1. Introduction
According to typological research, approximately a quarter of the world’s languages may be considered “evidential”, that is, possessing a paradigm of grammatical inflections and/or clitics encoding the speaker’s types of access to information (Aikhenvald, Reference Aikhenvald2004: 1). English is typically presented as a language lacking evidentiality (Aikhenvald, Reference Aikhenvald2004: 7–11; de Haan, Reference de Haan2001). A simple English sentence like “it fell” is typically translated with different grammatical forms in an “evidential language”. In Tibetan, sentence (1a) includes the direct perception suffix -song, which encodes that the speaker saw the object fall. In (1b), the inferential suffix -bzhag indicates that the speaker did not see what happened but inferred it from some clues. Sentence (1c) combines the factual marker -pa red and the hearsay enclitic =ze, indicating that the speaker knows about the state of affairs through someone else’s report.
The status of French is more ambiguous. While generally not classified as an “evidential language”, it is often cited as an example of a language with tenses that may express evidentiality (Aikhenvald, Reference Aikhenvald2004: 11). The conditionnel may notably encode hearsay in (2), and inferential evidentiality in (3).
Additionally, some scholars argue that the French passé composé and futur may encode inferential evidentiality, as in (4) and (5), and the imparfait hearsay evidentiality, as in (6).
While several scholars argue that French possesses grammatical inflections that may function as evidential markers (Dendale, Reference Dendale and Hilty1993; Caudal, Reference Caudal2012; Bres, Reference Bres2022 inter alia), these uses are typically seen as distinct from “genuine” evidentials (Lazard, Reference Lazard2001; Aikhenvald, Reference Aikhenvald2004: 11). After assessing whether the French passé composé, imparfait, futur, and conditionnel may be considered evidential markers, we will examine the criteria by which the French evidential system may be similar or different from the systems of languages widely cited to illustrate evidential paradigms, such as Tibetan or Turkish.
This article is structured as follows. Section 2 offers a brief presentation of the literature on the evidential uses of the French passé composé, imparfait, futur, and conditionnel. Section 3 presents the methodology. Section 4 is the core of the article and investigates the potential evidential functions of French tenses and their basic conditions of use. Section 5 discusses the arguments in favour of including French among the so-called “evidential languages”. Our conclusions are in Section 6.
2. Literature review: the evidential use of French tenses
Many scholars have identified evidential functions in several French tenses—most notably the passé composé, imparfait, futur, and conditionnel.
The passé composé is formed with the auxiliary avoir (‘to have’) or être (‘to be’) followed by the main verb in its past participle inflection. It has two main functions in Present-Day French: the perfective past and the resultative perfect. Few authors have discussed the evidential use of the French passé composé, but Guentchéva (Reference Guentchéva1994) is frequently cited as arguing that this tense carries an inferential meaning. More specifically, she argues that, in parallel with the conditionnel, certain uses of the passé composé can only be interpreted as expressing abductive inference. She provides example (7).
In (7), it is clear that speaker B’s statement is based on the observation of the result of the state of affairs. The appearance of the caretaker’s eyes leads the speaker to think that he has been crying. However, it remains to be shown whether this inferential reading is indeed encoded by the passé composé itself, or simply conveyed by the context (see §4.1).
As summarized by Squartini (Reference Squartini2001), numerous scholars have described non-assertive uses of the Romance imperfect, including a reportative evidential function (Bazzanella, Reference Bazzanella1990 inter alia). This type of use appears more constrained in French, but is possible in specific contexts, as in (8) (pronounced with rising intonation).
This use of the imparfait may initially seem strange, given that the state of affairs lies in the future, but is standard when the speaker has acquired the information in a past conversation. The past feature of the imparfait no longer refers to the time location of the state of affairs but to the moment of the acquisition of the information. Several scholars therefore argue that the French imparfait has an evidential use, although it is not generally considered a “pure” evidential (Barceló & Bres, Reference Barceló and Bres2006: 53–54; De Mulder, Reference De Mulder2012). We consider this hypothesis in §4.2.
As its name implies, the French futur primarily refers to future states of affairs, but other uses have been documented (Damourette & Pichon, Reference Damourette and Pichon1970[1911–1936]: 388–389 inter alia). The use we will focus on has received several names, such as “hypothetical” (Martinet, Reference Martinet1979: 109), “epistemic” (Dendale, Reference Dendale2018; Rocci, Reference Rocci2000), or “conjectural” (Bres & Azzopardi, Reference Bres, Azzopardi, Bres, Nowakowska, Sarale and Sarrazin2012). Most linguists agree that the evidential use of the futur simple, as in (9), is infrequent and largely restricted to the verb être or avoir (Rocci, Reference Rocci2000 inter alia). However, many also argue that this function can more readily be served by the futur antérieur (the compound version of this tense made up of the auxiliary avoir or être in the future and the past participle), as in (10).
We examine whether this use of the futur can be considered evidential in §4.3.
Finally, the conditionnel is another tense whose complex semantics has attracted the attention of many linguists. It is generally argued that it has three main uses (Dendale & Tasmowski, Reference Dendale and Tasmowski2001: 9; Kronning, Reference Kronning, Dendale and Tasmowski2001: 264–265; Haillet, Reference Haillet2003: 38 inter alia). It may express (i) time/aspect, encoding a future projection of the state of affairs from a past perspective, (ii) irrealis result, referring to the potential or counterfactual result of an explicit or implicit protasis, and (iii) evidentiality, indicating either a hearsay or an inferential access to information (see Patard, Reference Patard2017 for a synthesis of their origins and frequencies). All three uses occur in both the simple (vb.cond) and compound forms (have/be.cond + vb.pprt). In what follows, we use the terms “hearsay” and “inferential” for the evidential uses of the conditionnel, as these are standard labels in typological literature, even though the French tradition resorts to varied labels, such as “borrowed information” for the former (Dendale, Reference Dendale and Hilty1993), and “conjectural” for the latter (Bres & Azzopardi, Reference Bres, Azzopardi, Bres, Nowakowska, Sarale and Sarrazin2012). Example (11) illustrates its hearsay use, and (12) its inferential one.
French speakers possess a high metalinguistic awareness of the hearsay function of the conditionnel, as illustrated by the comment l’emploi du conditionnel est de mise (‘the use of the conditional tense is recommended’) in (11). The second use is less well-known and is limited to polar interrogative clauses (Bres, Reference Bres2022). Whether the different functions of the conditionnel should be treated as separate or as variations of the same basic meaning is highly debated. While considering these functions to be related is relevant from a diachronic and theoretical standpoint (Bres, 2020; Reference Bres2022), several tests suggest that treating them as separate provides a clear account of the scattered distribution of this French tense. We will reinvestigate the semantic characteristics of the two evidential uses of the conditionnel in §4.4.
3. Methodology
As this article reassesses the literature on the evidential uses of French tenses, it is primarily based on a critical reading of previous studies and the authors’ native competence in Standard French. The semantic analysis of the forms under study adopts a strict definition of evidentiality as it prevails in typological work. Anderson (Reference Anderson, Chafe and Nichols1986: 274–275) argues that four criteria are necessary to identify an evidential. Using more recent terminology, these criteria can be stated as follows (see also Boye, Reference Boye and Aikhenvald2018): an evidential must (i) encode the speaker’s type of access to information (i.e. direct perception, inference, or hearsay), (ii) have scope over the main predication, (iii) have the speaker’s type of access to information as an inherent semantic feature, and (iv) be grammatical.
Since French tenses are grammatical forms (i.e. inflections) with propositional scope, they satisfy criteria (ii) and (iv). Criteria (i) and (iii), however, require thorough semantic testing to determine whether evidential interpretations of these tenses indeed stem from their inherent semantic make-up or are simply conveyed by the context. If the passé composé, imparfait, futur, and conditionnel each occupied highly homogenous semantic spaces, it would be sufficient to observe in what contexts they are felicitous or infelicitous to reveal the determining semantic conditions of their uses. However, French tenses are multifunctional, so an in-depth semantic analysis is needed to determine whether their various interpretations correspond to separate meanings, or just pragmatic inferences associated with the diverse contexts of a single meaning. Our goal is to identify the necessary and sufficient semantic features of these tenses, in the same way a lexicographer determines the number of meanings that a word entry should include. For example, the verb leave can be used in the contexts of (13) or (14).
A common test for polysemy is sometimes called the “zeugma test”. Using a word only once with two separate meanings is typically perceived as unusual or humorous (Lakoff, Reference Lakoff1970; Cruse Reference Cruse1986: 61). The perceived oddity of (15) indeed confirms that the two interpretations of leave in (13) and (14) are due to distinct meanings of the word, rather than being simply conveyed by the context.
The same test can be applied to (16) and (17) to assess whether the upward and downward movements referred to by the verb go reflect distinct meanings of this verb or are merely contextual interpretations.
Using go only once with the two complements is perfectly natural and not perceived as odd, as shown by (18), so directionality is simply conveyed by the context. Relative to directionality, go is not polysemous, but semantically vague.
The “zeugma test” will help us assess whether French tenses qualify as evidentials under criteria (i) and (iii)—that is, whether their hypothesized evidential meanings reflect inherent semantic content or merely arise from context.Footnote 2
Following the semantic analysis, we conduct a quantitative investigation of their evidential uses based on a corpus of Present-Day French. To investigate possible genre-based variation, our corpus consists of three sections. The spoken subcorpus (approximately 352,000 words) was extracted from clapi, a corpus of conversations recorded between 1979 and 2017. The journalistic subcorpus (approximately 320,000 words) contains articles published in various newspapers in 2021. It was extracted via Europresse. Finally, the literary subcorpus (approximately 450,000 words) brings together late-20th-to-early-21st-century novels written by Frédéric Beigbeder, Michel Houellebecq, Jean-Marie Gustave Le Clézio, and Leïla Slimani. It was collated via Frantext. From each subcorpus, we extract all instances of the imparfait, conditionnel, and futur, and randomly sample 500 occurrences per tense—yielding a total of 4,500 examples. After semantically annotating these examples, we calculate the frequency with which each tense expresses evidential meaning and examine how text type may influence their use.
Finally, we draw on typological data to compare French with languages commonly cited as having fully grammaticalized evidential systems. Examples in Common Tibetan and Ladakhi are drawn from the first author’s fieldwork (2010–2023), while the Turkish data were verified by native-speaking linguists.Footnote 3
4. The evidential uses of the French tenses
4.1. Passé composé
Guentchéva (Reference Guentchéva1994) argues that the passé composé tends to appear in an abductive inferential context, as in (7) (see §2). While this tense may be more readily compatible with such contexts than other past forms, this correlation likely stems from its aspectual meaning, rather than from any inherent evidential meaning. When functioning as a resultative perfect, the passé composé emphasizes the resultant state of a completed state of affairs. In such cases, the speaker may not have witnessed the state of affairs but inferred its occurrence abductively—that is, by observing its effect and inferring its cause, as in (7). Conversely, the imparfait typically refers to a static, repeated, or ongoing state of affairs in the past, and more often suggests that the speaker has directly witnessed the state of affairs or acquired the information through hearsay. Therefore, the fact that il a pleuré (‘he cried’ in the passé composé) is more often used in inferential contexts than il pleurait (‘he cried’ in the imparfait) does not entail that the passé composé carries an inherent inferential meaning. The verb phrase a pleuré is also compatible with other evidential categories, such as direct perception (19), and hearsay (20).
Examples (7), (19), and (20) show that the passé composé can appear in any evidential context, which suggests that this tense is not inherently evidential. One may argue, nonetheless, that the passé composé is a polysemous tense, and that its inferential meaning is distinct from its other meanings. However, the “zeugma test” shows that this tense does not have a distinct evidential use. In (21), il a pleuré occurs in a direct perception context, as indicated by j’étais là (‘I was there’). The use of Marie aussi (‘Marie too’) should require an identical semantic content with the co-referential predicate a pleuré. However, Marie aussi appears in a context of inferential evidentiality, as the speaker did not witness the state of affairs but only its result, that is, her ‘red eyes’ (yeux rouges).
Native speakers do not perceive this example as odd, which confirms that any inferential reading of the passé composé arises from the context rather than its semantics, so it does not qualify as an evidential marker.
4.2. Imparfait
The imparfait appears to serve a distinct function in hearsay contexts. Its use in (22) refers to a future plan and suggests that the information was reported to the speaker.
Like many hearsay markers, this use of the imparfait is associated with the speaker’s lack of commitment to the actualization of the state of affairs. The proposition can indeed be cancelled without contradiction, as in (22), where the context clarifies that Bertrand is no longer leaving the following day. The same sentence using the present tense, as in (23), would be infelicitous.
Examples like (22) clearly correspond to a distinct function of the imparfait, because this tense appears exceptionally in a clause referring to a future state of affairs. This use of the imperfect is attested across Romance languages, and Squartini (Reference Squartini and Thieroff1995) notes that it is not limited to future contexts, but also applies to past states of affairs, as in (24) (in Standard Italian).
Example (24) may be used when the speaker does not know whether Paolo actually visited the house, but only has indirect knowledge about that potential state of affairs. This use is also possible in French, as in (25):
In (25), the speaker believed his girlfriend had worked, but later discovered that she had not, since she failed to appear at her workplace. In this context, the imparfait differs aspectually from its typical usage, as it is compatible with delimiting time adverbials—normally excluded by imperfective tenses—as illustrated in (26). An indication of direct perception such as je l’ai vue (‘I saw her’) would be odd in this context, which suggests a restriction to certain types of access to information (see Squartini Reference Squartini2001: 310).
Although fully acceptable in French, this use of the imparfait is infrequent, as shown by our corpus data. Out of the 1,500 verbs in the imparfait randomly extracted from the corpus, only one displays this meaning in the spoken subcorpus. As this subcorpus contains 4,199 verbs in the imparfait (across 352,000 words), the extrapolated normalized frequency of this use is approximately 24 instances per million words (henceforth ipmw) for spoken French.
Bazzanella (Reference Bazzanella1990: 450–452) argues that this use of the imperfect in Romance languages likely derives from the ellipsis of a past verbum dicendi (‘verb of saying’), so the inflection would be the result of tense agreement in reported speech. Although no longer requiring an explicit verbum dicendi, this construction appears to have retained a hearsay reading and the associated lack of speaker commitment. However, reported speech may also refer to internal thoughts (e.g. ‘I thought that…’), and this usage of the imparfait shows the same possibility. Consider (27):
In (27), the information was not accessed through hearsay. Instead, the speaker himself has decided to leave, so the information originates from his own mind. Speakers have direct access to their own thoughts, so it is difficult to argue that this use of the imparfait is limited to hearsay, or even to indirect evidentiality. It is indeed with direct evidentials that the speaker’s inner thoughts are typically encoded in “evidential languages” like Tibetan or Tariana (Mélac et al., Reference Mélac, Tournadre, Aikhenvald, Aikhenvald, Storch and Velupillai2025). In contexts such as (22) or (27), the imparfait indicates that the state of affairs was planned but may never be actualized—a meaning that is not evidential per se, since it does not specify how the speaker became aware of this planned state of affairs. One might argue that (27) illustrates a different meaning of the imparfait from the hearsay meaning in (22), but the acceptability of (28) shows that they are not perceived as semantically distinct.
Example (28) confirms that the imparfait does not have a separate hearsay meaning. In these contexts, the imparfait has an irrealis status, which may suggest an indirect access to information. However, it does not encode evidentiality per se, but rather indicates that the state of affairs was planned but not (yet) actualized.
4.3. Futur
In addition to its temporal meaning, the future tense serves an inferential function in several Romance languages (Squartini, Reference Squartini2001). In French, the inferential futur is rare, and appears to be scarcely understood by younger speakers. While many studies acknowledge the low frequency of the inferential futur simple (the lexical verb inflected for future tense), they often describe the inferential futur antérieur as common and widely accepted (Dendale, Reference Dendale2001; De Saussure & Morency, Reference De Saussure and Morency2012). This use of the futur antérieur is clearly distinct from its classical usage, as it refers to an inferred past state of affairs rather than a completed state of affairs in the future. As (29) shows, using only one and the same form to express both the temporal-aspectual meaning of the futur antérieur and its inferential meaning results in an infelicitous sentence.
The first meaning of the futur antérieur may be paraphrased by using the futur simple and the preposition avant, as in (30).
This time meaning is clearly distinct from its evidential one, which can instead be paraphrased using the inferential auxiliary devoir followed by the compound infinitive, as in (31).
Meyer (Reference Meyer and Neveu2012) analysed all instances of the futur antérieur in the Tintin comic series (Hergé, 1930–1976), and found that 48 out of 71 are inferential. Our corpus contains 54 instances of the futur antérieur, none of which are inferential. Examples of the inferential futur are indeed considerably harder to find in 21st-century corpora than in earlier decades. Although a thorough variationist investigation would be necessary to draw firm conclusions, these corpora suggest a significant decline of the inferential futur in Present-Day Standard French. Both authors of this article are native French speakers; the first finds this use odd, while the second occasionally resorts to it. Given that many French speakers now have limited understanding of this use, the reliability of acceptability judgements in earlier studies remains uncertain. These works do not include acceptability questionnaires involving a panel of native speakers, so we can only trust the authors’ intuitions regarding the felicity conditions of this inferential futur. In order to reassess their claims, we have collected authentic examples of this use from previous works and from our own reading (total number of instances: 105). The examples from the Tintin series provide precious clues, because they are numerous and the situations of utterance are depicted.
A first, widely debated dimension is the classification of this use of the French futur as either epistemic (i.e. denoting the degree of likelihood of the proposition) or evidential (i.e. denoting the speaker’s type of information access to the proposition). We argue that the epistemic and evidential interpretations are not mutually exclusive; on the contrary, it is common for evidential markers—particularly inferentials—to carry epistemic implications. One diagnostic for determining whether a form is only epistemic or inferential+epistemic is to examine whether the speaker’s uncertainty can arise from vague personal memory of a situation that s/he has witnessed (Mélac, 2024; Reference Mélac2025). For example, replacing epistemic I think by inferential+epistemic must in (32) would be infelicitous, because must can only be triggered by an inferential context, while I think is compatible with other types of contexts, such as witnessed states of affairs.
All examples of the inferential futur we collected occur in contexts involving the speaker’s inferential processes, which suggests that this use is indeed evidential, and not just epistemic. Moreover, the inferential futur can appear in contexts of certainty, as in (33), suggesting that its evidential meaning is more stable than any epistemic interpretation (see also Caudal, Reference Caudal2012).
In (33), the phrase je suis persuadé que (‘I’m convinced that’) shows that the inferential futur does not necessarily presuppose the speaker’s lack of certainty, so its epistemic meaning is context-dependent, while its distinct inferential meaning is stable.
Dendale (Reference Dendale2001) argues that the inferential futur encodes a fast, instinctive conclusion, unlike inferential devoir (‘must’), which indicates complex reasoning. However, corpus data challenge this binary distinction. Example (34) shows that devoir can be used with a fast, instinctive deduction, and (35) that the futur can encode a slow, reflective inference.
Another often-cited condition for the use of the inferential futur is the possibility to verify the state of affairs in the future (Schrott, Reference Schrott1997: 294). This may reflect the origin of the inferential futur: its futurity component may have shifted from denoting time location to projecting the moment at which the speaker might confirm his/her inference. This “verifiability hypothesis” may be true as a tendency, but several counter-examples exist (see also De Saussure & Morency, Reference De Saussure and Morency2012). In (36), two men in a helicopter are flying over the jungle and looking for the main protagonists. One of the two men sees something moving in the river, and thinks he may have located them. They end up abandoning their search, and the pilot says (36).
We understand from the context of (36) that the speaker will not be able to verify what he actually saw as the pilot has decided to fly away from the location. One may argue that although the inferential futur may have scope over a proposition that the speaker cannot verify, the proposition is still theoretically verifiable (by God, for instance). We do not see, however, what proposition cannot theoretically be verified, so this hypothesis does not seem falsifiable, and cannot be retained to determine the necessary conditions of use of the inferential futur. Examples such as (36) confirm that the “future verifiability hypothesis” is not a strict constraint to the use of the inferential futur.
Lastly, Caudal (Reference Caudal2012) investigates the characteristics of the premises that may trigger the inferential futur. He concludes that this form cannot be used for premises based on clear direct perception or hearsay. Our collected examples confirm this hypothesis. To our knowledge, however, no investigation has really been conducted on the characteristics of the other end of the inferential process, that is, the types of logical conclusion that the inferential futur allows.Footnote 4 In all the examples we examined, the logical conclusion of the inference refers to a specific state of affairs, and never a general truth. The specific property of the conclusion is one of the few felicity conditions of the inferential futur, and may be an alternative explanation for the unacceptability of much-cited examples, such as (37).
The restriction of the futur to certain types of premises and conclusions confirms that this form is evidential, since its felicity requires that the speaker has acquired the stated information in a certain mode.
In conclusion, we argue that French does possess a future tense with an evidential meaning. The futur can notably be used when the speaker infers the existence of a specific, and not a generic, state of affairs. Other claims, such as the instinctiveness of the reasoning, or the future verification of the state of affairs are potential tendencies, but cannot be considered strict felicity conditions. Although this use of the futur is unequivocally evidential, its frequency should not be overstated. Its absence from our corpus supports the broader observation that the inferential futur is now rare in Standard French.
4.4. Conditionnel
As in many Romance languages, the conditionnel may be used as an evidential marker to express either a hearsay or an inferential access to information, in both its simple form (vb.cond) for a present or future state of affairs, and its compound form (have/be.cond + vb.pprt) for a past state of affairs. The hearsay conditionnel typically appears in declarative clauses, while its inferential use is only found in polar interrogative clauses (Tasmowski, Reference Tasmowski, Dendale and Tasmowski2001: 334, cited in Bourova & Dendale, Reference Bourova and Dendale2013: 188). We examine these uses separately.
4.4.1. Hearsay function
The description of the hearsay conditionnel often includes four main semantic features (Bourova & Dendale, Reference Bourova and Dendale2013: 184): (i) the proposition is reported from another source, (ii) its veracity is not certified by the speaker, (iii) it is considered uncertain, and (iv) it is awaiting confirmation. We now examine these features.
The first feature is obviously necessary to identify a hearsay evidential. However, some authors argue that the conditionnel does not meet this criterion. Rossari (Reference Rossari2009: 79) remarks that the so-called “hearsay” conditionnel is not necessarily based on an earlier utterance, but may also refer to information extracted from graphs or statistics. In (38), the journalist’s statement is based on poll charts.
We do not think, however, that such examples disqualify the conditionnel from hearsay evidentials. While charts are not words, they are symbolic representations that follow conventional semiotic rules. It is not surprising that acquiring information from such sources may be classed in the same evidential category as verbal reports. Speech verbs such as dire (‘say’) may also introduce information extracted from graphs, as in (39), which also confirms that linguistic forms denoting hearsay can easily extend to such sources.
Celle (Reference Celle2020) also questions the evidential status of the conditionnel, noting that it can be felicitous when the speaker relies on multiple clues, not solely hearsay. A journalist uses the conditionnel in (40), although the statement is based on partial data and several testimonies.
The semantics of the conditionnel may be broad, but we do not believe that examples like (40) disqualify it as a hearsay marker either. Although the statement draws on various clues, it partly relies on verbal testimony—a necessary condition for the use of the conditionnel. In the world’s languages, certain hearsay evidentials indicate that the speaker is reporting exactly what s/he was told while others are still felicitous with multiple accesses to information as long as the statement is partly based on some verbal source. In Ladakhi, lo corresponds to the former type, as in (41), while rag to the latter, as in (42).
Other authors have also questioned whether the conditionnel has an inherent hearsay function by arguing that its hearsay interpretation is only a rhetorical specialization of its irrealis result use. Rossari (Reference Rossari2009) argues that a statement in the conditionnel is interpreted as hearsay because of an implicit protasis in the present tense such as si X dit juste (‘if X is telling the truth’), as in (43).
Considering the hearsay meaning of the conditionnel to be simply rhetorically induced minimizes its semantic distinctiveness, as several tests indicate that the irrealis result and hearsay conditionnels behave differently. First, if the hearsay meaning of the conditionnel were merely an irrealis result use with an ellipted present-tense hearsay protasis, a different evidential reading should be obtained with a present-tense protasis denoting direct perception or inference. However, if one replaces a protasis denoting hearsay by one denoting a direct perception such as si je me souviens bien de ce que j’ai vu (‘if I remember correctly what I saw’) or an inference such as si mon analyse de la scène de crime est exacte (‘if my analysis of the crime scene is accurate’), it becomes infelicitous to use the conditionnel in the following apodosis. In a direct perception or inferential context, another tense such as the passé composé would be expected, as illustrated in (44–45).
Furthermore, in its compound form, the conditionnel presents the information as possible when used as a hearsay marker, and counterfactual in its irrealis result use. In (46), the clause révèle ce mardi France 5 (‘France 5 revealed this Tuesday’) makes the hearsay function of the conditionnel explicit, which implies a likely state of affairs provided that the source is reliable.
Example (47), however, includes an irrealis result use of the past conditionnel, which typically entails a counterfactual state of affairs.
Using a single instance of the conditionnel to convey both an irrealis result and a hearsay interpretation is infelicitous, as in (48), which confirms that these two meanings are distinct.
The oddity of (48) shows that providing a clear, comprehensive description of the French conditionnel necessitates to avoid conflating its hearsay meaning and its irrealis result meaning.
The second and third features proposed by Bourova and Dendale (Reference Bourova and Dendale2013: 184) (i.e. the uncertified veracity and uncertainty of the proposition) somewhat overlap, and point at a thorny issue, that is, the type of speaker’s commitment presupposed by the hearsay conditionnel. This use of the conditionnel typically does not commit the speaker to the truth of the proposition, as the latter may be presented as true (49) or untrue (50).
This flexibility regarding the truth status of the proposition has led some authors to argue that the hearsay conditionnel is neutral in terms of commitment to the truth of the statement (Dendale, Reference Dendale2018: 69–71). However, we believe “neutral” may not be the most appropriate term, and to illustrate the different types of speaker’s commitment, we will consider several possibilities.
An unmodified sentence in the indicatif présent such as il est chez lui (‘he is at home’) commits the speaker to the truth of the statement, as it may be followed by …c’est vrai ! (‘…that’s true!’), but not by …mais c’est faux ! (‘…but that’s false!’), as shown by (51).
If the sentence is modified by certain hearsay markers such as il a dit que (‘he said that’), it becomes possible to either confirm or contradict the proposition, as in (52), which shows that in this case the speaker’s commitment is indeed neutral.
However, if one replaces the indicatif présent in (51) by a conditionnel présent, as in (53), it is actually infelicitous to either confirm or contradict the proposition.
As suggested by this test, not only does the hearsay conditionnel typically presuppose the speaker’s lack of commitment to either the truth or falsity of the proposition, but it also typically presupposes his/her lack of certainty towards one of these two options. In (49) and (50), the confirmation and contradiction of the statement in the conditionnel seem to be allowed only because the sentences in the conditionnel are further modified, as they explicitly refer to the contents of rumours. The speaker does not use the conditionnel to reflect what s/he believes, but is adopting the perspective of people who simply heard those rumours and do not know whether they are true. Consequently, it seems more accurate to say that by default the conditionnel involves uncertainty (from the speaker’s perspective, or in very specific contexts from other people’s perspectives), rather than neutral commitment. The conditionnel is not as neutral as other hearsay markers, since without a specific context of perspective shift, it is not an appropriate choice in a context of total certainty that the proposition is true or false (see also Kronning, Reference Kronning2012). Some authors hypothesize that the lack of speaker’s commitment associated with the hearsay conditionnel is an “automatic result” of its evidential profile (Dendale, Reference Dendale2018: 68). Hearsay markers in general are indeed often associated with the speaker’s lack of commitment (AnderBois, Reference AnderBois2014 inter alia). However, this is just a frequent semantic association, since hearsay evidentials may entail or presuppose different types of commitment (Mélac, Reference Mélac2014: 365). Both (41) in Ladakhi and (52) in French include a hearsay form that is commitment-neutral, which may thus be directly confirmed or contradicted by the speaker him/herself. However, (42) in Ladakhi includes a hearsay inflection that presupposes a full commitment, quite like (54) in French, which involves a factive hearsay verb.
Contradicting the propositional content under the scope of the hearsay markers in (42) and (54) would typically be infelicitous. If the speaker does not want to commit to the truth of the proposition, another hearsay evidential will be favoured in Ladakhi or French. As we saw, the hearsay conditionnel does not encode a neutral or full commitment, but rather uncertainty from the speaker’s perspective (and marginally from another person’s).
Finally, according to Bourova and Dendale’s (Reference Bourova and Dendale2013: 184) list of features, the hearsay conditionnel indicates that the proposition under its scope is awaiting confirmation (see also Gosselin, Reference Gosselin2005: 169). This might also be a tendency, but if we stick to our semantic analysis of necessary and sufficient features, we cannot retain this trait. The conditionnel is felicitous when the report is unverifiable, so it does not necessarily denote that a confirmation is expected. Example (55) refers to the number of times that the emergency opening of certain types of paragliders was triggered, and the conditionnel is used, although no confirmation can be expected.
Our analyses thus confirm that among its various meanings the conditionnel possesses a hearsay meaning. The latter covers a large semantic space, since it is compatible with information extracted from non-verbal semiotic systems (e.g. graphs) and additional inferential processes, but is indeed inherent to the semantics of this tense. The hearsay conditionnel typically presupposes the speaker’s lack of certainty, but not necessarily that the information is awaiting confirmation.
From a frequency perspective, there are 966 verbs in the conditionnel in the spoken corpus, 733 in the press corpus, and 1,186 in the literary corpus. Out of a random sample of 500 such verbs from each corpus, 7 have a hearsay meaning in the spoken corpus, 97 in the press corpus, and 2 in the literary corpus. The extrapolated normalized frequencies of the hearsay conditionnel are roughly 37 ipmw in the spoken corpus, 444 ipmw in the press corpus, and 11 ipmw in the literary corpus. Although the hearsay conditionnel is in no way limited to the journalistic style, this study confirms that it is indeed far more frequent in this genre than in spoken or literary French, which somewhat justifies its “conditionnel journalistique” label (Gosselin, Reference Gosselin, Dendale and Tasmowski2001).
4.4.2. Inferential function
The conditionnel can only be interpreted as inferential in restricted syntactic configurations, that is, exclusively in polar interrogative clauses (Bres & Azzopardi, Reference Bres, Azzopardi, Bres, Nowakowska, Sarale and Sarrazin2012). In some other Romance languages, like Spanish, it is sometimes possible to use the conditional tense with an inferential meaning in a declarative sentence, as in (56), while another inferential marker, such as the futur or devoir (‘must’), would be required in French (Azzopardi, Reference Azzopardi2011: 398).
In example (57) in French, the speaker suspects that someone may have voted on behalf of his mother, who is in a care home where proxies have been drawn up without consent. The conditionnel appears in a polar interrogative clause, and the question is based on an inference (the main clue being the suspicious number of proxies).
The conditionnel in a polar interrogative clause may, however, have another interpretation, as in (58).
In (58), the conditionnel has an irrealis result meaning, and the proposition is presented as counterfactual, since Aimé Césaire was deceased at the time of the vote.
The “zeugma test” confirms that the inferential meaning is a distinct semantic component of the conditionnel in an interrogative clause. In (59), the conditionnel is inferential, and it would sound odd to refer anaphorically to the same verb and give it an irrealis result interpretation.
Bres (Reference Bres2022) suggests that in sentences like (57), the interrogative syntax—rather than the conditionnel itself—motivates the inferential interpretation. Replacing the conditionnel by the passé composé, as in (60), does not have a substantial semantic impact, because the resulting sentence may still be understood as inferential.
Making the sentence declarative, however, inevitably cancels its inferential reading, and the conditionnel will then only be interpreted as either hearsay or irrealis result, as in (61) (see also Squartini, Reference Squartini2004).
Example (61) shows that the conditionnel needs to appear in an interrogative clause to receive an inferential reading. However, an interrogative clause in general tends to suggest an inference, since a question usually aims to test a hypothesis, so the real semantic contribution of the conditionnel in this evidential interpretation remains to be determined. This contribution may be tested by examining interrogative clauses that cannot be based on inferences. An official form may ask about the nationality of the applicant, as in (62).
Example (62) illustrates a context in which the speaker has no available clues to infer a likely proposition. The conditionnel would be infelicitous, which indicates that this tense restricts the meaning of an interrogative clause. Although an interrogative clause may commonly be interpreted as an inference, example (62) shows that inferential evidentiality is not an intrinsic meaning of this clause type. The whole construction combining the conditionnel and the interrogative clause does, however, possess an intrinsic inferential meaning. Since clause type and tense inflections are both fully grammatical elements, this construction is another instance of the grammaticalization of evidentiality in French.
What distinguishes the use of the conditionnel from other inferential markers in French, such as the futur or devoir (‘must’), is partly the degree of likelihood it suggests. As this use of the conditionnel appears in a question, it is compatible with more or less likely hypotheses, while the futur or devoir tends to encode a higher degree of likelihood. The infelicity of adding an adverb like certainement (‘certainly’), while peut-être (‘maybe’) is felicitous, as in (63), confirms that this inferential construction is avoided in a context of high likelihood (Bres, Reference Bres2022).
Nevertheless, the inferential “interrogative conditionnel” cannot have scope over a proposition that the speaker sees as inconceivable, as it presupposes that some evidence led the speaker to consider the hypothesis. In (64), the addressee’s alias, which is a culturally relevant name in Switzerland, led the speaker to ask about his origin.
It would be infelicitous to use the conditionnel in a polar-alternative interrogative clause, as in (65), because that would suggest that the speaker gives an equal status to the proposition and the counter-proposition, while the conditionnel presupposes that at least minimal evidence supports the hypothesis (Diller, Reference Diller1977).
A question in the conditionnel is biased, as shown by the infelicity of (65), but it remains a question, so it differs from the use of inferential markers in declarative clauses by appealing to the addressee. Dendale and Kreutz (Reference Dendale and Kreutz2014) argue that questions in the conditionnel can differ from questions in other tenses because they do not enquire information per se but rather the addressee’s opinion. However, our data do not support this hypothesis as many interrogative clauses containing the inferential conditionnel, such as (64), do not ask for an opinion but for factual information that the addressee can provide. When the question is self-addressed as in (12) (see §2), the choice of the “interrogative conditionnel” instead of another inferential marker in a declarative clause is also motivated by the pragmatics associated with the interrogative clause. A self-addressed question remains different from an inferential statement in a declarative clause as the former suggests the speaker’s ongoing cogitation and search for an answer. It is infelicitous to use an adverbial that indicates a personal stance such as pour moi (‘for me’, ‘I think’), as in (66), because the interrogative clause suggests that the speaker was not able to form an opinion, unlike a declarative clause with devoir, which presupposes a tentative conclusion, as in (67) (see also Bres, Reference Bres2022).
In conclusion, French possesses a polysemous composite construction involving both the interrogative clause and the conditionnel which can denote inferential evidentiality. An interrogative clause with a conditionnel differs from a more typical interrogative clause in the restriction of the former to contexts where some specific clue has led the speaker to consider a hypothesis. The speaker’s hypothesis is itself questioned and not asserted, which sets the construction apart from other inferential markers in declarative clauses. The inferential “interrogative conditionnel” is relatively infrequent in the three corpora. Out of a random sample of 500 verbs in the conditionnel from each corpus, 1 has an inferential meaning in the spoken corpus, four in the press corpus, and 4 in the literary corpus. The normalized frequencies of this construction are roughly six ipmw in the spoken corpus, 16 ipmw in the press corpus, and 21 ipmw in the literary corpus. Although our tests show that this construction indeed has an intrinsic evidential meaning, the statistics indicate that it is relatively rare.
5. Discussion
We have shown that evidentiality is a necessary concept to describe the semantics of the French tense system. The conditionnel possesses a distinct hearsay meaning, and the futur also has a separate inferential meaning. In addition, the conditionnel and the interrogative clause type may combine to denote an inferential access to a proposition. Evidentiality in French is thus encoded by fully grammaticalized forms. This challenges the putative division between so-called “evidential languages” with fully grammaticalized paradigms (e.g. Tibetan, Turkish), and those, like French, which are assumed to lack a “genuine” evidential system. We will now consider the main arguments in favour of granting a special status to “evidential languages”, and assess whether French should indeed be excluded from this group.
(a) “Evidential languages” encode evidentiality with fully grammatical morphemes
This claim is confusing, because even languages with evidential inflections or clitics also employ lexical means to indicate the speaker’s type of access to information. In addition to its evidential grammatical morphemes, a Tibetan speaker may also resort to perception or speech verbs to specify how s/he knows about the state of affairs. Like Tibetan or Turkish, French possesses both lexical means and tense inflections that may encode evidentiality, so this claim should place French among the “evidential languages”.
(b) “Evidential languages” have paradigms that distinguish several evidential categories
This claim would include Tibetan, but exclude Turkish from the list of “evidential languages”, although Turkish is often cited to illustrate evidentiality (Aksu-Koç, Reference Aksu-Koç, Johanson and Utas2000 inter alia). Turkish possesses one indirect evidential inflection, namely -miş (and its allomorphs), as in (68).
The other inflection -di (and its allomorphs) is often described as a direct marker in Turkish, but semantic tests show it is not strictly evidential. Any interpretation of direct perception is actually generated by the context, since -di, contrary to Tibetan -song, is also felicitous with information accessed indirectly, such as historical facts, as in (69).
Our analyses have shown that French, on the other hand, may encode two evidential categories—inference and hearsay—with distinct, fully grammatical forms. According to this claim, French should thus be considered an “evidential language”, and not Turkish.
(c) Evidential languages express evidentiality obligatorily
Obligatoriness is a criterion frequently suggested to identify “genuine” evidential markers (Aikhenvald, Reference Aikhenvald2007). However, the evidentials of many languages are actually optional, as conceded by Aikhenvald (Reference Aikhenvald and Aikhenvald2018: 10). Mélac (Reference Mélac2022) proposes to break down “obligatoriness” into several dimensions. From a morpho-syntactic perspective, deleting French verbal inflections makes the sentence ungrammatical, which means that they are obligatory, just like Tibetan or Turkish verbal inflections. However, from a semantic perspective, evidentiality can be avoided by using other inflections in French, such as the présent simple or passé composé, which do not specify the type of access to information. This is actually also the case in Turkish, as the suffix -di is evidentially neutral, as in (69), and also in Tibetan, which possesses factual markers that can be used when the speaker does not wish to specify how s/he acquired the information s/he is sharing (Mélac, Reference Mélac2022). Since even so-called “evidential” languages can avoid encoding evidentiality, obligatoriness is not a reliable criterion for distinguishing them from supposedly “non-evidential” languages.
(d) Evidential languages have morphemes that encode evidentiality as their only meaning
Few, if any, “evidential languages” meet this criterion, as grammatical forms are typically both cumulative and polysemous. If the exclusive expression of time were required to identify a tense marker, there would be very few such markers in the world’s language, as tense markers usually express other notions cumulatively, such as aspect, mood, and/or person. The English inflection were encodes the past and also the 2nd person singular, or 1st/2nd/3rd person plural. The cumulative semantics of morphemes cannot entail that a language has not grammaticalized a notion—otherwise, English and French would be considered tenseless, as their tense morphemes commonly express notions beyond time. Not only do tense markers tend to be cumulative, but they are usually polysemous. For example, the suffix -ed in English can refer to the past, but can also denote the irrealis status of a projected state of affairs, as in “If you asked nicely…”. Yet excluding -ed from tense markers simply because it carries other meanings would be problematic. Similarly, Turkish suffix -miş is cumulative and polysemous. In addition to its indirect evidential use illustrated in (68), it can denote an unexpected state of affairs when used in a context of direct perception, as in (70).
Excluding polysemous forms from the list of evidentials would not make more sense than claiming that -ed is not a tense marker because it may also express the irrealis. The criterion of semantic exclusivity cannot therefore be retained to identify evidentials and, by extension, “evidential languages”.
(e) Evidential languages have morphemes that encode evidentiality as their primary meaning
“Primary meaning” is an ambiguous term which may be understood in three different ways: (i) the inherent meaning as opposed to context-induced interpretations, (ii) the original meaning as opposed to later semantic developments, and (iii) the most frequent meaning.
A proponent of interpretation (i) is Anderson (Reference Anderson, Chafe and Nichols1986: 274–275), although it is unclear whether he accepts that a morpheme may possess several “primary” meanings. Our tests show that the evidential interpretations of the conditionnel and futur are not merely contextual, but part of their inherent semantics. From this perspective, French qualifies as an “evidential language”. If a form cannot have more than one “primary” meaning, however, the distinction between inherent meaning and context-induced interpretations cannot help us identify this supposed “primary” meaning, so “primary” has to be understood differently.
Proponents of interpretation (ii) argue that “genuine” evidentials are different from markers with “evidential-like extensions” acquired at a later stage (Aikhenvald, Reference Aikhenvald2004: 11). The diachronic dimension may be crucial when investigating the grammaticalization of evidentiality, but considering the original meaning to be the predominant one can lead to false conclusions, because a meaning that was once central to a form’s semantics may later become marginal or disappear entirely. For example, the English auxiliary might was the past tense of the verb may, which encoded dynamic modality. In Present-Day English, it is mainly used as a present epistemic marker, and the semantic features of pastness and dynamic modality are extremely marginal. Typological works have shown that evidential inflections typically develop from markers that encoded other grammatical domains (Friedman, Reference Friedman and Aikhenvald2018; Mélac & Bialek, Reference Mélac and Bialek2024). For example, the Tibetan direct perception suffix -song first encoded a translocative motion before becoming an evidential marker. If a morpheme had to encode evidentiality from the start of its grammaticalization to be considered a “genuine” evidential, there would consequently be very few such markers. This criterion cannot help us categorize the evidential systems of languages as they are spoken today.
Finally, interpretation (iii) suggests that frequency may be used to distinguish primary meanings and secondary ones. As we saw, although the inferential futur may still be heard occasionally, it is extremely infrequent in Present-Day French, as the futur of none of the 1,500 verbs extracted from the corpus has such meaning. The hearsay meaning represents 1.4% of the instances of the conditionnel in the spoken corpus (7/500), 19.4% in the press corpus (97/500), and 0.4% in the literary corpus (2/500). The inferential meaning represents 0.1% of verbs in the conditionnel in spoken corpus (1/500), 0.8% in the press corpus (4/500), and 0.8% in the literary corpus (4/500). In sum, evidentiality is therefore a minor function of the futur and of the conditionnel from a frequency perspective. It is only in the journalistic genre that the conditionnel has a relatively frequent—though still not dominant—hearsay meaning. Although evidentiality is necessary to describe their distribution, French tenses are not primarily evidential from a frequency standpoint.
French does possess fully grammaticalized evidential forms, but their overall frequency appears to set this language apart from more typical “evidential languages”. As there are very few corpus-based frequency studies of languages with highly advanced evidential systems, substantiating this claim remains difficult. However, the Tibet Student Corpus (approximately 26,000 words) may provide some frequency data for Common Tibetan (Mélac Reference Mélac2014: 76–83; Mélac Reference Mélac2023). Tibetan possesses a variety of highly frequent evidential inflections (’dug, -song, -bzhag, etc.), and clitics (-za and -ze). Altogether the extrapolated normalized frequency of evidential inflections and clitics in spoken Tibetan is 49,923 ipmw (n=1,298 in the Tibet Student Corpus), whereas in spoken French, tenses encode evidentiality at an estimated frequency of 43 ipmw (a number extrapolated from the futur and conditionnel sample of our spoken French subcorpus). This corpus sample thus suggests that fully grammatical evidentials are 1,161 times more frequent in spoken Tibetan than in spoken French. In a language like Tibetan, the expression of the speaker’s type of access to information is overwhelmingly grammatical, whereas in French, it is more commonly conveyed through lexical or semi-lexical forms. Despite piecemeal data and substantial variations according to text types, a comparison of Tibetan and French confirms massive cross-linguistic differences in the frequency of evidentials. Contrary to popular belief, however, our analyses did not show any clear-cut morphological or semantic differences between the French evidential system and that of other languages which supposedly exemplify the notion. While the grammaticalization of a morpheme can only be completed after many generations of speakers, frequency may change relatively fast. French need not modify its grammatical system to match the evidential status of languages like Tibetan or Turkish; it would simply require a dramatic increase in the frequency of its existing evidential inflections.
6. Conclusion
This investigation of French tenses leads us to conclude that a strict dichotomy between evidential and non-evidential languages is untenable. Under a strict dichotomy, French would have to be classified as an evidential language, since it does not differ from typical examples such as Tibetan or Turkish in its grammatical structure. A detailed semantic analysis of the polysemy of French tenses shows that the conditionnel displays an inherent hearsay meaning, while the futur and the “interrogative conditionnel” both carry an inherent inferential meaning. French thus possesses verbal inflections that encode evidentiality, making its expression of this notion similar to that of so-called evidential languages from both a formal and semantic standpoint. However, the analysis of several corpora also leads us to conclude that the one criterion that still makes French starkly different from languages such as Common Tibetan is the frequency of its fully grammatical evidentials. While inflectional evidentiality is pervasive in Tibetan, it is rare in French.
Corpora
clapi http://clapi.icar.cnrs.fr
Europresse https://www.europresse.com/
Frantext (ISSN 2729-2754) https://www.frantext.fr/
Acknowledgements
We would like to express our deepest gratitude to all our consultants in Tibet and Ladakh, as well as our French-speaking colleagues who shared their intuitions about numerous sentences. We also thank Hülya Mısır for her invaluable expertise on Turkish. Finally, we are deeply grateful to the three anonymous reviewers and the coordinating editor, Dalila Ayoun, for their precise corrections and insightful feedback.
Funding
None.
Competing interests
The authors declare none.
Open access
