Pas de Problème: The Distribution and Nature of Double Negation in French and Other Romance Negative Concord Languages

Abstract

Most Romance negative concord (NC) languages in particular configurations give rise to double negation (DN) readings. In this article, I discuss an intricate DN pattern in French. After discussing some previous accounts, I provide an analysis of French pas (and in its slipstream also of the French expletive negative element ne) that takes pas to be a purely semantically but not formally (i.e., syntactically) negative element. I then argue that the reason why French pas is so different from other negative markers lies in its diachrony and show that other attested asymmetries between the formal properties of negative markers and neg-words in Romance must receive a diachronic explanation as well.

1. Introduction

Most Romance negative concord (NC) languages, in particular configurations, give rise to double negation (DN) readings (cf. Longobardi 2014; Zeijlstra 2022, a.o.). For instance, Italian is a so-called non-strict NC language (cf. Giannakidou 2000, 2006) where postverbal neg-words must be accompanied by a negative marker or a neg-word in preverbal position (1a–b), but where preverbal neg-words cannot combine with the negative marker to yield NC readings (1c).1

(1)	a.	Gianni non ha telefonato a nessuno	(Italian)
		Gianni neg has called to neg-body
		‘Gianni didn’t call anybody’
	b.	Nessuno (*non) ha telefonato a nessuno
		neg-body has called to neg-body
		‘Nobody called anybody’
	c.	Nessuno (*non) ha telefonato
		neg-body neg has called
		‘Nobody called’

Whenever in such languages a preverbal neg-word appears to the left of the negative marker non, the only reading that is available is a DN one, as illustrated again for Italian.2

(2)	NESSUNO non ha telefonato	(Italian)
	Nobody neg has called
	‘Nobody didn’t call.’

What we see here is an instance of configurational or structural DN. Even though neg-words and negative markers can yield NC readings, in some particular configurations they do not do so. Following Zeijlstra (2004, 2008), who bases himself on Ladusaw (1992), this is due to the fact that neg-words carry an uninterpretable feature [uNEG] that needs to be checked off at surface structure by a higher, c-commanding interpretable feature [iNEG] (though other approaches have been presented in the literature, e.g., Haegeman and Zanuttini 1996; Giannakidou 2000; Herburger 2001; De Swart and Sag 2002; De Swart 2010; Haegeman and Lohndal 2010). In non-strict NC languages, such as Italian, only two elements carry such a feature [iNEG]: the negative marker (non) and a covert operator (Op¬). Given that sentential negation must always be overtly marked least as high as at the vP level, it follows that postverbal neg-words need to be accompanied by the negative marker (and have their [uNEG] feature checked off by it), or by a preverbal neg-word (both being checked by [iNEG] on Op¬), as shown below:

(3)	a.	Gianni non[iNEG] ha telefonato a nessuno[uNEG]
	b.	Op¬[iNEG] Nessuno[uNEG] ha telefonato (a nessuno[uNEG])

Now, (2) follows as well. Since nessuno’s [uNEG] feature cannot be checked off by [iNEG] on non, it can only be checked off by the covert negative operator, but then a second semantic (i.e., anti-morphic) negation is present as well, hence the DN reading (4):

(4)	Op¬[iNEG] Nessuno[uNEG] non[iNEG] ha telefonato

However, things are different in French. French is an exceptional Romance NC language in that it also exhibits inherent DN readings. Where the Italian negative elements can all establish NC readings, it appears at least prima facie that the French negative marker pas is not able to do so. Whenever pas appears in a clause with one or more neg-words, a DN reading emerges, irrespective of their relative positions (5).3

(5)	a.	Personne (ne) mange pas4
		neg-body ne eats neg
		‘Nobody doesn’t eat.’
	b.	Je (ne) mange pas rien
		I ne eat neg neg-thing
		‘I don’t eat nothing.’
	c.	Personne (ne) mange pas rien
		neg-body ne eat neg neg-thing
		‘Nobody doesn’t eat anything.’

In this article, I discuss this intricate DN pattern in French. First, in Section 2, I present the relevant facts. In Section 3, after discussing some previous accounts, I provide an analysis of French pas (and in its slipstream also of the French expletive negative element ne) that takes pas to be a purely semantically but not formally (i.e., syntactically) negative element. In Section 4, I then argue that the reason why French pas is so different from other negative markers lies in its diachrony and show that other asymmetries between the formal properties of negative markers and neg-words must receive a diachronic explanation as well. Section 5 concludes the study.

2. French DN: The Facts

French is a language that exhibits two particular characteristics with respect to the expression of (sentential) negation. First, it is an NC language: a clause-internal combination of neg-words, which each can independently induce a semantic negation (as in (6a–b)), together only yield one semantic negation, as illustrated in (6c) below.5

(6)	a.	Personne (ne) mange	(French)
		neg-body ne eats
		‘Nobody eats’
	b.	Maria (ne) mange rien
		Maria ne eats neg-thing
		‘Maria doesn’t eat anything’
	c.	Personne (ne) mange rien
		neg-body ne eats neg-thing
		‘Nobody doesn’t eat anything’

Apart from that, French also displays so-called embracing negation, the phenomenon where a language has not one, but two negative markers (in the case of standard French: the preverbal negative marker ne and the postverbal negative marker pas that normally “embrace” the finite verb, see (7)). Note, though, that ne is mostly a property of formal French; in colloquial registers, it is almost always dropped, in line with so-called Jespersen’s Cycle (see Section 4).6

(7)	Denis (ne) mange pas	(French)
	Denis ne eats neg
	‘Denis doesn’t eat’

Interestingly, French ne and pas differ quite drastically from each other in the sense that, while ne may participate in NC constructions and can be combined with pas, thereby yielding a single semantic negation, no combination of pas and a neg-word may ever give rise to an NC reading (cf. De Swart and Sag 2002; Zeijlstra 2010b, 2022). Inclusion of pas in a sentence containing one or more neg-words always yields an additional semantic negation, as shown in (5), repeated below:

(8)	a.	Personne (ne) mange pas
		neg-body ne eats neg
		‘Nobody doesn’t eat.’
	b.	Je (ne) mange pas rien
		I ne eat neg neg-thing
		‘I don’t eat nothing.’
	c.	Personne (ne) mange pas rien
		neg-body ne eat neg neg-thing
		‘Nobody doesn’t eat anything.’

This leads to the following questions: What are the properties of French neg-words, ne, and pas, such that ne can combine with both neg-words and pas, while still yielding a single semantic negation? Additionally, why can pas and neg-words not be combined in such a way?

3. Explaining the Facts

In this section, I first discuss two previous analyses that have dealt with this problem and show that they still face several challenges. After that, I present my own proposal. In short, I argue that ne should not be analyzed as a concordal element or as a negative marker, but rather as a plain negative polarity item (NPI). Then, I argue that once ne is taken to be an NPI, it naturally be explained that pas and neg-words do not establish an NC relation. Adopting Zeijlstra’s (2004, 2008) approach that takes NC to be an instance of syntactic agreement (see also Haegeman and Zanuttini 1996; Ladusaw 1992; Haegeman and Lohndal 2010), every element that can participate in an NC relation should carry a formal negative feature; however, that does not entail that every semantically negative element must carry such a formal feature too. I then conclude that that pas does not carry such a formal negative feature.

3.1. Previous Analyses

De Swart and Sag (2002) provide an account for the inability of French pas to participate in NC relations that does not rely on agreement but rather on the fact that, unlike neg-words, negative markers are not regular generalized quantifiers (pace Collins and Postal 2013). Instead, De Swart and Sag argue that negative markers should be thought of as zero-quantifiers, i.e., quantifiers that bind no variable. Being quantificational and negative in nature, negative markers can then participate in NC relations (as is the case in most languages). However, since negative markers in NC constructions do not bring in any new quantificational contribution (as they do not bind variables), languages may also choose to leave negative markers out of the NC system. French would then be an example of a language that forbids one of its negative markers, namely pas, to participate in NC relations.

Such an analysis of French pas faces a number of problems though. Apart from the general problems that it inherits from De Swart and Sag’s treatment of NC in terms of quantifier resumption (see Zeijlstra 2022 for discussion), the analysis appears too general, as it suggests that a salient subset of NC languages allows negative markers to be banned from NC constructions. However, to the best of my knowledge, only French exhibits this particular kind of behavior. No other NC languages or varieties, even those close to French (e.g., Quebecois), share this property. So, it appears to be just a characteristic of French that pas cannot participate in NC constructions. However, it is unclear, under De Swart and Sag’s proposal, why this is so. More importantly, the question arises as to how this property of French pas is lexically encoded. What is it that French pas has that all other negative markers in NC languages lack (or the other way round)?

Penka (2007, 2011) addresses these above-mentioned questions and argues that it is not so much a special property of French pas that it cannot establish NC relations with a neg-word, but rather a special property of French neg-words. Penka adopts the analysis of NC pursued above and takes NC to be an instance of syntactic agreement, but argues that such an analysis cannot extend to French: regardless of the featural status of pas ([iNEG] or [uNEG]), pas should be able to participate in at least some NC relations, contrary to fact. In order to solve this problem, Penka proposes that neg-words in French do not carry a feature [uNEG], but rather a feature [uNEGØ] that states that neg-words can only be checked by a covert negative operator (which, in French, she takes to carry the corresponding feature [iNEGØ]). French ne, in her system, carries a general feature [uNEG] that does not specify the phonological status of its checker. Pas, finally, is an overt negator, thus carrying [iNEG]. Now, the patterns follow: neg-words and ne can be checked by a single abstract operator, yielding an NC reading (9a); ne can have its [uNEG] feature checked against pas’ [iNEG] feature (9b); and, finally, a combination of (one or more) neg-words with pas yields a DN reading, as pas cannot check the neg-words’ [uNEGØ] features ((9)c), and thus an additional negative operator is required:

(9)	a.	Op¬[iNEGØ] Personne[uNEGØ] (ne[uNEG]) mange rien[uNEGØ]
	b.	Marie ne[uNEG] mange pas[iNEG]
	c.	Op¬[iNEGØ] Personne[uNEGØ] (ne[uNEG]) mange pas[iNEG] rien[uNEGØ]

Although Penka’s proposal is an improvement of the original [iNEG]/[uNEG] system, as it can handle the French patterns illustrated above, it is still problematic for at least two reasons. First, independent motivation is lacking for the existence of [iNEGØ] and [uNEGØ] features next to [iNEG] and [uNEG], especially since other NC languages do not appear to exhibit such features, and also since syntactic features should generally be blind for the phonological status of the elements carrying them. However, even more problematic is that Penka’s analysis makes a wrong prediction—namely that if ne can be licensed by Op_¬, it should be able to negate a sentence by itself, contrary to fact (see (10), though see footnote 6). Given these problems of earlier analyses, the explanation for French pas’s behavior should lie elsewhere.

(10)

*Marie ne mange

(French)

The observation about (10) is important, as it shows that ne cannot be an NC item. Since NC is the phenomenon where elements that may induce a semantic negation by their own together only yield one semantic negation, ne, by definition, cannot appear in NC constructions (see also (Breitbarth 2009) for a similar observation for West Flemish en/ne). If NC is taken to be an instance of syntactic agreement, instantiated by an underlying feature system, it follows as well that ne cannot have any formal negative feature at its disposal. In fact, ne does not even count as a negative marker as it is not able to render a sentence negative by itself.

3.2. Proposal

Let me start my own proposal by addressing the question as to what properties ne exactly should have. Two properties immediately come to mind. First, ne is semantically non-negative, as it is unable to induce a semantic negation by itself, but may appear in sentences that are negative (see Corblin 1994, 1996). Hence, it can at best act as a scope marker (cf. Corblin et al. 2004; Godard 2004). Second, ne’s distribution is outside negation (mostly) restricted to non-negative (Strawson-)downward entailing or non-veridical contexts7, such as exceptives (11a), comparatives (11b), complement clauses of expressions of fear (11c), avoidance (11d), denial or doubt (11e), conditionals (11f), and temporal before-clauses (11g), as shown below (all examples have been taken from (Rooryck 2017, pp. 3–4)) and is much more restricted outside such usages. That renders it an NPI, as any element that is restricted such a limited distribution that includes negative contexts, by definition, is an NPI.8

(11)	a.	Jean (ne) voit que Marie
		Jean ne sees comp Marie
		‘Jean only sees Marie’
	b.	Jean est plus malin que Pierre (ne) l’est
		Jean is smarter Pierre ne it is
		‘Jean is smarter than Pierre is’
	c.	Il a barricadé la porte de peur/crainte qu’on (n’) entre chez lui
		he has blocked the door of fear that they ne enter with him
		‘He blocked the door for fear that people might come in’
	d.	Jean a évité que Lucienne (ne) tombe
		Jean has avoided that Lucienne ne fall.SUBJ
		‘Jean prevented Lucienne from falling’
	e.	Nie/doute-t-il que je (ne) dise la vérité?
		denies/doubts he that I ne tell.SUBJ the truth?
		‘Does he doubt/deny that I am telling the truth?’
	f.	Je viendrai à moins que Jean (ne) soit là
		I will-come to less that Jean ne is.SUBJ there
		‘I will come unless Jean is there’
	g.	Il est parti avant que nous (n’) ayons mange
		he is left before that we ne have eaten
		‘He left before we ate’

The contexts where ne may appear without being supported by a neg-word or by pas, and without giving rise to a semantic negation, are all contexts that are known to license weak NPIs and where the usage of ne is either scope-marking or expletive.9 Even though it may not appear in every known NPI licensing context (ne, for instance, is not accepted in if clauses or restrictors of universal quantifiers, as are, for instance, modal NPIs such as English need or its Dutch counterpart hoeven), this does not run against the analysis. Not every NPI must be licensed in every NPI-licensing context (see Van der Wouden 1994; Zeijlstra 2022). Note that this furthermore implies that constructions containing ne and pas cannot count as NC constructions either; such constructions only show that pas is able to license NPI ne—not surprising, given that pas is a semantic negation and thus an NPI licenser.

Focusing on French neg-words, as these can independently induce a semantic negation, but together only yield one semantic negation (as shown in (6)), these neg-words should be taken to be equipped with an uninterpretable formal negative feature that may be licensed by an abstract negation. Application of this mechanism to French neg-words is illustrated for the sentences in (12). Note that ne is not equipped with such a feature, as it is an NPI, not an NC element.10

(12)	a.	Op¬[iNEG] Personne[uNEG] (ne) mange	(French)
		neg-body ne eats
		‘Nobody eats’
	b.	Op¬[iNEG] Jean (ne) mange rien[uNEG]
		Jean ne eats neg-thing
		‘Jean doesn’t eat anything’
	c.	Op¬[iNEG] Personne[uNEG] (ne) mange rien[uNEG]
		neg-body ne eats neg-thing
		‘Nobody doesn’t eat anything’

However, now the question arises again why, instead of the abstract operator Op_¬, pas cannot be the checker of a neg-word’s [uNEG] feature. Why would (13) be impossible with an NC reading in standard French?

(13)	*Jean (ne) mange pas[iNEG] rien[uNEG]	(French)
	Jean ne eats neg neg-thing
	Int: ‘Jean doesn’t eat anything’

Even though pas is a clear semantic negation, pas’ semantic status does not entail that pas must carry a formal negative feature as well. In fact, as pas appears to be unable to license any other negative element (except for ne), pas can actually not be said to carry an interpretable formal feature [iNEG]. Pas is only semantically, and not formally negative: it has the meaning of a negation but lacks a formal or syntactic negative feature.

Consequently, whenever a neg-word and pas co-occur in one and the same clause, it must be the abstract negative operator Op_¬ that checks off the neg-word’s [uNEG] feature, and the sentence thus contains two semantic negations: one introduced by Op_¬ and one by pas. When one neg-word precedes and one neg-word follows pas, there is still one abstract negative operator that checks off both neg-words’ [uNEG] features, whereas pas introduces a semantic negation of its own. This is illustrated below:

(14)	a.	Op¬[iNEG] Jean (ne) mange pas rien[uNEG]
		Jean ne eats neg neg-thing
		‘Jean doesn’t eat nothing’
	b.	Op¬[iNEG] Personne[uNEG] (ne) mange pas rien[uNEG]
		neg-body ne eats neg neg-thing
		‘Nobody doesn’t eat anything’

French pas is thus semantically negative, but not formally equipped with any negative feature. In other words, morpho-syntactically, pas is not negative. French neg-words, on the other hand, form the mirror image of pas. They are semantically non-negative, but only formally equipped with a negative feature; they carry an uninterpretable formal negative feature, which needs to be checked off by an element that carries an interpretable formal feature. The fact that pas does not act as an intervener in the agreement relation between Op_¬ and the two neg-words then follows straightforwardly. Since pas is morpho-syntactically not marked for negation, any morpho-syntactic process must be blind to pas’ negative semantics.

However, the idea that certain negative markers in NC languages do not carry a feature [iNEG] still needs independent motivation, as, otherwise, it may be as much of an ad hoc explanation as the other accounts discussed above. Why would this not be the case in other languages? In the next section, I argue that certain particular properties of the diachronic development of negation in French motivate such an analysis where pas lacks any kind of formal negative feature.

4. The Diachronic Development of Negation in French

4.1. Jespersen’s Cycle

It is a well-known fact that diachronic changes with respect to the negative markers that languages employ may have strong effects on the emergence of particular types of NC and DN. Let us therefore look at what kinds of changes in the domain of negative markers can be attested in the first place. Already in 1917, the Danish grammarian and philosopher Otto Jespersen observed a general tendency in the expression of negation in various languages:

The history of negative expressions in various languages makes us witness the following curious fluctuation; the original negative adverb is first weakened, then found insufficient and therefore strengthened, generally through some additional word, and in its turn may be felt as the negative proper and may then in course of time be subject to the same development as the original word.

(Jespersen 1917, p. 4).

Jespersen supports this claim with a number of examples from different languages in which such a development can be found, including not only Romance languages but also Germanic and Slavic languages, Greek, and several varieties of Arabic (see a.o. Eckardt 2006, 2012; Van Gelderen 2008, 2009, 2011; Devos et al. 2010; Devos and van der Auwera 2013; Van der Auwera 2010, 2011; Chatzopoulou 2012; Willis et al. 2013; Gianollo 2016; Zeijlstra 2016 (from which (15) is taken); see also Greco (this volume)). Focussing at Romance, the way that the expression ‘I don’t say’ changed from Old Latin to Colloquial French or French-based creoles is as follows:

(15)	a.	*Ne dico	Proto-Latin11
	b.	*Ne dico oinom/oenum	Proto-Latin
	c.	Non dico	Classical Latin
	d.	Jeo ne di	Old French
	e.	Je ne dis (pas)	Middle French
	f.	Je ne dis pas	Modern French
	g.	Je (ne) dis pas	Colloquial French
	h.	Je dis pas	Quebecois
	i.	Je pa di	Haitian French Creole

What we see is that, in Latin, the weaker form ne was strengthened by an additional minimiser ^*oinom/oenum (‘one (thing)’), which merged into Classical Latin non. In Old French, it got weakened to ne and emphasised by a second element pas, which originally meant ‘footstep’ and nowadays, in spoken colloquial French, fully replaces the original negative marker ne. In some French-based creole languages such as Haitian French Creole, pas has phonologically weakened yet again, displaying the same behaviour as Old Latin and Old French ne. Other examples, also from languages outside the Indo-European language family, and additional discussion can be found in, for instance, Eckardt (2006, 2012), Van Gelderen (2008, 2009, 2011), Devos et al. (2010), Devos and van der Auwera (2013), Van der Auwera (2010, 2011), Larrivée (2011), Larrivée and Ingham (2011), Chatzopoulou (2012), Willis et al. (2013), and Poletto (2016, 2020). See also Greco (this volume).

Now, two types of questions arise. First, why is it that such changes happen? What causes the original negative marker to weaken, such that it appears to require the additional support of a second negative marker? Additionally, relatedly, why does this new negative marker take over the role of the original one? Second, what is the source of these negative markers? In English, the negative marker stemmed from nawiht, a negative quantifier meaning ‘nothing’. However, in French, pas originally meant ‘footstep’, a meaning that at first seems quite remote from the meaning of ‘not’. Let us look at both questions in more detail.

Jespersen attributes the negative cycle to the tension between the phonological weakness of the old negative marker and the need to strongly mark the sentence as being negative. If a negative marker, such as English or French ne, becomes phonologically too weak to fully function as the negator of the sentence, a second, supporting negative marker jumps in. Jespersen’s explanation has raised some criticism, both empirically and theoretically. Empirically, second negative markers can also emerge in languages in which the original negative marker does not count as phonologically weak, as was the case in Greek (cf. Kiparsky and Condoravdi 2006). However, more importantly, phonological weakening in general does not drive morpho-syntactic change, but rather results from it. Even phonologically weak elements can be very robust markers of a particular semantics (e.g., past-tense morphology is quite often phonologically rather weak, but, nevertheless, does not fail in fulfilling its function). Partly for these reasons, Kiparsky and Condoravdi (2006) argue that something else must be at hand. These scholars argue that every language must be able to express both plain negation and emphatic negation. To get a feeling for the distinction, look at the minimal pair in (16).

(16)	a.	I didn’t read
	b.	I didn’t read anything

Truth-conditionally, the two sentences in (16) are identical: If somebody did not read anything, that person did not read; and if somebody did not read, she did not read anything. Yet, (16b) feels more emphatic than (16a), because (16b) puts a strong focus on the absence of anything that could have been read by the speaker. English is a language that expresses plain negation by means of a negative marker (often n’t, sometimes not) and has a number of additional means at its disposal to further emphasise negation (such as indefinites of the any-series, cf. Haspelmath 1997). In general, languages can add two types of elements to emphasise what would otherwise be a plain negation: minimisers and generalisers. Minimisers are elements that denote (extremely) small amounts of something. A sip in I didn’t drink a sip is a good example of a minimiser that gives rise to an emphatic effect in negative contexts. The reading that emerges is that the speaker did not even drink the smallest amount possible. Generalisers are elements that, in contrast, extend the domain of quantification. Anything in (16b) is a good example. It says that, of an extended set of things that could have been read, the speaker did not read even one of them. However, why would the requirement that negation can be expressed both emphatically and non-emphatically trigger syntactic change? Why could the two strategies in a particular language not remain constant over time?

Following work by Dahl (2001), Kiparsky and Condoravdi (2006) allude to a general semantic weakening mechanism, arguing that what is marked emphatically loses its emphatic power in the course of time. If a particular emphasising strategy is used more and more often, it tends to inflate and become less emphatic. As a result, it will be used even more frequently, given that the originally emphasised expression can also be used to describe situations that require less of a negative emphasis. If this trend continues, at some point, the original strategy for emphasising negation ends up being the plain way of expressing negation, and becomes obligatory. The old negative marker then loses its function of expressing plain negation and, being superfluous, will disappear. The new marker (and, originally emphasiser) of negation, then, has taken over the role of the plain negator, and new supportive elements to express emphatic negation may emerge, thus continuing the cycle.

A particular prediction of Kiparsky and Condoravdi’s proposal is that negative markers may have different diachronic sources: They could have developed out of minimisers or out of generalisers (generally, quantifiers). With this in mind, let us look at how the various changes in the system of negative markers triggered the emergence the NC system in French.

4.2. The Emergence of NC in French

As shown by Muller (1991), Larrivée (1996, 2010), Tovena et al. (2004), Eckardt (2006) a.o., and also discussed in detail in Roberts and Roussou (2003), French ne (and, before that, non) started out as a single negative marker that, in terms of Kiparsky and Condoravdi, was not able to yield emphatic negation anymore, and therefore required additional material to express increased negative force. Due to the unavailability of existing neg-words, in this language, positive generalisers, such as personne (‘person’) or rien (‘thing’), or positive minimisers, such as pas ‘footstep’, point (‘point’), or mie (‘crumb’), were added to the negation to strengthen it.

While, originally, in Old French (appr. 900–1300) those expressions could only be literally interpreted, as shown in (17) below for pas (‘step’) and rien (‘thing’), taken from Eckardt (2003), later on, they could also yield a more general interpretation. For instance, in (18), pas does no longer mean ‘footstep’, nor does mie mean ‘crumb’.

(17)	a.	Ne vus leist pas aler avant12	(Old French)
		ne you is-allowed step to-go forward
		‘You may not go forward a step’
	b.	D’avanture ne sai je rien …13
		of adventure ne know I thing
		‘Of adventure, I don’t know anything …’

(18)	a.	N’i ot rei, prince ne baron, Qui pas m’ i pöust contrester14	(Middle French)
		ne there has king prince ne baron who ‘pas’ me there could contradict
		‘There is no king, prince or baron who could even/at all contradict me there’
	b.	Quel part qu’il alt, ne poet mie cäir15
		which part that-he goes ne can ‘mie’ fall
		‘Wherever he goes, he cannot fall a bit’

As Eckardt shows, minimisers that originally meant ‘step’, ‘thing’, ‘crumb’, etc., were used to strengthen the negation. However, in line with Kiparsky and Condoravdi (2006), such strengtheners became obligatory over time, lost their literal reading, and got semantically bleached into expressions meaning ‘even’, ‘at all’, or ‘a bit’. As is well known from the literature on NPI-hood, minimisers or other indefinite expressions that are obligatorily focused tend to behave as NPIs (see Chierchia 2013). The new strengtheners all became NPIs with a bleached interpretation. The originally positive expressions personne and rien were reanalysed as ‘anybody’ and ‘anything’ accordingly.

At this stage, virtually every occurrence of ne should be accompanied by such NPIs that were either indefinites or pure strengtheners. Such NPIs could, in the earlier stages of the change, appear in all kinds of downward entailing contexts, and thus count as weak NPIs (see (19)), but were later on restricted to purely negative contexts—presumably an effect of the relative high frequency of negation as their licenser (cf. Roberts and Roussou 2003; Eckardt 2006)—and became strong NPIs.

(19)	a.	Je doute que personne y réussisse.	(Middle French)
		I doubt that anyone there succeed
		‘I doubt that anyone might succeed at it’
	b.	Suis je pas bele dame e gente, digne de servir un preudome?
		am I ‘pas’ beautiful dame and gentle worthy to serve a gentleman
		‘Am I not a beautiful dame and gentle worthy to serve a gentleman?’

Now, the following cocktail emerged. French ne could not be used without an additional strengthener, even though the strengthening effect has become obsolete itself, and these reinforcers of negation only appeared in sentences that were fully negative themselves; therefore, they all contained ne. In addition, the competition between the various possible strengtheners disappeared, basically leaving pas (and sometimes point, especially in the southern varieties) and the future neg-words as the only available candidates to be an additional negation.

Therefore, at this stage, language learners had to analyse pas as an obligatory part of the negation system. It only appeared in negative sentences and served no other function than co-expressing negation. This way, language learners assigned formal features [uNEG] to both pas and ne (as neither of the two could be taken to be the main negator). In the same vein, with ne no longer being able to act as a main negation itself, elements such as personne and rien were taken by language learners to be part of the negation system and got assigned [uNEG] as well—also because multiple occurrences of such elements would not add more semantic negations. When interpreting a sentence such as Je ne dis rien (I neg say thing ‘I don’t say anything’), a language-learning child could no longer assume that ne was the only source for the negation. The behavior of ne in the language input was no longer compatible with that of a plain negative marker. Consequently, the child should take rien (and personne, etc.) to contain formal negative features as well.

Crucially, now arriving in the Middle French era, the language was a strict NC language where ne, pas, and the neg-words could freely be included in a single clause and thus yield an NC reading, as the following example form Molière (taken from Zeijlstra 2010b) shows:

(20)	On ne veut pas rien faire ici qui vous déplaise …16	(17th cent. Fr.)
	we neg want neg neg-thing do that you displeases …
	‘We don’t want to do anything that displeases you …’

The emergence of strict NC in Old/Middle French can thus be understood as the result of the need of a negative marker to be strengthened. This strengthening initially led to an emphatic negation reading, but the joint meaning contribution of the negative marker and the original strengthener shortly afterwards became the plain negation, the original strengthener being a second negative marker. Due to this, the language turned into an NC language, given that, now, two negative forms corresponded to one negative meaning. Strengtheners that contained an indefinite (such as elements meaning ‘person’ or ‘thing’) also became part of the negative system for the same reasons and got reanalysed as neg-words (after already having undergone a change from positive indefinites, via weak NPIs into strong NPIs). The French developments thus clearly show how the cause underlying Jespersen’s Cycle triggered the emergence of NC as an epiphenomenon. Naturally, however, this cannot be the full picture. As outlined earlier, French further developed from a strict NC language into some kind of a partial NC language where pas no longer participates in NC.

4.3. From Strict to Partial NC

So where do we stand? While Old French only had the negative marker ne at its disposal, during the period of Middle French, ne got more often combined with the additional minimiser pas (as well as various competitors, such as point ‘point’, etc.), until the embracing negation ne … pas became the standard way of expressing sentential negation. As described above (see also Roberts and Roussou 2003; Kiparsky and Condoravdi 2006; Eckardt 2006), in Middle French, ne was thus no longer felt to be able to express sentential negation by itself and should always be accompanied by an indefinite or minimiser. However, in cases where there was already an indefinite reinforcer present, there was no need for pas to further strengthen ne. Additionally, this was exactly the case with neg-words. Hence, combinations of ne and neg-words did not need an additional reinforcer pas, and, therefore, combinations of ne with a neg-word and combinations of ne with pas have both been frequently attested, but combinations of ne with both a neg-word and pas less so. This does not entail, though, that it must have been forbidden by then to reinforce combinations of a neg-word and ne by an additional minimiser pas.

Early Middle French was a strict NC language. Additionally, later on, such combinations have also been attested. However, as the famous follow-up of the example in (20) shows, such combinations of pas with ne and a neg-word became increasingly rare/stigmatized:

(21)	a.	… de pas mis avec rien tu fais la récidive,
		et c’est, comme on t’a dit, trop d’une negative
		… with pas put with rien you do the recidival,
		and it.is as one you.has said too.many of.one negative
		‘… with “pas” put together with “rien”, you commit recidival,
		and it is as one says a negation too many’

Most cases of negation thus involved either the combination of ne and pas or ne and a neg-word. The reason for this rarity of combinations of ne, pas, and a neg-word is functional in nature. The need to reinforce ne was absent in cases where ne already had been reinforced by an additional indefinite, and could only lead to a very strong emphasis (see (Zeijlstra 2010a) for a discussion of a similar development in Dutch and German).

Now, given that in the course of time, the large majority of instances of negation either involved ne and pas, or ne and a neg-word, ne by itself was no longer able to render a sentence negative and started to be optionally dropped. Consequently, ne was no longer able to act as a cue for language learners to assign it a feature [i/uNEG], and had to be reinterpreted as an NPI. Hence, neg-words and pas were the only elements that could induce a semantic negation. Since neg-words could co-occur in the same clause and establish an NC relation, the loss of ne did not trigger a difference in their featural status. French neg-words remained carriers of [uNEG]. However, pas, which hardly ever co-occurred with any other neg-word in a clause, could only be analyzed as the sole marker of negation in a sentence, and consequently no longer be assigned a formal negative feature. Language learners had to take it to be a semantic negation and had no information present in their language input that would force to assign it a feature [iNEG] too. This way, the French partial NC system was born. The same development that led to pas becoming a NC item also caused its absence from NC in later stages.

4.4. A Note on Non-Strict NC Languages

If it is indeed the case that different types of NC systems are the result of the interplay of different diachronic forces, this should also hold for other cases of ‘mixed’ NC systems, including non-strict NC systems such as the one in Italian. One may wonder why languages would opt for the negative marker being semantically negative but not for neg-words. Again, the cause of non-strict NC is historical. At least for the Romance non-strict NC languages, it turns out that, at earlier stages, they were strict NC languages, as illustrated for Old Spanish in (22) and for Old Italian in (23).17

(22)	Qye a myo Cid Ruy Diaz, que nadi no diessen posada18
	that to my Lord Ruy Diaz, that neg-body neg gave lodging
	‘that nobody gave lodging to my lord Ruy Diaz’

(23)	Mai nessuno oma non si piò guarare19
	neg-ever neg-even-one man neg himself can protect
	‘Nobody can ever protect himself’

Given the fact that the language input during L1 acquisition contained expressions of the forms in (22)–(23), the negative marker must have been assigned a formal feature [uNEG] (see (Arshad 2023) for an depth study into the L1-acquisition of NC in Italian). However, at some point, speakers began to omit the negative marker no in such constructions (cf. Zanuttini 1997; Martins 2000). This change is not surprising, since the negative marker in these constructions did not contribute in any way to the semantics of the sentence, not even as a scope marker; it is semantically fully redundant. A negative marker preceding a neg-word, by contrast, can still function as a scope marker for negation (and therefore yield sentential negation), but the fact that there is an abstract negative operator located in a higher position than the preceding neg-word in examples such as (22)–(23) already follows from the presence of these preverbal neg-words themselves. It is therefore not surprising that this negative marker got dropped when it was preceded by a neg-word. Note that this effect is still visible in certain varieties of Catalan. In (Central Variety) Catalan, a strict NC language in the sense that preverbal neg-words can be followed by the negative marker but can also be optionally left out, as the example in (24) from Zeijlstra (2022) below shows (see also Espinal et al. 2016).

(24)	Ningú (no) ha vist res.	(Central Variety Catalan)
	nobody neg has seen neg-thing
	‘Nobody has seen anything’

Now, if the L1 input included constructions such as (25) and an increasing relative frequency of examples such as (26), at a certain point, the absence of cases of no(n) following nessuno/nadie should become thus high that they can no longer form a cue for the negative marker to carry [uNEG]. As a result, no(n) must be reanalysed as [iNEG], leading to the standard non-strict NC patterns. Almost every instance of no(n) corresponds to a semantic negation. Hence, the reinterpretation of no(n) leading to the change from strict to non-strict NC thus again follows from the interplay of diachronic change and language learnability.

(25)	Op¬	nessuno/nadie	no(n)	V
	[iNEG]	[uNEG]	[uNEG]

(26)	Op¬	nessuno/nadie	V
	[iNEG]	[uNEG]

5. Conclusions

In the beginning of this paper I raised the following question: what are the properties of neg-words, ne and pas, such that ne can combine with both neg-words and pas, while still yielding a single semantic negation, whereas pas and neg-words may not be combined in such a way?

In this paper, I have argued that French pas is semantically negative, but not equipped formally with a negative feature. In other words, morpho-syntactically pas is not negative. Neg-words, on the other hand, form the mirror image of pas: they are semantically non-negative, but only equipped formally with a negative feature. They carry an uninterpretable formal negative feature, which needs to be checked off by an element that carries an interpretable formal feature.

This also explains why pas and neg-words cannot yield any NC reading: since pas is formally non-negative, it can never check any neg-word’s [uNEG] feature. These features can thus only be checked off by an abstract negative operator, carrying an interpretable negative feature [iNEG], which has to be assumed to be present in the sentence as well. Consequently, combinations of neg-words and pas always induce two negations.

I have furthermore demonstrated that French ne is a plain NPI in the sense that it must always be licensed by some downward entailing licenser (though not every downward entailing operator proves to be a valid licenser). In negative sentences, ne can thus be licensed by the purely semantic negation pas or by the abstract negative operator Op¬ that carries an interpretable formal feature [iNEG].

Finally, I have argued that these assumptions concerning the nature of ne, pas, and neg-words are not postulated in order to make the system work, but rather follow from the development of French negation (along the lines of Jespersen’s Cycle) and from independent principles of language acquisition. I have also shown that the emergence of non-strict NC systems, such as the one in Italian, can be accounted for in similar terms.

This study has a few theoretical consequences. First, it provides more evidence for theories of NC in terms of syntactic agreement, such as Zeijlstra’s (2004, 2008) approach. The French facts follow immediately once formal properties (such as the feature system underlying NC) are thought to be distinct from semantic properties (such as NPI licensing). Second, it also shows that there is no 1:1 correspondence between formal and semantic properties. Some lexical items, such as pas, can have a particular semantic property, which is not reflected in their formal representation, and vice versa, some lexical items, such as neg-words in most languages, have formal properties that are not reflected in their lexical semantics. Finally, it shows that attested cross-linguistic variation, here in the domain of negation, can be thought as a result of the interplay of functional forces applying to language change and formal properties in terms of language acquisition.