2.1 Albanian

Modern Albanian distinguishes between a standard propositional negation nuk (which can be replaced by s’) and the negation mos, which is related to the imperative, optative, or subjunctive mood, in root clauses. In declarative contexts, nuk cannot be replaced by mos, see (7), and in imperative contexts, mos cannot be replaced by nuk, see (8), as observed by Turano (Reference Turano2000).

1. (7) Nuk/ *mos vajta në bibliotekë.

   neg_decl/neg_proh go-1sg.pstin library

   ‘I didn't go to the library.’

2. (8) Mos/ *nuk më ndihmo!

   neg_proh/ neg_declcl.1sg help-2sg.imp

   ‘Don't help me!’

Importantly for us, mos can receive expletive uses in the embedded clause of fear verbs, among other contexts (where it occurs as an optional element).

1. (9) Kam frikë se mos më vdes babai.

   have-1sg fear that neg_proh cl.1sg die-2sg father-the

   ‘I fear that my father will die.’

2.2 Greek

Modern Greek distinguishes between a standard propositional negation dhen and the negation mi(n), found in non-declarative contexts like imperative clauses or with the subjunctive (marked by the particle na). As observed by Chatzopoulou (Reference Chatzopoulou2017), these two negations cannot freely alternate; see (10) and (11).

1. (10) o Jánisdhen/ *min írthe.

   the-nom Janis-nom neg_decl/ neg_proh came-3sg.pp

   ‘John did not come.’

2. (11) Mi/ *dhen féris ton Jáni!

   neg_proh/ neg_decl bring-2sg.pnp the-acc Jani-acc

   ‘Don't bring John!’

Min can receive an expletive use in the embedded clause of fear verbs, among other contexts (where it does not occur as an optional element).

1. (12) Fováme na min érthi.

   fear-3sg sbjv neg_proh come-3sg

   ‘I fear that he may come.’

2.3 Classical Latin

Classical Latin (roughly from 150 BC to 300 AD) distinguishes a standard propositional negative marker non, used with the indicative mood, and the negation ne, used with the imperative or subjunctive moods.

Latin prohibitive negation, which we discuss in subsequent sections, receives a (seemingly) expletive use with fear verbs, and with a wider set of verbs (where it does not occur as an optional element).

1. (13) Haec timeone impediantur.

   this fear-1sg neg_proh prevent-1sg.sbjv.pass

   ‘I fear that I may be prevented from (doing) this.’

   (lit., ‘I fear may I not be prevented from (doing) this.’)(Lewis Reference Lewis1879)

In view of this set of data, we turn to the investigation of the connections between prohibitive and expletive negation focusing on Latin and French.

4.1 Parataxis

Jespersen (Reference Jespersen1917) was one of the first to suggest that Latin ne-clauses are in some way treated as independent sentences when occurring under prohibitive or apprehensive verbs. Under his analysis, such clauses retain both the syntax and the semantics of a root negative imperative. He was followed by different scholars, including Ageno (Reference Ageno1955) and Parry (Reference Parry, Willis, Lucas and Breitbarth2013), for whom the Latin ne-clause is paratactically juxtaposed to the verbal clause (e.g., Timeo; ne veniat, ‘I fear; may he not come!’).

What Jespersen was suggesting is that verbs like prohibeo and timeo receive a parenthetical use. They serve a presentative function having almost no interpretative effect upon the negative imperative, which has a main point status (see Simons Reference Simons2007).

Prohibitive verbs name the illocutionary act performed by the negative imperative (“X forbids: prohibition”). The meaning of the main verb is thus redundant with that of the ne-clause, with respect to the whole construction. In the same line of thought, one can add that the presentative function that apprehensive verbs serve is meant to provide evidential motivation (see also Krifka Reference Krifka2017; Simons Reference Simons2007) to the utterance of the negative imperative (“X fears: apprehension”).

This line of analysis is based on the old assumption that languages follow a diachronic trajectory of development from parataxis to hypotaxis (see for instance Bennett Reference Bennett1910, Wallin Reference Wallin1910, Meillet and Vendryes Reference Meillet and Vendryes1924). According to this assumption, a language displays, at early stages of its development, paratactic constructions that ultimately develop into subordinated constructions.

The matching of illocutionary flavours between matrix imperatives and embedded subjunctives in Latin makes a paratactic analysis intuitively attractive. However, the paratactic analysis fails to capture the Latin facts adequately. In fact, even though subjunctive ne-clauses are introduced by no embedding device, they can also display characteristic properties of embedded clauses. They display sequence of tense, as in (19), repeated here as (34), as well as coreference between the matrix and embedded subject, see (35) (McCloskey Reference McCloskey, Campos, Herburger, Portner and Zanuttini2006).

1. (34) Caesar suis imperavit ne quod omnino telum in

   Caesar poss.dat order-3sg.prfneg_proh some none weapon-acc to

   hostes reicerent.

   enemies-acc throw-3pl.sbjv.ipfv

   ‘Caesar ordered them not to throw back any weapon.’ (Caesar, Gal, 1.46)

2. (35) [Sententiam ne diceret_i]recusavit_i.

   opinion-acc neg_proh say-3sg.sbjv.ipfv refuse-3sg.perf

   ‘He refused to give his opinion.’ (Cicero, Off, 3.100)

In these cases we cannot evoke parataxis. For this reason we propose a middle ground analysis between a paratactic and hypotactic analysis (see Section 5).

4.2 Speech-act embedding in Latin

In a series of recent studies, Krifka has promoted the idea and developed a formal model for speech-act embedding (Krifka Reference Krifka, Roper and Speas2014, Reference Krifka2017, Reference Krifka, Hartmann and Wollstein2023; see also discussion in Crnič and Trinh Reference Crnič, Trinh, Riester and Solstad2009; Kaufmann Reference Kaufmann2012, and Woods Reference Woods, Kim, Umbal, Block, Chan, Cheng, Finney, Katz, Nickel-Thompson and Shorten2016). He argues that in English, just as in German, the verb say can be used in two different configurations, as in (36). The verb only embeds a proposition in (36a); in (36b), say is claimed to embed a speech-act (of assertion).

1. (36)

   1. a. Mary said that she hates John.

   2. b. Mary said she hates John.

For Krifka, speech-act embedding is a device that can be applied to embedded clauses that show root clause property, such as the absence of sequence of tense in English, as in (36b), or verb-second in German, as in (37b).

1. (37)

   1. a. Mary sagte, das sie John hasst.

   2. b. Mary sagte, sie hasst John.

The structure corresponding to proposition embedding is given in (38a) and that corresponding to speech-act embedding in (38b) (Krifka Reference Krifka, Roper and Speas2014).

1. (38)

   1. a. ⟦[_VP Mary [$_{{\rm V}} $' say [_CP that [_IP she hates John ]]]]⟧

   2. b. ⟦[_VP Mary [$_{{\rm V}} $' say [_ForceP she hates John ]]]⟧

We follow Krifka's proposal and propose that, in Latin, ne-clauses are embedded speech-acts.Footnote ⁵ However, embedded ne-clauses cannot display root clause property, as they do not allow for morphological imperatives, which remains an open issue for our speech-act embedding analysis.

In our perspective, we assume that ne is a specialised negation, with a clause-typing function. More precisely, we propose that ne heads the Force projection (Rizzi Reference Rizzi and Haegeman1997). The structures corresponding to the fragments (39) are given in (40).

1. (39)

   1. a. Ne me territes. (‘Don't try to frighten me.’)

   2. b. Iubeo, ne me territes. (lit., ‘I order, don't try to frighten me.’)

   3. c. Prohibeo, ne me territes. (lit., ‘I forbid, don't try to frighten me.’)

2. (40)

   1. a. ⟦[_ForceP Ne me territes ]⟧

   2. b. ⟦[_VP Iubeo [_ForceP ne me territes ]]⟧

   3. c. ⟦[_VP Prohibeo [_ForceP ne me territes ]]⟧

An important insight in Krifka (Reference Krifka, Roper and Speas2014) is that speech-act embedding verbs are speech-act reports; that is to say, they describe a locutionary act. A question-embedding verb may describe a locutionary act in an explicit manner (e.g., ask) or an implicit manner (e.g., wonder). To Krifka, wonder is a speech-act report insofar as it describes the psychological attitude one has when asking oneself questions.Footnote ⁶ We follow Krifka in assuming that directives explicitly denote a locutionary act, just as desideratives implicitly do. However, we believe that imperative embedding verbs need not be speech-act reports. Indeed, ne-clauses may also be embedded under intention reports, like efficio (‘I try’) or uito (‘I avoid’).

1. (41) Efficio tacitum ne mihi funus eat.

   endeavor-1sg.prs shut.up-sup neg_proh my funerals-acc be-3sg.sbjv

   ‘I take heed that my funeral rites pass not off in silence.’ (Ovid, Tr, 5)

2. (42) Quem ego uitaui ne uiderem.

   him-acccl.1sg avoid-1sg.pfv neg_proh see-3sg.sbjv.ipfv

   ‘And this man, I avoided seeing him.’ (Cicero, Att, 3)

Our proposal is thus that, in Classical Latin, directives and desideratives have the ability to embed imperative speech-acts. Speech-act embedding may correspond, if the hypothetical earlier stage of parataxis were to be confirmed in ancestors of Latin, to an intermediary stage of development between parataxis and hypotaxis.

4.3 Proposition embedding in French

With the rise of the overt-complementation system in Old French, the Latin prohibitive ne undergoes syntactic reanalysis. We believe that expletive ne in French is a descendent of Latin ne. Note that the French expletive ne is homonymous with the standard negation ne, but we make the assumption that these two markers were morphologically distinct until the standard negation ne lost its negative meaning to pas due to the Jespersen cycle. Recall that Latin ne is semantically redundant with the intrinsic negativity of the negative priority embedding verb. We argue here that expletive negation is semantically in harmony with the intrinsic negativity of the (attitudinal) embedding verb (see infra).

During the transition from Latin to French, the use of the complementizer que (originating from the Latin indicative complementizer quod; see Vincent Reference Vincent, Harris and Vincent1988) develops and becomes more systematic.

Directives and desideratives (and by extension, their negative counterparts) systematically subcategorize for que-clauses (or infinitival complements) by the end of the Old French period. They thus cease to embed imperative speech-acts and start behaving, syntactically, as attitude reports.

Our proposal is that with the development of the que from Latin to French, prohibitive ne loses its clause-typing function in embedded clauses. The key change is that prohibitive ne fossilizes into a modal negation, undergoing downward-reanalysis from ForceP to MoodP, under negative priority attitudes; see (45). Crucially, there is a continuity of meaning of harmonic ne in Latin and French, despite its syntactic reanalysis. To get from (43) to (44) essentially no semantic change is required.

1. (43) Timeo, ne malefacta mea sint inventa

   fear-1sg neg_proh misdeeds-acc mine be-3pl.sbjv discovered-ptcp-pst

   omnia.

   all.

   ‘I fear that my past misdeeds will be discovered.’ (Lewis Reference Lewis1879)

2. (44) Je crains que mes anciens méfaits ne soient découverts.

   1sg fear-1sg that my ancient misdeeds ExN be-3pl.sbjv discovered.

   ‘I fear that my past misdeeds will be discovered.’

3. (45) ⟦[_VP Je crains [_CP que [_IP mes anciens méfaits ₁ [_MoodP ne + sont ₂-sbjv [_VP t₁ t₂ découverts ]]]]]⟧

With ne reanalyzed as a negation marker base-generated in the Mood projection, our diachronic investigation resonates with accounts that reach a similar conclusion based on synchronic evidence. In particular, Zovko-Dinković and Ilc (Reference Zovko Dinković and Ilc2017)'s synchronic analysis of expletive negation in Croation and Slovenian (building on Abels (Reference Abels2005)'s analysis for Russian) also concludes that the negation which receives an expletive reading is base-generated in MoodP.

From a semantic perspective, we propose that ne in MoodP conveys the speaker's preferential attitude towards the negation of the proposition. Just as in imperatives clauses, ne conveys a general meaning of dispreference: the speaker wants not-p to be realized. We argue that ne is reanalyzed in MoodP as it enters ‘Modal Harmony’ with embedding verbs that express negative preferences.

The notion of ‘Modal Harmony’ was introduced by Lyons (Reference Lyons1977), Huitink (Reference Huitink2012), and Giannakidou and Mari (Reference Giannakidou and Mari2018), among others, who explored the relation between modal verbs and adverbs.

1. (46) John must definitely be at home.

Modal Harmony relies on the idea that two modal elements bear the same meaning; in (46), for instance, the modal and the adverb bear the same epistemic flavour.Footnote ⁷ Unlike with Negative Concord approaches, there is no notion of semantic dependency (as in Espinal Reference Espinal1992, Reference Espinal2000; see discussion in section 6.1). The main idea is that, together with the embedding verb, expletive ne yields a unitary semantic meaning of dispreference at large. Note that, like adverbs, expletive negation can be omitted in Modern French, without any change in the interpretation of the sentence.

1. (47) Je crains qu’ il soit arrivé quelque chose à mon chien.

   1sg fear-1sg that 3sg be-3sg happened some thing to my dog

   ‘I fear that something happened to my dog.’

To sum up: in our analysis, directive and desiderative attitudes embed imperative clauses in Latin, and in this context prohibitive negation is a contentful element which has a clause-typing function. Imperative embedding is lost and proposition embedding becomes the only option in French for directives and desideratives and their negative counterparts. Prohibitive negation fossilizes into expletive negation with negative priority attitudes.

Second, our analysis provides a new handle on understanding how languages develop from parataxis to hypotaxis, by positing an intermediary stage of speech-act embedding, where there is embedding – as evidenced by consecutio temporum and pronominal coreference – without complementizer. Finally, we hope that we have shed new light on the intricacies between sentential and verbal mood as two intimately related phenomena (see Portner Reference Portner2018, a.o.).

6.1 Expletive negation as a negative concord item

Previously, van der Wurff (Reference van der Wurff, van Ostade, Tottie and van der Wurff1999); van der Wouden (Reference van der Wouden, Kanazawa and Pinon1994) and Espinal (Reference Espinal2000) have posited that expletive negation is a Negative Polarity Item. Relatedly, Espinal (Reference Espinal1992, Reference Espinal2000) and Zeiljstra (Reference Zeijlstra2004) have posited that expletive negation enters Negative Concord with the embedding context. Together with the main verb, expletive negation would yield a ‘single-negation’ semantic reading. To the previous authors, expletive negation is indeed embedded under ‘inherently negative’ contexts.

In the works of van der Wurff (Reference van der Wurff, van Ostade, Tottie and van der Wurff1999) and van der Wouden (Reference van der Wouden, Kanazawa and Pinon1994), it is argued that the negativity of the contexts of appearance of expletive negation relies on downward-monotonicity. Subsequent works like that of Espinal (Reference Espinal2000) propose that the negativity of contexts where expletive negation occurs relies on nonveridicality (see also Jin and Koenig Reference Jin and Koenig2019), defined as the property of operators that do not entail the truth of the proposition they take as argument (see Zwarts Reference Zwarts1995, Giannakidou Reference Giannakidou1998, Reference Giannakidou1999, Reference Giannakidou, Musan and Rathert2011), with respect to an epistemic model (see Giannakidou and Mari Reference Giannakidou, Mari, Blaszczak, Giannakidou, Klimek-Jankowska and Migdalski2015, Reference Giannakidou and Mari2016, Reference Giannakidou and Mari2021). Simply relying on downward-monotonicity or non-veridicality would lead to an obvious over-generalization. For instance, downward-monotonic predicates such as the emotive-factive attitudes regret or be surprised (see von Fintel Reference von Fintel1999 for a discussion) do not allow for expletive negation. Similarly, belief predicates do not allow for expletive negation in spite of being non-veridical (they do not entail p).

Finally, for Espinal, the main predicate is interpreted as semantically ‘negative’, while the negation marker in the subordinate clause is interpreted (out of a semantic mechanism of ‘logical absorption’) as a dependent concord item, which is semantically empty. Our analysis proposes an alternative explanation, grounded in the diachronic trajectory from prohibitive negation, by appealing to the notion of Modal Harmony rather than the notion of Negative Concord. According to our account, both the attitude and the negation bear the same modal meaning of negative preference. Since the negative preference remains once the expletive negation is omitted, it is hard to find contexts in which the negative contribution of the expletive ne is visible. However, there are cases – especially with temporal connectives – in which this component can still be seen, as in (51).

1. (51) Context: A doctor receives a patient who promises to stop smoking once he has recovered from his current pneumonia.

   1. a. Vous devriez arrêter de fumer avant que vous soyez

      2sg should-cond stopof smoking before that you be-2sg.sbjv

      guéri.

      healed

      ‘You should stop smoking before you get better.’

      Intended: the speaker wants to convey a relation of consecution

      between the main event and the subordinate event.

   2. b. #Vous devriez arrêter de fumer avant que vous ne

      2sg should-cond stop of smoking before that you ExN

      soyez guéri.

      be-2sg.sbjv healed

      ‘You should stop smoking before you get better.’

      Intended: the same meaning as (51a).

The use of expletive negation in (51b) is infelicitous for the reason that the avant que-clause describes a positively valued event. Indeed, the most natural interpretation of (51b) would be what Tahar (Reference Tahar, Baauw, Drijkoningen and Meroni2021b) labels an apprehensive use of the avant que-connective, conveying that the goal of the doctor is that his patient does not get better, which is conflicting with the intended interpretation.

6.2 Expletive negation as a modal particle

For Yoon (Reference Yoon2011) and Mari and Tahar (Reference Mari, Tahar, Franke, Kompa, Liu, Mueller and Schwab2020), expletive negation is not a standard negation marker but a modal particle (but see also Makri Reference Makri2013). For these authors, expletive negation is the grammatical reflex of a component related to the ordering source of the embedding verb. According to Yoon's analysis, expletive negation realizes the ordering source of the predicate, as it imposes an ordering on the (non-veridical) modal base M _B of the verb fear, ranking ¬p-worlds higher than p-worlds on a desirability scale.

1. (52) Scalar semantics for expletive negation with fear Yoon, Reference Yoon2011: p.161:

   1. a. If fear (x,p) is true in a context c, then M _B(x) ${\cap} $ p is not $\emptyset $ in c.

   2. b. The evaluative component of EN (x,p) expresses in context c

      as the following:

      M _B(x) – p >_Desirability M _B(x) ${\cap} $ p in c

While stating that expletive negation is the morphological realization of the ordering source of the embedding verb provides only an ad hoc solution which we cannot retain as such, our diachronic analysis supports the view that expletive negation conveys a meaning of negative preference.