From (semi-)oppositional to non-oppositional middles: the case of Spanish reír(se)

4.1 Conversive joyful verbs

Synchronic data suggest that subject-demoting and subject-deleting diatheses with human subjects are the oldest to have been marked with the mm, since they are almost systematically marked nowadays (de Benito Moreno 2015, 2022). This is particularly true for conversive verbs, such as alegrarse ‘to be happy, pleased’ and regocijarse ‘to rejoice’.^[17]

In the conversive diathesis, the object of the active diathesis is promoted to subjecthood, while the subject of the active diathesis is demoted to either an indirect object or a high rank oblique object (Kulikov 2011). As Kulikov (2011) notes, verbs in the conversive diathesis are often perception and emotion verbs, which take an experiencer and a stimulus. This is true for most emotion verbs in Spanish. In the active alignment, these verbs oscillate between an accusative-marked or a dative-marked object, depending on the agentivity of the sentence (Fernández-Ordóñez 1999). In the conversive pattern, the stimulus subject is demoted to a prepositional object (the preposition varies from verb to verb), while the experiencer object is promoted to the subject position, as can be seen in the changes in syntactic alignment from (14a) to (14b).

The semantic connection between alegrar(se) and reír(se) is evident from the syntagmatic associations illustrated in (15),^[18] where both verbs are coordinated, mostly for stylistic purposes.

The available historical data confirm that middle marking of alegrarse is quite old. In the whole corpus, with texts dating from 1250 to 1975, we find 1,295 unambiguous occurrences of alegrar in the conversive diathesis, of which 1,289 are middle-marked. That is, only 6 cases (0.5 %) are unmarked. Figure 3 depicts the frequency of the mm with this verb over 33 22-year periods from 1250 to 1975, showing that the few instances where the mm is not present appear before the 16th century.^[19] Because the amount of text in each period is rather variable, the total tokens in each period are also given in Figure 3.^[20]

Because there are only 6 examples of alegrar(se) with no mm, it is difficult to find patterns that can explain this absence.^[21] Three of these examples are non-finite forms, two of them with an impersonal subject (as in 16a), while two more are finite forms coordinated with the unmarked antipassive verb gozar ‘to enjoy’ (as in 16b).

Figure 3:

Frequency of the mm in intransitive alegrar(se) over time.

It is worth noting that alegrar(se) might show middle marking even in causative contexts (which were excluded from Figure 3). Only 4 occurrences were found in this context, one of which shows the mm (compare 17a–b). There is a single example of alegrar(se) as the object of a perception verb, which is also marked (see 17c).

Finally, it is worth noting that alegrar(se) combines with a variety of prepositional objects that encode the stimulus, which also might be omitted. Alegrar(se) also allows for the presence of a dative, which is not to be interpreted as the stimulus, however. On the contrary, it is a possessive dative that appears when the subject is a body part (such as the heart or soul) – the dative encodes the human possessor of the element that experiences the emotion of happiness, as in (18). Figure 4 shows the frequency of these different contexts. As can be seen, the omission of the stimulus is the most common situation – more than half of the time, there is no prepositional object at all. The most common prepositional objects take the preposition de ‘of’, but objects with prepositions con ‘with’, en ‘in’, por ‘by’ and possessive datives are also possible and common.

Figure 4:

Objects taken by intransitive alegrar(se).

The verb regocijar(se) ‘to rejoice’ has a very similar meaning to alegrar(se) and also participates in the conversive diathesis. This is a more modern verb, whose first attested examples date from the early 16th century. Although less frequent than alegrar(se) (the corpus only provides us with 108 valid examples of the conversive diathesis, ranging from 1525 to 1995), it is clear that its behavior is very similar to this verb, since it is consistently middle-marked (as alegrarse is from the 16th century onwards), as shown in Figure 5. The single unmarked example appears in a causative construction (see 19).^[22]

Figure 5:

Frequency of the mm in intransitive regocijar(se) over time.

Syntactically, regocijar(se) allows for the same possibilities as alegrar(se). However, the frequency of these contexts differs from the latter. As Figure 6 shows, regocijar(se) appears much more often with no objects than alegrar(se). When it takes an object, it is typically introduced by the same prepositions that appear with alegrar(se) (con, en, de and por), although the most frequent one is con, rather than de.

Figure 6:

Objects taken by intransitive regocijar(se).

In sum, historical data show that conversive verbs that encode the experiencing of a joyful emotion have virtually always been middle-marked since the first Spanish texts and that they combine with a variety of prepositional objects, preferring prepositions de ‘of’, con ‘with’, en ‘in’ and por ‘by’. With these verbs, middle marking is associated with a valency-reducing diathesis from the earliest texts onwards. Since the most frequent syntactic contexts in which these verbs appear involve the complete absence of an object or a prepositional object with prepositions de and con, we expect these to be the contexts where reír(se) is most frequently marked. Moreover, since alegrar(se) is the more frequent of these two verbs, we expect it to be the main source of analogy, which predicts that both having no object or having a prepositional object with de should favor middle marking in reír(se) more than having a prepositional object with con.

4.2 Antipassive mocking verbs

Object-demoting and object-deleting diatheses are more irregularly middle-marked in modern Spanish than subject-demoting and deleting diatheses (de Benito Moreno 2015, 2022). In the antipassive diathesis, where the direct object is demoted to a prepositional object (which can be, of course, omitted, see 20b), middle marking is highly dependent on the lexical item. Some Spanish verbs that are semantically similar to reír(se) show or have shown at some point in history an active-antipassive alternation. This is the case of burlar(se), mofar(se) or escarnecer(se), all of which broadly mean ‘to mock, to ridicule, to make fun of’.^[23] The most frequent one of these verbs is burlar(se), from which the CDH offers 1,159 intransitive examples.^[24] Of these, 444 instances are not middle-marked (38.3 %), while 715 are (61.7 %). Escarnecer(se) only provides 73 intransitive examples, of which just one is middle-marked (1.4 %).^[25] There are 53 unequivocal occurrences of intransitive mofar(se) in the corpus, of which 24 (45.3 %) are middle-marked.^[26]

Why the differences between these three verbs? Figure 7 helps us understand these asymmetries better, since these verbs are not equally frequent over time.^[27] While intransitive burlar(se) is found throughout most of the period under study, from the 14th century until nowadays (it is first attested in 1386), intransitive mofar(se) appears for the first time about half a century later, in 1438. Intransitive escarnecer(se), on the other hand, is already found in the 13th century (in 1260) and is no longer used from the 17th century onwards (last attestation in 1596). Middle marking is only sporadically found before the 16th century, at the end of which its frequency of use exploded. It became the rule around the middle of the 17th century, which explains why it was never dominant or even frequent with escarnecer(se).

Figure 7:

Frequency of the mm in intransitive mocking verbs burlar(se), escarnecer(se) and mofar(se) over time.

Despite the far lower number of total cases that we have for mofar(se), it seems clear that it follows the same path as burlar(se). It must be noted that their history as intransitive verbs in the antipassive diathesis is not necessarily parallel to their history as transitive active verbs. On the one hand, transitive burlar spans the same period as intransitive burlar(se), from the late 14th century (the first instance of transitive burlar(se) in the corpus is documented in 1400) until the end of the period under study (the last example in the corpus comes from 1971).^[28] Transitive mofar, on the other hand, is attested for the first time in 1438 in the corpus (the same date as intransitive mofar(se)), but is no longer found from the beginning of the 19th century onwards (the last example of the corpus dates from 1807). Finally, transitive escarnecer is documented for the first time in 1260 (like its intransitive counterpart), but never disappeared, its last documentation in the corpus dating from 1989. Interestingly, this means that mofarse is a non-oppositional middle verb for modern Spanish speakers, that escarnecer shows no diatheses other than the active, and that burlar(se) is an oppositional antipassive verb.

Only burlar(se) was found in causative constructions (n = 1) or as the object of perception and permission verbs (n = 2). While none of the latter show middle marking (see 21a), the former does (see 21b). This situation might have to do with the fact that the examples with perception and permission verbs were documented in the 16th and early 18th century, while the one in the causative construction was attested in 1975. (21a) shows a nice contrast between both reír(se) and burlar(se) in two different contexts – unmarked as objects of permission verbs with no object of their own and marked as fully inflected verbs with a prepositional object with de.

It is worth noting that the prepositional objects that these antipassive mocking verbs might take also present the prepositions that were the most frequent with conversive alegrar(se) and regocijar(se), that is, de, con, en and por (see Figure 8).^[29] They also allow for dative objects, which, in this case, do indeed encode the source or stimulus of the mocking event. However, there are two important differences between the conversive verbs examined in Section 4.1 and these antipassive verbs. First, the latter show far less variation regarding prepositions than the former, since they largely prefer one single preposition, namely, de ‘of’. Second, antipassive verbs tend to appear with a prepositional object, as opposed to conversive verbs, where the object is most often omitted^[30] – note that this difference is far smaller for escarnecer, where both contexts are almost as equally frequent.

Figure 8:

Objects taken by intransitive mocking verbs burlar(se), escarnecer(se) and mofar(se).

In order to examine the effect of different linguistic parameters (such as the verb lexeme, having an overt prepositional object with de or the inflection of the verb) on the presence of the mm over time, I have fitted a generalized linear fixed model on the data, which is summarized in Table 3.^[31] The dependent variable of the model is the presence/absence of the mm (the reference level is the absence of the mm, which means that the model predicts its presence). There are four independent variables:^[32] 1) the verb (burlar – reference level – and mofar, with escarnecer omitted since it shows very little variation and the model predicts its behavior during the whole time period, which makes little sense); 2) the verb form (finite – reference level – and non-finite); 3) the type of object (prepositional object – reference level, I have conflated all prepositional objects into a single level – and no object); and 4) year (which has been centered, so that the reference level is the mean year, that is, 1666). Since the goal of the model is to test the hypothesis that all these factors play a relevant role in the presence of the mm, I have fitted the maximal model. I have also added the interactions between all three independent categorical variables and year, in order to see whether the relationship between the levels of these factors is stable over time. Although the author of each occurrence might be a source of noise, it has not been included as a random factor, since it has too many levels (n = 197), which prevents the model from converging.^[33]

As shown in Table 3, all individual factors have a significant effect, although the size of this effect differs from factor to factor. The largest effect is found for the verb lexeme – as compared to burlar, mofar disfavors the presence of the mm. Non-finite forms of the verb also disfavor middle marking, when compared with finite forms – this effect is about 1.6 times smaller. Similarly, having no objects (as opposed to having a prepositional object) also disfavors the presence of the mm, an effect that is about 2.7 times smaller than the effect of verb lexeme. We should recall that all these estimates refer to the year 1666 as the reference level. The only factor that has a positive effect on the presence of the mm is time. Although this effect might seem small, we should bear in mind that the relevant scale is the year. That is, the estimate must be interpreted in the following way: every year, the probability of the presence of the mm increases by 0.022. Over the almost six centuries considered in this sample (589 years), the cumulative is of course the greatest.

None of the interactions (between year and the other individual factors) are found to be significant. Interpreting these results is easier when looking at Figures 9 and 10. Figure 9 shows that the presence of the mm increases steadily with both verbs over time, following similar paths – they behave similarly over time, which is why the interaction among these factors is not significant. However, as we can see, burlar(se) is always in the lead, often by a substantial degree – this is why the verb lexeme has a significant effect on the presence of the mm (recall that the reference level for year is 1666).

Figure 9:

Interaction between mocking verb and time.

Figure 10:

Interaction between form of the verb form and time.

Similarly, Figure 10 shows that the presence of the mm with finite and non-finite forms increases over time following similar paths – the interaction with time is not significant. However, finite forms consistently favor middle marking over time – the verb form is found by the model to be a significant factor. The distance between both curves is smaller than in Figure 9, because the size of this effect is also smaller.

Figure 11 shows the interaction between time and type of object, where both curves follow different paths. This means that the effect of having a prepositional object changes over time – while it disfavors the presence of the mm at the beginning of the period considered, this changes during the 15th and 16th century, when having a prepositional object starts to favor the presence of the mm. However, this effect is not significant, as can be seen by the fact that the confidence intervals overlap or are rather close.

Figure 11:

Interaction between type of object form and time.

To sum up, the data confirm that middle marking of antipassive mocking verbs is more recent than middle marking of conversive joyful verbs. The available data show that the behavior of these verbs diametrically changed over time, from never being marked to being marked systematically. In the case of mocking verbs, systematic middle marking was consolidated during the 18th century, which explains why escarnecer(se) was almost never middle-marked – its intransitive variant disappeared earlier. Mofar(se), on the other hand, lost its transitive counterpart, becoming a non-oppositional verb.^[34] The generalized mixed model confirms that the degree of the association between these verbs and the mm changed over time and shows that all the factors considered (verb lexeme, verb form and the presence/absence of a prepositional object) have a significant effect – the verb mofar, non-finite forms, and the absence of an object all disfavor the presence of the mm (when compared with the verb burlar, finite forms, and the presence of an objects, respectively). The effect of the latter is the only one that changed over time. If these verbs acted as a source of analogy for reír(se), we expect a similar pattern, that is, that non-finite forms and the absence of a prepositional object disfavor middle marking. Moreover, because prepositional objects with de are the ones that most frequently appear with antipassive mocking verbs, we expect them to favor the presence of mm.

4.3 Reír(se) (and other intransitive laughing verbs)

As compared to oppositional verbs associated with a change of valency, middle-marked intransitive verbs are the ones that take the mm less systematically in modern Spanish (de Benito Moreno 2015, 2022). This is consistent with these verbs being a late development, as shown in Bogard (2006).^[35] Reír(se) ‘to laugh’ is an intransitive verb that takes the mm optionally, with no change of valency and no clear change of meaning. In fact, some authors have noted that semi-oppositional middle verbs are either diastratically or diaphasically (i.e. stylistically) marked (Cartagena 1972; Klein 1987) or that they encode some kind of expressive, affective or intensive nuance (Gili y Gaya 1961; Klein 1987; Lázaro Carreter 1964; Seco 1962). Note that these proposals are complementary, in that expressiveness and affectivity are associated with the language of immediacy rather than with the language of distance (see Koch and Oesterreicher 2007 for these concepts).

The mm, however, is compulsory in some contexts with reír(se). Martín Zorraquino (1979) uses the contrast in (22) to illustrate that the mm is obligatory when the verb takes the same meaning as burlar(se) ‘to mock, to make fun of, to laugh at’. The parallelism is rather interesting, because in these cases, reír(se) takes a prepositional object with de ‘of, from’, as does burlar(se) (see Section 4.2).

Synchronic data suggest that, while it is true that the mm is optional in contexts like (22a), the norm (in Coseriu’s 1973 [1952] sense) is for it to be marked, at least in speech. A study of 151 semi-directed interviews (ca. 200 h) from the corpus COSER with elderly, uneducated and non-mobile rural speakers from different regions of Peninsular Spanish has shown that reír(se) is almost always marked, even if there is no prepositional object with de (69 marked examples of 73, that is, 94.5 %) (de Benito Moreno 2015, 2022).^[36] Data from a questionnaire with video stimuli conducted in 43 different rural towns in Spain with speakers of a similar profile (elderly, uneducated and non-mobile) confirm these findings, showing that reír(se) is far more frequent with the mm, also when there is no prepositional object with de (118/140, 90.8 %) (de Benito Moreno 2015).^[37] When the prepositional object with de is present, the verb is always marked, in both corpora (COSER: n = 10; questionnaire: n = 16).

In the Spanish cities considered in PRESEEA, the situation is rather similar. From a total of 71 examples with no prepositional object with de, 61 (85.9 %) are middle-marked; again, the mm systematically appears if there is a prepositional object with de (n = 10).^[38] A qualitative analysis of these examples suggests that the absence of the mm is often induced by the presence of a mm in the periphrasis where reír(se) is found. Six out of ten non-middle-marked examples appear in reflexive periphrases and all examples that appear in reflexive periphrases are unmarked, except for (23c), which has a prepositional object with de.^[39] The PRESEEA data, however, are too scarce and the sample is too unbalanced to run a statistical model to check for this association.

In order to analyze the historical evolution of the mm with reír(se), data from a number of corpora have been collected (see Section 3). After manually analyzing the 3,212 occurrences obtained and excluding those that did not contain the verb reír(se),^[40] as well as idioms, repeated examples and transitive uses of the verb, a total of 2,400 valid intransitives were analyzed. Note that, although reír(se) shows some transitive uses, its intransitive uses are not to be understood as a derived diathesis – on the contrary, reír(se) is an intransitive verb that accepts so-called internal accusatives (see 24). As expected, these are rather rare in quantitative terms (only 72 were found).

Of the 2,400 intransitive examples, 55 appeared within a periphrasis that is optionally middle-marked and where the mm appeared only once and was attached to the finite verb (see 25). As noted in Section 3, these examples are intrinsically ambiguous, since Spanish allows clitic climbing to the finite verb of a periphrasis – there is no way of telling to which of the two verbs the mm is attached. Thus, these examples have also been excluded from the remainder of the analysis.

Instances of reír(se) within a causative construction were never marked (n = 188) in the corpus (see 26). Within control structures with perception or permission verbs, reír(se) was almost never marked – there are only 2 marked examples out of 29 (7 %) in the corpus (see 27).^[41] Finally, no middle-marked examples were found within an obligatorily middle-marked periphrasis (n = 13) (see 28). Consequently, I have also excluded these examples from the remainder of the analysis. This leaves us with 2,115 examples of intransitive reír(se).

Figure 12 shows the frequency of the mm with reír(se) over time.^[42] One first conclusion that can be drawn from this figure is that reír(se) already shows middle-marked examples in the 13th century, that is, earlier than the antipassive verbs analyzed in Section 4.2. This means that those verbs could not have been a model for the analogical emergence of middle marking in reír(se) – only joyful conversive verbs could have played such a role.

Figure 12:

Frequency of the mm in intransitive reír(se) over time.

The first thing to catch one’s eye in Figure 12, however, is the rather chaotic picture it draws, especially when compared to the verbs analyzed above. Of course, in some periods there are very few total cases, producing extreme results, but no clear tendency can be observed in the data. It is worth noting that, even during the two most modern periods, which include the dates when the spoken data reported above were collected, the global frequencies of the mm are far more moderate (between 50 and 65 %) than the ones found in these spoken corpora. This supports the aforementioned observations that the mm with a verb such as reír(se) might be associated with affective and expressive nuances and be affected by diaphasic factors, the mm being more frequent in the language of immediacy. However, this also means that understanding its historical evolution is far harder, since we only have access to written data from older stages, which mostly comprise samples of the so-called language of distance (in Koch and Oesterreicher’s 2007 sense).

Figure 13 shows the global proportions of the mm depending on whether reír(se) takes a prepositional object and what preposition such an object takes.^[43] Because of the extreme differences in sample size between the CDH and the other corpora, the figure also takes the corpus into account. As can be seen, the most frequent case is for reír(se) not to take an object (like joyful conversive verbs) – in these cases, middle marking is about as frequent (CDH: 248/557, 44.5 %; other corpora: 457/918, 49.8 %) as its absence (CDH: 309/557, 55.5 %; other corpora: 461/918, 50.2 %). When reír(se) takes a prepositional object, it is most often introduced by the preposition de (like alegrar(se) and antipassive mocking verbs) and in the overwhelming majority of the cases the verb is middle-marked (CDH: 250/278, 88.9 %; other corpora: 280/283, 98.9 %). Objects introduced by con are both less frequent overall and less frequently middle-marked. The frequency of marking differs substantially from one dataset to the other (CDH: 9/23, 39.1 %; other corpora: 29/33, 87.9 %), but note the low overall figures. All other possibilities are attested less than 5 times, which precludes us from reaching any conclusions.^[44]

Figure 13:

Proportion of the mm by object taken by reír(se).

In order to shed some light on the interplay between the different linguistic factors considered (namely, the type of object, the characteristics of the subject, and the verb form) and time in the presence of the mm, I have fitted a generalized linear fixed model to the data, which is summarized in Table 4. Again, the dependent variable of the model is the presence/absence of the mm (the reference level is the absence of the mm, which means that the model predicts its presence). There are four independent variables:^[45] 1) the type of subject (animate – reference level –, inanimate and impersonal); 2) the verb form (finite – reference level – and non finite); 3) the type of object (no object – reference level –, prepositional object with de and prepositional object with con);^[46] and 4) year (which has been centered, so that the reference level is the mean year, that is, 1887). Since the goal of the model is to test the hypothesis that all these factors play a relevant role in the presence of the mm, I have fitted the maximal model. I have also added the interactions between all three independent categorical variables and year, in order to see whether the relationship between the levels of these factors is stable over time. Although the author of each occurrence might be a source of noise, it has not been included as a random factor, since it has too many levels (n = 197), which prevents the model from converging.^[47]

The model shows some interesting tendencies. First, time on its own does not have a significant effect on the presence of the mm. This contrasts with the situation of antipassive mocking verbs, where time was an extremely relevant factor, but is consistent with Figure 12, where no clear pattern arises over time. The largest significant effect is the positive association between prepositional objects with de and middle marking, as compared to the absence of an object. However, the positive association between prepositional objects with con and the presence of the mm does not reach the established level of significance. Regarding verbal form, non-finite forms (as compared to finite forms) disfavor the presence of the mm, but the effect size is again small. Inanimate subjects (as compared to animate subjects) also disfavor the presence of the mm – this is the second largest effect identified by the model. The negative association between impersonal subjects (as compared to animate subjects) and middle marking does not reach the established significance level. We might recall that all these estimates refer to year 1887 as the reference level. Again, towards a better interpretation of the results of the interactions, I provide and explain the relevant visual representations below.

Figure 14 shows the interaction between type of object and time, which is the only one that is found to be significant in our model. This has to do with the fact that the direction of their effect changes over time when compared to the effect of the reference level (which is stable over time). That is, both prepositional objects with de and with con disfavor middle marking at the beginning of the time period under study when compared to the absence of object, but this effect is reversed over time (earlier for objects with de than for objects with con). The figure also shows that middle marking, when an object with de is present, has virtually reached full systematicity. Again, recall that this finding contrasts with the data from spoken corpora, suggesting that the presence of the mm is dependent on the communicative situation.

Figure 14:

Interaction between type of object of reír(se) and time period (based on model 1).

Figure 15 depicts the interaction between type of subject and time, showing that the association between reír(se) and the mm: 1) remains stable over time when the subject is animate; 2) decreases over time when the subject is inanimate; and 3) increases over time when the subject is impersonal. However, none of these tendencies is significant, as shown in Table 4 above, but also in the wide confidence intervals of the curves. If a larger sample could demonstrate these tendencies, such findings would be very relevant, since impersonal subjects are actually animate (and human) – this would mean that reír(se) with impersonal human subjects has become more “regular” over time, assimilating to its behavior with any other animate subject. In Section 3, I argued that the absence of the mm is less expected with impersonal subjects because the mm is originally a reflexive pronoun – these findings would suggest that the mm gradually departs from that status over time, becoming an element that belongs with the verb instead of an element that refers to the subject. The decrease of frequency with inanimate subjects might suggest that the presence of the mm has become associated with meanings exclusive to animate subjects, such as agentivity. The role of agentivity has been mentioned for some semi-oppositional intransitives of Spanish (see Martín Zorraquino 1979: 299; Sánchez López 2002: 121), even for reír(se) – according to Fernández Ramírez (1986), middle-marked reírse encodes a more intentional act than unmarked reír. However, these hypotheses must still remain as such until more robust data can either confirm them or reject them.

Figure 15:

Interaction between type of subject of reír(se) and time period (based on model 2).

Figure 16 plots the interaction between verbal form and time, showing that: 1) the association between reír(se) and the mm remains virtually stable over time regardless of the verbal form; and 2) finite forms favor the presence of the mm (as opposed to non-finite forms). As was claimed above, the reason for this pattern might be that non-finite forms have no formal person agreement, which might disfavor the presence of a particle that compulsorily shows subject agreement, as the mm does.

Figure 16:

Interaction between the verbal form of reír(se) and time period (based on model 3).

In sum, the historical data of reír(se) show that the presence of the mm with this verb is affected by the same factors that favor the presence of the mm with antipassive verbs (non-finite forms, and the presence of a prepositional object, specifically with de). Moreover, contexts that are atypical of both conversive joyful verbs and mocking antipassive verbs (such as inanimate subjects) disfavor the presence of the mm with reír(se). Although antipassive mocking verbs cannot have been the original source of analogy, as shown by the fact that reír(se) shows middle marking before them, the fact that they are affected by similar factors suggests that there has been an effect of analogy. But in which direction? In my opinion, the fact that reír(se) shows a more complex scenario than antipassive mocking verbs suggests that the former has also been influenced by the latter. As shown in Section 4.3, all variants showed an increasingly positive association with the mm for antipassive mocking verbs, and although some variants led the way, the result was the regularization of middle marking in all contexts. This is not the case for reír(se), where the variants that led the way with antipassive mocking verbs show a stronger association with the mm, whereas not all variants show such an increase. Actually, systematicity has only been reached when a prepositional object with de is found, which are the contexts where the meaning of reír(se) is closer to that of mocking verbs. Middle marking with reír(se), then, shows a semantic specialization that is not found in the other verbs analyzed here, which is a strong piece of evidence that argues for analogy as the mechanism behind these changes. However, such an analogy is more complex than what it might appear and was probably bi-directional – the emergence of middle marking in antipassive mocking verbs was most likely influenced both by conversive joyful verbs and by reír(se).

Crucially, differences were found between written and spoken data, which shows that the frequency of the mm is higher in speech in all linguistic contexts. This offers support to the observations of some authors that the mm with some semi-oppositional verbs encode affective, expressive and intensive meanings and is more common in some registers. Checking this hypothesis with our data is not easy, since most of our historical data belong to the so-called language of distance, although a more detailed study that takes into account genres and discourse traditions might shed some light on this in the future. There is, however, some indirect evidence that supports this hypothesis. Spanish has a number of non-oppositional middle-marked verbs meaning ‘to laugh’, such as desternillarse and descojonarse, and ‘to mock’, such as pitorrearse, guasearse and cachondearse. All of these verbs encode a more intense laughing event than reír(se) and at least some of them (such as descojonarse, pitorrearse and cachondearse) are associated with the language of immediacy. Interestingly, these verbs emerged in the language as non-oppositional, showing a strong association between the mm and these more intense laughing contexts.

The behavior of a verb such as carcajear(se) ‘to roar with laughter’ is also illustrative of this pattern. Again, this verb encodes a more intense laughing event than reír(se) (it is derived from the word carcajada, roughly meaning ‘guffaw’). Carcajear(se) is, as opposed to the other verbs just mentioned, a semi-oppositional verb, and allows for the presence or absence of the mm with no change of valency or meaning. This verb is not only used in the language of immediacy, meaning that it only differs from reír(se) in the intensive nuance. Although far less frequent (and modern) than reír(se), corpus data offer some interesting information here. In the CDH documents (up until 1975), only 11 instances of the verb are found, one in the 16th century, one in the 17th century and 9 in the 20th century. Of these 11 examples, the oldest 3 (documented in 1533, 1620 and 1932) are unmarked (and appear with no prepositional object), while the 8 examples found in the third quarter of the 20th century are middle-marked (only one takes a prepositional object, with por). In the CREA (Versión Anotada) documents (1975–2000), 32 examples are found, of which 27 (84 %) show the mm. The proportion is basically the same when only contexts with no prepositional object are considered (20/24, 83 %).^[48] In the CORPES XXI documents (ranging from 2001 to 2020), 100 out of 106 (94 %) examples are middle-marked. Again, the proportion is virtually the same when no prepositional object appears (79/85, 93 %).^[49] Even examples in causative constructions or within reflexive periphrases might be middle-marked (see 29). That is, the added intensity encoded by carcajear(se) seems to be responsible for a higher frequency of the mm in the language of distance, frequencies that are rather similar to the ones found for reír(se) in speech. All in all, this explains the behavior of reír(se), which seems to be virtually non-oppositional in speech, where affective, intensive and expressive forms are more common, while alternating more frequently between the unmarked and the middle-marked form in the language of distance.^[50]