2.1 A cartographic approach

Initiated by Pollock’s (1989) seminal paper on an enriched inflectional system, research in generative grammar has gradually developed into the study of a more and more fine-grained clausal structure. The empirical basis for such an enterprise lies in the observation that cross-linguistically, different functional categories such as Aspect, Mood, Tense, display very similar organizations, as shown by word order. This has led to the highly articulated clausal functional sequence proposed in Cinque (1999). On the basis of observable word order variations that involve the left edge of the clause, one line of research within the generative framework, developed by Cinque (1990), Rizzi (1997) and subsequent work within the Cartographic approach, proposes a rich left periphery, with functional projections hosting hierarchically constrained discourse-related elements (on a motivation for syntactic positions for discourse-related elements, see Rizzi 2014). In such an approach, the left periphery is composed of at least a C (or Force) head, which encodes the type (or force) of the clause, a (possibly iterative) Topic head, a Focus head and a Fin(iteness) head related to finiteness which occurs at the junction with the temporal domain. Note that in this paper, I will use the label C, rather than Force, to refer to the highest functional projection of the left-periphery:

A growing bulk of research has shown that from one language to the next, discourse-related elements, that is, topics and foci, occur in clause-initial positions, and that these positions are occupied through movement (see Rizzi 1997 for discussion and references).

While Hungarian (15a) can freely move several tropicalized elements to a clause-initial position, only one focus-marked constituent is allowed to move to a position structurally below the topics.¹ In Gungbe (5b), the order of preposed constituents is identical; in addition, topic and focus are signaled by specific morphological marking.

It is thus proposed that different elements target specific functional projections, rather than moving to an underdetermined edge position. While empirical methods enable to identify the relevant projections and their hierarchical organization, the movement of constituents to (the specifiers of) specific functional projections is claimed to be a syntactic operation triggered by the need to check features. These features, which act as scope markers, are present on the functional heads, and identified by the interpretation they are associated with, via movement of elements whose featural make up includes the relevant feature. Interpretations are thus assigned on a strictly structural basis.²

2.2 The syntax of modality

In this framework, a syntax of modality will condition parts of the interpretation of modal constructions. While we do take resulting interpretations as clues, the underlying assumption is that scope relations (and interpretations) are the result of syntactic operations which have their source in feature-checking mechanisms.

In some languages, modal expressions take the form of modal verbal/auxiliary-like elements which encode different kinds of modality (or modal strength). Below are examples from English and French:

While (6a), (7a) express possibility, (6b) and (7b) express necessity. The difference is often lexically encoded. On the other hand, in many Indo-European languages including English (8), but also in Hungarian (9) or Chinese (10), modal expressions may be ambiguous between different modal categories, for example deontic and epistemic readings.³

However, it has been observed that the different readings correlate with different syntactic constraints. Picallo (1990) argues that Catalan modals, which are ambiguous between the two readings (11), occupy two structurally distinct positions, which can be identified thanks to their interaction with negation (12):⁴

On the basis of syntactic evidence, such as the binding of anaphoric clitics, the construal with auxiliaries and the distribution of multiple modals, among others, Picallo argues that epistemic pot occurs in I(NFL). This can be translated in recent approaches as T or Agr, the position of inflected elements, and root (ability) pogut occurs as a modal in a VP adjoined position.

Modals have thus syntactic positions associated with their different interpretations, and enter into different scope interactions with e.g. negation.

Cinque (1999), on the basis of a cross-linguistic study of the distribution of various adverbials, comes to the conclusion that languages are remarkably similar in the distribution of their various syntactic markers, including modal expressions. In such a structural approach, different modals occur in dedicated, hierarchically organized functional projections, which correspond to their featural content.

These different functional projections are distributed over the spine of the clause ranging from left-peripheral positions for Speech Act to positions between TP and VP for “lower” modals. Given that some languages may have ambiguous modal markers, it is reasonable to assume that each version actually corresponds to a different bundle of modal properties. In other words, English may realizes [EPIST(emic); POSS], but also [DEON(tic); POSS]. Ambiguity, in this view, is not a matter of individual lexical items, but of feature bundling. The features will build up the structure and determine structural positions. The structural organization of the clause, in turn, conditions the scopal properties of its elements; thus, scope relations are a by-product of other, syntactically driven, phenomena. As already mentioned above, this approach, proposed in Beghelli (1995) and Szabolcsi (1997), puts the burden of scope mainly on syntax, in that syntactic scope will condition semantic scope.

2.3 Negation

The section presents a brief introduction to the syntax of negation within generative grammar, as well as a discussion of the various interpretations of negation, at the sentence level and beyond.

2.4 Interactions between modal markers and negation

With the above assumptions about syntactic structures, we now come back to the interaction between modals and negation.

Modality markers do not all interact identically with negation. Let us start with root interpretations. Deontic must can either take negation in its scope, as in (22a), or scope under negation, as in (22b).

English can also has a root (deontic or ability) reading. In all its interpretations, can scopes under negation:¹³

Similar scope patterns emerge in French, where devoir, pouvoir (although they behave like verbs and not “auxiliaries”) exhibit the properties of their English counterparts:¹⁴

French devoir (‘must’) can both be interpreted above negation (24a) and below it (24b). As illustrated in (25), pouvoir (‘can’) systematically scopes under negation.

One possible line of analysis, as proposed e.g. in Homer (2010; 2015) accounts for the scope variations as a result of covert movement of the modal itself. Indeed, the author argues that deontic must and should are PPIs which may move out of the domain of negation in an anti-licensing environment. As can is not a PPI, it cannot escape negation.

However, epistemic interpretations of the same modals reveal a different pattern. Epistemic must seems to occur exclusively outside the scope of negation, while epistemic can occurs (preferably?) in its scope:

The same observation holds for French epistemic devoir and pouvoir:

In the hierarchy proposed in Cinque 1999, which is constructed on the basis of cross-linguistic evidence, epistemic modality is encoded higher in the functional sequence than deontic/ability modality markers, potentially above NegP where negation would be interpreted. Under an approach which requires matching feature bearing elements to raise (overtly or covertly) to the position where feature checking may apply, modals endowed with an [EPIST] feature will move to (or Agree with) the functional head encoding epistemic modality. One expects thus epistemic modals to be systematically interpreted outside the scope of negation. While this seems to be the case with must/devoir, can/pouvoir pose a potential problem. Moscati (2006) observes that given the right contextualization, negation can indeed scope over epistemic modality. He gives the following Italian example, which actually correspond to the attested examples given in (26b, 27b):²⁰

As discussed above, Moscati argues that in English, covert neg-raising raises negation to the CP(-domain), allowing it to scope over high modal markers. He assumes that negation is inserted in NegP, a projection which immediately dominates the VP. But the expression of negation may covertly raise to the highest clausal position available. This position is spec ForceP, which is involved in clause typing. Moscati proposes that given the relevant conditions, the Force (C) head can encode a NEG typing feature. Neg-raising will (covertly) target spec ForceP. Negation is thus able to scope over all the clausal material, including high modals. He proposes that negation may occur in the following configurations:

Thus, while both a PPI and a standard covert movement analysis may account for the scope alternations of deontic must/devoir (note that they are not incompatible approaches), a covert neg-raising analysis enables us to explain the wide scope interpretations of negation w.r.t epistemic modality markers.

Focussing on the English-type negator, Moscati claims that English not raises covertly to spec ForceP. We need to further assume that the [neg] clause-typing feature itself occurs on the head itself, and licenses possible negative material in its specifier. I will propose below that the negative feature in the C (Force) head may also be checked when a relevant head moves to C. This is the case with negative interrogatives, among others:

While (30a) exhibits overt movement of a head-like negator (see e.g. Zeijlstra’s proposal) to C, (30b), which may have an identical interpretation, will resort to covert movement of the negator as discussed above. Both, I assume, are triggered by the need to check a clause-typing [NEG] feature as proposed by Moscati.

2.5 Hypotheses

Given the properties of negation and modality observed here, namely that:

1. modal markers bear syntactic features that participate in feature-checking

2. negation may raise overtly (or covertly) to high, left-peripheral positions,

3. scope relations are to be read off (interpretable) syntactic structures,

the following hypotheses are formulated:

(H1) The peculiar behavior of optatives and hortatives (which include modal properties of the bouletic kind) w.r.t negation is accounted for by the interaction between the two types of elements. In a framework which assumes a direct mapping of syntax onto interpretation, this suggests that the syntactic encoding of bouletic modality involves mechanisms/positions, which interfere with negation. (H2) Metalinguistic negation readings are also constrained, at least in some cases, by syntactic structures.

The remaining of the paper will propose an analysis of hortative and optative markers, as expressions of bouletic modality, which is compatible with these hypotheses.

3.1 Optatives

Optatives express a wish or a desire of the speaker that some situation be brought about. As also noted in Condoravdi & Lauer (2012), speaker’s wishes do not need to have any addressee. Optatives are typically cases in which there is no addressee on which the fulfilment or satisfaction of the wish may depend. That optatives are expressions of bouletic modality seems to be supported by the fact that negating an (overtly expressed) wish of the speaker’s is infelicitous:

Compare with e.g. expressions of deontic modality, where negating the speaker’s wish is unproblematic:

3.2 Hortatives

Although hortatives are also expressions of bouletic modality, they differ in several respects from optatives. A striking feature of hortatives is that they share properties with imperatives. Their interpretation is that of a directive speech act in which, among others, “the agent(s) of the action may not be the addressee(s) of the utterance, but someone else: the speaker or a third party” or “the speaker herself may be included among the agents” (Jarry & Kissine 2014: 26). They encode bouletic modality, just as imperatives do in that they express the speaker’s desire that p obtain. Condoravdi & Lauer (2012) argue that imperatives “commit the speaker to a particular kind of preference and to be bounded by a condition that limits his active involvement in making the content true” (2012: 45). Morphologically, hortatives may also look like imperatives and may share the same inflectional paradigm. This is however not systematically the case, suggesting that while they belong to the family of directives, they still are distinct from imperatives proper. Typically, imperatives may encode a variety of uses, such as directives (command, request), but also wishes, permissions, invitations or advice (see Condoravdi & Lauer 2012). Hortatives, on the other hand, are restricted to wishes and suggestions (48) while doing part of the same job as imperatives, that is, committing the Speaker to a preference (49):

3.3 Interim conclusion

The discussion of the syntactic properties of two expressions of bouletic modality has led us to the conclusion that bouletic modality is encoded in a functional projection which dominates the functional head encoding clause-typing (C⁰). While optatives were shown to occur overtly in the bouletic head, hortatives were argued to appear lower, possibly in C, due to their imperative component. It was tentatively proposed that they are “frozen” in this position. However, their bouletic feature is checked with the bouletic head, thus creating a chain which scopes over the rest of the clause. Interaction with negation has shown both overt constraints (in French) and covert constraints, in the form of scope relations, which confirm both the difference in overt syntax and the similarity in covert syntax between the two types of bouletic modality expressions. The syntactic properties of bouletic modal markers as expressions of the highest left-peripheral modality are in line with their interaction with descriptive negation. The distinction between these two expressions of bouletic modality, which was identified both in their syntactic distribution and their interaction with descriptive negation, is structurally encoded. I therefore propose a higher granularity of the left periphery, where, in addition to positions dedicated to epistemic modality and speech act mood as proposed in Cinque (1999), a speaker-related position associated with bouletic modality is included:

Expressions of bouletic modality as associated with optative and hortative markers will occur in the left periphery, licensed by the corresponding features:

The analysis proposed here converges with Hypothesis I: descriptive negation is blocked from scoping above expressions of bouletic modality due to syntactic constraints.

4.1 Hortatives

Consider the following examples of hortative contexts:

As discussed in the literature (see references in 2.1.2 above), metalinguistic negation may scope over different sizes of constituents. (62) may involve different levels of metalinguistic negation, as attested by different continuations:³⁴

(63a, b) are cases of metalinguistic interpretations at different sub-sentential levels. In (63c), which bears a stress on the negative marker, negation targets the propositional level. That we are dealing with metalinguistic negation is attested by the fact that under this interpretation, NPIs are not licensed:³⁵

Similarly, metalinguistic interpretations of negation with hortatives are also accessible in French. Recall that descriptive sentential negation licenses the indefinite de (example (58b) repeated here as (65a)). This contrasts with (65b):

The presence of the indefinite de in (65a) versus un (‘a’) in (65b) clear indicates that metalinguistic negation is at play.

However, the metalinguistic negation interpretation does not include the bouletic operator. In other words, there is no situation in which the speaker’s wish (that p) is negated, even as metalinguistic interpretation. This appears to challenge the claim, at least in its strongest form, that metalinguistic negation scopes over the speech act. Before attempting to account for this, we compare the data with those of optative clauses.

4.2 Optatives

Recall that optatives differ from hortatives in their syntactic realization, as attested by the ban on high adverbials. On the other hand, very much like hortatives, optatives scope over descriptive negation. Given the properties of metalinguistic negation, we again expect that the latter should scope over the modal in similar contexts. Consider now the following:

Metalinguistic negation can target the complement of the modal (i.e. it can be understood as expressing that Otilia wishes Polona were NOT late, but e.g. very late), but it cannot cancel the bouletic modality on any ground. In other words, metalinguistic negation may affect the embedded units, but never the modal itself. This could be a purely semantic constraint, but it is difficult to argue that there is no way one could object to a wish.³⁶

Similarly, (67) cannot be interpreted as metalinguistic negation scoping over the modal. In other words, there is no interpretation which cancels the wishing (while the wish can be objected to very easily):

That (66) and (67) do involve metalinguistic negation can also be shown by the NPI test (68). Here again, the scope of metalinguistic negation does not include the bouletic operator:

Why would a metalinguistic negation interpretation be blocked in these contexts, while it may be accessible with other types of modals? A piece of the puzzle is contributed by data from French. Recall that descriptive negation was shown to be restricted to the complement of the bouletic modal maker, and the explanation proposed was the fact that negation cannot raise to the position which the inflected modal targets above CP.

Note that metalinguistic negation can still target the complement of the modal operator:

I would like to explore the idea that the syntactic behaviour of negation and modality uncovers some important property of metalinguistic negation. Crucially, the key to the problem seems to be the fact that the ban on the overt raising of ne to a high position, a syntactic constraint, correlates with the impossibility of having metalinguistic negation interpretations which would outscope the modal marker.

4.3 The syntax of metalinguistic negation

The analysis builds on previous work showing that Metalinguistic negation (MlN), while relying on inferential processes to retrieve the metalinguistic interpretation, is triggered in contexts which require access to a contrast/set. It runs as follows:

(i) As discussed above, it is argued in the literature (see Horn 1989; Carston 1996; Larrivée 2011, etc) that metalinguistic interpretations are inferences that are drawn with respect to some contextual background. In this sense, MLN is standardly considered as a (purely) pragmatic phenomenon.

(ii) it has also been observed (see a.o. Moeschler 2010; Larrivée 2011; 2016) that MlN readings require overt or covert corrective material. As proposed in Larrivée (2011):

“M[eta]Linguisitc] N[egation] upsets the expected old/new information dynamics: it ranges over discourse-old material, and transforms a discourse-old information into a discourse-new focus. Modifying the expected flow of IS is why MLN gives rise to garden-path effects, and why more generally it is marked” (2011: 3).

In all of the cases discussed here, it is indeed possible to have a continuation which overtly states some corrective material:

(iii) as was also mentioned in Section 3 above, it has further been observed that MLN interpretations are triggered by some “signal”, often encoded as a phonological marking. Indeed, Horn (1989) notes the association of the metalinguistic negation interpretation with a specific intonation, in the form of a particular stress pattern. In the examples discussed in this paper, a wide scope metalinguistic negation interpretation is actually accessible in the presence of a heavy pitch accent on the negative marker (not, pas).

(iv) a strong accentuation in the form of stress is, among others but as a priority, associated with contrast. Selkirk (2012) claims that a L+H* pitch accent is most readily associated with Focus, and focus marking of a unit introduces a contrast. Typically, languages such as English, which have no syntactic marking of Focus, signal the contrast by a heavy stress:

(v) by (i) to (iv), and following Larrivée (2016) and Puskás (2012) who also give arguments to the fact that contrast marking is the MLN inference trigger, I will conclude that what enables to have access to MLN readings is the contrast (stress) marking on some element of the sentence.

(vi) I will adopt here a rather consensual view of the notion of contrast which claims that the role of a contrast is to introduce alternatives. Such a view has, for example, been proposed in order to account for the semantics of Focus (see e.g. Rooth 1992), but it extends to other types of contrast (as discussed in e.g. Jacobs 2001; Büring 2003; Frey 2006; Molnár 2006).

(vii) The stress associated with Focus thus signals the fact that alternatives will be triggered.

(viii) Other intonational patterns may also be associated with contrast, such as the “hat contour” discussed in Jackendoff (1972) and Büring (2003). While focus accent is associated with the selection of one alternative, to the exclusion of others, the “hat contour” argued by Büring to be the characteristic pattern of Contrastive Topic, evokes several, non-mutually exclusive alternatives (see also Molnár 1998 on the idea that contrastive topics introduce weak alternatives).

(ix) under the view that contrast has the property of giving access to alternatives, we can also conclude that MLN is conditioned by the availability of (contrasting) alternatives. In this sense, all MLN readings involve a contrast. Since metalinguistic negation appears in corrective contexts, its function of is to give access to alternatives.

(x) in the syntax, contrast (topic, focus) is encoded in some let-peripheral position. There is indeed a large bulk of work which has established cross-linguistically that the overt syntactic positions associated with contrast are located above the inflectional head(s). Ranging from Italian (Rizzi 1997) to Hungarian (Horváth 1986; E.Kiss 1987; Puskás 2000; Lipták 2001; 2011 a.o.) and Gungbe (Aboh 1998) among others, researchers have shown that the syntactic realization of focused constituents which encode (exhaustive) contrast, but also contrastive topics associated with non-exhaustive contrast involve overt or covert movement to dedicated left-peripheral functional positions. As discussed in Section 2.2.1, focused constituents move to a dedicated left-peripheral position located in the CP-domain, overtly in languages like Hungarian (72a) or Italian (72b):

Contrastive topics, which also bear a contrast feature (signaled by the diacritic √), can also appear in the left –periphery:

Languages like English do not (necessarily) have overt movement to these left-peripheral positions. Both focus and contrastive topics are prosodically marked, but do not undergo overt movement:

It has been proposed (see Drubig 1994; Wagner 2006) that in languages like English which do not display word-order variations, focused constituents undergo covert movement to the Focus position. I assume that similar covert movement raises contrastively topicalized material to a left-peripheral position (see Lipták 2011). This means that in situations in which contrast is signaled by stress but there is no overt movement to a left-peripheral position, as is the case for French pas and English not, there is covert movement to a contrast position.

(xi) since contrast occurs when some element is moved overtly or covertly to a left-peripheral position, and since MLN readings arise when there is contrast, (from ii and ix), I will propose that MLN is triggered if there is some element which raises overtly or covertly to a left-peripheral contrast position. Note, however, that assuming that all MLN readings are triggered by movement to the left –periphery is probably too strong. There are cases of MLN readings which involve low (local) negation, such as (75) below:

The argument of movement to a high contrast position is here more difficult to defend. The cases I am focusing on involve sentential negation, that is, negation at the clausal level. Therefore, I am not concerned with every kind of MLN, but only with those which are triggered by the presence of a stress on the (sentential) negator.

(xii) I claim that the syntactic implementation of negative configurations is a crucial part of metalinguistic processes. It goes as follows. As was discussed above, I assume, following Moscati that there is a high NEG position in the left periphery. In negative clauses, the C head is endowed with a negative clause-typing feature (Moscati 2006). In languages like French, the negative scope marker may either raise overtly if the verb it is attached to raises to C, or may undergo covert movement. In languages like English, the negative marker not raises covertly to the left-peripheral position. This brings negation to a sentential scope domain.

As such, negation will scope over the Focus or Contrastive Topic projection, where focused, resp. contrastively topicalized, material is licensed.

(xiii) alternatives are assumed to be of the same kind as the contrast-marked unit (see e.g. Büring 2003). I will propose further that syntactically as well, there has to be isomorphism. Consider the following cases of overt contrasts:

Both French (76) and Hungarian (77) show that what contrasts with the focused unit has to be of the same syntactic form (PPs in this case).

(xiv) Therefore, whatever NEG scopes over will be contrasted with identical segments.

(xv) However, syntactic movement of negation is blocked at the C(P) level; no negative marker is licensed above the clause-typing head. I have argued that this is what blocks descriptive negation from scoping above the bouletic modality markers discussed here.

(xvi) if the negative marker reaches a high left-peripheral position, alternatives will be of the same kind as what it takes in its scope. Since negation can target a focus position in the left periphery, alternatives are available. They will correspond maximally to the portion of the clause which is in the focus domain. Recall, however, that the bouletic modality markers of this study were argued to occur in SpeakerP, a high functional projection of the clause above C, the clause typing head.

(xvii) therefore, alternatives cannot include the bouletic modal marker.

(xviii) from (xi), (xiv) and (xvi), we can conclude that contrastive interpretations which scope over bouletic modality will not be available.

(xix) Since (proposition level) MLN readings build on the accessibility to alternatives, there will be nor reading in which bouletic modality is in the scope of negation, even in its MLN reading.

Therefore, metalinguistic negation readings at the sentence level are also syntactically constrained. The above analysis supports hypothesis II.