interesting analytical result from this procedure was that there is a significant increase in the degree of orality of theater plays over time, reflecting a gradual weakening of the norms of written BP.

Having described the operationalization of the dependent and predictor variables, we can now go on to discuss the results from the logistic regression model, calculated in R (R Development Core Team 2019), which predicted the likelihood for an InSituWh token to be used as a rhetorical question in contexts in which the proposition is discourse-new dependent on the predictor variables listed in Table 5. With a c index of concordance of 0.87, the model explains about 83 percent of the variation in the data regarding the dependent variable. The results from the regression analysis are summarized in Table 6.

The regression analysis evinces correlations between the dependent and the predictor variables, most of which reach statistical significance. The strongest of these effects is the variable NEXTMOVEBYUTTERER. Figure 4 visualizes this effect. In comparison to contexts in which the speaker stops her turn after uttering the interrogative (level ‘None’),4 the likelihood for the interrogative to receive a rhetorical interpretation is highest in contexts in which the speaker goes on to deliver an answer, whether using an assertive or questioning intonational contour (‘Answer’ and ‘Answer-question’), followed by explanations of her motive to utter the interrogative (‘Explanation’). Rephrasing the question after uttering the interrogative (level ‘Question’) does not lead to a significantly higher likelihood for the interrogative to receive a rhetorical interpretation.

The analysis also found significant effects for the predictor variables VERBCLASS, SIMPLEWH and LOGORALITY, as well as a marginally significant effect for ADDRESS. As predicted, InSituWh is more likely to be used in rhetorical function with a discourse-new proposition with psychological verbs (the most frequent of which were querer ‘want’, and pensar ‘think’), as well as with simple as opposed to complex wh-elements (cf., e.g., quem ‘who’ vs. de onde ‘from where’). In addition, InSituWh is significantly less likely to be used in texts that scored high on the variable LOGORALITY, thus representing less formal writing styles that represent spoken discourse more accurately. Although with a p value of .0578 this effect does not reach statistical significance, it might be that InSituWh is less likely to be used in rhetorical function with a discourse-new proposition when there is an address term present. The analysis did not find a significant effect for the predictor variable CONNECTOR, measuring the use of the discourse-connecting conjunction e ‘and’ and adverb então ‘then’.

By uncovering the correlations between the rhetorical interpretation of InSituWh and contextual parameters it is finally possible to verify the findings regarding the change from non-rhetorical to rhetorical meanings on the basis of these objective quantifiable contextual criteria. Figure 5 plots the diachronic changes in the distribution of InSituWh with respect to NEXTMOVEBYUTTERER.

Surprisingly, Figure 5 demonstrates an increase in the frequency of InSituWh tokens in contexts in which the speaker does not give an answer to her or his own interrogative, from 25 percent in 1925–1949 to almost 40 percent in the 21st century. This result, which a logistic regression analysis shows to be statistically significant,5 appears to contradict the finding of an increase of rhetorical readings of InSituWh because not an increase, but a decrease of the use of InSituWh in contexts with high conditional relevance would be expected. However, a closer look at the data demonstrates that the two historical trends are not incompatible. Consider Figure 6, a heat plot of the diachronic relationship between discourse function and the next move by the speaker. The colors of the tiles in Figure 6 indicate which context-meaning association is strongest in each time period. For instance, in 1950–1974, elaboration uses of InSituWh are most frequently followed by another question.

Figure 6 demonstrates that between 1925 and 1974, the rhetorical readings of InSituWh typically arise in contexts in which the speaker follows up on her or his interrogative with a “questioning answer” (as in the examples (16) and (17)) or an explanation of her or his motivation to utter the interrogative. However, especially in the 21st century texts InSituWh also frequently acquires a rhetorical reading in contexts which are atypical for rhetorical question interpretations, i.e. contexts in which the speaker does not give an answer to the interrogative (18) or follows up on the interrogative with a second (rhetorical) question (19).6

The last step in the analysis aimed at corroborating in an inferential statistical analysis that the change from discourse-old to discourse-new propositions documented in Figure 2 does not result from the fact that InSituWh became more frequent in information question functions, but rather reflects the increase in the frequency of InSituWh in rhetorical question functions documented in Figure 3.

A generalized additive regression model (GAM, cf. Wieling et al. 2011) was calculated in which the parameter of activation of the proposition (variable ACTIVATION_P, with the levels ‘False’ and ‘True’) was predicted from the year of publication of the play (numerical predictor variable YEAR). This baseline model (α) was then compared to a second GAM (β), which additionally included the random effect RHETORICAL referring to whether or not the InSituWh token expressed a rhetorical question.7 In doing so, the analysis allowed evaluating to which degree the effect of YEAR on ACTIVATION_P is contingent on the predictor RHETORICAL. The assumption was that if the effect of YEAR disappeared when controlling for RHETORICAL, it could be assumed that RHETORICAL is a better predictor of ACTIVATION_P than YEAR, implying that the observed change in activation is only due to the simultaneous increase in rhetorical uses of InSituWh.

The results from the two regression models are summarized in Figure 7. Both in the baseline model α (left plot) and the controlled model β (right plot) there is a negative correlation between the parameter of activation of the proposition and the year of publication. The baseline model judges this correlation to be stronger in the baseline model than in the controlled model; in the left plot, the probability of activation of P descends from 0.5 to –1.1 (a decrease of 1.6), whereas in the right plot, it merely descends from –0.1 to –0.5 (a decrease of 0.4) when taking into account the whole time span, which would indicate that the change is almost entirely explained by the use of InSituWh in rhetorical function. However, the controlled model also differs from the baseline model in that it displays a more informative curve; according to the controlled model, the decrease in the probability that the proposition is activated is restricted to the time period between 1950 and 2016. In this time period, the trend is also relatively strong in the controlled model; the probability that the proposition is activated descends from about 0.8 in 1950 to –0.5 in 2016 (a decrease of 1.3).

Although due to the limited amount of data, the analysis is not very reliable (note the huge error bars indicated by the dotted lines in Figure 7), it does suggest that the decrease in the probability for the proposition of InSituWh tokens to be activated is at least partially due to the fact that InSituWh became more frequent in rhetorical question uses.