2.1. Postmodality: From modal harmony to ‘concomitant of subordination’

The grammaticalization approach represents a fundamental framework in the analysis and modeling of linguistic change. It was Bybee, Perkins & Pagliuca (Reference Bybee, Perkins and Pagliuca1994) who first applied the framework to developments in the domain of modality, including the relevant morpho-syntactic category of mood (see in particular their Chapter 6.9). Their seminal work on grammaticalization has served as a landmark for other approaches that share this general outlook on language change but that enrich the general grammaticalization perspective by suggesting more specific mechanisms of change, especially with respect to the evolution of mood systems in the languages of the world and especially in the realm of complement clauses.

Bybee et al. are first and foremost concerned with diachronic developments and, therefore, do not engage in a theoretical debate on what the semantics of mood might be from a synchronic point of view. Although they do not deny the existence of meaning-bearing mood contrasts, they frame the semantic question from the broader perspective of grammaticalization. In their view, an all-encompassing semantic value inherent in mood grams does not exist, and, therefore, they conclude that a gram ‘might be meaningful in one context but not in another’ (Bybee et al. Reference Bybee, Perkins and Pagliuca1994: 213). Interestingly, Bybee et al. fall back on the label ‘modal harmony’ when describing the role of mood markers in complement clauses in Latin and other languages of the world. This notion is taken up especially by the Latinist scholar Magni, who applies it systematically to the role of (the subjunctive) mood in complement clauses in the classical Latin system (see Magni Reference Magni, Baldi and Cuzzolin2010: 245–246, in line with Bybee et al. Reference Bybee, Perkins and Pagliuca1994: 219–220).

The notion of ‘modal harmony’ (or ‘modally harmonic’) goes back to Lyons (Reference Lyons1977) and was elaborated in Coates’ (Reference Coates1983) study on ‘should’ in which it denotes the correlation of a modal (in this context, ‘should’) with a co-occurring modal item that expresses the same degree of modality, i.e. does not add any further modal nuances to the modal meaning (see Coates Reference Coates1983: 69, cited in Bybee et al. Reference Bybee, Perkins and Pagliuca1994: 214). An illustrative case in point is example (1):

In this example, the modal ‘should’ combines with the matrix verb ‘expect’ and, apparently, does not make any further semantic contribution – it is semantically ‘void’ but ‘harmonizes’ with the modal features (e.g. future–orientation, preference–semantics; see below) of the main verb.

Interestingly, Magni (Reference Magni, Baldi and Cuzzolin2010), following Bybee et al.’s considerations on the uses of the Latin subjunctive (see synopsis in Bybee et al. Reference Bybee, Perkins and Pagliuca1994: 223), uses the concept as a key term for her own analysis, applying it both to the description of the underlying base structure of the Latin mood system and to the modeling of its diachronic dynamics in the course of developments. If, in the original Latin system, the uses of the subjunctive in the complement clause ‘echo’ its occurrences in main clause contexts (and are therefore ‘harmonic’), the subsequent evolutions allegedly make the system move away from this original ‘balance’ by admitting non-harmonic uses in different contexts of use. This leads to a reanalysis of the subjunctive mood that is characterized in the following way (Bybee et al. Reference Bybee, Perkins and Pagliuca1994: 219, taken up by Magni Reference Magni, Baldi and Cuzzolin2010: 267 and synthesized as a general conclusion on p. 268):

This reanalysis, linked to a generalization of the subjunctive in complement clauses, is tantamount to the loss of its specific meaning or, in other words, to its ‘semantic bleaching’ according to the terminology of the grammaticalization theory. But caution is in order when it comes to such far-reaching generalizations.

Magni (Reference Magni, Baldi and Cuzzolin2010), like Bybee et al. (Reference Bybee, Perkins and Pagliuca1994: 222, 223), takes the developments in indirect interrogative sentences (if- and wh-complement clauses) as evidence for their hypothesis. In particular, they cite the cases of ‘dubito an’ (‘I doubt whether’) or ‘nescio an’ (‘I do not know if’), in which the subjunctive becomes the rule. Although the analysis of these contexts is beyond the scope of this paper, it must be stressed that in a modal–semantic approach, the occurrence of the subjunctive can be perfectly accounted for: from the perspective of modal semantics, indirect questions present alternatives of how the state of affairs referred to by the underlying propositions might be – e.g. in a sentence like ‘I doubt whether Peter arrived yesterday’, alternatives can be distinguished in which Peter arrived and others in which this is not the case. The fact that alternatives are evoked strongly motivates the selection of the subjunctive mood (see below) and can be interpreted as an instance of a coherent mood expansion to new contexts within the complement clause domain (on mood expansion, see especially Sections 4.2, 4.3). Another problem with Magni’s approach is the fact that she does not take into account further important developments in the complement clause domain, especially those related to evaluative and doxastic predicates whose selectional patterns change profoundly, as we will see in Section 4.3. All in all, we can profit from Magni’s excellent description of the mood system in Latin, without, however, agreeing with her analysis of the unfolding diachronic dynamics.

2.2. Regrammation

Jan Lindschouw avails himself in his study on evolutionary trends in the domain of mood in French and beyond of Andersen’s analytical framework on ‘regrammation’ (Andersen Reference Andersen2006). According to Lindschouw (Reference Lindschouw and Arteaga2011a: 125) and in line with Andersen (Reference Andersen2006), ‘regrammation’ can be conceived ‘as an extended version of grammaticalization’. More specifically, it can be defined as a process that ‘reorganizes grammatical systems into new grammatical systems provided that the relationship between the form and the meaning of the grammatical item in question undergoes a change within a closed paradigm (…), and that the scope of the item is reduced’ (Lindschouw Reference Lindschouw and Arteaga2011a: 125).

Although Lindschouw focuses, in the first place, on concessive clauses, his analysis is designed to uncover and capture general principles and mechanisms of change underlying the mood systems of Romance languages. Lindschouw ascribes clear-cut semantic values to the indicative and subjunctive moods in Romance based on the fundamental contrast between assertion and non-assertion (as suggested by Hooper Reference Hooper and Kimball1975, Haverkate Reference Haverkate2002, Korzen Reference Korzen, Birkelund, Boysen and Søren Kjærsgaard2003; see also Lindschouw Reference Lindschouw, Fagard, Prévost, Combettes and Bertrand2008, Reference Lindschouw and Arteaga2011a, Reference Lindschouw2011b). In line with this basic distinction, he characterizes the general evolutionary tendency as a ‘lessening’ of the non-assertive potential of the subjunctive during the period from Old French to modern French (Lindschouw Reference Lindschouw and Arteaga2011a: 123). The overarching process of regrammation falls into three mechanisms that are at work (see Lindschouw Reference Lindschouw and Arteaga2011a: 123–124, 146):

• – A process of desemanticization

• – A process of obligatorification

• – A process of specialization

Assuming that ‘the result to a large extent reflect the general evolution of the mood system’ (Lindschouw Reference Lindschouw and Arteaga2011a: 124), Lindschouw takes the sub-system of concessive clauses as a testing ground (Lindschouw Reference Lindschouw and Arteaga2011a, 2011b).

Basically, the author claims that the value of the subjunctive in concessive clauses is reduced from two values (labeled as ‘presupposition’ and ‘irrealis’) to the single value of ‘presupposition’ (see Lindschouw Reference Lindschouw and Arteaga2011a: 146). This process is interpreted as an instance of ‘regrammation’. Conversely, the indicative mood is reorganized, extending its functional domain from the single value of ‘realis’ to three values, adding ‘presupposition’ and ‘irrealis’ to its semantic range.

In view of the fact that the subjunctive becomes obligatory in its contexts of occurrence, the variation with the indicative disappears and the subjunctive is confined to residual and specialized contexts of occurrence (see also Hopper & Closs Traugott [1993] Reference Hopper and Traugott2003: 116–118). Overall, the indicative is gaining ground at the expense of its competing form in the course of developments.

This general evolution manifests itself in two specific developments in the domain of concessive clauses: The concessive conjunctions ‘bien que’ (‘although’) and ‘encore que’ (in pre-position) no longer admit mood variation, the selection of the subjunctive mood becoming obligatory (‘obligatorification’) (e.g. as in ‘Bien qu’/Encore qu’il soit invité à la fête, il n’est pas content’) (see Lindschouw Reference Lindschouw and Arteaga2011a: 142 [Tables 7.7, 7.8]; for a detailed picture of the evolutions, see Lindschouw 2011b: 177–198, 199–224). On the other hand, a new conjunction ‘même si’ (‘even if’), enters the stage (at the end of the sixteenth century) that uniquely selects the indicative, covering exclusively the irrealis domain (‘même s’il pleuvait’, ‘even if it rained’) (Lindschouw Reference Lindschouw and Arteaga2011a: 144–145; 2011b: 230–237). For reasons of space, we will not take a critical look at the details of Lindschouw’s analysis but will focus briefly on the subjunctive-as-presupposition explanation that is also put forward by the author and by other scholars in order to account for the selection of the subjunctive in factive–emotive contexts in Romance (see e.g. Lindschouw’s example ‘Je suis content que tu ailles mieux’ in Lindschouw (Reference Lindschouw, Haßler and Mutet2022: 417) but also Wandruszka’s analysis of the Italian mood system, which analyzes the same type of contexts under the label ‘thematic’/’factive subjunctive’ (see Wandruszka Reference Wandruszka, Renzi and Salvi2001: 472–481).

First of all, being a presupposition is not a grammatical value or function. Linked to lexical items like cognitive (‘to know’, ‘to remember’), factive (‘to regret’), and causative verbs (‘to cause’), it forms part of the semantic structure of the lexical item.

More importantly, the presuppositional feature does not entertain a particular, let alone exclusive, relationship with the selection of the subjunctive mood. Presuppositional verbs like ‘savoir’/’ignorer’, ‘se souvenir’/’oublier’, and ‘entendre’/’voir’ select the indicative and not the subjunctive. On the other hand, the subjunctive is selected in non-presuppositional contexts like:

In this case, the truth of the proposition is not presupposed (as part of the common ground) but the evaluation and/or attribution of a truth value is postponed. So, in the example at hand, the proposition p might be wrong (p.ex. Que Bouzigue soit un ivroge, c’est faux. C’est sa petite amie qui nous a menti). So presuppositional contexts are not inherently associated with the subjunctive nor can these contexts be reduced to a (single) grammatical/functional value. Therefore, the diachronic dynamics that sees concessive conjunctions and factive–emotive predicates increasingly having the subjunctive has to be analyzed in a different way.

2.3. The evolution of mood in a variationist approach

Bybee et al.’s (Reference Bybee, Perkins and Pagliuca1994) thoughts on language change in the domain of modality are also the starting point for Shana Poplack and her co-authors. In their recent publications on the evolution of mood, however, Poplack, Cacoullos, Dion, Berlinck, Digesto, Lacasse & Steuck (Reference Poplack, Cacoullos, Dion, de Andrade Berlinck, Digesto, Lacasse, Steuck, Ayres-Bennet and Carruthers2018: 248) identify the processes of lexical routinization and structural conventionalization as driving forces of evolutional changes that are due to the desemanticization of mood.

The defining property of lexicalization is that ‘the lexical identity of the governor contributes to the subjunctive selection independent of any meaning to be expressed’ (Poplack et al. Reference Poplack, Cacoullos, Dion, de Andrade Berlinck, Digesto, Lacasse, Steuck, Ayres-Bennet and Carruthers2018: 238). If ‘lexical bias’ is considered the major predictor of subjunctive selection, then neither contextual factors nor the semantic properties of governors play a significant role in the mood variant choice (see Poplack et al. Reference Poplack, Cacoullos, Dion, de Andrade Berlinck, Digesto, Lacasse, Steuck, Ayres-Bennet and Carruthers2018: 238).

The phenomenon referred to as conventionalization amounts to the ‘increasing association between structural context and variant, and concomitant reduction or loss of ‘“free” ! (semantically-motivated) choice amongst them.’ (Poplack et al. Reference Poplack, Cacoullos, Dion, de Andrade Berlinck, Digesto, Lacasse, Steuck, Ayres-Bennet and Carruthers2018: 222). This mechanism of change relates mood selection to language-specific factors, such as the role of negation, interrogative sentence modality and the conditional morphology of the matrix verb (‘eu gostaria que’, ‘I would like to’) as well as to the presence of epistemic and evaluative adverbs (maybe, apparently, etc.).

In their study, Poplack et al. highlight three aspects that according to their interpretation come to light in the sample of spoken Romance data:

1. – Governors display either a high or a low level of association with the subjunctive mood (as a crucial part of their lexical profile).

2. – A small number of governors account for a huge proportion of subjunctive forms (e.g. only three governors [‘pensar’, ‘acreditar’, and ‘querer’] account for 50% of the subjunctive occurrences in Portuguese, and only two governors [‘falloir’ and ‘vouloir’] for almost 80% of the subjunctive forms in French (see Poplack et al. Reference Poplack, Cacoullos, Dion, de Andrade Berlinck, Digesto, Lacasse, Steuck, Ayres-Bennet and Carruthers2018: 240 [Figure 6]); for French, see also Siepmann & Bürgel Reference Siepmann and Bürgel2015: 175).

3. – In addition, it turns out that the subjunctive is recruited from a very limited set of embedded verb forms (see Poplack et al. Reference Poplack, Cacoullos, Dion, de Andrade Berlinck, Digesto, Lacasse, Steuck, Ayres-Bennet and Carruthers2018: 243 [Figure 8]).

These data are put forward as evidence for the overall conclusion that ‘in all four (Romance) languages, lexical bias was found to be the major predictor of subjunctive selection’ (Poplack et al. Reference Poplack, Cacoullos, Dion, de Andrade Berlinck, Digesto, Lacasse, Steuck, Ayres-Bennet and Carruthers2018: 248).

We do not deal with Poplack et al.’s ranking of the Romance languages according to ‘grammaticalization measures’, with Latin as the starting point; Spanish as the most conservative; and Portuguese, Italian, and French as the most advanced languages (see Poplack et al. Reference Poplack, Cacoullos, Dion, de Andrade Berlinck, Digesto, Lacasse, Steuck, Ayres-Bennet and Carruthers2018: 244 [Table 10], 246 [Figure 9]), given that a differentiated and critical review of their interpretation of the data would require a separate paper. Some critical aspects, however, have to be mentioned inasmuch as they concern the line of argumentation in this paper.

First, as will become clear throughout this article, comparing governing lexical items and their frequencies in Romance languages without taking into account their modal properties, as well as their belonging to modal domains, is somewhat like comparing apples with oranges. For the sake of illustration, take Poplack et al.’s Figure 6, representing the governors most responsible for mood selection in the Romance languages at hand (Poplack et al. Reference Poplack, Cacoullos, Dion, de Andrade Berlinck, Digesto, Lacasse, Steuck, Ayres-Bennet and Carruthers2018: 240):

Figure 1. From Poplack et al. Reference Poplack, Cacoullos, Dion, de Andrade Berlinck, Digesto, Lacasse, Steuck, Ayres-Bennet and Carruthers2018: 240.

If you depart from the most frequent governors in each language under scrutiny, you get the idea that the subjunctive is associated with belief verbs in Portuguese (‘pensar’, ‘acreditar’) and in Italian (‘pensare’, ‘credere’, ‘sembrare’: ‘mi sembra’) but with the deontic verb ‘falloir’ in French and with the verbs ‘creer’ and ‘decir’ in Spanish. The authors do not point out that, in the cases of Portuguese and Spanish, negation is the triggering factor for the subjunctive (in Spanish always, in Portuguese in most cases), whereas in affirmative contexts the indicative is mandatory (in Spanish) and strongly preferred in Portuguese (see Becker Reference BeckerTo appear). These facts strongly contrast with Standard Italian where belief predicates generally select the subjunctive in all contexts (‘il congiuntivo dubitativo (epistemico)’; see Wandruszka Reference Wandruszka, Renzi and Salvi2001: 434–437). Finally, the seemingly very frequent verb ‘decir’ (in Spanish and elsewhere) is polysemic and only selects the subjunctive when referring to an order (‘dicere’ with a deontic meaning) but not as a verbum dicendi, i.e. in contexts of reported speech. In the remainder of this paper, we will demonstrate the paramount relevance of the semantics of these verbs and, even more crucially, the presence of specific modal features that are responsible for the selection of the subjunctive. Against Poplack et al.’s thesis concerning the crucial role of lexical bias for the selection of the subjunctive mood and, linked to it, the desemanticization of the form in the course of developments, we argue that the subjunctive has a consistent core (abstract) meaning that pervades all contexts of occurrence and that has remained intact until the contemporary language stages in Romance. Although we do not agree with Poplack et al.’s stance on the subjunctive and its evolutionary dynamics, we will refer to their data as they illustrate certain trends we are going to highlight (without sharing their interpretation).

A final caveat as to the term grammaticalization, as used by Poplack et al., has to be made at this point: If, as Poplack et al. suggest, the overall evolution pointed toward an increasing strengthening of lexical biases, the underlying mechanism could hardly be referred to as ‘grammaticalization’. As is known, the outcome of grammaticalization is a new grammatical or a more grammaticalized form that has its place in a determined paradigm (e.g. a verb form like the present subjunctive that is included into the overall verbal paradigm with its different sub-paradigms, e.g. the present tense). In contrast to that, ‘lexicalization’ (and, linked to it, the central role of lexical bias in mood selection) describes the loss of grammatical status due to conventionalization. This is clearly the case of conventionalized governor–mood–combinations or ‘patterns’. In these combinations, the mood selection pattern becomes necessary part of the lexical item and its entry in the mental lexicon. The ‘freezing’ of mood information (or, in a generative perspective, mood selection as a subcategorization–feature of the matrix verb or governor) has nothing to do with grammaticalization given that in the course of the latter process a new grammatical form emerges that possesses its own autonomous meaning or function.

This subsection has shown that the all-encompassing concept of grammaticalization as well as more specific concepts, such as ‘postmodality’ and ‘harmony’, ‘regrammation’, and ‘presupposition’ as well as ‘grammaticalization’, ‘lexicalization’, and ‘conventionalization’ turn out to be insufficient in order to account for the evolutionary trends and dynamics in the domain of mood, especially with respect to complement clauses. In the next subsection, we are going to sketch an approach that is both general and fine-grained enough to analyze and motivate the developments in the realm of mood.

4.1. On the basic structure of the Latin system

‘On the origins of the Latin subjunctive, there is a long literature (Calboli Reference Calboli1966–1968, Reference Calboli1983), and no satisfactory solution’, as Magni (Reference Magni, Baldi and Cuzzolin2010: 204) summarizes the state of the art in her outline of the history of mood in Latin (in Baldi & Cuzzolin’s [Reference Baldi and Cuzzolin2010] handbook on historical Latin syntax). There is a certain consensus among scholars that the functional differentiation of the Latin system is due to the conflation of forms and meanings of the Indo-European subjunctive and optative moods. From a semantic point of view, it is a conspicuous fact that the core values revolve around the notions of will and wish (see Delbrück Reference Delbrück1893–Reference Delbrück1900: 365–369, Magni Reference Magni, Baldi and Cuzzolin2010: 204–206). In this vein, also Pinkster (Reference Pinkster2015: 388) affirms that ‘in the Latin subjunctive both with respect to form and with respect to meaning the Indo-European optativus, coniunctivus and iniunctivus have merged’. As we will see, these core notions remain the prototypical meaning zone of the subjunctive until the contemporary stages of Romance languages.

Magni, in her description of the Latin mood system, draws on Givon’s functionalist characterization of the semantic values of the subjunctive, falling back on the basic distinction between two types of subjunctive, namely the ‘subjunctive of weak manipulation’ and the ‘subjunctive of low epistemic certainty’ (Magni Reference Magni, Baldi and Cuzzolin2010: 250, in the vein of Givón Reference Givón1994: 286; see also Givón Reference Givón1995: 126–136). This characterization implies two meaning zones that constitute the two poles of a continuum with a transitional midscale region (see Magni Reference Magni, Baldi and Cuzzolin2010: 250). Interestingly, this description identifies two prototypical areas but does not oppose two complementary types of subjunctive as, for instance, does Quer in his dichotomic description of mood in Romance (according to which an intensional subjunctive is opposed to a polarity subjunctive; see Quer Reference Quer1998: 32–41).

The first type of subjunctive, the subjunctive of weak manipulation, manifests itself in volitive and desiderative uses, expressing wishes and desires, hope, and permission, both in main clauses (4) and in subordinate clauses (5).

Note that the category of strong manipulation, applying in main clause contexts, turns out to be the imperative (for examples, see Magni Reference Magni, Baldi and Cuzzolin2010: 235). In subordinate contexts, strong manipulation correlates with constructions that select for the infinitive (Givón Reference Givón1994: 283, quoted in Magni Reference Magni, Baldi and Cuzzolin2010: 248). Magni confirms this general tendency for Latin verbs such as iubeo (‘I order’), veto (‘I forbid’), and sino (‘I allow‘) that are ‘more frequently followed by an infinitive but may also admit subjunctive complemention’ (Magni Reference Magni, Baldi and Cuzzolin2010: 248).

The second type of subjunctive, the subjunctive of low epistemic certainty, includes potential and dubitative readings, denoting possibility (‘may’) or contingency (with impersonal expressions denoting existence) but also doubt in combination with questions. The following examples, (6) and (7), illustrate the possibility and dubitative readings:

Pinkster (Reference Pinkster2015) basically echoes this general bipartition by opposing deontic and potential uses of the subjunctive. However, in contrast to Magni, he does not follow Givón’s conceptualization of the mood system by identifying two poles of a scale that are connected by an intermediate and transitional zone of readings denoting preference/aversion and ‘epistemic anxiety’ (see Magni Reference Magni, Baldi and Cuzzolin2010: 250, in the vein of Givón Reference Givón1994: 286). Pinkster subsumes deliberative questions and counterfactual uses of the subjunctive under the deontic and the potential type (see Pinkster Reference Pinkster2015: 483–496, 497–509). Interestingly, he also focuses on one particular group of verbs that is generally neglected in descriptions of the Latin mood system but that plays an important role for our discussion, namely the group(s) of axiological and factive–emotive expressions. While the axiological sub-group, with expressions like ‘iustum est’ (‘it is just that’), has always been linked to the subjunctive from the earliest times, the factive–emotive sub-group exhibits a wide range of variation with respect to mood selection. So mirative expressions like ‘mirror/mirum est’ (‘to be surprised’/’it is surprising’) select more often the indicative, ‘but the subjunctive is found as well’ (Pinkster Reference Pinkster2015: 626), while the ‘subjunctive is more common’ with verbs like ‘queror’ (‘I complain’) and ‘glorior’ (‘I boast’) (Pinkster Reference Pinkster2015: 626). Also, the verb ‘gaudeo’ (‘I am glad’, ‘I rejoice’) can be found with the subjunctive, as the following example illustrates.

This mood variation, which can already be traced already to classical Latin, as the quotation demonstrates, would also remain a central characteristic of the mood system in Romance on an intra- as well as an inter-linguistical level (see particularly the study of Quer Reference Quer2009: 1784–1785; 2010: 172, and Section 4.3 below).

Despite different nuances/shades in the description of the basic mood values, Magni (Reference Magni, Baldi and Cuzzolin2010) and Pinkster (Reference Pinkster2015) strongly agree on the fact that the mood in subordinate clauses is ‘harmonic’ with its occurrences in main clauses (see e.g. Pinkster Reference Pinkster2015: 621). It is especially Magni (in the vein of Palmer 1986: 132) who establishes systematic correspondences between main and subordinate clause occurrences of the Latin subjunctive. The following quotation summarizes these correlations in a concise way:

Other systematic parallels have already been listed (see above). Conspicuously, Magni postulates a diachronic relationship between main and subordinate uses of the subjunctive that goes back to the earliest times of Latin:

In this context, she even goes so far as to identify a single core value of the subjunctive:

To sum up, we can say that the subjunctive is equally used in main and subordinate clauses in Latin and that it codes deontic and epistemic modalities, which manifest and substantiate themselves in specific characteristic readings. The origins of the subjunctive point to a diachronic priority of deontic (or, to be more precise, deontic, and bouletic) readings, a priority that also manifests itself in the prototypical status these readings display in most of the contexts marked by the subjunctive mood. However, we are still in need of more historical data that corroborate the diachronic priority of the deontic (and bouletic) modality. It has also become clear that the Latin mood system is characterized by strong and systematic correspondences between main and subordinate uses of the subjunctive mood. Yet the assumptions as to the diachronic relationship between main and subordinate clause uses remain speculative, although they have made their way into theories of grammaticalization (on the origins of subordination, see e.g. Diewald Reference Diewald1997: 11–12).

Finally, the concept of ‘modal harmony’ remains an imprecise concept as it seems to cover modal correspondences (between a main verb and a modal verb or a mood marker) as well as the phenomenon of ‘semantic bleaching’ in subordinate contexts. As has already been pointed out, Magni (Reference Magni, Baldi and Cuzzolin2010) claims that the Latin subjunctive has expanded to ‘non-harmonic’ contexts in the course of developments, such as indirect questions, becoming in this process a mere concomitant of subordination. But, as already indicated, this position is untenable for several reasons:

1. 1. If this scenario of use expansion of the subjunctive were the case, we would expect the subjunctive mood to have become the general mood of subordination. And this is clearly not the case.

2. 2. The zones of expansion, as we will see, are fully in line with the semantics of the subjunctive on the condition that we capture its semantic peculiarities by an appropriate descriptive framework. As we will see, the occurrence of the subjunctive in evaluative as well as in belief contexts can be accounted for within a modal–semantic descriptive approach to the subjunctive.

3. 3. This last aspect leads to the more general claim: Even in subordinate clauses, the subjunctive does have a clearly defined semantics that can be spelled out by sufficiently abstract, modal–semantic features (see Section 5).

4.2. Some major changes from Latin to Romance

In this subsection, we sum up the most crucial developments concerning the mood system from Latin to Romance in the domain of main and complement clauses. As we are concerned with the general evolutionary dynamics and trends of the mood system(s), our perspective is limited to structural aspects and does not deal with fine-grained chronological issues.

First, the ‘balanced’ distribution and use of the subjunctive in main and complement clauses are getting lost in favor of a general expansion of the subjunctive in subordinate and especially in complement clauses. In contrast to that, the use of the subjunctive mood in main clauses is relegated to residual contexts in Romance.

Epistemic readings of the subjunctive disappear from main clause contexts. An exception is residual dubitative questions in Old French, which combine interrogative sentence mood with subjunctive marking:

As the translation to modern French illustrates, the subjunctive imperfect is substituted by a combination of a modal verb (‘pouvoir’) and conditional morphology.

The subjunctive in main clauses generally yields a bouletic or optative reading, as in the following example:

In a modern translation, it would be necessary to mark the sentence mood with the complementizer ‘que’.

In modern French, the optative reading of the subjunctive has also largely disappeared, with the exception of archaic and/or conventionalized formulae like ‘vive la République’ (‘long live the Republique’) or ‘plût au ciel que’ (‘may it please heaven that’, see Riegel et al. Reference Riegel, Pellat and Rioul2009: 565). However, in modern Italian and Portuguese, the past subjunctive with an optative reading is still available in very formal registers, as attested by standard grammars (see Schwarze Reference Schwarze1995, Wandruszka Reference Wandruszka, Renzi and Salvi2001: 276 for Italian; Hundertmark-Santos Martins Reference Martins and Teresa1982, for Portuguese). See the following examples:

Another context characterized by the use of the imperfect subjunctive in combination with an exclamative sentence mood is also quoted in Buridant’s reference grammar of Old French (Buridant [2000] Reference Buridant2007: 334):

The sentence combines the deontic reading of ‘devoir’ with the typical counterfactual reading of the past subjunctive, yielding an expression of regret. The speaker wishes the alternative p (p: lead the army) to hold true for the sake of moral justice. The counterfactual meaning would be expressed by the conditional in modern French.

Second, the modalizing functions of the subjunctive in main clauses in Latin but also in some residual contexts in Old French, as example (14) illustrated, are superseded by modal verbs in Romance whose range of use expands from the deontic and alethic to the epistemic domain from the thirteenth century onward (for French, see e.g. Kronning Reference Kronning1996, Goldschmitt Reference Goldschmitt2007, Becker Reference Becker, Casanova and Rigual2013). One of the first attestations of an epistemic reading of the modal verb ‘devoir’ in Old French (which may still have a flavor of dynamic modality) is the following:

Typical uses of the Latin subjunctive in modalizing functions are illustrated by the following examples. In (15), the imperfect subjunctive form ‘quaererem’ would be rendered by a past conditional form of devoir (‘must’) in French (‘j’aurais dû le demander’, ‘I should have asked him’) and in (16), the present subjunctive ‘interroget’ corresponds to the present form of the French modal ‘devoir’ (‘doit procéder à un interrogatoire’, ‘must hold’):

Third, the conditional, an innovation of the Romance languages, also replaces important modalizing functions of the Latin subjunctive in main clauses. It serves such different functions as the polite attenuation of a request (‘j’aimerais bien’), the unfolding of a hypothetic or counterfactual scenario (‘je ferais’, ‘j’aurais fait’), or the apodosis of a conditional sentence (‘si p, je ferais/j’aurais fait q’). All these were contexts covered by the Latin subjunctive, for instance:

Fourth, in contrast to what can be observed for the main clause domain, finite complement clause structures that may vary between the indicative and the subjunctive clearly expand in Late Latin and Romance, an evolution that spans several centuries and is documented by Cuzzolin (Reference Cuzzolin1994) for Late Latin and by Becker (Reference Becker2014) for Late Latin and Romance. This evolution in the complement clause domain affects especially factive–emotive predicates as well as doxastic (i.e. ‘belief’) predicates and is linked to the decline of the classical Latin accuativus-cum-infinitivo (AcI) constructions (for a general description, see Calboli Reference Calboli and Calboli2002).

The transition from AcI to finite complement constructions takes place quite early with factive–emotive predicates. As Cuzzolin (Reference Cuzzolin1994: 143) points out, already in Plautus, in the third century BC, the ratio between the AcI and finite complement clause constructions amounts to 7:3. For the sake of illustration, we juxtapose the two constructions:

In the case of doxastic predicates (‘credere’ and ‘putare’), the transition from AcI to finite complement clauses takes place in a much later period. Based on his study of Late Latin texts written by Augustine of Hippo and Hieronymus on the threshold of the fifth century a.d., Cuzzolin reports that AcI constructions still largely prevail over finite ‘quod’-sentences, with a ratio of 1:3.3 (i.e. 89 to 296 occurrences) in favor of the infinitival construction type (see Cuzzolin Reference Cuzzolin1994: 283–284).

The first records in Romance attest to the complete disappearance of the AcI construction from vernacular texts. The following two examples contrast the availability of the two constructions in Late Latin texts.

4.3. The expansion of the subjunctive in complement clause contexts in Romance

A particular evolution of the mood system in Romance is the expansion of the subjunctive mood to two new modal domains, namely to the evaluative and the doxastic domains. Interestingly, the expansion to these two domains is characterized by a strong (intra- as well as inter-linguistic) mood variation in synchrony as well as in diachrony. Apart from that, both domains lie beyond the domains of priority modality (i.e. deontic and bouletic modality), which have been identified as the prototypical core zone of the subjunctive mood. In view of this expansion, the question arises as to how we can account for the semantics of the subjunctive. In other words, does the semantics of the subjunctive change due to these evolutions, or do we have to give up the idea of a unitary basic semantics that revolves around the notion of priority? In this section, we are, therefore, concerned with some basic principles and trends of mood differentiation in the two domains under scrutiny.

We start with the evaluative domain, which is constituted by factive–emotive predicates. In synchrony, this domain is characterized by the highest degree of variation, both intra- and inter-linguistically: Quer (Reference Quer2009, Reference Quer, Becker and Remberger2010, Reference Quer and Haßler2022) emphasizes the strong divergence between the Spanish varieties concerning mood selected by evaluative and emotive predicates. DeMello (Reference DeMello1996) has already pointed out that the selection of the indicative is much more frequent in Latin American Spanish than in European Spanish. According to Busch’s (Reference Busch2017) study, this trend is particularly characteristic of expressions like ‘es bueno’ (‘it is good’), ‘es una lástima’ (‘it’s a pity’), and ‘me alegro de que’ (‘I am happy that’). The same variation as in the Spanish varieties obtains among other Romance languages. Thus, the use of the subjunctive mood is very consistent in Italian from the earliest time onward, as in example (23), whereas in Romanian the indicative is selected systematically in combination with factive–emotive predicates, as in example (24):

To motivate this strong inter- and intra-linguistic variation, the diachronic evolution of mood selection in French is an illustrative showcase, given that mood selection in complement clauses shifted from the indicative to the subjunctive from Old to Classical French (for a detailed description, see Becker Reference Becker, Becker and Remberger2010, Reference Becker2014). The main landmarks of the evolution can be summarized in the following way:

1. 1. The starting point of the diachronic development is the systematic selection of the indicative mood in Old French as in the following example:

   
   Interestingly, we also encounter instances of the indicative in Old Spanish:
   

2. 2. In a second stage, the subjunctive only emerges in intensional contexts that evoke alternative worlds. In the following example, for instance, future alternatives – preferred worlds in which the addressee leaves the place, in contrast to dispreferred alternatives in which he remains – are evoked and evaluated:

   
   Note that the presence of alternatives is also manifested in the fact that the complement clause in the future-oriented context at hand can be rendered as a conditional clause (‘I am happy if p’).

3. 3. The third stage, spanning from the sixteenth to the first half of the seventeenth century, is characterized by a considerable variation as regards mood selection, depending on factors such as the role of expressivity linked to semantic features like ‘mirativity’ as in (28) and (29) but also the impact of intensional operators like negation and the necessity-operator (for more examples, see Becker Reference Becker2014: 354–359).

   The author of the two examples quoted in (28) and (29) is the same; in other words, the mood variation we find in his text is clearly exploited in order to achieve pragma–semantic effects. In (28), with the subjunctive, the author places the emphasis on his surprise as to the silence of his interlocutor about a relevant topic (highlighted by the expression ‘not a single word’). He would have expected: p (‘you write me about it’), but non-p turned out to be the case. In the subsequent example (29), with the indicative, the author communicates his expectations as to the course of the interaction (implying the question: Why don’t you ask me that?), without, however, emphasizing a strong counter-expectation as in the previous example:

   
   

4. 4. In the next stage, from the second half of the seventeenth century onward, the subjunctive tends to generalize into the standard mood in evaluative contexts. However, the temporal and the modal variables still compete, as is nicely illustrated by the following pair of examples taken from the same author:

   
   

   The subjunctive is selected in the present tense context whereas the indicative still has the upper hand in past contexts. This contrast illustrates that there is still the possibility of the time variable winning out over the modal variable when the complement clause refers to past states or events. By resorting to the subjunctive, the speaker in (30) focuses on the evaluation (‘having comedians is a preferred state’), whereas in (31) the past state of not having been able to say more about the topic is highlighted as a fact by the use of the indicative.

5. 5. In the end, with the consolidation of the classical prescriptive norm imposed by the Académie Française at the end of the seventeenth century, the subjunctive becomes the exclusive mood in evaluative contexts, at least in texts monitored by the standard norm.

How can we account for the synchronic and diachronic variation in evaluative contexts? There is no doubt that, from a modal semantic perspective, the evaluative domain is at the crossroads of two competing principles that determine mood selection in Romance.

On the one hand, the principle of ‘veridicality’ obtains given the fact that the propositional content of complement clauses selected by factive–emotive verbs is presented as being true. Obviously, the principle of veridicality prevails in earlier stages of French and, to a minor degree, also in Spanish. It also seems to be the explanatory factor that accounts for mood selection in informal language use. As Poplack et al.’s data clearly show, the French predicate ‘content’ (‘satisfied’) only combines with the subjunctive in less than 20% of the sentences with this predicate in their informal corpus of spoken French (see Poplack et al. Reference Poplack, Cacoullos, Dion, de Andrade Berlinck, Digesto, Lacasse, Steuck, Ayres-Bennet and Carruthers2018: 239 [Figure 5]).

On the other hand, the peculiarity of factive–emotive predicates lies in the fact that they imply both alternatives (alternative worlds) and an underlying relevant scale for each predicate. This scale acts as an ordering source that discriminates, ranks, and highlights specific worlds. Take, for instance, the predicate ‘content’ (‘satisfied’) as coming with a scale that compares and ranks states of affairs in accordance with the well-being of the subject whose internal state is evaluated. Internal states of well-being (of feeling ‘content’) are ranked higher than those of less well-being. This specific characteristic of factive–emotive predicates approaches them to expressions belonging to the domains of ‘priority modality’. This aligns with the understanding that factive–emotive predicates are associated with (negative) inferences, thereby introducing alternatives or alternative worlds. For example, the predicate ‘regret’ (‘regret a state of affairs p’) entails a preference for worlds where p does not hold, as discussed in detail by Giannakidou & Mari (Reference Giannakidou and Mari2021: 283). Similarly, the predicate ‘content’ (‘to be content with a state of affairs p’) involves p-worlds and non–p-worlds as conditions under which the predicate applies. It is important to recall that this condition-like interpretation of the complement clause p was crucial for the mood shift within these contexts during the evolution of the French mood system, as illustrated in example (27). We can, therefore, conclude that the factive–emotive predicates constitute a group that is situated between the ‘priority’ and the realis domain. As we have already seen, whenever alternative states of affairs are evoked, compared, and ranked according to a scale (an ordering source), the subjunctive enters the stage. With factive–emotive predicates at the cross-roads between realis/veridicality and the ordering of alternatives, the selection of either mood is perfectly motivated. This explains variation within varieties of a language (the case of Spanish), in comparison between languages (Romanian and Italian) and from a diachronic perspective (the case of French).

The doxastic domain is the second domain of expansion. The predicates concerned, expressions of belief, describe convictions held by an individual, a group of individuals, or a whole community. The notions of (inter-)subjectivity and doxastic accessibility are relevant for this domain. The bearer of a propositional attitude may hold beliefs that he shares with others for good reasons (e.g. based on appropriate evidence) or his beliefs may belong only to himself. In the first case, the beliefs are intersubjectively accessible; in the second case, they are only accessible in a subjective way. Beliefs can be conceived as worlds that are doxastically accessible by a belief-holder and in which states of affairs hold that are compatible with what the belief-holder believes in a world and at a time. Note that the doxastic domain is intertwined with the epistemic domain strictu sensu (i.e. when the two domains, as often happens, are not conflated). The epistemic domain refers to the speaker’s or a matrix subject’s knowledge and involves the degree of commitment of a subject who evaluates in Nuyts (Reference Nuyts and Frawley2006: 21) words ‘the chance that a certain hypothetical state of affairs under consideration will occur, is occurring, or has occurred in a possible world’. The epistemic domain is therefore associated with a probability scale that ranks states of affairs according to whether they are ascertained/established knowledge, more or less probable or impossible outcomes. Although the relationship between knowledge and belief is a philosophical issue that cannot be raised in this paper, it is safe to say that these two cognitive states are contiguous and intrinsically interrelated. As already noted in Section 4.1, the epistemic domain was already prone to mood distinctions in Latin but did not pertain to the prototypical core domains of the subjunctive mood (see Magni Reference Magni, Baldi and Cuzzolin2010: 233). Thus, the marking of low epistemic certainty was confined to a small number of epistemic expressions like ‘fieri (non) potest’ (‘it is (not) possible’) and some very specialized uses in contexts of interrogative sentence mood, especially in self-addressed questions like

It is, therefore, not astonishing that the system of mood distinctions was extended to the contiguous domain of belief predicates. However, what would become a crucial aspect for the nascent Proto-Romance system is the fact that Late Latin dramatically reduced the selection of AcI constructions in favor of complement clauses introduced by the complementizers ‘quod’, ‘quia’, and ‘quominus’. This development made it necessary to establish principles of mood selection and distinction that could be applied in the domain of finite complement clauses. The extension of the system of mood alternations to the doxastic domain is linked to a functional enrichment of the mood system as it transcends the divide between the principle of veridicality, on the one hand, and of ordering (relevant for the subjunctive associated with the domains of deontic and bouletic [i.e. priority] modality), on the other. In the doxastic domain, the mood opposition is now exploited in order to mark and oppose different statuses of doxastic accessibility in relation to belief contents. Roughly speaking, while the indicative marks intersubjective accessibility, the subjunctive signals that a certain state of affairs is only accessible in a partial and subject-centered way. The following example, taken from the Old French Roman de la Rose, illustrates the subjective character of the doxastic accessibility relationship entertained between the belief holders and the propositional content of the belief state:

In example (33), the self-assessment as beautiful, a de se belief of each of the subjects, is marked as highly subjective, or, in other words, as an ensemble of belief worlds only accessible to each individual in question.

The expansion of the mood distinction to the doxastic domain reveals that mood distinctions can be further explored for new modal domains. The evolution of mood distinctions from Old French to modern French and the comparison of these stages to other Romance languages in synchrony casts light on possible principles of mood distinction in the domain of doxastic modality in Romance.

Conspicuous is the case of Standard Italian, in which the subjunctive can be interpreted as a general marker of doxastic accessibility, distinguishing beliefs from knowledge. This analysis is even corroborated by Poplack et al.’s data from spoken Italian that point to almost 100% of subjunctive occurrences for the doxastic predicate ‘ritenere’ (‘to believe’, ‘to hold the belief that’) and of 80% for the prototypical belief predicate ‘credere’.

Mood in Old French belief contexts – as has already been illustrated by example (33) – distinguishes between subjective/not-intersubjective (in the case of speakers’ disclaimers) accessibility versus intersubjectivity. This corresponds to the trend observed by Loengarov (Reference Loengarov2003) for less formal registers in Italian. Here, the indicative is selected if the speaker believes that the content of his propositional attitude can be accessed intersubjectively on the basis of good evidence (see Loengarov Reference Loengarov2003: 195: ‘basa il suo giudizio su argomenti che considera come validi o che sono comunemente accettati’). These observations may also explain the remaining (20% of) indicative uses in combination with ‘credere’ in Poplack et al.’s data.

In a further development, particularly in Middle French, the subjunctive has been reinterpreted as a marker of counterfactuality, especially in the scope of negation or other modal operators, such as the interrogative operator, and especially in rhetorical questions (for more details and examples, see Becker Reference Becker2014: 236–244). In particular, the subjunctive can be encountered in contexts where a speaker vigorously contrasts his own beliefs (taken, of course, to be true) with the ‘false’ beliefs of his opponents. Jean Calvin’s theological writings are sprinkled with such polemic contrasts, in which the use of the subjunctive mood as a marker of counterfactuality turns out to be a highly expressive device:

Similarly, in Brazilian Portuguese, counterfactuality (or mistaken beliefs) is marked by the subjunctive in doxastic predicates, as evidenced by the following example:

In modern French and Spanish, the subjunctive only appears in combination with negation or the interrogative operator. In this stage, the subjunctive has changed into a marker of disbelief (i.e. ‘not believe that p is the case’), as the following example shows:

The mood configuration in Romanian marks the final stage of a possible development in other Romance languages: In this Romance language, the subjunctive does not occur in belief contexts, not even when the relevant proposition is in the scope of negation.

In the case of Romanian, the doxastic domain is fully covered by the indicative. This amounts to saying that the principle of relative veridicality (defined as veridicality linked to a matrix subject’s epistemic model; see Giannakidou Reference Giannakidou1997: 110; Reference Giannakidou1999: 388) is opposed to the subjunctive principle of priority (and linked to the notion of ‘ordering’). In this respect, the mood system of modern Romanian) strongly converges with the mood system of modern Greek (see Giannakidou Reference Giannakidou, Simons and Galloway1995: 101).

We can summarize the principles of mood selection in the doxastic domain and align them on a continuum that ranges from a strong prevalence of the subjunctive to the exclusive selection of the indicative. This continuum mirrors the range of mood variation in Romance. But it also represents, at least partly, possible clines of mood evolution, as has been shown, for instance, for French:

1. 1. The subjunctive acts as a general marker of doxastic accessibility and contrasts knowledge (ascertained facts) and beliefs (Standard Italian) (beliefs vs. knowledge).

2. 2. The subjective acts as a marker of subjective (or, at least, non-intersubjective, that is, the speaker does not want to guarantee the truth of p) accessibility to belief worlds (Old French, non-Standard Italian) (subjective vs. intersubjective).

3. 3. The subjunctive is a marker of counterfactuality, i.e. the speaker believes that p is not the case (Middle French, Brazilian Portuguese) (true vs. false beliefs).

4. 4. The subjunctive is a marker of disbelief (belief vs. disbelief) (modern French and Spanish).

5. 5. Belief-states are not differentiated by mood; the subjunctive is banned from the doxastic domain (beliefs are covered by the principle of relative veridically; the subjunctive is reserved to the domain of priority, i.e. deontic and bouletic, modality) (Romanian and modern Greek).

To conclude this subection, we can say that the doxastic domain is characterized by a highly differentiated system of mood alternations from a synchronic as well as a diachronic point of view. The mood alternations attest to the fact that the subjunctive has been successful in extending its functional range beyond its core value of priority, linked to the ordering and ranking of alternatives. The strong variation can be explained by different possible principles of mood differentiation that range from the marking of doxastic modality (by the subjunctive) to the signaling of belonging to the domain of (relative) veridicality (by the indicative). This raises the question of whether the plurality of principles concerning mood differentiation runs counter to a unitary approach with respect to the (abstract) semantics of the subjunctive mood.

In the last subsection, we take stock of the overall evolutionary dynamics of mood from Latin to Romance and try to tackle the question of how to account for the semantics of mood. Has the expansion led to a polysemy of the subjunctive mood or even to a blurring of its semantics, i.e. has it changed into a conventionalized and lexically biased concomitant of subordination?