3.1 A brief overview of Irish clausal syntax

Irish is noted for its VSO order and for its rich system of complementisers. For the sake of brevity, I will start with the description of a negative sentence:²

The element responsible for encoding sentential negation in (6) is ní. According to McCloskey (1996b), ní is a negative complementiser, which is realised on C° (cf. also Moscati (2006)). This means that the standard syntactic position for the expression of sentential negation (and the only possible one) is left-peripheral.

I will adopt McCloskey’s (2017) recent analysis of Irish clause structure, according to which its VSO order is derived by movement of the finite verb to a PolP. The structure for (6) is thus as follows:

The adoption of the structure in (7) has several motivations. Firstly, as noted by McCloskey (2017), there are two possible subject positions in Irish. One is reserved for indefinite subjects, which can appear below vP peripheral adverbs (8a) and is characteristic of existential constructions, and the other is higher and reserved for definite subjects (8b):

Thus, indefinite subjects do not move out of vP, while definite subjects move to the canonical subject position in SpecTP.

The second relevant aspect of the structure in (7) concerns verb movement. If in all cases the subject never moves higher than SpecTP, one must account for the fact that the finite verb is always higher than the subject. Crucially, McCloskey (2017) proposes that the landing site of the finite verb is a polarity phrase, akin to Laka’s (1990) ΣP. Consider the following example:

Mura is a negative conditional complementiser realised in C° (a negative version of má, ‘if’). From that position, it scopes over both conjuncts in the sentence above. Nonetheless, while the conditional import applies to both conjuncts, the negative import only applies to the first one: (9) means ‘if I don’t go and they do win’, not ‘if I don’t go and they do not win’, which would be expected if the negative component of mura had the same scope of its conditional component. For this reason, McCloskey (2017) proposes that negation is realised in a position that is situated below C° in Irish, and that this position is a PolP. Negative complementisers are in turn the result of Agree between an [iNeg] feature on Pol°, which is identified as the locus of polarity and hence of negation in Irish, and a [uNeg] feature on C°. PolP is necessarily lower than CP and higher than TP. Since we also know that the verb is in turn higher than TP, it follows that Pol° is the best candidate as the landing site of the finite verb.³ The derivation of an Irish finite clause like (6) goes as follows:⁴

3.2 Irish ‘relativising’ complementisers

The second salient aspect of Irish syntax is the realisation of Ā-dependencies. Irish morphologically distinguishes between movement and resumption. In the case of Ā-movement to the left periphery, the complementiser αL is employed. The letter L stands for ‘lenition’, and the complementiser is so glossed because it lenites the initial consonant of the following verb (11a). In the case of resumption, the homophonous complementiser αN—where N stands for ‘nasalisation’—is used, which nasalises the initial consonant of the following verb (11b). The negative marker ní is a member of the class of complementisers across which no Ā-dependency takes place. While in positive matrix sentences this position is not realised, in positive subordinate clauses it is realised as go, as exemplified in (11c). This class of complementisers is employed in complementation.

McCloskey (2002; 2009), proposes the following structures for (11a) and (11b) respectively:

This is the model I will assume throughout this paper and which can be summarised in this fashion following McCloskey (2002): (i) a C° whose specifier is filled by Internal Merge is realised as αL; (ii) a C° whose specifier is realised by External Merge is realised as αN; (iii) a C° whose specifier is unfilled is realised as go. In (12a), duine moves to the left periphery, which triggers the appearance of the αL complementiser. In (12b), duine is base-generated in the left periphery, and the resumptive chain between it and the pronoun é triggers the appearance of the αN complementiser.

McCloskey assumes that in the αN case the relation between the element at the top of the dependency and that at the bottom is mediated by a null operator. This move has to do with the necessity of explaining some cases of ‘mixed chains’ instantiating a type of successive cyclic movement in which a clause headed by an αL complementiser dominates another headed by the αN complementiser. These cases are problematic, since the element that triggers the appearance of the αL complementiser cannot have moved from its base position in the clause headed by αN (since this form of the complementiser excludes that movement has taken place). The problematic structure is exemplified below:

This problem is easily solved by proposing that what moves is not the element at the top of the chain, but rather the null operator, which mediates the relation between the various elements that participate in the chain itself, as illustrated below, where the resumptive é agrees with Op, and this later moves to the Specifier of the higher CP headed by αL:

The fact that nothing moves in the αN case means that Irish resumptive pronouns can be used when movement is barred by island effects, as noted by McCloskey (1979) and Oda (2012):

It must be noted, though, that the form of the complementiser is not merely associated with the ‘passing’ of the element that moves across it. In fact, McCloskey (2002) associates each of the lexical forms of Irish complementisers with a specific featural makeup: a complementiser which bears both an Operator feature and an EPP-feature is realised as αL, while αN is taken to bear an EPP-feature, which requires its Specifier to be filled, but not an Operator feature which would be associated with movement. Finally, go is the lexical realisation for a complementiser that has neither feature—that is, its Specifier must not be filled to satisfy the EPP and, consequently, no Operator feature is there to trigger movement to that position. In this way the form of complementisers and the movement—or lack thereof—of a constituent across it receives a unified featural explanation (cf. Table 1).

From what was said so far, it should come as no surprise that embedding and relativising complementisers are in complementary distribution:

This fact has important consequences for the syntax of negation, too. These so-called relativising complementisers have negative, past, and negative past forms, which are identical with those of embedding complementisers of the go type, as exemplified below:⁵

We can draw some conclusions about the expression of sentential negation in Irish from the fact that embedding and relativising complementisers are in complementary distribution. We never get a hypothetical ‘negative marker+complementiser’ sequence like the following:

This is of course expected: if negation is of the same category as complementisers, and if go and a complementisers are in complementary distribution, it follows that ‘negation’ in Irish cannot occur analytically, that is, independently of embedding or relativising complementisers.

Summing up, standard sentential negation in Irish is invariably realised on complementisers, never as an independent projection. Since both embedding complementisers of the go type and relativising complementisers of the αL/αN type head their projection CP, it follows that negation is also realised as a head in Irish—or rather on a head. But apart from this syncretic quality, Irish sentential negation has other interesting syntactic and semantic properties, which I illustrate in the next section.

5.1 DemNeg+XP

What remains to be explained is the syntactic adjacency of the two elements in the αL case, then, as well as the adjacency between the demonic complex and the complementiser. As illustrated above, according to McCloskey (2009; 2018) this follows from the syntactic adjacency of the multiple specifiers of CP. Since in a cartographic perspective this solution is not viable, what I would like to propose is that in the αL case what one gets is the structure below:

Dheamhan is base-generated in a higher projection that encodes the left-peripheral ‘sentential polarity’. Of course, I take this to be the same position in both cases. I thus propose that dheamhan is an incarnation of the highest syntactic expression of polarity, whose positioning in the left-periphery was previously proposed by Greco (2019; 2020a, b). This assumption has several consequences. First, under this analysis DemNeg is a monoclausal structure. The fact that the lower clause is headed by the positive complementiser a (i.e., has the positive, not the negative form of the complementiser) means that it is dheamhan that is responsible for negating the sentence, and thus that it is not a left-peripheral metalinguistic negator or an ‘external negator’ (Horn 1985; 1989; Martins 2014; 2020). Still, the fact that a sentence with DemNeg does not need a further negative marker for sentential negation to arise does not mean that it cannot. An anonymous reviewer notes that the present analysis would be made stronger if dheamhan could co-occur with the lower standard negation on Pol°/C°. As Jim McCloskey (p.c.) notes, this cooccurrence is indeed routinely possible and, crucially, regularly gives rise to a double negation reading, not to a concord reading, as exemplified below:

This point is also relevant because it means that, after all, Acquaviva’s (1996) generalisation that Irish obeys the ‘one sentence, one negation’ rule is not so stringent: Irish indeed cannot have two expressions of negation below PolP, but it has no problem coupling standard negation with an additional emphatic negator.

Several facts indicate that dheamhan is indeed an expression of descriptive/sentential negation. This is confirmed, apart from the ability of giving rise to double negation with a lower marker, by the fact that it licenses negative indefinites (40a), polarity-sensitive minimisers like duine ná deoraí, ‘person nor stranger’, i.e., ‘anybody at all’, and exceptives (40b):

Metalinguistic negators, on the one hand, are like DemNeg in being compatible with a lower expression of negation, with which they yield a double negation at the logical level (41). On the other hand, they are incapable of licensing polarity items (42). Following Horn’s (1985; 1989) classic analysis, this is because metalinguistic negation is a pragmatic operator functioning at discourse level, and as such it cannot license PIs, which need a semantic rather than pragmatic licensing (cf. also Martins 2020):

In this connection, with respect to Acquaviva’s (1996) observations, note that monotone decreasing quantifiers do occur in clefts with a lower negative complementiser (and a higher positive copula):

In (43), sentential negation is realised on the negative complementiser, which in McCloskey’s (2017) terms means that it is encoded on the Pol° occupied by the finite verb. The second aspect that is relevant to the structure in (38) is the position that is assigned to the XP. As mentioned above, McCloskey (2009; 2018) takes this element to sit in SpecCP. This is also the approach he takes in general for the structure of Ā-dependencies (cf. McCloskey 2002). There are several reasons to propose that the XP raises to SpecFocP. Firstly, there are some similarities between DemNeg and Germanic Negative Inversion. Negative Inversion is taken by some authors, in particular Haegeman (2000), to be the result of movement of the negated constituent to a Contrastive or Corrective Focus position, along the following lines:

The same type of interpretation can be associated with the demonic constituent:

In (45), the NP that follows dheamhan is not referential; rather, it seems to denote a scalar endpoint, or perhaps even to have a kind reading, deriving an interpretation that is similar to the scalar interpretation of NPs in negative inversion. This idea is also reinforced by the use of minimisers in this context, which also denote a non-referential scalar endpoint. Such minimisers are polarity-sensitive items licensed only under nonveridical operators:

Ostrove (2013) notes that Irish only expresses constituent negation with a negative cleft sentence. This is the natural consequence of the Irish restriction on the expression of negation below PolP since, as noted above (Acquaviva 1996), a lower expression of negation would be ungrammatical in Irish:

Now, crucially, DemNeg can indeed express constituent/contrastive negation, as in (48–49). It follows, then, that in these cases the import of DemNeg (or, rather, the interpretation of the contrastive constituent) is not scalar, but contrastive:¹⁴

DemNeg can also be employed in denials, as in the following example (Ó Siadhail 1980: 49; our italics):

In (50), B’s denial focusses on the interlocutor’s claim that the people who are referred to were poor, and while denial as an external negator scopes over full propositions, the part of the preceding utterance to which the speaker objects (the predicate) is highlighted by raising it to a Focus position. In a way, DemNeg+XP is perfectly identical to Bare DemNeg: in both cases the negator scopes over the whole sentence, but with DemNeg+XP a constituent has moved to the SpecFocP of that sentence, independently of the presence of dheamhan.

Further evidence for analysing the XP following dheamhan as sitting in SpecFocP comes from fragment answers that are derived from DemNeg, in which ellipsis takes place right after the XP:¹⁵

Kramer & Rawlins’ (2011) analysis of answer particles as sitting in a left-peripheral ΣP/PolP chimes in with the fact that DemNeg can be used as an answer particle to yes-no questions, rather than as an answer to constituent questions. This is especially interesting with respect to the position defended here that DemNeg is a sentential negator, and not just a constituent negator, not even in the DemNeg+XP case.

A further argument from cartography comes from the fact that DemNeg appears below topical material:

Following Benincà & Poletto’s (2004) analysis of the left periphery, topics are consistently situated above the Focus layer, which is exactly the positioning we find with respect to DemNeg. These facts corroborate the hypothesis that dheamhan sits in a Focus position, in line with the semantic import of the construction and with the parallels with similar negative focalisation strategies like Germanic Negative Inversion.

What can be concluded from the data above is that DemNeg, despite its emphatic character, is not a metalinguistic negator, but rather an expression of semantic (sentential) negation. The variability in the interpretation of DemNeg constitutes evidence for this: the interpretation of DemNeg is not limited to metalinguistic negation or denial, and the variability itself is due to the interpretation that is given to the focussed XP (that is, whether a scalar, contrastive, or metalinguistic interpretation is involved).

5.2 Bare Demonic Negation

(53) illustrates what Bare DemNeg looks like so far:

The presence of the αN complementiser with Bare DemNeg represents an important piece of evidence in favour of the base generation of dheamhan. Given its syntactic properties, αN is the complementiser that is used not only when the XP at the top of the Ā-dependency is base-generated in the left periphery, but also when it is reprised by a resumptive element. That is because, quite obviously, in the αN case the XP does not leave a trace and it can thus usually be coindexed with a resumptive pronoun. Usually, but not always: McCloskey (2002) notes an interesting interaction between the semantics of adverbial phrases and αN/resumption. Durative and frequency adverbials only take αL, as in (54). Others (locative, manner, and temporal) can freely alternate between the use of αL and αN, as exemplified in (55–56). Reason adverbials, on the other hand, can only take the αN complementiser (57):

McCloskey (2002) explains the variable behaviour of adjuncts with the variation in the availability of a null pronominal category for the corresponding adverbial. (Null) pronominals corresponding to locative and temporal adverbials are indeed very common (Cinque 1990), and the αN case is simply the result of the coindexation of the locative/temporal adverbial with such a null pronominal. Thus, the αL/αN alternation in these cases is merely a reflection of the optionality in the choice of movement vs resumption. On the other hand, there are no (null) pronouns corresponding to other adverbial types, including durative and frequency ones, which leaves movement as the only option in this case, triggering the appearance of the αL complementiser.

The most interesting observation by McCloskey (2002) concerns the obligatoriness of αN with reason adverbials. He notes that Rizzi (1990; 1996) argues that the interrogative (i.e., left-peripheral) forms of reason adverbials are always base-generated in SpecCP. This view was updated with the introduction of IntP (cf. also Rizzi & Bocci 2017: 5), in whose specifier reason interrogatives like why are base-generated. Base-generation of reason adverbials in this position differentiates them from other wh-elements, and the fact that Irish reason adverbials only occur with αN confirms this view. The αN complementiser is thus triggered not so much by resumption, but rather by External Merge of an item in the specifier of CP, regardless of whether the constituent above αN is later resumed. This would mean that there is no need to assume the presence of a resumptive element, not even a null one, whenever αN is realised. It is for this reason that McCloskey (2009; 2018) takes dheamhan, too, to be base-generated in the left periphery, without need to ask with what it is coindexed.

This should obviously not lead one to think that dheamhan is an adjunct: what this negator and reason adverbials or adjuncts have in common is merely the fact that both are externally merged above CP. Notably, the fact that in DemNeg+XP the presence of an XP triggers the appearance of the αL complementiser means that its morphology, which is of the αN type when bare dheamhan is present, is overwritten by the subsequent movement operation. The exact description of this mechanism of overwriting depends upon the model of the relation between syntax and morphology one adopts (realisational or postsyntactic theories of morphology, like nanosyntax and Distributed Morphology, seem especially well equipped to deal with this specific case), but clearly the morphology that signals Internal Merge (the αL complementiser) takes precedence over that for External Merge (the αN complementiser) when both operations take place (that is, when dheamhan is merged and the XP is later moved). This is also in line with Adger & Ramchand’s (2005) idea that αL has one more feature than αN. In their model, the completive complementiser go is featureless, αN has a feature [Λ] which is interpreted as predicate abstraction, and αL has that feature plus an [Id] feature, which is interpreted as a variable that ensures the pairing of the complementiser with a lower pro category. Similarly, McCloskey (2002) bases his analysis on the idea that αL has both an EPP feature and an Op-feature, αN only has an EPP feature, and go has neither. The fact that αL is under both analyses understood to be featurally richer than αN also explains why the αL morphology overwrites the αN morphology: since αL is more specific, the Elsewhere Condition selects it and rewrites αN as αL, because grammar finds in the lexicon an element that is more specific for the featural makeup of the C° node in that case. Likewise, αN is selected as it is more specific than go when an EPP/[Λ] feature is present on C°.

Taking stock of this observation, we can now draw a conclusion of some importance. DemNeg is a true semantic negator, as evidenced by its interpretational and functional versatility and licensing capabilities. It is also clearly base-generated in the left periphery (indisputably so in the case of Bare DemNeg). Ergo, semantic/sentential negation can be externally merged in its scope position and does not have to move there from a VP-internal negative projection.

In the case of non-emphatic sentential negation (negative complementisers), one could still maintain a (relatively) lower origin for negation. De Clercq’s (2013: 114) PolP corresponds not only to Zanuttini’s (1997) NegP1, but also to what she calls ‘sentential negative polarity’, the position responsible for making a sentence negative, even in languages in which the sentential negative marker appears below TP. She also assumes that PolP, and not FinP, is the lowest projection (i.e., the edge) of the left periphery. As mentioned, McCloskey (2017) also links the expression of negation in Irish to an Agree relation between the interpretable polarity features on PolP and the unintepretable ones on CP, of which the negative form of the complementiser is the morphological reflex.

But in the case of DemNeg, no relation seems to take place between the negator and PolP. Rather, the polarity of Pol° seems to be not so much positive as neutral, or unmarked: once dheamhan, a descriptive negator in Horn’s (1989) terms, is merged in the left periphery nothing else is needed in the lower projections to negate the sentence. Given McCloskey’s (2017) analysis, negative complementisers are not the locus of negation, but merely a manifestation of its presence in the position that is immediately below, PolP. No such marking is necessary when another expression of negation like dheamhan is present, and thus there is no need to enrich Pol° with a [iNeg] feature to cause the appearance of negative morphology on its usual bearer, C°. Hence, neither Pol° nor C° present any trace of negative morphology, since Agree with DemNeg has no reason to take place. Moreover, proposing that Pol° needs to agree with dheamhan would have the additional unwelcome consequence of treating it as the probe in the case of standard negation and as the goal in that of DemNeg. The fact that, as shown in (39), it is possible to merge an additional negation in Pol° and that it regularly gives rise to double negation in composition with DemNeg confirms this analysis.

The proposal that DemNeg is a left-peripheral incarnation of sentential negation also makes sense of its pragmatic properties. Ramchand & Svenonius (2014: 169) propose that Fin° (Fin* in their model) is the head which “needs to combine with a situational description to create a proposition by binding off the situational variable s, anchoring it (expressing a relationship to) the utterance situation s*”. Just as in their model situations constitute an enrichment of events, propositions are further enriched with discourse features. This is interesting with respect to its relation to DemNeg especially for this reason: DemNeg is a finite negator. Irish only has one nonfinite negator, gan (base-generated in Pol° according to McCloskey (2017)). Secondly, the authors add that “the proposition is a relationship between a situation and an assertor and contains information about the speaker and speaker attitude as well as encoding of familiarity and novelty of the information to the members of the utterance situation”. That is where the ‘emphatic’ character of DemNeg comes in: emphatic negation is usually associated with a stronger commitment on the part of the speaker, and in this strict sense DemNeg may be described as a propositional negator (where the term ‘proposition’ is understood in the attitudinal sense of Ramchand & Svenonius 2014), rather than a sentential one: one that selects FinP once this has combined with the situational variable, to yield a pragmatically enriched negation. The view that DemNeg is linked to a precise pragmatic attitude on the part of the speaker also explains why it is only compatible with finite assertions: dheamhan is incompatible with questions or imperatives, because it is a specialised negator that expresses a strong commitment to the negative truth value of the lower proposition. It stands to reason that clause types that specify no truth value are incompatible with dheamhan. This is not due to the fact that dheamhan is negative, though, but with the fact that it is emphatic, which in turn is the consequence of its left-peripheral position. In this sense, DemNeg represents a confirmation of Greco’s (2020a: 777) idea that the interpretation of negation is dependent on the position in which it is realised, and that a left-peripheral negator will have certain pragmatic and speaker-oriented properties because of the domain in which it occurs.

The concept of propositional negator may also justify the fact that Irish, a parsimonious language when it comes to the expression of negation, has two semantic expressions of negation just a phrase apart. Although identical in semantic terms, these two exponents have a different function at the pragmatic level. The expression of negative emphasis is enough to motivate the presence of DemNeg in the scant negative paradigm of the Irish language. Moreover, one can view the development of DemNeg as a case of syntacticisation of discourse (Haegeman & Hill 2013): dheamhan, as was mentioned above, derives from do dheamhan, ‘to a demon’ (the initial consonant of dheamhan still bears the lenition—graphically h—that was caused by the preposition do, ‘to’, now mostly absent from the construction). Although we have no record of the pragmatic prehistory of DemNeg (the earliest use I could trace, in the 17^th century, is already modern-looking), what happened in the development of DemNeg is that a hell-word, not dissimilar from English like hell, received a fixed position in syntax because of its strict association with negative meaning, thus becoming syntacticised. Importantly, especially from a cartographic perspective, syntacticisation requires an expression to be identified with a syntactic position, and such a position must obviously be available in syntax. The fact that DemNeg developed from a hell-word (i.e., an external, pragmatic negator) into a sentential/propositional negator also constitutes evidence that a potent expression of polarity exists in the syntactic structure above FinP, such that speakers could at a certain point in its history identify DemNeg with the lexical realisation of that position.

5.3 Bare Demonic Negation and fragment answers

I would like to conclude this analysis with some considerations on the strategy that derives fragment answers from Bare DemNeg, which has a general relevance for the syntax of Irish negation. Consider the following elliptical use of Bare DemNeg. As mentioned above, DemNeg can serve as a basis for emphatic negative fragment answers to both constituent questions and yes-no questions in Irish:

Now, while with answers based on DemNeg+XP fragment ellipsis cuts off after the focussed constituent, with answers derived from Bare DemNeg dheamhan does not appear on its own, but is followed by the pronoun é, ‘it’, which as far as I know does not occur in non-elliptical versions of Bare DemNeg. I identify this pronoun with those which Ostrove (2013: 42) takes “to be some kind of anaphoric element which is co-referential with the embedded clause (McCloskey 1996a)”, in other words a sentential pro-form. The question is what this pro-form precisely co-refers with in demonic fragments. The answer comes from the very interesting interaction of demonic answers with negative antecedents:

Crucially, bare demonic fragments to negative questions or statements of the dheamhan é type confirm the negation of the previous utterance (i.e., they do not convey double negation). This means that the negative polarity of the previous utterance is not reprised along with the rest of the propositional content by the pro-form é, which is rather unmarked (i.e., non-negative) as in the non-elliptical cases of DemNeg seen above. This tells us much about which constituent the pro-form co-refers with. The general understanding for Irish syntax is that what follows the position of the finite verb is one large postverbal constituent that includes all the clausal material to the exclusion of the finite verb (for an overview of this long-held claim, cf. McCloskey (2011; 2017)). This structure is illustrated below for the sentence Ní fhaca mé duine, ‘I did not see a person/man’:

If, following McCloskey (2017), the position of the finite verb is PolP, then the ‘large constituent’ that is its sister is TP/IP. Importantly, PolP is also the lowest position in which polarity, and hence negation, is encoded in Irish. Since pro-forms can only co-refer with constituents, and since the TP forms one constituent to the exclusion of the finite verb, the reason the pro-form é does not cause double negation with dheamhan is that it co-refers only with TP, not with PolP+TP. Now, since polarity (positive or negative) is absent from (the lower) TP in Irish, being instead encoded on PolP (the landing site of the finite verb) or in a higher position, it follows that in demonic fragments the pro-form é can never include the polarity of the previous utterance, but only the polarly unmarked situation encoded in TP, prior to composition with PolP. This being so, the negation of DemNeg will never compose with a sentential negation present in the previous utterance, and thus double negation does not arise in fragment answers based upon Bare DemNeg. This brings the present analysis of DemNeg once more in line with Greco’s (2019; 2020a, b) treatment of Surprise Negation, another kind of left-peripheral ‘special’ negation which sits in the CP layer and is followed by movement of TP to the SpecFocP, the position I propose for the TP pro-form é in demonic fragments.

Perhaps the most interesting conclusion one can draw from (60) is the following: since Pol° is the lowest position in which negation can be expressed, this would mean that (sentential) negation in Irish is nothing but ‘constituent negation of TP’, an intriguing result for those approaches that seek to unify sentential and constituent negation. And since no expression of negation can be found in Irish below PolP (not just emphatic negation), the idea that negation does not necessarily originate in the thematic layer in all languages receives further corroboration.