2.1 The role of the distributive patterns of the matrix verb

Traditionally, in prescriptive grammars and also in some semantic studies, it is argued that the type of predicate from the main clause is the leading factor responsible for selecting the mood in the complement clause. More precisely, verbs expressing cognitive states and mental events such as fear, obligation, order, astonishment, pain, advice and expectation must be followed by the subjunctive whereas verbs expressing cognitive states and mental events such as certainty, probability, judgement, thought, belief and opinion must be followed by the indicative (Bescherelle 1990; 2019). The idea behind this distributive assumption is that the type of predicate from the matrix clause, that is evaluative/emotive vs. intellectual/cognitive, determines what type of cognitive states or mental events is expressed, as argued by studies with a focus on Greek (e.g. Giannakidou 1994; 1995; 1998), Italian (e.g. Giorgi & Pianesi 1997; see Mari 2016 for a different proposal), Spanish (e.g. Villalta 2000; 2006; 2008; Rubinstein 2012), Catalan (e.g. Quer 2001; 2006; 2009; 2010), French (e.g. Gosselin 2010; 2015; 2016; Schlenker 2005) or Romanian (Farkas 1985; 1992; Brasoveanu 2006).

Nevertheless, studies have also found that these classes of predicates are not necessarily uniform both with respect to the behaviour of certain individual verbs and cross-linguistically. According to Portner (2018), in most languages, the following predicates are indicative governors in Romance languages: predicates of knowledge and belief (e.g. know, believe), predicates of assertion (e.g. say, declare), predicates of inquiry (e.g. ask), perception predicates (e.g. see, hear), commissive predicates (e.g. promise) and predicates of fiction and mental creation (e.g. imagine, dream). In contrast, inquisitive predicates (e.g. wonder), preference predicates (e.g. want, prefer, hope), directive predicates (e.g. order, require), the negative counterparts of predicates of knowledge/belief and of predicates of assertion (e.g. doubt), and modal predicates (e.g. be possible) are subjunctive governors. The class of emotive factive predicates (e.g. regret) takes the subjunctive in certain languages (e.g. French) as in (5) while taking the indicative in others (e.g. Romanian) as in (6). Another example is that of the verb believe, which is predicted to be an indicative governor but which, in Italian (i.e. credere), can alternate between the subjunctive and the indicative (Mari 2016) as in (7). According to Mari (2016), polysemous verbs feature an expressive (a doxastic dimension, that is asserting certainty) and an inquisitive interpretation (which features both a doxastic and an epistemic dimension, the latter presupposing lack of knowledge). In (7), the mood alternance for the Italian credere (‘believe’) is explained by its polysemy: expressive-credere selects the indicative because it signals that the content of the opinion is personal and cannot be verified against a fact of the matter whereas the inquisitive-credere selects the subjunctive because it conveys, on the one hand, doxastic certainty at the level of the assertion, and, on the other hand, epistemic uncertainty at the level of the presupposition.

Working on Hexagonal French, Gosselin (2010: 433) classifies the predicates from the matrix clause according to whether they can be (almost) exclusively followed by one of the two moods and whether the choice of the mood generates or not a change in their meaning or that of the entire sentence. Gosselin’s classification consists of: (i) verbs which must, almost exclusively, be followed by one or the other mood (e.g. croire ‘believe’ must be followed by the indicative and the subjunctive is not acceptable, whereas douter ‘doubt’ must be followed by the subjunctive and the indicative is not acceptable), (ii) verbs that may be followed by one of the two modes with or without a change in meaning of the verb itself (e.g. comprendre ‘understand’ as in (8) and (9), and espérer ‘hope’ as in (10)), and (iv) verbs which require the indicative in affirmative sentences but change their requirement in negative, interrogative or imperative sentences (e.g. supposer ‘suppose’).

Gosselin’s classification, and more recently Baunaz & Puskás (2014, 2022) and Baunaz (2017), points to the fact that the predicates from the matrix clause may be classified in three main sets: those which require the subjunctive, those which require the indicative and those which accept an alternance between the two modes, with or without entailing a change in meaning. Regarding the first two classes, Baunaz & Puskás’s framework assumes that it is the syntactic feature make-up of the predicates that determines their selectional properties: verbs selecting the subjunctive present an emotive feature, and it is this feature which licences the subjunctive. In contrast, verbs selecting the indicative lack the emotive feature and are neutral with respect to emotivity. For them, the alternance of mood observed in the case of comprendre (‘understand’, ‘accept’) (example (8) vs. (9)) comes from the emotivity feature: when comprendre involves an intellectual component only (as in (8)) it selects the indicative, whereas when it involves some kind of empathy (as in (9)), it includes an emotive feature and it selects the subjunctive. As for the third class, the verbs and the conditions under which a verb accepts the alternance between the indicative and the subjunctive are numerous and varied.

While Soutet (2000) provides approximately 30 alternance verbs, Petkovic (2020) collects 75 alternance verbs for her empirical study. As observed by Gosselin, alternance may be completely free or allowed in some cases by the type of sentence (negative, interrogative). Amsili & Guida (2014) investigated the case of alternance in negative sentences. For them, the alternance between the indicative and the subjunctive in negative contexts is particularly frequent with some epistemic verbs (e.g. penser ‘think’, croire ‘believe’, considerer ‘consider’). In contrast, factive and semi-factive predicates such as savoir ‘know’ or realiser ‘realise’ (which require the indicative), as well as emotive factive predicates such as regretter ‘regret’ and apprécier ‘appreciate’ (which require the subjunctive), do not allow alternance when they occur under the scope of negation. Amsili & Guida (2014) carry out an elicitation experiment with a verb conjugation task, in which they test the role of four epistemic verbs selecting the indicative and of four emotive factives selecting the indicative in affirmative contexts, and their role for mood choice in negative contexts. They confirm that the verbs selecting the indicative and those selecting the subjunctive behave as predicted, and find that in negative contexts, the verbs tested are followed by the subjunctive in 75% of the cases (80% in a stricter analysis) and by the indicative in 25% of the cases (20% in a stricter analysis). Thus, the use of certain epistemic and emotive factives under the scope of negation privileges the use of the subjunctive but not exclusively.

In contrast to these semantic studies, Poplack et al. (2013) argue that there are no subjunctive governors, such as those listed above as belonging to the set of verbs selecting the subjunctive. In their study, they compare the prescriptions of grammars of French, published between 1530–1999, to the actual use of the subjunctive in spoken Quebec French during three periods of time (1940–1950, 1982, and 2005–2007). They find an important discrepancy regarding the type and the number of matrix verbs between grammars’ prescriptions and speakers’ actual use. For instance, while grammars predict more than 304 verbal governors for the subjunctive, their analysis of spoken data (spontaneous everyday speech) shows that the verbs falloir ‘be necessary’ expressing necessity, vouloir ‘want’ expressing volition and aimer ‘like’ expressing desire are the most frequent and the most productive governors of the subjunctive.² They write:

“Almost without exception, no governor of even moderate frequency (i.e. five occurrences) could be shown to trigger the subjunctive categorically. This is true not only of the hundreds prescribed throughout the grammatical tradition, but also of the few dozen that ever actually governed a subjunctive in usage. There are thus no subjunctive governors: only potential governors, each of which features its own characteristic associations” (2013: 190).

In other words, as put by Comeau (2020: 42), the subjunctive would be constrained not by semantic classes of verbs, but rather by a small set of main clause elements (verbal and/or non-verbal). Comeau (2020) puts forward data from Acadian French, spoken in the Baie Sainte-Marie region, that contrasts with Poplack et al.’s (2013) data. He argues that the subjunctive is not a linguistic variable in this variety of French: he finds that the subjunctive is governed by the meaning or the semantics of the matrix verb: the verbs aimer que ‘to like’, espérer ‘to hope’, falloir que ‘to be necessary’, guetter que ‘to wait’ and vouloir que ‘to want’ are exclusively followed by the subjunctive (100%). The only exception is that of point croire que ‘to not believe’, which was found to be followed by the subjunctive in 29% of the cases. Comeau explains this result by the fact that, in this negative context, the subjunctive expresses the semantic meaning of doubt (which is its main semantic function since its apparition in Latin, also confirmed for Old French, cf. Jensen 1974). This result is interpreted to show that the subjunctive isn’t always subject to inherent variability. For Comeau (2020), the alternation, in some varieties of French, between the indicative and the subjunctive moods may be explained by two main factors: the semantic class or the meaning of the matrix verb (such as, verbs expressing volition, doubt, etc.) and the tense concordance. More specifically, the loss or the weak frequency of past tenses within the subjunctive mood (e.g. the imperfect subjunctive) allowed for the subjunctive to vary with other moods (the indicative or the conditional) to ensure tense concordance.

In sum, most of the state of the research on Hexagonal French, as well as some other varieties of French (e.g. Acadian French spoken in the Baie Sainte-Marie region, but also in other Romance languages), suggests that the matrix verb (through its meaning) determines the mood choice in the complement clause. In this paper, we focus on three sets of matrix verbs established according to prescriptive grammars and linguistic studies: those expected to select only the subjunctive, those expected to select only the indicative and those expected to accept an alternance of moods. We will call this assumption the prescriptive distributional hypothesis.

2.2 The role of the tense of the matrix verb

As mentioned in the Introduction, studies at the syntactic-semantic interface have argued that the subjunctive mood in complement clauses is deficient or anaphoric, which means that it is dependent on the temporal properties of the matrix verb (Landau, 2004; Manzini 2000; Quer 1998, 2006) while the indicative mood in complement clauses is independent. As put by Socanac (2017: 59–60), in the case of indicative complement clauses, the embedded predicate can generally denote all types of temporal relations with respect to the reference time of the matrix predicate, whereas in the case of subjunctives, the embedded predicate is typically restricted to a posterior-referring interval of time with respect to that of the matrix predicate. In other words, the tense of the matrix predicate constraints the tense used in the subjunctive complement clause in that the latter cannot denote an event that took place prior to the one denoted by the matrix predicate. This is illustrated in examples (13)-(14) from Socanac (2017: 60).

The question is whether the subjunctive mood is dependent on other properties of the matrix predicate, such as its expression of past vs. present tense, and its denotation of perfective vs. imperfective eventualities via the category of grammatical aspect. Tense is defined as the grammatical category, obligatorily marked on the verb in tensed languages, locating eventualities in time (situation time TSit or event time E) with respect to the utterance time UT (or moment of speech S) and to a topic time TT (or reference moment) (Klein 1994, and respectively Reichenbach 1947). The distinction between past (E precedes S) and present tense (E is simultaneous to S) may be seen as parallel to the realis vs. irrealis distinction (used to describe the indicative vs. subjunctive mood division). As put by Mithun (1999) and Palmer (2001), realis depicts situations as actualized, as having occurred or actually occurring, accessible through direct perception. In contrast, irrealis presents a situation as belonging to the realm of thought, accessible through imagination. Such a comparison between tense and mood would then result in the following association: the past tense is associated with the indicative while the present tense with the subjunctive.

To the best of our knowledge, the role of tense from the matrix clause for mood choice has been mentioned in two previous empirical studies: Poplack et al. (2013) and Gudmestad & Edmonds (2015). First, Poplack et al. (2013), who analyzed three corpora of spoken Quebec French belonging to the 19^th century, the 20^th century and, respectively, the 21^st century, found that the tense of the matrix verb played a significant role for mood choice in the 19^th and the 20^th century but this effect disappeared in the 21^st century. More precisely, in their set of data, they found that a matrix verb in the present indicative tense was more frequently followed by the subjunctive (with a probability higher of 0.68 in the 20^th century) than by other modes. Regarding past time, the authors report that a matrix verb in the Imparfait was less frequently followed by the subjunctive (a probability of 0.38 in the 19^th century), and a matrix verb in the Passé Composé was also less frequently followed by the subjunctive (a probability of 0.42 in the 20^th century) than by other modes. According to the authors’ analysis, the role of the tense of the matrix clause was not found to be statistically significant in the 21^st century corpus. Second, Gudmestad & Edmonds (2015) investigate experimentally mood choice complement clauses in contemporary Hexagonal French. For this, they carry out two elicitation experiments to test three types of factors: the semantic category of the matrix verb, its temporality (past vs. present vs. future) and hypotheticality. They find that native speakers used the subjunctive in 78.2% of the cases when the matrix verb expressed future time, in 53% of the cases when it expressed present time and in 50.3% of the cases when it expressed past time.

In sum, previous research shows a link between the temporality of the matrix clause and mood choice, even if they are not perfectly consistent across centuries. Thus, we may formulate a tense hypothesis, according to which, in Hexagonal French, the tense of the matrix verb influences the mood choice in the complement clause. Specifically, the present tense features the use of the subjunctive whereas the past tense features the use of the indicative. If we follow Poplack and colleagues’ findings for spoken Quebec French in the 19^th and the 20^th century, we may expect to find there still is a relation between tense and mood. In contrast, if Hexagonal French has the same behavior as spoken Quebec French in the 21^st century, then we may expect that there is no influence of tense on mood choice in Hexagonal French.

2.3 The role of the aspect of the matrix verb

Grammatical aspect refers to “different ways of viewing the internal temporal consistency of a situation” (Comrie 1976: 3). For Comrie, perfective aspect indicates “the view of a situation as a single whole, without distinction of the various separate phases” and is, thus, associated with eventive meaning. Imperfective aspect, on the other hand, is used to denote progressive and ongoing events. In French, in the past time domain, the Passé Composé is regularly described as perfective³ whereas the Imparfait is imperfective⁴. As for the present time domain, there is no grammatical distinction between the perfective and imperfective, as the Présent expresses both aspects.

The question of the relation between grammatical aspect (of the matrix clause) and mood (of the complement clause) arises. To our knowledge, there are no studies which have investigated or which mention this relation, with the exception of Poplack et al. (2013). As written in section 2.2., Poplack and colleagues report that having an Imparfait or a Passé Composé in the matrix clause may have an influence on mood choice. More specifically, they report a probability of 0.38 in the 19^th century corpus and of 0.65 in the 20^th century corpus for having a subjunctive in the complement clause when the matrix verb is in the Imparfait. As for the Passé Composé, they report a probability of 0.42 in the 20^th century corpus for having a subjunctive in the complement clause. Furthermore, other scholars have discussed how the mood from the matrix clause determines the choice of the grammatical aspect used in the complement clause, such as Giannakidou (2009) targeting Greek and Socanac (2017) targeting Balkan Slavic languages such as Bulgarian and Croatian. According to Giannakidou (2009: 1887), in the context of non-past tenses, Greek verbs associated with the imperfective aspect appear in indicative complement clauses introduced through the indicative marker oti whereas perfective verbs appear in subjunctive complement clauses introduced through the subjunctive marker na, as shown in examples (15) and (16), respectively. Indeed, the na particle and the perfective morpheme (present on the verb) are considered to be the overt marks of the subjunctive.

Thus, Greek and Bulgarian reveal that there is an interaction between mood and grammatical aspect (cf. also Hacquard 2009 regarding the interaction between aspect and modality). However, studies have not considered whether the grammatical aspect of the matrix verb influences the mood choice in the complement clause. In fact, to the best of our knowledge there are no studies investigating this specific configuration and its application to (European) French. A rapid look at examples containing predicates selecting the subjunctive as in (17) and (18), predicates selecting the indicative as in (19) and (20), or alternating predicates as (21) and (22) shows that the grammatical aspect of the matrix verb does not seem to influence the mood choice.

This result is concordant with Poplack et al.’s (2013) findings for the 21^st century: the grammatical aspect of the matrix verb was not statistically associated with the use (or not) of the subjunctive in the complement clause. Nevertheless, if we draw on their findings for the 19^th and the 20^th century corpora, we may expect that grammatical aspect orients the speaker towards a subjunctive or an indicative complement clause. We thus formulate the aspect hypothesis according to which the aspect of the matrix verb influences the mood choice in the complement clause.

3.1 Hypotheses and predictions

As discussed in Section 2, from the unclear literature (which provides, on the one hand, contrastive findings about the role of the matrix verb – its distributive patterns, and on the other hand, too little evidence about how the tense and the aspect of the matrix verb influence mood choice), we expect that the matrix verb, via its distributional patterns, its tense and its grammatical aspect, influences mood choice one way or another. On the basis of the three hypotheses formulated earlier, we provide a number of exploratory predictions in Table 1.

For the semantic predicate hypothesis to be validated, we expect that the distributive pattern of the matrix verb determines mood choice according to the three groups of verbs defined: subjunctive_only, indicative_only and alternance. In other words, in both Experiment 1 and 2, we expect to find that the verbs from the subjunctive_only set are followed almost exclusively (>90%)⁵ by the subjunctive, verbs from the indicative_only set are followed almost exclusively (>90%) by the indicative, and verbs from the alternance set are followed either by the subjunctive or by the indicative according to a variable pattern. Assuming that there is no other factor playing a role in mood choice, the baseline for a variable pattern would correspond to a 50%-50% distribution between the subjunctive and the indicative. This type of result would mean that the three-way classification of predicates is meaningful. If the semantic predicate hypothesis does not hold (as found for spoken Quebec French), then we expect to find no significant associations between the verbs tested and mood choice in the complement clause.

For the tense hypothesis to be validated, in Experiment 1, we predict either a main effect of tense or an interaction effect between tense and the distributive pattern of the matrix verb. More precisely, we expect to find that a present tense in the matrix clause favors the use of the subjunctive whereas a past tense significantly favors the indicative. In the case of the interaction effect, we expect that this effect applies only to alternance verbs. In other words, the tense of the matrix verb orients mood choice only in the absence of a highly constraining semantics of the predicate (i.e. which requires one of the two moods). If the variety of French tested follows the pattern found for spoken Quebec French of 21^st century, then we expect to find no statistically significant effect of tense on mood choice.

For the grammatical aspect hypothesis to be validated, in Experiment 2, we predict either a main effect of aspect or an interaction effect between the distributive pattern and the aspect of the matrix verb. With respect to the direction of the effect, we have identified two exclusive options. They depend on whether the influence of aspect is similar to that observed for Greek and Bulgarian for the choice of mood in the matrix clause (cf. discussion section 2.3) or to that observed for spoken Quebec French of the 19^th and of 20^th century. According to the first option, a perfective tense in the matrix clause favours the use of the subjunctive whereas an imperfective tense significantly favours the indicative. According to the second option, a perfective aspect favours the use of the indicative whereas an imperfective aspect significantly favours the subjunctive. In the case of the second option, the influence of grammatical aspect would be coherent to that of tense, if we assume that tense, grammatical aspect and mood concord: past/perfectivity/indicative and present/imperfectivity/subjunctive⁶. Finally, Option 3 assumes that if the variety of French tested follows the pattern found for spoken Quebec French of the 21^st century, then we expect to find no statistically significant effect of grammatical aspect on mood choice.

3.2 Experimental design

In order to investigate the constraints imposed by the matrix verb ‒ its distributive pattern, its tense and its grammatical aspect ‒ on mood choice in the complement clause, we carried out two elicitation experiments with a 2x3 mixed design.

In Experiment 1, we manipulated two variables: the distributive pattern of the matrix verb, with three levels: predicates selecting only the subjunctive, predicates selecting only the indicative and predicates accepting both moods, and the tense of the matrix verb, with two levels: present and past time, as shown in Table 2. In Experiment 2, we manipulated two variables: the distributive pattern of the matrix verb as in the first experiment, and the grammatical aspect of matrix verb, with two levels: perfective and imperfective aspects, as shown in Table 2. In both experiments, the first variable was a within-group variable: participants read experimental items with verbs from all of the three semantic verb categories tested. The second variable was a between-group variable in Experiment 1 and a within-group variable in Experiment 2. In Experiment 1, half of the participants read experimental items in which the matrix verb was inflected in the Présent (of the Indicative) and half of the participants read experimental items in which the matrix verb was inflected in the Passé Composé. In Experiment 2, participants read some experimental items in which the matrix verb was inflected either in the Passé Composé and expressed the perfective aspect and others in which the matrix verb was inflected in the Imparfait and expressed the imperfective aspect.

3.3 Experiment 1

3.3.3 Results

We coded participants’ answers consisting of the inflected form of the verb from the subordinate phrase for mode: subjunctive, indicative and conditional. There were also cases when participants wrote two answers for a single experimental item: the form of the verb in the subjunctive and in the indicative. We coded these cases as both subjunctive and indicative. Data was analyzed in three steps. First, we carried out a quantitative descriptive analysis of the frequencies of mood types in the subordinate clause (indicative, subjunctive, both indicative and subjunctive, and conditional). Second, we fitted a generalized mixed model on the data points consisting of cases when participants answered either the subjunctive or the indicative. Third, we analyzed only the subset of alternance verbs in order to examine their overall behavior with respect to triggering a subjunctive or an indicative form in the complement phrase.

Quantitative descriptive analysis of the whole set of data

The results of the first quantitative descriptive analysis, carried out on the total number of data points (2010 observations) and investigating the role of the verb class and the verb tense expressed in the main clause, are provided in Figure 1. They revealed differences among the three semantic categories of verbs tested.

Figure 1

Frequencies of mood types in the complement clause- Experiment 1.

First, verbs belonging to the alternance category present a different behavior according to the verbal tense from the main phrase: they trigger more indicative forms (61%) than subjunctive forms (26%) when the matrix verb is the past tense, and more subjunctive forms (53%) than indicative forms (46%) when the matrix verb is in the present tense. Similarly, the conditional mood is used more frequently in the subordinate phrase when the matrix verb is the past tense (13%) than when it is in present tense (1%).

A closer look at the mood selection behavior of individual verbs from this set, provided in Table 4, reveals that it depends on the tense of the matrix verb. Specifically, with present tense, some verbs exhibit an almost 50/50 mood choice pattern, such as concevoir, comprendre and se plaindre. S’assurer is the only verb that presents such a behavior independently of the tense of the matrix verb. The verb accepter, in contrast, showed a strong preference for the subjunctive independently of the tense of the matrix verb. The verb admettre selects the indicative more frequently than the subjunctive, and this pattern is stronger when the matrix verb is in the past tense (81%) than in the present tense (63%). Finally, verbs such as imaginer and supposer reveal that that the indicative and the subjunctive are in a genuine competition with the conditional. Our manipulation of the tense of the matrix verb further revealed that this competition appears predominantly in the domain of past tense (as shown in Table 4).

Table 4

Individual verbs from the alternance set: Experiment 1.

Matrix verb	Past tense								Present tense
	Indicative		Subjunctive		both		Cond		Indicative		Subjunctive		both		Cond
	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.
Accepter ‘accept’	1	3%	31	97%	0	0%	0	0%	2	6%	33	94%	0	0%	0	0%
Admettre ‘admit’	26	81%	4	13%	0	0%	2	6%	22	63%	11	31%	0	0%	2	6%
Comprendre ‘understand’	31	97%	1	3%	0	0%	0	0%	15	43%	20	57%	0	0%	0	0%
Concevoir ‘understand’	28	88%	4	13%	0	0%	0	0%	14	40%	20	57%	1	3%	0	0%
Imaginer ‘imagine’	7	22%	2	6%	1	3%	22	69%	30	86%	4	11%	0	0%	1	3%
Nier ‘deny’	23	72%	9	28%	0	0%	0	0%	13	37%	21	60%	0	0%	1	3%
Rêver ‘dream’	23	72%	7	22%	1	3%	1	3%	2	6%	33	94%	0	0%	0	0%
S’assurer ‘make sure’	17	53%	15	47%	0	0%	0	0%	18	51%	17	49%	0	0%	0	0%
Se plaindre ‘complain’	25	78%	7	22%	0	0%	0	0%	18	51%	17	49%	0	0%	0	0%
Supposer ‘suppose’	14	44%	2	6%	1	3%	15	47%	26	74%	9	26%	0	0%	0	0%
Total	195	61%	82	26%	3	1%	40	13%	160	46%	185	53%	1	0%	4	1%

Second, verbs belonging to the subjunctive_only set of verbs triggered most frequently but not exclusively the subjunctive in the complement clause, and this both when the matrix verb was the present tense (93% subjunctive, 7% indicative and 0.3% conditional) and past tense (93% subjunctive, 4% indicative and 3% conditional). Similar to the verbs from the alternance group, having a past tense in the main clause increases the use of the conditional in the subordinate clause.

A closer look at the behavior of individual verbs from this category, provided in Table 5, reveals that there are some verbs, such as apprécier, approuver and redouter, which select the competing mood (here, the indicative) more frequently when the matrix verb is in the present tense (14%, 23%, and respectively 14%) than when it is in the past tense (3%, 19% and respectively 6%). For approuver and craindre, the subjunctive is in competition not only with the indicative but also with the conditional, and the effect is stronger when the matrix verb is in the past tense (19%, and respectively 9%) (Table 5).

Table 5

Individual verbs from the Subjunctive_only set: Experiment 1.

Matrix verb	Past tense								Present tense
	Indicative		Subjunctive		both		Cond		Indicative		Subjunctive		Cond
	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.
Apprécier ‘appreciate’	1	3%	31	97%	0	0%	0	0%	5	14%	30	86%	0	0%
Approuver ‘approve’	6	19%	20	63%	0	0%	6	19%	8	23%	26	74%	1	3%
Attendre ‘wait’	0	0%	32	100%	0	0%	0	0%	0	0%	35	100%	0	0%
Craindre ‘fear’	3	9%	26	81%	0	0%	3	9%	2	6%	33	94%	0	0%
Exiger ‘require’	1	3%	30	94%	1	3%	0	0%	1	3%	34	97%	0	0%
Ordonner ‘order’	0	0%	32	100%	0	0%	0	0%	2	6%	33	94%	0	0%
Permettre ‘allow’	0	0%	32	100%	0	0%	0	0%	0	0%	35	100%	0	0%
Proposer ‘offer’	0	0%	32	100%	0	0%	0	0%	0	0%	35	100%	0	0%
Redouter ‘dread’	2	6%	30	94%	0	0%	0	0%	5	14%	30	86%	0	0%
Souhaiter ‘wish’	0	0%	32	100%	0	0%	0	0%	2	6%	33	94%	0	0%
Total	13	4%	297	93%	1	0%	9	3%	25	7%	324	93%	1	0%

Third, verbs belonging to the indicative_only set of verbs present the most stable behavior among the three sets of verbs tested. As expected, they trigger almost exclusively the indicative mood in the complement clause, and this applies both when the matrix verb is in the past tense (99% indicative, 0% subjunctive and 1% conditional) and present tense (99% indicative, 1% subjunctive and 0.3% conditional). Table 6 provides a finer-grained view of the stability found for the indicative_only category. It is worth mentioning the case of the verb croire which selected the subjunctive in 6% of the cases independently of whether the matrix verb is in the past or the present tense (Table 6).

Table 6

Individual verbs from the Indicative_only set: Experiment 1.

Matrix verb	Past tense				Present tense
	Indicative		Cond		Indicative		Subjunctive		Cond
	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.
Affirmer ‘state’	32	100%	0	0%	35	100%	0	0%	0	0%
Constater ‘notice’	32	100%	0	0%	34	97%	1	3%	0	0%
Croire ‘believe’	30	94%	2	6%	33	94%	2	6%	0	0%
Déclarer ‘declare’	32	100%	0	0%	35	100%	0	0%	0	0%
Observer ‘observe’	32	100%	0	0%	34	97%	1	3%	0	0%
Penser ‘think’	32	100%	0	0%	34	97%	0	0%	1	3%
Raconter ‘tell’	32	100%	0	0%	35	100%	0	0%	0	0%
Remarquer ‘remark’	32	100%	0	0%	35	100%	0	0%	0	0%
Trouver ‘find’	32	100%	0	0%	35	100%	0	0%	0	0%
Voir ‘see’	32	100%	0	0%	35	100%	0	0%	0	0%
Total	318	99%	2	1%	345	99%	4	1%	1	0%

Analysis through logistic mixed-effects models

In order to better understand how the distributive patterns of the matrix verb and its tense influence, independently and in interaction, mood choice, we constructed logistic mixed-effects models on the data in which participants’ answers were a subjunctive or an indicative (this means, we excluded from the analysis the cases when participants used a conditional and when they explicitly provided both the indicative and the subjunctive). Precisely, models were fitted on 1948 data points.

To build the logistic mixed-effects models we used the R software (R Development Core Team, 2010, version 3.1.2). We included both participants and items as random factors in all analyses. Models were tested using the glmer() function of the lmer4 package of R, and model comparisons were assessed using the anova() function, which calculates the chi-square value of the log-likelihood to evaluate the difference between models, following Baayen’s (2008) procedure. Following Field et al. (2014), we built the models by going from the simplest model to the one of interest. Precisely, we first tested a model that only encompassed items and participants as random factors (i.e., random intercepts). We then compared this model to one including the distributive patterns of the matrix verb as a fixed factor and finally one that incorporated both the category and the tense of the matrix verb, as well as their interaction, as fixed factors.

Adding the distributive patterns of the matrix verb as fixed factor to the simplest model, which only included items and participants as random effects, significantly improved the model (Δχ² = 62.42, Δdf = 2, p < .001). Adding the tense of the matrix verb as a fixed factor further improved the model’s fit (Δχ² = 23.11, Δdf = 1, p < .001). Finally, adding the interaction between the distributive patterns factor and tense of the matrix verb also further improved the model’s fit (Δχ² = 28.91, Δdf = 2, p < .001). According to this final model, we find in our data a significant main effect of the semantic category of the matrix verb, a significant main effect of its tense and a significant interaction between them. Precisely, the significant main effect of the semantic category of the matrix verb confirms that, compared to the alternance category of verbs (which trigger the indicative in 57.1% and the subjunctive in 42.9% of the cases), subjunctive_only verbs trigger the use of the subjunctive (94.2%) (β = 5.13, SE = 0.72, p < .001). The significant main effect of the tense of the matrix verb reveals that a present tense significantly increases the probability to have a subjunctive form in the complement clause (49.2%) rather than a past tense (41.9%) (β = 1.58, SE = 0.24, p < .001). Finally, the significant interaction between the distributive patterns and the tense of the matrix verb indicates that verbs from the subjunctive_only category which are inflected in the present tense strongly decreases the probability to use the subjunctive mood (92.8%) compared to when they are inflected using a past tense (95.8%) (β = – 2.14, SE = 0.42, p <. 001).

Analysis of the subset of alternance verbs

To examine the role of the tense of the matrix clause for orienting the use of an indicative or a subjunctive mood in the complement clause independently of the semantics of the main verb, we carried out an analysis of the subset of alternance verbs only. As before, we built the models by going from the simplest model to the one of interest. Adding the tense of the matrix verb as a fixed factor to the simplest model, which only included items and participants as random effects, significantly improved the model (Δχ² = 38.46, Δdf = 1, p < .001). According to the model including the tense of the matrix verb, inflecting alternance verbs in the present tense significantly increases the probability to use the subjunctive mood (53.6%) compared to when they are inflected in the past tense (29.6%) (β = –1.56, SE = 0.23, p < .001).

3.4 Experiment 2

The imperfective vs. perfective manipulation was done only in the domain of past tense because, in French, aspect cannot be grammatically marked on the verb in the domain of present tenses (as it is the case of English, for example)⁸. Since Experiment 1 revealed that having a past tense matrix verb enhances the indicative and the conditional, the current experiment tests whether grammatical aspect has an additional influence on mood choice (compared to the effect already achieved by the tense of the matrix verb).

3.4.3 Results

As in Experiment 1, we coded participants’ answers consisting of the inflected form of the verb from the complement clause for mood: subjunctive, indicative and conditional. Cases when participants wrote two answers for a single experimental item (i.e. the form of the verb in the subjunctive and in the indicative) were coded as both subjunctive and indicative. Data was analyzed as in Experiment 1. First, we analyzed the frequencies of mood types in the complement clause. Second, we fitted a logistic generalized mixed model on the data points consisting of cases when participants answered either the subjunctive or the indicative. Third, we analyzed only the subset of alternance verbs in order to examine their overall behavior with respect to triggering a subjunctive or an indicative form in the complement clause.

Quantitative descriptive analysis of the whole set of data

The results of the quantitative descriptive analysis, carried out on the total number of data points (2016 observations) and investigating the role of the verb class and the verb tense expressed in the main clause, are provided in Figure 2. They revealed differences among the three sets of verbs tested.

Figure 2

Frequencies of mood types in the complement clause- Experiment 2.

First, verbs belonging to the alternance category present a different behavior according to the grammatical aspect of the main clause. Specifically, depending on whether the grammatical aspect of the matrix clause is the imperfective or the perfective, they trigger more indicative forms (60%) with the perfective than with the imperfective (51%). Also, more subjunctive forms (36%) are triggered when matrix verbs are in the imperfective than in the perfective (26%). Participants’ choice of the conditional mood is not affected by the grammatical aspect of the matrix verbs: participants used in 13% of the cases independently of whether the matrix verb was the imperfective or the perfective.

Taking a closer look at the mood selection behavior of individual verbs from the alternance category (as shown in Table 8), Experiment 2 confirms the findings from Experiment 1. First, the verbs from this set present a variable pattern with respect to their selection of the indicative or the subjunctive. Second, the results confirm that there are verbs which manifest a strong preference for the indicative, such as admettre, expliquer, ignorer and se plaindre, and this independently of the grammatical aspect of the matrix verb. In contrast, comprendre and s’assurer have a strong preference for the indicative only when the matrix verb is in the perfective (100% and, respectively, 76%). Other verbs select most frequently the subjunctive, such as accepter and suggérer, and this independently of the grammatical aspect of the matrix verb. In contrast, the verb rêver selects more frequently the subjunctive when the matrix verb is in the imperfective (91%) than when it is in the perfective (40%). Finally, verbs such as dire, imaginer, reconnaître and supposer confirm that the conditional is an important competitor mood in the past tense domain, and this pattern does not seem to be affected by the grammatical aspect of the matrix verb (with the exception of reconnaître, for which the frequency of the conditional drops when a perfective is used in the matrix clause).

Table 8

Individual verbs from the alternance set: Experiment 2.¹⁰

Matrix verb	Imparfait						Passé Composé
	Indicative		Subjunctive		Cond		Indicative		Subjunctive		both		Cond
	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.
Accepter ‘accept’	1	4%	22	96%	0	0%		0%	25	100%	0	0%	0	0%
Admettre ‘admit’	20	80%	4	16%	0	0%	23	100%	0	0%	0	0%	0	0%
Comprendre ‘understand’	12	48%	11	44%	1	4%	23	100%	0	0%	0	0%	0	0%
Dire ‘say’	14	61%	0	0%	9	39%	9	36%	2	8%	0	0%	14	56%
Expliquer ‘explain’	23	100%	0	0%	0	0%	25	100%	0	0%	0	0%	0	0%
Ignorer ‘ignore’	24	96%	1	4%	0	0%	18	78%	2	9%	0	0%	3	13%
Imaginer ‘imagine’	10	43%	5	22%	8	35%	11	44%	4	16%	0	0%	9	36%
Nier ‘deny’	18	72%	6	24%	1	4%	17	74%	5	22%	0	0%	1	4%
Reconnaître ‘recognize’	18	78%	1	4%	4	17%	20	80%	4	16%	0	0%	1	4%
Rêver ‘dream’	2	9%	21	91%	0	0%	13	52%	10	40%	1	4%	1	4%
S’assurer ‘make sure’	6	26%	14	61%	3	13%	19	76%	6	24%	0	0%	0	0%
Se plaindre ‘complain’	18	72%	7	28%	0	0%	18	78%	5	22%	0	0%	0	0%
Suggérer ‘suggest’	0	0%	25	100%	0	0%	1	4%	21	91%	0	0%	1	4%
Supposer ‘suppose’	4	16%	3	12%	18	72%	6	26%	3	13%	0	0%	14	61%
Total	170	51%	120	36%	44	13%	203	60%	87	26%	1	0%	44	13%

Second, verbs belonging to the subjunctive_only set triggered most frequently but not exclusively the subjunctive in the complement clause, and this both when the grammatical aspect of the matrix verb was the imperfective (88% subjunctive, 10% indicative and 1% conditional) and the perfective (92% subjunctive, 7% indicative and 1% conditional). In other words, when verbs from the subjunctive_only set are inflected in the perfective, they select less often the competing mood (here, the indicative) (7%) than when they are inflected in the imperfective (10%).

A closer look at the behavior of individual verbs from this set, provided in Table 9, reveals that some verbs, such as douter, redouter and s’inquiéter, select the subjunctive less frequently (less than 90%) than expected (100%), and this effect is stronger when the matrix verb is in the imperfective. Approuver is shown to have an alternance behavior regardless of the grammatical aspect of the matrix verb: it selects the indicative in 44% of the cases when inflected as imperfective and 48% of the cases when inflected as perfective, while it selects the subjunctive in 52% of the cases when inflected as imperfective and 48% when inflected as perfective.

Table 9

Individual verbs from the subjunctive_only set: Experiment 2.

Matrix verb	Imparfait						Passé Composé
	Indicative		Subjunctive		Cond		Indicative		Subjunctive		Cond
	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.
Apprécier ‘appreciate’	0	0%	23	100%	0	0%	1	4%	24	96%	0	0%
Approuver ‘approve’	11	44%	13	52%	0	0%	11	48%	11	48%	1	4%
Attendre ‘wait’	1	4%	24	96%	0	0%	0	0%	23	100%	0	0%
Craindre ‘fear’	2	9%	19	83%	1	4%	1	4%	24	96%	0	0%
Désirer ‘desire’	0	0%	23	100%	0	0%	1	4%	24	96%	0	0%
Douter ‘doubt’	4	16%	19	76%	2	8%	3	13%	20	87%	0	0%
Exiger ‘require’	2	8%	23	92%	0	0%	1	4%	22	96%	0	0%
Ordonner ‘order’	2	9%	20	87%	1	4%	0	0%	24	96%	1	4%
Permettre ‘allow’	0	0%	23	100%	0	0%	0	0%	25	100%	0	0%
Proposer ‘offer’	0	0%	23	100%	0	0%	1	4%	24	96%	0	0%
Redouter ‘dread’	6	24%	19	76%	0	0%	2	9%	20	87%	1	4%
Refuser ‘refuse’	0	0%	25	100%	0	0%	0	0%	23	100%	0	0%
S’inquiéter ‘worry’	2	9%	20	87%	1	4%	3	12%	21	84%	0	0%
Souhaiter ‘wish’	2	8%	21	84%	0	0%	0	0%	23	100%	0	0%
Total	32	10%	295	88%	5	1%	24	7%	308	92%	3	1%

Third, verbs belonging to the indicative_only set triggered most frequently but not exclusively the indicative mood, and this applies both when the matrix verb is in the imperfective (76% indicative, 8% subjunctive and 16% conditional) and in the perfective (78% indicative, 7% subjunctive and 14% conditional), as shown in Table 10. As it was the case for alternance verbs, the conditional is particularly frequent in the case of the indicative_only set of verbs. A closer look at the behavior of individual verbs from the indicative_only set shows that the indicative is in strong competition with both the subjunctive and the conditional when the matrix verb is décider, déclarer, espérer and se douter, regardless of their grammatical aspect. Estimer and croire diminish their preference for the indicative and increase their preference for the conditional, and this independently of whether they are inflected is in the perfective or the imperfective. The other verbs tested, such as constater, penser, raconter, trouver and voir, remain relatively stable with respect to their selection of the indicative, and this pattern holds independently of their grammatical aspect.

Table 10

Individual verbs from the indicative_only set: Experiment 2.

Matrix verb	Imparfait						Passé Composé
	Indicative		Subjunctive		Cond		Indicative		Subjunctive		Cond
	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.	abs. freq.	rel. freq.
Affirmer ‘state’	24	96%	0	0%	0	0%	21	91%	1	4%	1	4%
Constater ‘notice’	23	100%	0	0%	0	0%	23	92%	1	4%	0	0%
Croire ‘believe’	23	92%	0	0%	2	8%	20	87%	0	0%	3	13%
Décider ‘decide’	4	17%	1	4%	18	78%	2	8%	4	16%	19	76%
Déclarer ‘declare’	19	83%	1	4%	3	13%	23	92%	1	4%	1	4%
Espérer ‘hope’	7	28%	12	48%	6	24%	6	26%	12	52%	5	22%
Estimer ‘consider’	8	35%	0	0%	15	65%	11	44%	0	0%	14	56%
Observer ‘observe’	23	92%	2	8%	0	0%	22	96%	0	0%	1	4%
Penser ‘think’	23	100%	0	0%	0	0%	25	100%	0	0%	0	0%
Raconter ‘tell’	25	100%	0	0%	0	0%	23	100%	0	0%	0	0%
Remarquer ‘remark’	24	96%	0	0%	1	4%	23	100%	0	0%	0	0%
Se douter ‘suspect’	5	20%	11	44%	9	36%	16	70%	4	17%	3	13%
Trouver ‘find, consider’	23	100%	0	0%	0	0%	24	96%	1	4%	0	0%
Voir ‘see’	23	100%	0	0%	0	0%	24	96%	1	4%	0	0%
Total	254	76%	27	8%	54	16%	263	78%	25	7%	47	14%

Analysis through logistic mixed-effects models

As in Experiment 1, in order to investigate how the distributive patterns of the matrix verb and its grammatical aspect influence mood choice, we constructed logistic mixed-effects models on the data in which participants’ answers were a subjunctive or an indicative. The models were built following the same procedure as in Experiment 1 on 1808 data points.

Adding the distributive patterns of the matrix verb as a fixed factor to the simplest model, which only included items and participants as random effects, significantly improved the model (Δχ² = 43.91, Δdf = 2, p < .001). Adding the grammatical aspect of the matrix verb as a fixed factor further improved the model’s fit (Δχ² = 6.26, Δdf = 1, p < .001). Finally, adding the interaction between the distributive patterns and tense of the matrix verb also further improved the model’s fit (Δχ² = 11.33, Δdf = 2, p < .001). According to this final model, we find in our data a significant main effect of the distributive patterns of the matrix verb, a significant main effect of its grammatical aspect and a significant interaction between them. Precisely, the significant main effect of the distributive patterns of the matrix verb confirms that, compared to the alternance category of verbs (which trigger the indicative in 64.3% and the subjunctive in 35.7% of the cases), subjunctive_only verbs trigger the use of the subjunctive (91.5%) (β = 3.4730, SE = 0.826, p < .001). The significant main effect of the grammatical aspect of the matrix verb reveals that the perfective aspect (here, the Passé Composé) significantly increases the probability to have an indicative in the complement clause (53.8%) compared to when the aspect is the imperfective (here, the Imparfait) (50.8%) (β = 0.9624, SE = 0.2473, p < .001). Finally, the significant interaction between the distributive patterns and the aspect of the matrix verb indicates that verbs from the subjunctive_only category which are expressed with the perfective aspect decrease the probability to use the indicative mood (7.2%) compared to when they are expressed using an imperfective aspect (9.8%) (β = –1.3158, SE = 0.4010, p < .001).

Analysis of the subset of alternance verbs

As before, we analyzed the subset of alternance verb to investigate the role of grammatical aspect independently of the semantics of the main verb. We built the models by going from the simplest model to the one of interest. Adding the grammatical aspect of the matrix verb as fixed factor to the simplest model, which only included items and participants as random effects, significantly improved the model (Δχ² = 16.547, Δdf = 1, p < .001). According to the model including the grammatical aspect of the matrix verb, having alternance verbs expressed in the Imparfait significantly decreases the probability to select the indicative mood (58.6%) compared to when they when they are expressed in the Passé Composé (70%) (β = –1.0047, SE = 0.248, p < .001).

4.1 The role of the distributive pattern of the matrix verb

The semantics of the predicate hypothesis was tested in Experiment 1 and 2. As shown in Table 11, the results of our experiments confirm it partially. More specifically, we find that in each of the three sets of verbs tested (indicative_only, subjunctive_only and alternance), there are verbs which behave as predicted by grammars and semantic studies but also verbs which do not follow the predicted pattern. In other words, our results reveal a partial discrepancy between the theoretical prescriptions in terms of mood choice and the speakers’ actual use of the indicative and the subjunctive. For example, Gosselin (2010) argues that verbs such as douter ‘doubt’, craindre ‘fear’ or souhaiter ‘wish’ must be followed by the subjunctive and the indicative is not acceptable. Our results show that speakers do use these verbs with the indicative (13% for douter ‘doubt’ when is inflected in the Passé Composé and 16% when inflected in the Imparfait, 6% for craindre ‘fear’ when inflected in the present tense, 9% in the Imparfait and 4% in the Passé Composé, and 6% for souhaiter ‘wish’ when inflected in the present tense and 8% when inflected in the Passé Composé). Another example is that of supposer ‘suppose’, imaginer (‘imagine’) and admettre ‘admit’ which are included in the alternance category because they require the indicative in affirmative sentences but change their requirement in negative, interrogative or imperative sentences (Gosselin 2010). Our findings show that these verbs accept the subjunctive in affirmative sentences as well (31% for admettre when inflected in the present tense and 16% when inflected in the Imparfait, 22% for imaginer when inflected in the Imparfait, 16% when inflected in the Passé Composé, 16% for supposer when inflected in the Imparfait and 13% when inflected in the Passé Composé). These results confirm the hypothesis put forth by Mari (2016), and then Giannakidou & Mari (2021), that fictional predicates license both the indicative and the subjunctive, and this due to their “systematic polysemy” consisting of expressive and inquisitive interpretations. In addition, in the case of imaginer and supposer, the indicative is not only in competition with the subjunctive but also with the conditional when inflected in the past tense (35% for imaginer with the Imparfait, 36% with the Passé Composé, 72% for supposer with the Imparfait, 61% with the Passé Composé). Crucially, to the best of our knowledge, this result has not been predicted by theoretical studies.

Furthermore, our results indicate that the variety of French tested does not follow the same pattern found for spoken Quebec French by Poplack et al. (2013). Indeed, throughout the two experiments, we have found that there are subjunctive governors which were followed exclusively by the subjunctive, such as attendre ‘wait’, permettre ‘allow’, proposer ‘propose, offer’, ordonner ‘order’, or almost exclusively by the subjunctive (>90% of the cases), such as apprécier ‘appreciate’, craindre ‘fear’, exiger ‘demand, require’, ordonner ‘order’ and souhaiter ‘wish’, when they are inflected in the past tense. This pattern is less systematic when these verbs are inflected in the present tense, with the exception of attendre ‘wait’, permettre ‘allow’, proposer ‘propose, which are exclusively followed by the subjunctive (Table 5). Regarding the category of indicative_only, Experiment 1 showed that all the verbs tested behaved as predicted by grammars and semantic studies, with rates of indicative between 99%-100%, with the exception of croire ‘believe’, which was followed by the indicative in 94% of the cases. Unexpectedly, Experiment 2 revealed a strong interaction between grammatical and distributive patterns of the verb: the rate of the indicative for most of the indicative_only verbs dropped significantly compared to Experiment 1 while the rates of the subjunctive and the conditional increased. This pattern is observed for decider ‘decide’, espérer ‘hope’, estimer ‘consider’ and se douter ‘suspect’ (cf. Table 10).

In sum, we argue that these sets of verbs are more heterogenous that initially supposed and that they should be rethought. We follow Comeau (2020), who argues that the subjunctive is governed by the meaning or the semantics of the matrix verb. In other words, the role of the matrix verb goes beyond simply belonging to fixed lexical constructions but this has been hidden by a methodological issue generally adopted in corpus-based studies. As put by Comeau:

“By limiting the analysis to those contexts which can select the subjunctive (the main clause verbal and non-verbal elements), we do not clearly see whether these elements belong to larger semantic classes in the language or whether they are the sole members of such classes. To accurately assess the semantic aspects of these main clause elements, one would have to consider all main clause elements which select a finite embedded clause within a corpus. By considering whether there are other main clause verbs which fall into the same semantic classes (volition, doubt, etc.), then we might be able to see whether it is the larger semantic class of verbs which can express a similar meaning that selects the subjunctive or only a particular subset of this class (namely, those verbs which have been found to select the subjunctive).” (2020: 42).

We thus put forward that the formulation of homogenous, semantically justified, classes of matrix verbs should be targeted, as the simplified method of applying a label (subjunctive-selecting or indicative-selecting) is not nuanced enough to account either for the empirical data or for the fine-grained analyses put forward in formal semantics (e.g. Mari 2016, Giannakidou & Mari 2021; cf. discussion in section 4.4). We leave this objective for future research. At the same time, it is crucial to remember that the distributive pattern of matrix verbs as indicative governors, as subjunctive governors and, respectively, as alternance verbs is strongly influenced by the tense and the grammatical aspect used in the matrix clause (as we discuss in sections 4.2 and 4.3).

4.2 The role of tense

The tense hypothesis was tested in Experiment 1. This experiment, in which the matrix verbs were inflected either with a past or a present tense, showed the following findings. First, the subjunctive is in strong competition with the indicative but also with the conditional, and this both in what concerns the alternance and the subjunctive_only sets of verbs. Importantly, this effect is observed almost exclusively when the matrix verb is in the past tense. Two typical examples are imaginer (from the alternance set), which was followed by the conditional in 69% of the cases when it was inflected in the past tense compared to only 3% when in was inflected in the present tense (Experiment 1, Table 4), and approuver (from the subjunctive set), which was followed by the conditional in 19% of the cases when it was inflected in the past tense but only in 3% when it was inflected in the present tense (Experiment 1, Table 5).

Second, we found that, in general, there is a significant effect of the tense of matrix verb: past tense significantly boosts the indicative whereas present tense boosts the subjunctive. However, this effect comes from the alternance set of verbs. We thus find that when verbs from this set are inflected in the present tense the probability to have the subjunctive in the complement clause significantly increases (54%) compared to when they are inflected using a past tense (30%). For example, the verbs comprendre ‘understand’, concevoir ‘understand’, nier ‘deny’, rêver ‘dream’ frequently select the subjunctive (57% of the cases for comprendre and concevoir, 60% for nier, 94% for rêver) when they are inflected in the present tense whereas they select this mood less frequently (0% for comprendre and concevoir, 3% for nier, 3% for rêver) when they are inflected in the past tense. In contrast, the opposite effect is found for verbs from the subjunctive_only set. We find that when verbs from this set were inflected with a present tense the probability to have the subjunctive in the complement clause decreased compared to when they were inflected using a past tense. For example, the verbs apprécier (‘appreciate’) and redouter (‘dread’) select the subjunctive in 86% of the cases when they are inflected in the present tense whereas they selected this mood in 97% and 94%, respectively, when they were inflected in the past tense.

In sum, Experiment 1 reveals the crucial effect of tense on the mood choice and, crucially, its interaction with the distributive patterns of the matrix verb (as discussed above).

4.3 The role of grammatical aspect

The grammatical aspect hypothesis was tested in Experiment 2. This experiment, in which all matrix verbs were either in the past perfective or the past imperfective, showed the following results. First, it confirmed that the subjunctive is in strong competition with the indicative but also with the conditional, and this both in what concerns the alternance and the subjunctive_only sets of verbs. Second, we found a significant effect of grammatical aspect of the matrix verb: across all three sets of matrix verbs, the perfective aspect boosts the use of the indicative. Third, there was a significant interaction between the distributive pattern and the aspect of the matrix verb: for verbs from the subjunctive_only category expressing the perfective strongly decreases the probability of using the main competing mood (the indicative) compared to when they express the imperfective. For example, the verbs craindre ‘fear’ and redouter ‘dread’ select the indicative in 4% and 9%, respectively, of the cases when the matrix verb is in the perfective and 10% and 24%, respectively, of the cases when the matrix verb is in the imperfective. Fourth, for the alternance distributive pattern only, the use of an imperfective matrix verb significantly decreases the probability to use the indicative mood compared to when they are inflected in the perfective. For example, the verbs rêver ‘dream’ and s’assurer ‘make sure’ select the indicative in 9% and 26%, respectively, of the cases when the matrix verb is inflected in the Imparfait whereas they select the indicative in 52% and 76%, respectively, of the cases when the matrix verb is inflected in the Passé Composé.

At a general level, our results confirm the aspect hypothesis and show that the variety of French tested does not follow the Greek/Bulgarian pattern (cf. discussion in 2.3) but that of the spoken Quebec French of the 20^th century. The main effect of grammatical aspect is concordant with that of tense and they map mood choice straightforwardly. This mapping may be related to the fact that past/perfective/indicative seem to form one class with respect to veridicality whereas the present/imperfective/subjunctive form a different one. We leave this open for future research.

Furthermore, what has not been predicted is the direction of the interaction effect between grammatical aspect and the distributive patterns of the predicate, and more specifically, the effect on the subjunctive_only set of verbs compared to that of alternance verbs. As observed by one anonymous reviewer, our findings reveal that the subjunctive is boosted with present tense on alternance verbs while present tense decreases its probability with subjunctive selecting verbs. Likewise, the probability of the indicative decreases with alternance verbs when inflected in the imperfective while the perfective decreases the probability of the indicative with subjunctive verbs. We address this mood choice puzzle in the following section.

4.4 The interrelations between tense/aspect and the distributive patterns of the matrix verb

As suggested by one anonymous reviewer, one plausible explanation for this puzzle pertains to the ambiguity of different predicate types which belong to the subjunctive_only or the indicative_only distributive patterns. Indeed, as previously shown by Mari (2016) and Giannakidou & Mari (2021), some verbs (such as predicates of belief and fictional predicates) are in fact polysemous (cf. also Gosselin’s 2007 classification of verbs that present a change in meaning when followed by the indicative or the subjunctive, as discussed in section 2.1 with reference to the verb credere ‘believe’). Mari’s (2016) distinction between expressive (a doxastic dimension) and inquisitive (which features both a doxastic and an epistemic dimension) interpretations holds also for the bouletic predicate espérer ‘hope’, as shown in examples (38) and (39). In (38), the hearer infers that it is already decided where the Olympics will take place and s(he) does not know it. Here, espérer has an inquisitive use, i.e. the matter is settled, and it is followed by a subjunctive. In contrast, in (39), the hearer infers that it is not already decided where the Olympics will take place. Here, espérer has an expressive use, i.e. the matter cannot be verified, and it is followed by an indicative.

In this context, the effect of tense and grammatical aspect on mood choice, and its interaction with the distributive patterns of the matrix verb, may be better understood. In the case of subjunctive_only verbs, the present/imperfective decreases the use of the subjunctive to convey that in the complement clause there is no fact of matter at stake. For example, in our data from Experiment 2, the verb redouter ‘dread’ is classified as a subjunctive_only but it triggered the indicative for 8 observations, among which 6 observations corresponded to cases when the matrix verb is in the Imparfait. In contrast, in the case of alternance verbs, the present/imperfective increases the use of the subjunctive to convey that in the complement the matter is settled. For example, admettre ‘admit’ is classified as alternance verb (thus, we expect a variable pattern regarding the selection of the indicative and the subjunctive) but it triggered 16% subjunctive responses (4 observations), which all correspond to cases when the matrix verb is in the Imparfait.

The pairs of examples (40)–(41) and (42)–(43) illustrate the two possible interpretations of redouter and admettre, when they are followed by the subjunctive and, respectively, the indicative. Examples (40) and (42), the verbs redouter and admettre have an inquisitive use (i.e. a matter of fact is at stake) and the subjunctive is used in the complement clause to convey that the matter is settled. In contrast, in (41) and (43) the verbs redouter and admettre have an expressive use (i.e. there is no fact of the matter at stake) and the indicative is used in the complement clause to convey that the matter is not settled.

As such, our results reveal that verbs are much more polysemous than initially believed, and that this polysemy (linked to inquisitivity vs. expressivity) often determines their mood choice behavior.