2.1 ‘Counterfactuality’

I would like to argue that the widespread term ‘counterfactuality’ is confusing because it refers to two different interpretations conveyed by distinct constructions, only one of which can be said to be truly counterfactual. I thus make the distinction between two epistemic interpretations pertaining to what the speaker knows or believes at the time of utterance: (i) the contrary-to-fact interpretation of proposition p—the genuine counterfactual interpretation—(cf. (1)), and (ii) the interpretation according to which p’s realization is unlikely (cf. (2)). Examples (1) and (2) are repeated below:

(1) if JFK had not been assassinated, he would obviously have been re-elected. (The Spectator)

⇒ ¬p (counterfactuality)

(2) And if you left me I would suffer a great deal. (R. Jaffe, After the reunion)

⇒ ¬πp (unlikelihood)

I henceforth note the ‘true’ counterfactual interpretation ¬p (‘it is not the case that p’) and the unlikelihood interpretation ¬πp (‘it is not probable that p is the case’). For the sake of convenience, and in accordance with a frequent usage in linguistics, I will continue to use the term of ‘counterfactuality’ with quotation marks when referring to both interpretations, but use ¬p and ¬πp to refer individually to one or the other interpretation.

At first sight, one may think that the distinction between ¬p and ¬πp reflects a morphological difference between ‘two-past’ and ‘one-past’ contexts (see again (1) and (2)). However data from Romance languages show that ‘simple’ imperfects (i.e. imperfective pasts) may also license ¬p, notably within indicative conditionals of the type [if pluperfect/imperfect, imperfect]. Here are examples from Italian, French and Spanish:³

(3)

Se	arrivavi	prima,	vedevi	il film	dall’inizio.	[Ita]
If	arrive.ipfv.pst	earlier	see.ipfv.pst	the movie	from the beginning

‘If you had arrived earlier, you would have seen the movie from the beginning.’ (Ippolito 2004)

(4)

Si	je	n’ étais pas intervenu,	ça	se finissait	au couteau.	[Fre]
If	I	not intervene.prf.ipfv.pst,	it	end up.ipfv.pst	with stabbing

‘If I hadn’t intervened, it would have ended up with stabbing’ (Patard 2007)

(5)

Si	lo	sabía,	venía.	[Spa]
if	it	know.ipfv.pst	come.ipfv.pst

‘If I had known that, I would have come.’ (Martínez-Atienza 2012)

These examples attest that the interpretation ¬p is not only met in ‘two-past’ contexts, but may also be found in ‘one-past’ contexts. I will argue in section 4 that ‘one-past’ counterfactuality can only be conveyed by imperfects (imperfective pasts), hence its absence in a Germanic language like English.

Finally, I must signal that, contra a common conception in the formal literature, I do not consider that, in irrealis examples like (6) and (7), the fake tense licenses ¬p (counterfactuality):

(6) if I had longer arms I would push the clouds away. (Task Force)

(7) I wish he was dumb. I wish he was deaf. I wish he was blind. (C. Dickens, Master Humphrey’s clock)

My argument is that fake tense conveys ¬πp (unlikelihood) just like in the previous example (2). If, however, an irrealis reading is obtained, it is because fake tense enables it: an unlikely situation may prove counterfactual without contradictions, and, crucially, because it is inferred from the speaker’s world knowledge and the stative aspect of the predicate (cf. [I have longer arms], [he be dumb], [he be deaf] etc.) (see R. Martin 1991, Gosselin 1999, Iatridou 2000 or Arregui 2007 for more detailed analyses). In contrast, contexts of epistemic ignorance/future reference and/or telic predicates generally license unlikelihood (and not irrealis) (cf. (8) and (9)). These are examples known as ‘Future Less Vivid’:

(8) If you missed the last bus, you would walk home.

(9) Mary wishes that Peter could come to her birthday party next week.

These observations show that the modal interpretation of fake tense in ‘one-past’ contexts—whether irrealis (cf. (6) and (7)) or future-less-vivid (cf. (8) and (9))—is not coded in the past morphology but is co(n)textually implicated. I will detail the inferential process in section 4.2.

2.2 A Functional and Diachronic Perspective

Contrary to the recent works in formal semantics, I will not attempt to define the truth conditions of fake tenses, nor utilize the Kripkan notion of possible worlds. The reason for that is double. First, I claim that it is quite possible to give an operational description of the semantics and pragmatics of fake tense without using the formalism of modal logics, provided the description is based on an adequate formal apparatus including a precise metalanguage. Second, I consider that the semantics of modal constructions in natural languages (such as the ones responsible for fake tense and ‘counterfactuality’) is not reducible to their truth conditions, which may either be too inclusive (some aspects of their truth conditions do not pertain to their semantics proper) or too exclusive (some aspects of their semantics cannot be formalized in terms of truth conditions). Crucially for the present issue, the epistemic category of unlikelihood (¬πp), which is fundamental in my view to account for examples such as (2) (see section 4), cannot be easily captured (if at all) by quantification over possible worlds.⁴ I thus adopt a more functional point of view and will be concerned with how linguistics structures serve to communicate information. The present paper will consequently aim at understanding how (fake) past tenses may contribute to evoke ‘counterfactual’ situations in discourse.

I further approach fake tense from a diachronic perspective with the usage-based underlying idea that “language structure emerges from language use” (Tomasello 2003: 5). Hence, the paper seeks to answer the question: how have fake tense and counterfactuality emerged from the (normal) usage of past tenses? This very research question comes up with a few common but nontrivial observations. First, in language with temporal and modal usages of past tense, the default and prototypical interpretation is the temporal one. Second, cross-linguistic studies show that the past tense alone generally fails to convey ‘counterfactual’ interpretations, unless it is combined with modal markers such as if and would in English counterfactual conditionals (cf. James 1982, Fleischman 1989, Van linden and Verstraete 2008). Third, as emphasized by James (1982) (see also Dahl 1997 and Hogeweg 2009), temporal uses of past tenses are ‘regular and productive’ and thus predictable while modal uses are ‘irregular and idiosyncratic’ and consequently less prone to predictions. Indeed, there seems to be considerable variation in the way languages express ‘counterfactuality’ with past tense morphology (see for instance Van Linden and Verstraete 2008, Boogaart and Trnavac 2011).

From these observations, I hypothesize that (i) fake tenses have developed from ‘real’ tense uses and (ii) surfaced with the creation of (idiosyncratic) constructions whose meaning is not compositional but derives from conventionalized implicatures. Thus, the particular structures of these constructions, which I assume to be responsible for fake tense and ‘counterfactuality’, are (at least partly) determined by the initial conditions at the time of their emergence, i.e. the TAM paradigms of the language in question and their ongoing evolutions. According to these hypotheses, one cannot strictly predict from the synchronic TAM paradigms which verbal forms (whether subjunctive or past indicative) will serve to express ‘counterfactuality’ (see for instance Ippolito 2009 or Karawani 2014 for such predictions). Indeed, the diachrony of the language in question should also be taken into account (cf. the diachronic analysis of French and Latin data in section 5). However, predictions can be made from the proposed inference-based analysis as to the semantics of past tenses with which ‘counterfactuality’ may (or may not) obtain. Thus, I will argue in section 4 that (i) ¬p (the ‘true’ counterfactual interpretation) is only licensed by unbounded pasts and that (ii) ¬πp (the unlikely interpretation) cannot be found with perfective pasts. The proposal is based on a neo-Reichenbachian conception of past tenses which is elaborated in the next section.

3.1 Past Tenses Encode Past

Following the temporalist view, my first hypothesis is that past tenses encode anteriority to the time of speech; their meaning is fundamentally temporal, not epistemic (contra Iatridou (2000) and the Cognitive Grammar view). This approach is supported by the facts recalled in the preceding paragraph: past is the default and prototypical interpretation and is found in contexts that are regular, productive and predictable, whereas the modal interpretations are only obtained with additional modal markers, in contexts that are irregular, idiosyncratic and thus much less predictable.

Another fact is that the modal readings carried by past tenses may be cancelled by the subsequent context. Cancellation proves possible in conditionals yielding ¬πp (cf. (10)), as has been amply shown in the logical literature (see Anderson 1951 or Stalnaker 1975), as well as in ‘imperfective paradox’ contexts yielding ¬p (cf. (11)):

(10) If the patient had the measles, he would have exactly the symptoms he has now. We conclude, therefore, that the patient has the measles. (Iatridou 2000)

(11) Paul se noyait.ipfv.pst lorsqu’ un sauveteur a plongé pour le sauver. Mais Paul s’ est noyé quand même. [Fre]

‘Paul was drowning when a rescuer dived in to save him. But Paul drowned all the same.’

In (10), the preterit conveys a sense of unlikelihood (measles is considered as unlikely), which is cancelled afterwards by the speaker’s conclusion (We conclude, therefore, that the patient has the measles). Similarly, in (11), the French imperfect insinuates that the drowning finally did not happen, but this interpretation is overridden by the following sentence (Mais Paul s’ est noyé quand même). The possibility to cancel the modal readings in such contexts suggests that the modal meaning is not asserted but conversationally implicated⁵ (see again Anderson 1951 or Stalnaker 1975).

I would like to argue now that the cancellability of the modal interpretation (hence the fact that it is an implicature) is incompatible with the non-temporal and monosemous approach defended by Iatridou (2000) or cognitive grammarians. In these theories, past tenses encode a non-temporal abstract meaning (cf. Iatridou’s exclusion feature) from which one could contextually derive a past or modal interpretation. On this view, the final interpretations of past tenses are specific instantiations of the abstract meaning, elaborated thanks to contextual information. Then, when the (specific) modal interpretation is canceled by the following context, one may expect that the abstract meaning encoded by the past tense is reinterpreted differently. How then would it contribute to the revised interpretation? The theory would predict that the past tense should either yield the default reading, i.e. the past interpretation (cf. Brisard 2010), or possibly induce an underspecified reading whereby the domain on which the exclusion feature operates would remain undetermined.

However this is not exactly what happens in the quoted examples. In the revised reading of (10), there is no exclusion to be interpreted, be it from the ‘actual time’ (the speaker’s ‘present reality’) or from the ‘actual world’ (the speaker’s ‘actual reality’): the topic time/world just coincides with the actual time/world and does not exclude it (We conclude, therefore, that the patient has the measles). This leads us to the nihilist conclusion that the past tense has no semantic content but only carries implicatures, which one may not want to. Example such as (11) is also problematic for the aforementioned theories, but for a different reason. In (11), the imperfect initially licenses both a past and modal interpretation (the denoted drowning is past and counterfactual: it did not happen in the past) before being reinterpreted as expressing past only (the drowning effectively took place in the past). On the non-temporal view, such an example would suppose that, at first, the past morphology is interpreted twice, licensing past AND modality, as though there was two layers of past tense marking as in ‘two-pasts counterfactuals’ (e.g. If Paul had drowned, Marie would have been inconsolable). But then how come a single tense morpheme is used to convey past AND modality? The theory should explain this exception to the complexity principle of iconicity, i.e. why is a complex meaning expressed by a simplex morphology (and not by a complex one)? Furthermore, the possibility to cancel the modal interpretation in contrast to the past interpretation (see below) rather hints at a different semantic status of the past and modal interpretations. While the modal reading appears as a cancelable implicature derived from the context, the default past reading definitely has another status which has to be accounted for.

Indeed, pastness conveyed by simple past morphology proves very difficult (if not impossible) to eliminate. When a non-past reinterpretation is possible, the validity of the situation is in fact extended beyond the past domain. As a result, the denoted situation may be interpreted to hold both in the past and in the present (12) or to have a generic validity (13):

(12) The patient was sick, but, in fact, he still is.

(13) John was sick, but, in fact, he is always sick.

In either case, the situation is still valid in the past, i.e. the past interpretation is maintained. Then, if modality can contextually be cancelled but not past reference, one should come to the conclusion that the past interpretation is not just a contextual elaboration of non-temporal core meaning à la Iatridou or à la Langacker, but that it is encoded by the past tense. I shall elaborate the proposed semantic description of past tenses in the following section that outlines the main points of a previous description (Patard 2014).

3.2 Meaning and Default Interpretation of Past Tenses

In the wake of Klein (1994) or Gosselin (1996) and adopting a Reichenbachian terminology (Reichenbach 1947), I take past tenses to express relations between three moments: the time of speech S, the reference time or reference interval R, and the eventuality or situation time E. Every past tense first encodes a deictic relation with the time of speech: R precedes S, i.e. the past tense refers to a moment or interval that is anterior to the time of speech. A second relation is optionally expressed by past tenses and corresponds to their aspectual import, it is the relation between R and E. In the case of imperfects, R is included within the duration of E (the so-called ‘imperfective aspect’). In the case of preterits, the relation between R and E is left unspecified, which may be dubbed ‘neutral aspect’ (see Smith 1991/1997). The aspectual interpretation of the situation is then contextually determined, most often by the actionality of the predicate. The schematic meaning encoded by imperfects and preterits is summarized in (14) and (15):

(14) imperfects

[R < S]	[past time reference]6
[R ⊂ E]	[imperfective aspect]

(15) preterits

[R < S]	[past time reference]
[R 0 E]	[neutral aspect]

In the proposed schematic meaning, R is an abstraction, an underspecified time point, but it can get specific instantiations in context. For instance, in the prototypical use of past tenses, R specifically refers to:

1. a past topic time (Tt) (see Klein 1994), i.e. the past moment about which the speaker says something or asks a question;

2. the aspectual vantage time (Ap) from which the internal time of the eventuality is considered (e.g. from an inner perspective in the case of imperfective aspect).

In other contexts, R may instead instantiate:

1. an epistemic evaluation point (Ep):⁷ this is another sort of vantage point from which the speaker vouches for the validity of the proposition or V (it be true that [p]) because of what she knows or believes at the time. V (it be true that [p]) can be viewed as a meta-predicate scoping over the predicate expressed in the proposition, it hence refers to a situation (one could say a ‘meta-situation’) that is located on the timeline and is seen from a certain aspectual perspective (just like ‘normal’ situations expressed by predicates). V can be regarded as a stative eventuality as it involves no change over time and possesses no intrinsic endpoint.

By default, the speaker commits to V at speech time S: the speaker normally believes in the present that what she says is the case (={it is true that [p]}).⁸ This is illustrated in example (16) with a prototypical past interpretation.

(16) Max ran fast.

The corresponding aspecto-temporal interpretation is diagrammed in figure 7.1.

View Full Size

Figure 7.1

Aspectual and temporal interpretation of V (it be true that [p]) and E (Max run fast) in the utterance ‘Max ran fast’.

• Download Figure
• Download figure as PowerPoint slide

The default interpretation in (16) is {it is true that [Max run.past fast]}. Thus, E (Max run fast) received the default interpretation of the preterit combined with a telic situation, i.e. the past and perfective interpretation: {R < S} {R = E}. As for V (it be true that [p]), it is given a present and imperfective characterization, i.e. it is assessed from an evaluation point Ep that coincides with the time of speech ({Ep = S}) and that is included within V ({Ep ⊂ V}).

However, in other contexts, the speaker does not commit to V at the time of speech but the validity of the proposition is endorsed in the past, which means that the evaluation point Ep is not present but past. Such contexts are discussed in the next paragraph.

3.3 Marked Interpretation of Past Tenses

Consider the following examples from Dutch where the preterit kwam is used in a past indirect speech to refer to a telic situation:

(17) Marie zei dat Jan morgen kwam.pst. [Dut]

‘Mary said that John was coming tomorrow.’

View Full Size

Figure 7.2

Aspectual and temporal interpretation of V (it be true that [p]) and E (Jan morgen komen) in the utterance ‘Marie zei dat Jan morgen kwam’.

• Download Figure
• Download figure as PowerPoint slide

In (17), the interpretation of the past tense is almost the reverse of the one obtained in (16): here V receives a past interpretation ({it was true that [p]}) whereas E gets a non-past interpretation (signaled by the adjunct morgen ‘tomorrow’). Contrary to (16), the past R denoted by the past tense coincides with the evaluation point Ep (R= Ep): it is not the actual speaker who commits to the proposition’s validity V but the past speaker (Marie). In other words, the past tense does not bear on the situation E anymore but applies to V.⁹ This means that the past tense has a proposition-external scope, which is manifested by fake tense: despite the past morphology, the situation is not grounded in the past domain, but may occur in the present or future (because R ≠ Tt). It is easy to show that this type of context also licenses fake aspect. Take the French example with an imperfect:¹⁰

(18) Marie a dit que Jean restait.ipfv.pst demain jusqu’ à quatre heures.[Fre]

‘Marie said that John was staying tomorrow until four o’clock.’

Here, the imperfective morphology (restait) does not impose an internal perspective on the situation. That is why it is acceptable to use a delimitative adverbial licensing a perfective reading (jusqu’ à quatre heures ‘until four o’clock’). This shows that the past R denoted by the past tense is also dissociated from the aspectual vantage point Ap (R ≠ Ap), thus allowing for a perfective interpretation of imperfective morphology.

Fake tense and fake aspect may be indicative of the fact that past tenses do not scope over the proposition but over V, the validity of p. I will argue in section 4.3 that the unlikely conditionals illustrated in (2) are comparable to the preceding uses with regard to the functions of R: in those contexts, R crucially functions as an evaluation point Ep while its reference to a topic time (Tt) or an aspectual vantage point (Ap) is optional. In those contexts, it is the interpretation of R as Ep that originally triggers the unlikely implicature ¬πp.

4.1 Scalar Implicatures

The proposed analysis builds on the idea formulated in previous work (see for instance Ziegeler 2000, Ippolito 2004 or Verstraete 2006) that past tenses may locally give rise to scalar modal implicatures that are derived from Grice’s quantity maxim: ‘Do not make your contribution more informative than is required’. Past tenses are seen as less informative than alternative tense forms that could be used in the same contexts, thus forming with them a scale of informativeness. The marked under-informative use of the past tense then triggers a Q-based implicature (Horn 1984, 1989), i.e. an ‘upper-bounding’ inference according to which the hearer should not interpret beyond what is said. The implicatum is not necessarily that anything beyond what is said does not hold (the strong implicature ¬p), but—and this is the main difference with the authors quoted above—the implicatum may also be that the unmarked alternative tense provides too much information considering what the speaker knows, i.e. the speaker is not certain about whether p is valid or not (the weak implicature p or ¬p).¹¹ It is the latter that is at the origin of the implicature ¬πp conventionalized in the unlikely use of past tenses. I will further argue that the difference between the strong implicature ¬p and the weak implicature p or ¬p is dependent on the types of informativity scale the past tense forms with the alternative tense.

4.2 Strong Implicature ¬p

The strong implicature ¬p, which is the genuine counterfactual interpretation, is only obtained with imperfects, i.e. imperfective pasts, because it is triggered by the imperfective aspect. Indeed, imperfective pasts, or more generally unbounded pasts, form an informativity scale with ‘bounded pasts’ in regards to the completion of the situation denoted by the predicate. What I call ‘bounded pasts’ are either perfective pasts (like the French passé simple) that give a global perspective on the unfolding of the situation including its boundaries, or perfects referring to anterior situations (like the French passé composé) that rather focus on the resulting state beyond the final boundary. In either case, bounded pasts assert or presuppose that the situation has reached its natural endpoint and is completed. By contrast, unbounded pasts (like imperfects) focus on subintervals of the situation excluding its boundaries (R ⊂ E in our terminology), thus giving an inner perspective on the situation that is blind to its outcome. As a consequence, nothing is said about whether the situation has reached its natural endpoint. Bounded pasts are thus more informative than unbounded pasts as to the completion of the denoted situation. Hence the scale of informativity:

(23) Scale of informativity I (completion of the situation)

⟨ unbounded past, bounded past ⟩

By virtue of (23), the use of an imperfect may contextually implicate the negation of the content conveyed by an alternative bounded past. Thus, when using a less informative imperfect, the speaker signals that the use of a bounded past would be over-informative and that (s)he is not in a position to endorse the completion of the past situation. The speaker then implicates ¬p_{bounded past}: it is not the case that p, with p using a bounded past. Hence the following characterization of the strong implicature ¬p:

(24) Strong implicature

⟨ unbounded past, bounded past ⟩

unbounded past ⇒ ¬p_{bounded past}

As a Q-based implicature, ¬p_{bounded past} is locally derived in marked contexts where the unmarked alternate could not have been used appropriately (Horn 2004: 16). The markedness of the context in our case is due to the unusual combination of an imperfect with a telic predicate. In their unmarked use, imperfects apply to atelic predicates (states or activities) and entail that the situation was the case in the past (Fre Pierre courait ‘Pierre was running’ → Pierre a couru ‘Pierre ran’). When combined with a telic predicate (achievement or accomplishment), one comes up with ‘the imperfective paradox’ (Dowty, 1979): the imperfect does not entail that the situation occurred (Fre Pierre courait le marathon ‘Pierre was running the marathon’ ↛ Pierre a couru le marathon ‘Pierre ran the marathon’), thus implicating that the situation did not occur (⇒ Pierre n’ a pas couru le marathon ‘Pierre did not run the marathon’). Hence, the counterfactual implicature is only triggered in the marked context of ‘imperfective paradox’:

(25) Strong implicature

⟨ unbounded past, bounded past ⟩

unbounded past ⇒ ¬p_{bounded past} / ‘imperfective paradox’

The following utterance from Spanish is a typical example of Romance imperfect triggering implicature ¬p:

(26) Salía.ipfv.pst (vs salí.pfv.pst) del trabajo, cuando el jefe me llamo. [Spa]

‘I was leaving work when the boss called me.’ (Ferraro and Ortiz Lopez 2002)

The imperfect salía (‘was leaving’) signals that the scalar alternative sali (‘left’), which is a perfective past, is too informative. As a consequence, salía implicates ¬salí: the speaker finally did not leave work.

This implicature is also found in the so-called ‘counterfactual use’ of the French imperfect:

(27) Une seconde de plus il [le taureau] l’ éventrait.ipfv.pst (vs éventra.pfv.pst). [Fre]

‘One more second and the bull would have gored (/ gored) him.’ (G. Flaubert, Un cœur simple)

Éventrait (‘was goring’) here indicates that the alternative perfective past éventra (‘gored’) conveys too much information and therefore implicates ¬éventra. I will argue in section 5.2 that the implicature ¬p has conventionalized in such use.

The proposed analysis leads to the prediction that pasts that are not unbounded—i.e. perfective pasts, anterior perfects but also preterits (that can get a bounded interpretation)—cannot convey implicature ¬p because they cannot produce an ‘imperfective paradox’. However, this does not mean that their use in counterfactual contexts is excluded, only they cannot contribute to counterfactuality. However, for this very reason, one may predict that unbounded pasts will be preferred to bounded pasts in counterfactual contexts.

4.3 Weak Implicature p or ¬p and Implicature ¬πp

Implicature ¬πp is contextually derived from the weak implicature that the speaker is not certain about the validity of p, i.e. does not know whether p or ¬p. This weak implicature is itself obtained with both imperfects and preterits because it is triggered by the past meaning. I will argue that the weak implicature is licensed by the proposition-external scope of the past morpheme (manifested in fake tense and fake aspect, see supra 3.3) in contexts presenting similarities with indirect speech.

I should remark at this point that perfective pasts do not allow for a proposition-external-scope reading. Let us consider the following French translation of example (17):

(28) Marie a dit que Jean venait.ipfv.pst (*vint.pfv.pst) demain. [Fre]

‘Marie said that Jean was coming (*came) tomorrow.’

The imperfect venait does not scope over p (both tense and aspect are fake) but over p’s validity: the coming of Jean is endorsed by Marie in the past. The French past perfective vint is unacceptable in the context because it immediately scopes over p (the situation is past and viewed in its entirety), which is incompatible with the future adjunct demain ‘tomorrow’. This example illustrates that perfective pasts are never fake, they can never get a higher scope above the proposition. I suggest the reason is because V (p’s validity) is a state-like situation (see supra) the evaluation of which (Ep) requires an imperfective viewpoint (Ep ⊂ V, see supra again). This would predict that imperfective pasts (like imperfects) and aspectually-unspecified past (like preterits) can get a proposition-external-scope reading, but not bounded pasts (like the French perfective past or the French perfect). Consequently, only imperfects and preterits should allow for the weak implicature p or ¬p (the speaker is not certain about the validity of p) and hence for the implicature ¬πp it may lead to.

Non-perfective pasts with a proposition-external scope form an informativity scale with present tenses as to the present evaluation of p’s epistemic validity. Non-perfective pasts are less informative because p’s validity is not endorsed in the present but in the past (Ep < S) while, with present tenses, the speaker vouches for the situation’s present validity (Ep = S). Hence the informativity scale:

(29) Scale of informativity I (evaluation of p’s validity)

⟨ non-perfective past, present ⟩

By virtue of (29), the use of a non-perfective past indicates that a present tense would be too informative, that the speaker is not in position to commit to p’s validity in the present or, put differently, that (s)he is not certain about p’s present validity. Thus, a less informative past implicates p_present or ¬p_present, that the present validity of the situation is uncertain. Crucially, p_present or ¬p_present might contextually implicate unlikelihood: because the speaker does not endorse the situation’s validity in the present, (s)he may suggest believing that the situation is unlikely (¬πp_present).¹² Hence the following characterization of the weak implicature p or ¬p:

(30) Weak implicature p or ¬p

⟨ non-perfective past, present ⟩

non-perfective past ⇒ p_present or ¬p_present (⇒¬πp_present)

Like the strong implicature, the implicature p_present or ¬p_present is only locally obtained in contexts where the non-perfective past is marked, but where the unmarked alternate could not have been appropriate to convey the targeted meaning. The markedness of the configuration here lies in the unusual use of a past tense to refer to a non-past situation in contexts where past is fake (and so is aspect). These marked contexts are characterized by a proposition-external scope of the past tense which bears on the epistemic validity of p (cf. section 3): the past R denoted by the past tense functions as an evaluation point Ep (R= Ep), which means that the validity of the situation is endorsed in the past and not, as it is expected in normal communication, in the present. We have seen previously that the scoping over V corresponds to a proposition-external scope of the past tense as met in indirect speech (which may be manifested by fake tense and fake aspect). Hence the final characterization of the weak implicature p or ¬p:

(31) Weak implicature p or ¬p

⟨ non-perfective past, present ⟩

non-perfective past ⇒ p_present or ¬p_present (⇒¬πp_present) / proposition-external scope

Such configuration is typically found in evidential contexts presupposing a past source. Here is an example from Italian:

(32) Domani Paolo lavorava.ipfv.pst (vs lavora.prs) fino alle sei del pomeriggio. [Ita]

‘Tomorrow Paolo was supposed to work until six p.m.’

With the use of the imperfect lavorava, the speaker refers to some past evidence attesting the validity of a non-past situation. More formally, the imperfect refers to a past evaluation point (R = Ep) from which p’s validity is considered (Ep < S and Ep ⊂ V). Thence, as the imperfect does not directly scope over p, tense is fake: the situation is localized in the future thanks to the adjunct domani (‘tomorrow’). And so is aspect: the delimitative adjunct fino alle sei del pomeriggio (‘until six p.m.’) allows for a perfective interpretation of the imperfect. Formally, one gets: R ≠ Tt ∧ Ap, which means that R does not coincide with topic time and the aspectual vantage point. The aspectuo-temporal analysis of example (32) is shown in figure 7.3.

View Full Size

Figure 7.3

Aspectual and temporal interpretation of V (it be true that [p]) and E (Paolo lavorare fino alle sei del pomerrigio) in the utterance ‘Domani Paolo lavorava fino alle sei del pomeriggio’.

• Download Figure
• Download figure as PowerPoint slide

By using the imperfect lavorava, the speaker signals that the alternative assertion with the present tense (lavora) is beyond what (s)he can vouch for and hence implicates that the validity of p_lavora is uncertain: it is not sure whether Paolo is going to work until 6 p.m. This type of context is also well attested for the imperfect Spanish (see Leonetti and Escandell-Vidal 2003) and, to a lesser extent, for the French imperfect (see Patard 2012).

In conclusion, the ‘counterfactual’ interpretation of past tenses can be viewed as two distinct implicatures of scalar origin that are invited in discourse to meet specific communicative aims: to refer to a counterfactual situation or to convey the speaker’s belief that the situation is unlikely. These implicatures are drawn from different informativity scales and obtained in different contexts of markedness. Strong implicature ¬p (or genuine counterfactuality) is triggered by imperfects in contexts where a bounded past is expected (the so-called ‘imperfective paradox’). As for implicature ¬πp (or unlikelihood), it is derived from the weak implicature p or ¬p that is itself triggered by imperfects or preterits in contexts where a present tense is expected. In those contexts, tense (and aspect) may be ‘fake’ because of the proposition-external scope of the past tense.

In the next section, I will substantiate the proposed analysis by examining data concerning the French and Latin imperfect. I will further suggest a historical scenario whereby the conventionalization of the ‘counterfactual’ implicatures has led to the aspectuo-temporal bleaching of French imperfect in certain constructions, thus expanding the extent of fake tense and aspect to ¬p contexts.

5.1 Diachronic Model

Adopting a diachronic model inspired by Heine’s model for semantic change (Heine 2002), I hold that the semantic evolution of a grammatical item from one meaning to another is marked by two types of intermediate stages or contexts: bridging contexts and switch contexts. As Heine further assumes, I regard these contexts as traces of past semantic evolution(s) that may still be observed in synchrony, where they correspond to different uses of the same linguistic form.

The first intermediate stage (stage 2) is that of bridging contexts. It corresponds to contexts where the conventional meaning of the grammatical item carries an inference which is invited by the speaker, i.e., which constitutes the very meaning the speaker wishes to convey with the grammatical item. Such contexts are characterized by semantic ambivalence in as far as the grammatical item is associated with two contiguous meanings: (i) the source meaning encoded by the linguistic form and (ii) the target meaning derived by means of inferencing. This step is a necessary condition for the subsequent semantic change. For the imparfait, bridging contexts corresponds to contexts where pastness and imperfectivity allow for ‘counterfactual’ inferences (¬p and ¬πp), so that the imparfait is contextually associated to ‘counterfactuality’.

A more advanced stage (stage 3), and arguably the crucial step in semantic change, is that of switch contexts. It corresponds to contexts where (some aspects of) the source meaning (have/) has been backgrounded and consequently become inconsistent with the interpretation of the utterance. As a result, the inferred target meaning is the only meaning that remains focused in the interpretation. For the imparfait, switch contexts correspond to new contexts where pastness and imperfectivity have become partially or totally inconsistent: ‘counterfactuality’ has prevailed over tense and aspect. In other terms, tense and/or aspect are becoming fake since they are no longer interpreted.

Finally, the conventionalization of the target meaning is completed and the semantic change fully achieved in contexts where the source meaning is not interpreted anymore (stage 4). This reflects the fact that the target meaning is now considered to be the meaning encoded by the construction in the competence of the speakers. This corresponds, for the imparfait, to new contexts where the past and imperfective interpretation is not attainable anymore; ‘counterfactuality’ has become the very meaning of the construction. The different stages hypothesized are diagrammed in figure 7.4.

View Full Size

Figure 7.4

From past and imperfectivity to ‘counterfactuality’

• Download Figure
• Download figure as PowerPoint slide

The following sections examine two ‘counterfactual’ constructions based on the French imparfait: the so-called ‘counterfactual’ use of the imparfait (also known as the use of ‘thwarted imminence’) and the conditional use in hypotheticals. The aim is (i) to assess in Modern French at which stage of conventionalization ‘counterfactuality’ stands in these constructions and (ii) explore the diachronic evolutions that may account for the situation in synchrony.

5.2 Counterfactual Use of the imparfait

5.3 Conditional Use of the imparfait

5.3.2 Diachrony

One can trace the origin of the French construction from the Latin period. Indeed, from Early Latin, the use of indicative pasts in the protasis of ‘counterfactuals’ (including perfects, imperfects and pluperfects) is possible, albeit marked (Ernout and Thomas 1953, Sabanééva 1996, Haverling 2010, 2013). The past tense generally serves to ground the situation in the past domain and it is said to underline the reality/factuality of the situation while ‘counterfactuality’ is expressed by the subjunctive in the apodosis. Interestingly, the imperfect, in the usually quoted examples, carries specific modal undertones and has an ambivalent scope. Typically, it evokes or presupposes some previous speech or thought which the speaker seems to call into question on the basis of what (s)he knows (one could use really in the translation to emphasize the speaker’s attitude):²⁰

(43) —Quid faceret?—Si amabat.ipfv.pst, rogas quid faceret? adseueraret dies nocteisque, in custodia esset semper. [Lat]

‘—What could he do?—You ask what he could do if he (really) loved her?—He should have watched over her nights and days, guarded her all the time.’ (Plaute, IIIth century B.C. quoted by Haverling 2010)

The imperfect also shows a double scope:

• a proposition-internal scope: it bears on the predicate and thus gives a past (R = Tt) and imperfective (R = Ap) characterization of the situation (here the master’s love for Palestra denoted by amare)

• a proposition-external scope: it also applies to p’s validity V (R = Ep) by situating it in the past or, in other words, the reality’s situation is not endorsed by the actual speaker but in a presupposed anterior speech (about the master’s love for Palestra).

View Full Size

Figure 7.5

Aspectual and temporal of V (it be true that [p]) and E (amare) in the utterance ‘Si amabat, rogas quid faceret?’.

• Download Figure
• Download figure as PowerPoint slide

The corresponding aspecto-temporal interpretation is diagrammed in figure 7.5.

The external scope interpretation (over V) triggers implicature ¬πp: the speaker takes their distance with p’s reality and, as a consequence, the situation is interpreted to be unlikely. This reading is confirmed by the apodosis in the subjunctive that describes what the master should have done if he really loved Palestra, namely he should have watched over her night and day, guarded her all the time. As the imperfect carries both the source meaning and the target implicature, these contexts can be viewed as reflecting the second stage of bridging contexts in my diachronic model.

According to Biraud et Mellet (2000), the interpretation licensed in (43) is typical of the Latin imperfect in indirect speech. According to the authors, the imperfect in past indirect speech usually stresses the presence of an anterior speaker while the subjunctive or the infinitive implies that the actual speaker is more committed to the content of the reported speech. Therefore one may think that the imperfect has been attracted to such hypothetical contexts (in comparison with the perfect) to express the fact that the speaker distances herself with the reality of the situation described in the protasis.

Indeed, implicature ¬πp (unlikelihood) appears to be conventionalized in the VIth century with the emergence of new contexts with the same surface structure [si indicative imperfect, subjunctive] but where the source meaning is no longer available (R ≠ Tt ∧ Ap). Implicature ¬πp is the only interpretation that remains in the foreground. These new contexts hence correspond to the third stage of switch contexts in my diachronic model. (44) gives an illustration:

(44) Melius mihi est mori quam vivere. Si iubebas.ipfv.pst, […] accederemus ad prilium cognuscebas, cui ex nobis sit utelitas an ignavia. Lat

‘[…] If you commanded it, we would engage in combat you would see which one is brave, which one is a coward.’ (Chronicle of Fredegar, VIIth century quoted by Sabanééva, 1996)

The aspecto-temporal interpretation is diagrammed in figure 7.6.

View Full Size

Figure 7.6

Aspectual and temporal of V (it be true that [p])) and E (iubere) in the utterance ‘Si iubebas, accederemus ad prilium’.

• Download Figure
• Download figure as PowerPoint slide

The interpretation of the imperfect is the same as in the Modern French construction: tense and aspect have become fake (the described command received a non-past and perfective interpretation) and the imperfect only signals that the situation is considered unlikely (the speaker doubts that the hearer will command to engage in combat).

This semantic evolution of construction [si indicative imperfect, subjunctive] must be connected to the parallel development, in Late Latin, of the subjunctive pluperfect in non-past counterfactuals (Thomas 1938, Sabanééva 1996). According to Sabanééva (1996: 144), the use of the imperfect could have been favored in conditionals to express what he calls ‘possibility’ (what I name ¬πp), i.e. the situation is unlikely but still possible, by opposition to the subjunctive pluperfect which carries strict counterfactuality (¬p). This would explain why the imperfect in the protasis of unlikely conditionals has become unmarked in Old French and eventually become quasi-obligatory by Classical French.

5.4 Summing up the Diachronic Evolution

The following table maps the examples commented above with the different stages of the imparfait’s evolution from past and imperfectivity to ‘counterfactuality’ hypothesized in our diachronic model (see supra figure 7.4).