Metalinguistic negation from an informational perspective

This paper revisits the definition of metalinguistic negation (MLN) illustrated by e.g. They don’t have kids, they have children. A new definition is proposed that rests on two properties. The first is that MLN is a corrective speech-act. The second is that the sentence used to perform the speech-act has a paradoxical Information Structure: it is discourse-old material, along with the corrected a segment that is however treated as discourse-new by virtue of being focused and contrasted to the correcting segment. These properties are used to explain established features of MLN. MLN’s speech-act status accounts for the distinctive behaviour of relevant connectives. The paradoxical Information Structure distinguishes MLN from other uses of negation, relates it to other corrective constructions and metalinguistic phenomena (e.g. in conditionals and questions) and accounts for the alleged marked status of metalinguistic configurations. How MLN can be mapped by a cartographic approach is speculated upon.


Introduction
The purpose of this paper is to contribute to the ongoing debate about the definition of metalinguistic negation (henceforth, MLN). MLN standardly refers to "a formally negative utterance which is used to object to a previous utterance on any grounds whatever, including the way it was pronounced" (Horn 1989(Horn /2001. 1 It corrects the pronunciation of items, their morphological make-up (1b), their sociolinguistic (1a, 1c) and pragmatic effects (1d, e).
(1) a. They don't have kids, they have children. b. She's not interesting, she's interested. c. I don't want to buy pants, but trousers. d. He isn't a proto-fascist demagogue, he's the POTUS. e. She hasn't seen many of Murnau's films, she's seen all of them. 2 1 Similarly, "the disagreement between the parties to the discourse is not over matters of fact but over what words they ought to use in referring to the things they are talking about" (McCawley 1991: 190). A reviewer asserts that "what the negation in (1) corrects is not the proposition that they have kids, but the metalinguistic proposition that their children can properly be called kids." Another reviewer suggests that this may be captured by "an implicit or abstract "APPROPRIATE" operator" commanding the corrected segment. The ad hoc operator might technically work for MLN, but makes the wrong prediction for other metalinguistic configurations discussed in this paper. In Since they moved to New York, they are stressed out, it is not the (referential) appropriateness of the expression stressed out that is being considered, but rather the (sociolinguistic) fact that it is an Americanism. Appropriateness seems to derive from MLN (if an item is corrected, it is probably because it is judged inappropriate by the speaker), rather than metalinguistic value itself. 2 In this paper, following the recommendation of reviewers, I am steering clear of cardinals, of the type She doesn't have three children, she has four. A reviewer proposes that there is a consensus in the literature that cardinals "express 'exactly n' and do not merely implicate an upper bound, contra Horn's original Gricean Beyond this, how MLN should be defined is unclear. Such a definition should do at least three things, relating to the morpho-syntactic reflexes of MLN, how it differs from other negative uses, and what it has in common with other metalinguistic uses. A definition of MLN should account for its morpho-syntactic reflexes. Of these, the most stable cross-linguistic behaviour is that of the connective that introduces the correction. In some languages, there is a connective specialised for the MLN use, as German sondern and Spanish sino (Anscombre & Ducrot 1977;Horn 1989Horn /2001Schwenter 2002). As for languages that do not have a specialised connective for the MLN reading, the adversative connective that they use for both MLN and denial nonetheless can behave differently under each interpretation. This is the case with English but introducing a full clause, that is acceptable with denial, but infelicitous with MLN; (2) a. They don't have kids, (*but) they have children. MLN b. They do not have kids, (but) they have plenty of pets. Denial And reverse MLN acceptability with a fragment: (2) c. They don't have kids, *(but) children. MLN Another cross-linguistically stable behaviour is that MLN cannot be expressed by a lexically incorporated negative.
(3) a. It's not probable, but certain that they will show up. b. *It's improbable, but certain that they will show up.
(4) a. John borrows money not from anyone in his office but from his family. b. *John borrows money from no-one in his office but from his family. (5) a. They don't have kids, they have children. MLN b. That's not true! They do not have kids. Denial c. At least, they don't have kids. Descriptive (6) a. She's not interesting, she's interested. MLN b. She's not interesting, but she's not extraordinarily boring either. Denial c. Pity she's not interesting. Descriptive analysis; cf. e.g. Koenig (1991), Fretheim (1992), Horn (1992), Geurts (1998), and especially Bultinck (2005)." Cardinals would therefore behave differently from other scalars in discourse: the question whether someone solved three theorems apparently cannot be answered without knowing if the speaker means exactly three or at most three, whereas there is no such requirement on answering the question whether someone solve some theorems. Cardinals are also less resilient in understatements: the statement Lise has three children when Lise has four is more likely to be judged false than I assume that there are three major uses, following the initial proposal by Ducrot (1972). MLN is a reactive, corrective act. The corrected segment is rejected, while the rest of the sentence plus the correcting segment are the case: (7a) asserts a desire to buy what is called trousers. Denial is a reactive move that rejects an entire sentence. A descriptive negation is an initiative move (Foolen 1991) that similarly rejects the material under its scope. Unlike MLN, neither denial nor descriptive uses necessarily imply that parts of the sentence are the case, and a desire to buy is asserted by neither (7b) and (7c). This ternary distinction between descriptive, denial and metalinguistic uses is in contrast to the binary opposition between descriptive and metalinguistic proposed by some researchers. For Horn (1989Horn ( /2001 and Seuren (1988: 191) among others, metalinguistic and denial are part of the same category, that is opposed to descriptive uses. I maintain (with Schwenter & Johnson 2011 contra Martins 2014, as well as contra Batllori 2016 andDéprez, Tubau, Cheylus &Espinal 2015, to cite but a few recent papers) that MLN and denial are different uses, that are associated with different effects. As discussed in the previous paragraph, MLN implies that a part of the corrected sentence is the case, something that cannot necessarily be assumed with denial. MLN and denial have a divergent relation to connective but as in (2)  Finally, on top of explaining morpho-syntactic reflexes of MLN and distinguishing MLN from denial and descriptive uses, a definition of MLN should identify the properties that the different instantiations of MLN share. I assume that such common properties do exist, moving away from the scepticism expressed by e.g. Geurts (1998) and Davies (2016). Such properties should also be found with other metalinguistic configurations, as in conditionals, comparatives and questions.
If Iran is going nucular, you must be a George Bush admirer.
The identification of a property common to metalinguistic uses 3 should allow an answer to be formulated to the questions of why such usage is felt to be marked in terms of frequency and interpretation. The paper is organised as follows. It reviews, after a brief evocation of the definition of MLN in terms of truth-conditions, its characterisation as a corrective construction, recently proposed by Jacques Moeschler (Moeschler 2013; and collaborators (Blochowiak, Grisot & Moeschler 2016;Albu 2017). While the characterisation seems intuitively correct, it does not deal with the well-established morpho-syntactic reflexes of MLN. I argue that these reflexes follow from analysing MLN as a corrective configuration. I then articulate the Information Structure of the speech-act. It is organised as a discourseold sequence from which the corrected segment is turned into discourse-new information as a result of its focal status. The change in informational flow common to metalinguistic uses explains their markedness. How this paradoxical informational configuration can be mapped by a cartographic approach is speculated upon in the final section.
A number of novel proposals are put forward. A new definition is provided of MLN that appeals to its speech-act status and informational configuration. The definition aims to account for morpho-syntactic reflexes of MLN, notably its relation to connectives, and for its relation to different instantiations of corrective and metalinguistic structures. I hope to move the debate forward by showing that correction applied to a focused segment of a discourse-old sequence is the defining characteristic of MLN. We start below with a discussion of two existing definitions.

Is metalinguistic negation truth-conditional?
It was proposed early on that unlike other uses of negation, MLN is unconcerned with matters of truth (i.a. Horn 1985). It focuses on elements that are not part of the truth-conditions of a sentence. Such elements comprise morphological, sociolinguistic and pragmatic dimensions, as shown by the initial illustrations, repeated below for convenience.
(1) a. They don't have kids, they have children. b. She's not interesting, she's interested. c. I don't want to buy pants, but trousers. d. He isn't a proto-fascist demagogue, he's the POTUS. e. She hasn't seen many of Murnau's films, she's seen all of them.
None of these elements is normally part of the truth conditions of a sentence. In a modular view of language à la Fodor, morphological, sociolinguistic and pragmatic dimensions should not be accessible to the semantic module. This separation is nicely illustrated by the following literary example.
(16) "And then you won't have no one left to teach." "anyone," he says. "You won't have anyone left to teach. Not no one." Laura looks at him with an expression she might also give someone who's speaking to her in Latin. "It's a double negative," he says. "Won't and no one." (Nathan Hill, 2016, The Nix) In a vericonditional usage, the sequence Not no one in the third line would lead to cancellation of the two negators and yield an existential interpretation represented by someone. Yet, in the MLN usage illustrated by the passage, not no one is not intending to communicate the equivalent of the incoherent You won't have anyone left to teach. Someone.
The intention is to signal that the word no one is sociolinguistically inappropriate in that context and should be replaced by anyone. One way to deal with such cases is for all the dimensions of a sentence, including the morphological, pragmatic and sociolinguistic ones, to be entered into the semantic module (and be "semanticised"). This procedure may be applied to all sentences, which considerably complicates the semantic representation of a proposition, or only for those ML cases, which might appear ad hoc. Another way to approach MLN is not to deal with it in the semantic module at all; MLN then becomes a non-truth-conditional negation. Since MLN is expressed by the same negative markers that convey truth-conditional uses in most languages (but see e.g. Kroeger 2014), negation must be cross-linguistically ambiguous, the consequence of which has been extensively discussed (especially in the late 1980's, see Carston 1988;Seuren 1988;Burton-Roberts 1989; and the overview in Pitts 2011). The proposal of a non-truth-conditional definition of MLN seems unconvincing (for a general recent assessment, see Davis 2016). MLN may be marked, but it is interpreted, and interpreted without much greater difficulties than other uses of negation (Noh, Choo & Ko 2013;Blochowiak, Grisot & Moeschler 2016). Foolen finds that it is "intuitively acceptable to ask whether The job isn't pretty hard is true or false. Although the utterance may have as its primary function the rejection of (part of) the positive counterpart as stated by the previous speaker, the ground for rejection is the presumed falsity of that positive counterpart. " (1991: 225). Similarly, when choice of lexical expression is being rejected, as in (1d) above, questions of truth may be at stake: an exchange as to whether a certain person is a demagogue or not does not seem devoid of meaning. 4 Finally, if MLN had no relation to truth-conditions, it should not be found with explicit evocations of truth (Carston 1998: 326, 336). In other words, a sequence like It's not true that she has seen many of Murnau's films, she's seen all of them should just not be felicitous, contrary to facts. The existence of such cases constitutes a serious problem for any approach that calls into question the vericonditional dimension of MLN.
This section has provided arguments against defining MLN as non-vericonditional. Lack of truth conditions is not present in all cases, and vericonditional analyses have been proposed -for Geurts & Maier (2005), the corrected segment involves reference to a linguistic expression. Having given reasons to move away from a non-truth-conditional definition of MLN, I now turn to a more recent proposal.

MLN as a corrective use?
Reference to truth conditions need not be the only approach to defining MLN. Other proposals have been recently made that do not depend on them. Of interest is the current work by Jacques Moeschler and his collaborators (Moeschler 2013;Blochowiak, Grisot & Moeschler 2016;Albu 2017). Moeschler's goal is to provide a model distinguishing MLN from denial and descriptive uses. The model is based on two major properties. One is the outcome of the use of negation (see in particular Albu 2017, who had made the point in the earlier manuscript versions). The proposal is that the outcome of descriptive uses is negative, that of denial is negative with potential positive continuation, and that of MLN positive. In the illustrations below, the descriptive use evokes the negation of having children, as does denial (with the added possibility of a positive continuation, that of having pets for instance), while MLN evokes the positive fact of having what is appropriately called "children".
(5) a. They don't have kids, they have children. MLN b. That's not true! They do not have kids (, but they have plenty of pets). Denial c. At least, they don't have kids. Descriptive The second property is the behaviour of the connective relating the negative clause and its continuation. Connectives in some languages specifically relate to MLN. This is the case with German sondern, Spanish sino, but also Romanian ci (Bîlbîie & Winterstein 2011;Albu 2017). Moeschler seeks to relate a typical connector to each of the three negative uses. 5 (17) French (Moeschler 2016) a. Abi n' est pas belle, au contraire, elle Descriptive Abi neg1 be-pr-3sg neg2 pretty on the contrary, 3-sg-nom est quelconque. be-pr-3sg ordinary 'Abi isn't pretty, on the contrary, she's ordinary.' b. Non, Abi n' est pas belle, mais elle Denial no Abi neg1 be-pr-3sg neg2 pretty but 3sg-nom est gentille. be-pr-3sg nice 'No, Abi isn't pretty, but she's nice.' c. Abi n'est pas belle, puisqu'elle est MLN Abi neg1 be-pr-3sg neg2 pretty since 3sg-nom be-pr-3sg superbe. drop-dead gorgeous 'Abi isn't pretty, since she's drop-dead gorgeous.' These properties provide an elegant, symmetrical model of the three usages of negation. In this model, MLN is a corrective use of negation. That seems intuitively correct. Unexpectedly for the proposal however, correction is not always explicitly required of MLN (Larrivée 2011), as shown by the following attestations from French. However the fact that the corrected segment is a PPI which in the immediate scope of negation can only be used with a MLN, suggests that a correction is in order. The correction is left to be inferred from the context: that the speaker in (18a) indeed had tears in their eyes and that the one in (18b) is thinking about their wife constantly. That these can be accommodated by Moeschler's model as they would have to be by any model does not cast important doubts on the proposal.
Doubts do however arise as to the ability of the model to fully engage with wellrecognised morpho-syntactic reflexes of MLN. This particularly concerns patterns of connective behaviour and scopal conditions. Regarding connective behaviour, remember that with MLN, but cannot introduce a clausal correction (2a), unlike with denial (2b). Also, but must introduce a fragment MLN correction (2c).
(2) a. (3) a. It's not probable, but certain, that they will show up. MLN b. *It's improbable, but certain, that they will show up MLN c. -It's probable that the Spencers will show up.
Denial -Come on, it's not probable that they will. d. -It's probable that the Spencers will show up.
Denial -?*Come on, it's improbable that they will. e. After all, it's improbable that they will show up. Descriptive This is related to the observation that MLN is incompatible with a lexically incorporated negative. 6 (4) McCawley (1991: 201) a. John borrows money not from anyone in his office but from his family. b. *John borrows money from no-one in his office but from his family Not only does MLN have clausal scope, it has high clausal scope. In Indonesian, the MLN marker (bukan) is always in a higher position than the co-occurring descriptive marker (tidak) (Kroeger 2014). In Mandarin Chinese, MLN is more readily available when the negative attaches to a higher verb such as an auxiliary or a modal (Wible & Chen 2000). In American Sign Language, MLN can only be expressed by a negative headshake that ranges over the whole clause, with or without localised manual signing, that cannot on its own communicate MLN (Fleckenstein & Yoon 2016). Again, why MLN associates to high clausal scope is not immediately explained by Moeschler's proposals.
In sum, the proposal by Moeschler and colleagues to characterise MLN as a corrective use of negation seems intuitively correct. MLN does yield a positive outcome and is associated with particular connective behaviour. However, the characterisation is not clearly related to the well-established syntactic reflexes of MLN, and the reasons for the observed patterns of connectives and the scopal condition remain to be provided. The next section develops an alternative definition of MLN and considers how it can help account for its syntactic reflexes.

MLN as a corrective speech-act
The previous section has shown that it makes sense to characterise MLN as a corrective use. But what does it mean for a negative to have a corrective use? That is, at what level of representation should this characteristic apply? An answer starts to emerge when we reconsider the scope of MLN.
While having high scope, MLN cannot focus over everything. It never corrects the speech-act value of a sentence. 7 Consider the following elicitations that attempt to convey the rejection of an assertive in favour of an exclamative and of an interrogative respectively. (20) A. Abi is beautiful. B. *Abi isn't beautiful, boy is Abi beautiful! B'. *Abi isn't beautiful, is Abi beautiful?
They are infelicitous to the point of being barely intelligible. That is because negation never scopes over the speech-act value of the sentence that contains it. 8 It comes under the scope of the speech-act, and is restricted to assertion. 6 A reviewer finds (4b) not as bad as McCawley did. The following might be more convincing minimal pairs.
(i) A. Did you really say to the Spencers that the Duchess has kids? B. ?I didn't say to anyone that she has kids, but children.
(ii) A. Did you really say to the Spencers that the Duchess has kids? B. *I said to no one that she has kids, but children.
(iii) A. Did you really say that the Duchess has kids to the Spencers? B. *I said that she has kids, but children to no one.
The issue of acceptability is of course related to that of elicitations (see i.a. the discussion in Grisot 2017). 7 Thanks to Genoveva Puskas for discussions on this. 8 Thus, Not "Abi is beautiful." but "Boy, is Abi beautiful!" is a plausible sequence; but the negative is not part of the sentence that it corrects; and it comes under the global assertive act of correction that it cannot itself challenge.
The assertive value of the speech-act in which MLN is found explains an observation by Neeleman & Vermeulen (2011) that both corrected and correcting segments are not presupposed but asserted. This is shown by the redundancy of the following.
(21) a. #John invited Pia and he invited not BILL but PIA. b. #John invited not BILL but PIA and he invited Pia. c. #John didn't invite Bill and he invited not BILL but PIA. d. #John invited not BILL but PIA and he didn't invite Bill.
The fact that John invited not Bill but Pia is redundant with respect to John invited Pia in (21b) means that both assert that Pia was invited. This confirms that MLN is an assertive speech-act, where the assertion scopes over the corrective function. I therefore propose that a single speech-act is involved in the following equivalent structures.
(19) a. They have children, (*but) they don't have kids. c. They have children, (*but) not kids.
Their function is to assert a correction brought about by an inappropriate designation in the antecedent context. A strong argument in favour of this is the fact that even in biclausal MLN, the second clause cannot convey a different speech-act. Witness the infelicity of the following, This observation shows that a single assertive speech-act is communicated by MLN, which isn't necessarily the case with denial uses. The observation also makes visible, again, the differential behaviour of connectives. But how is one to explain the impossibility of but for bi-clausal MLN and fragment reverse MLN, and its compulsory use in fragment MLN? I propose here an explanation based on speech-act status. With bi-clausal cases like They don't have kids, they have children (and the reverse They have children, they don't have kids), the addition of a but is impossible with MLN. This is due to the fact that adding a connective would allow the clauses to be conceived as two separate speech-acts, as is the case with denial in (23) above, whereas a single corrective speech-act is involved with MLN. By contrast, fragment cases such as They don't have kids, but children require the use of but. I suggest that this is because but is necessary to signal the integration of the fragment into the single act. That leaves the ungrammaticality of *They don't have kids, children to be explained. In any case, the ungrammaticality, in view of the acceptability of MLN They don't have kids, but children, means that there is no competing fragment configuration that could be associated with denial alone. Competition between MLN and denial arises with reverse fragment between e.g. They have children, not kids which is only MLN and They have pets, but not kids, which is only denial. Each construction is associated with a given use by virtue of competition between denial and MLN. The reasoning is akin to the blocking analysis proposed by Bîlbîie & Winterstein (2011). While a blocking analysis does not explain why one connective goes with one configuration, it does account for the complementary distribution. There are two dimensions of explanation therefore. One is the unity of the corrective speech-act of MLN, that explains the impossible explicit connection between two clauses and the compulsory connection with a fragment. The second is the competition between MLN and denial that accounts for the complementary distribution around reverse fragment constructions. The proposal goes further than existing analyses by offering an explanation for the observed connective behaviour. The proposals are summarised in Table 1.
This section has proposed that MLN is defined as a corrective speech-act. It asserts the correction of an inappropriate designation in the antecedent context. MLN is always found in the scope of an assertive speech-act, and while it must have maximum scope over the material that it is to correct, it cannot correct the speech-act of the sequence that contains it. What's more, the speech-act represented by MLN is a unitary one. Even in bi-clausal cases, the clauses are integrated and cannot independently express a different speech-act. The unity of the speech-act expressed by MLN is used to account for the behaviour of connectives: but cannot connect the two clauses of MLN, as it might lead them to be interpreted as two different speech-acts, where only one is evoked by MLN; but must connect the corrected clause to the correcting fragment to signal the fragment's integration in the speech-act. Competition with denial explains the complementary distribution in reverse fragments. In the next section, the informational organisation of the speech act is discussed and identified as a second distinctive property of MLN.

Correction and Information Structure
In the previous section, it has been proposed that MLN is a corrective speech-act, a status that explains its principal morpho-syntactic reflexes. However, MLN comes in a variety of formats (i.a. McCawley 1991), as illustrated by the following.
(24) I want to buy not pants, but trousers.
(25) I want to buy trousers, not pants.
A. I want to buy pants. B. You mean you want to buy trousers. More generally, metalinguistic uses are felt to be marked. What is the common property of metalinguistic uses that makes them marked? I propose in this section that the answer to these three questions has to do with Information Structure. Contrastive structures, like MLN and other metalinguistic configurations, all involve correction. However, correction is not enough. McCawley points out that the not X but Y construction does not live solely on MLN environments. 9,10 There are denial uses of those corrective constructions.
(32) So at each moment, all the sun-made right triangles were exactly the same shape -not the same size, but the same shape. (www.anselm.edu/homepage/ dbanach/thales.htm) The difference between denial corrective structures and MLN uses can be illustrated by comparing the attestation above to the elicitation below.
(33) A. Are these triangles acute? B. Come on now, these triangles are not acute, but oblique. 9 Although they do suppose an alternative, as demonstrated a contrario by the following syntactic pun.
(i) Wolverhampton has won the annual English heritage most beautiful city award not once, not twice, not three times… obviously. (laughter.) Wolverhampton's first railway station (laughter) was at Henfield's heath. (I'm sorry I haven't a clue, BBC4, 5 dec 2016, episode 4, series 66) 10 See also Mignon (2008)

on the differences between metalinguistic and denial fragment constructions in
French; it may be as Mignon suggests that while MLN correction is not an elliptical structure -see Neeleman & Vermeulen (2011) -denial is.
It relates to the informational status of the sequence: whereas the rejected material in (32) is discourse-new -it is not a piece of information that is accessible to the reader before that point in the text -it is discourse-old with MLN in (33) -by virtue of the information being made available immediately before in the exchange. MLN, contrastive structures and metalinguistic configurations share the largely acknowledged property of ranging over discourse-old material that is explicitly present in the antecedent context. 11 This is what is suggested by the label "metalinguistic" itself. MLN is variously described as echoic (Horn 1989(Horn /2001Krifka 1991;van der Sandt 1991;Dancygier 1992, i.a.), metarepresentational (Carston 1996, i.a.) or an instance of mention rather than use (Jiang 2015). A shared characteristic of MLN, MLN contrastive structures and other metalinguistic configurations is that they scope over discourse-old information, which does not necessarily apply to denial. The second shared property is the informational status of the corrected segment. The segment is in focus, as assumed without discussion by various recent studies (Geurts & Maier 2005;Kawamura 2007;Kolokonte 2008). Focus is discoursenew information, that is information that is not accessible to the hearer at that point in the discourse, or following Halliday, that is "not being recoverable from the preceding discourse" (1967: 208). Among the various types of Focus (Kiss 1998;Rooth 2016), the case that we are dealing with is contrastive focus. The entity under contrastive Focus is interpreted against a set of alternatives; it is new information against the background of 11 Discourse-old is used as synonymous to given or activated, and refers to the property of a linguistic sequence that represents information made available to the hearer by virtue of explicit use in the antecedent context (explicit activation), or of accommodating constructions or inferential relations (accommodated and inferred activation) (Dryer 1996;Schwenter 2005;Larrivée 2012). While the paucity of MLN in actual usage has not made it easy to conduct corpus studies of it (but see indications in Silvennoinen 2017), the assumption that it refers to an explicitly used sequence is supported by the following, where only the explicitly discourse-old case is fully acceptable (the acceptability judgements characterising the response in relation to the initial assertion).
(i) A. Iran is going nucular. OK B. Iran isn't going nucular, but nuclear.
(ii) A. I'm concerned that Iran is going nuclear. ??B. Iran isn't going nucular, but nuclear. vs OK B'. Well, Iran isn't going nuclear.
(iii) A. I'm glad that we're visiting Iran next month. ?*B. Iran isn't going nucular, but nuclear. vs OK B'. Thank God Iran isn't going nuclear then.

See Larrivée (2015) on types of discourse-old information.
It is however true that there are occasional attestations of MLN that do not reject material explicitly used before (Horn 2014;Davis 2016). Manipulated for effect, these tend to be found in advertising and political speeches. The crucial point is that the information is accommodated as discourse-old. An interesting subcase of this is the reference to linguistic practices of (groups of) speakers, as with the Carston example in (15). In the following elicitations the out-of-the-blue use of a metalinguistic description ((iv)) and negation ((v), (vi)) refers to the linguistic practice of a group or of an individual.
(iv) After all, Obama is not the kind of politician who fears nucular power.
(v) Now, in Québec, note that we don't go to the corner store, we go to the dépanneur. (volunteered by Brandon Waldon) (vi) The royal family are very proper. They don't have kids, they have children. (volunteered by a reviewer) Without such a reference, there is no accommodation. Consider the following out-of-blue statement: (vii) After all, you wouldn't want to suffer from maclar degeneration.
Unless someone is known to the hearer and speaker to confuse maclar with macular, it is difficult to see how such a statement is relevant in the context, and the statement should be infelicitous, or at the very least quite odd. the rest of the proposition that represents the Question Under Discussion (Roberts 1996;Rooth 1996;Kadmon 2001). 12 MLN has the corrected segment as contrastive Focus, 13 that is explicitly contrasted to another alternative represented by the correction.
(1) c. I don't want to buy pants, but trousers (7) a. I don't want to buy pants, I want to buy trousers.
The same situation obtains with contrastive cases, (24) I want to buy not pants, but trousers.
(25) I want to buy trousers, not pants.
Things are less immediately obvious with some conditionals, comparatives, and interrogatives that highlight a previously used form without providing an explicit correction.
(28) A. I want to buy pants. B. If you want to buy pants, you must be an American in need of trousers.
(29) A. I want to buy pants. B. It might be better around here to ask for trousers rather than pants.
(31) A. I want to buy pants. B. Pants?
Likewise for the uncorrected correctives of the type below.
(34) Sometimes I just can't get out of bed… Well not sometimes… For the past week… Depression sucks … (http://whisper.sh/whisper/05268ebb2815c94f5 15e330c15af1859ade0cc/Sometimes-I-just-cant-get-out-of-bed-Well-not-sometimes-For-the-p) Yet in either case, contrastive Focus is new information following the interpretation of unpredictability of the entity chosen from a set of alternatives at that point in discourse. 13 Strong evidence of the Focus status of the corrected segment is provided by languages with overt Focus morphological marking. A Focus marking language is Yoruba (Niger-Congo family) in which the Focus of negation is followed by the negative and a focusing particle (Fábùmni 2013: 2).
However, in both cases, a correction is suggested by attention being driven 14 to the sociolinguistic dimension of the form pants as a designation of an outer garment worn on the lower part of the body, and by the PPIs being in the direct scope of negation. 15 So MLN is a speech-act that proposes the correction of a segment in Focus from a sentence that is explicitly discourse-old. It is different from denial negation, that does not have Focus on a sentential segment, and from both denial and descriptive, that do not necessarily range over discourse-old material.
At this stage, a paradox emerges. MLN ranges over discourse-old information. That critically includes the segment to be corrected, which is therefore also discourse-old. However, the segment is in Focus, and Focus is understood as being discourse-new information. 16 Therefore, both discourse-new and discourse-old status characterise the corrected segment. How is such an apparent contradiction possible? It is possible because the corrected segment is discourse-old as part of the sequence as a whole, and discourse-new as a corrected segment. In the sequence They don't have kids, they have children, the MLN in the first clause ranges over information that has been explicitly used in the antecedent context. All of the clause is therefore explicitly discourse-old. The corrected segment kids is however focused, and therefore, from discourse-old, is turned into discourse-new information. This atypical discursive move in large parts explains the possible gardenpath effects that MLN makes possible (i.a. Attal 1990). 17 The first clause of the following may be initially interpreted as a rejection of silliness inviting positive evaluation of the interlocutor; the interpretation disappears in view of the negative qualification offered by the second, correcting clause.
Such garden-path effects are due to the fact that a descriptive use might be the default interpretation of a negative sentence, that requires specific cues to be interpreted otherwise, only provided after the fact in (38). MLN upsets the normal flow of information. 18 In They don't have kids, it would normally be assumed in an unspecified context that it is having children that is negated; however, rejection of the appropriateness of the form kids alone to assert that a couple indeed has children is definitely marked. The cause of markedness of metalinguistic usage is found in its peculiar Information Structure. My proposal is that the defining characteristic of the correction offered by MLN relates to Information Structure. MLN ranges over a discourse-old sequence, and turns the part that it objects to into new information by virtue of its focal status. Turning old information into new is what distinguishes it both from descriptive and denial. It is what brings together the different metalinguistic corrective constructions (24)-(27), and metalinguistic uses in questions and conditionals (28)-(31). In other words, MLN upsets the expected discourse-old/discourse-new information dynamics: it ranges over discourse-old material, and sees the discourse-old information transformed into discourse-new focal material. Modifying the expected flow of Information Structure is why they may give rise to garden-path effects, and why more generally metalinguistic usage is marked.

An excursus into maps
The previous section has proposed that the markedness of MLN is due to an informational paradox: MLN ranges over discourse-old material, and turns one segment of it into discourse-new information by having it in Focus. Whether such an Information Structure can be mapped into a syntactic structure, and the extent to which it should, are two questions I want to speculate upon in this section.
Informational dimensions can be coded in the left periphery of the sentence assuming Rizzi's (1997) split-CP hypothesis. The proposed architecture is as follows: The higher projection ForceP has to do with the clause type (for MLN, assertion), and the lower FinP relates to Finiteness distinctions. Between these, I assume that the higher TopP can host discourse-old information, that FocP can host contrastive focuses such as MLN corrected segment, and that the lower TopP is the position that non-contrastive focuses would target (see Kolokonte 2008).
Thus, such an architecture could help capture the informational structure of MLN. This is especially useful for a language like Yoruba where overt morphological and positional manifestations of Focus are found (see footnote 13). Checking the morphological Focus feature would be a credible motivation for overt movement of the focused phrase to a FocP position. In the absence of such overt marking, as in English and French, the motivation might be felt to be lacking. However, using covert movement would be desirable to capture the properties of MLN, in particular its informational characteristics, that distinguish it from other negative uses.
How would one go about a left peripheral analysis of the informational status of MLN in a sequence such as They don't have kids, but children? Several options can be considered regarding negation, the corrected segment in Focus, and the rest of the discourse-old clause.
i. Only the negative operator moves to FocP, and both the corrected segment and the remaining, discourse-old material stay in situ. ii. The negative operator and the corrected segment move to FocP, and the discourse-old material stays in situ. iii. The negative operator and the corrected segment move to FocP, and the discourse-old material moves to the higher TopP. iv. The negative operator and the corrected segment stay in situ, and the discourse-old material moves to the higher TopP.
Note that I am excluding here the stative option where nothing moves, and where the difference between MLN and other negative uses is not accounted for. A movement analysis raises certain expectations. The expectation is that both Focus and Topic material should move. After all, the point of moving material to the left periphery here is specifically to mark the special informational status of MLN sequences. Leaving Focus or Topic material in situ seems to defeat the purpose of doing a cartographic analysis at all. I'm also assuming that marking the informational status of segments makes it desirable for all the discourse-old material to end up in the same projection. In that respect, the following structure " is splitting the discourse-old proposition in two, with part of the discourse-old material left in the VP. While this allows to maintain the superficial order, it seems undesirable to separate segments that share the same informational status at a level of representation the objective of which is to make that status visible.
In this approach, the most coherent analysis therefore seems to be iii, illustrated by the sketchy representation below.

iii. [ ForceP Assertion [ TopP They have [ SpecFocP not [ FocP kids [ TopP [They don't have kids, but children
It makes it explicit that the rest of the proposition in TopP is unambiguously discourse-old information, and the corrected segment kid is in Focus, under the direct scope of negation in SpecFocP. More would need to be said about the treatment of the correcting segment but children. A plausible proposal is that of Bianchi & Zamparelli (2004), for whom the two conjuncts figure in the Focus projection.
The explicitness of the model allows some of the potential difficulties to be spelled out. One is the fact that the Focus is not always a discrete continuous sequence. There is no clearly identifiable corrected segment in the following, where the existence of a referent, and the responsibility for an action are concerned respectively. It could be assumed in these cases that it is the clause as a whole that is in Focus, in which case the opportunity is lost to note explicitly that it was also discourse-old (for the role of Topichood for the King of France cases, see Brandtler 2006;Abrusán & Szendrői 2013).
Here, it might be useful to explore how such sequences are dealt with in overt Focus marking languages. Maybe in fact there are reasons to treat those as denial rather than Metalinguistic. This is definitely an area for future work. Yet another issue is that of syntactic islands. Movement to the Focus position should be constrained by general rules on syntactic operation. One such well-known rule is the ban of movement out of syntactic islands. With normal negative focus, the targeted element cannot be inside e.g. adjuncts headed by before. As a result, a sequence like John didn't start talking before he had greeted everybody cannot mean that the talking started at a time when not everybody had been greeted. Yet, there seems to be nothing objectionable to the metalinguistic focus being in an island. (40) A. 'Bush acted before Iran went nucular.' B. 'Bush didn't act before Iran went nucular; he acted before it went nuclear.' Why there is no barrier to Focus movement in MLN remains to be explained. Thus, MLN can be mapped onto a cartographic analysis. Such an analysis records the informational partition between explicitly discourse-old information and the Focus status of the corrected segment. Foreseeable difficulties include cases where there is no specific item that is objected to that can be raised to a Focus position, and movement escaping island effects. Having raised these open questions, I summarise in the next section the novel proposals put forward in this paper.

Conclusion
This paper proposes a definition of metalinguistic negation (MLN) illustrated by They don't have kids, they have children. MLN is defined as a corrective speech-act that ranges over discourse-old information, the corrected part of which is transformed into discoursenew information by virtue of being in Focus.
The definition is used to answer four main questions.
i. Why is MLN relating to a special behaviour of connectives? (2) a. They don't have kids, (*but) they have children. c. They don't have kids, *(but) children. The characteristic properties that separate MLN from other types of negative uses is that it is a reactive, corrective speech-act. The speech-act dimension accounts for the mysterious behaviour of connectives. I have argued that the impossibility for a metalinguistically negated sequence to be followed by connective but introducing a full clause (2) a. They don't have kids, (*but) they have children.
is due to the fact that such coordination would allow the sequence to be understood as two speech-acts; yet, a single speech-act is represented by MLN, as shown by the lack of autonomy of its component parts that cannot express independent speech-acts. MLN proposes a correction, that neither denial (5b) nor descriptive (5c) uses necessarily suppose. This underlies the paradoxical informational organisation that relates MLN to other corrective constructions. MLN ranges over discourse-old information, as with corrective structures. As with corrective structures, MLN has a corrected segment that is part of the discourse-old sequence, and yet treated as discourse-new by virtue of being in Focus, and contrasted to the correcting segment. The focus on the corrected and correcting segments is what MLN have in common with other metalinguistic configurations. The paradoxical Information Structure of MLN, corrective constructions and metalinguistic configurations account for their equally marked status. They turn the corrected segment of discourse-old information into discourse-new focal material. By upsetting the normal flow of information, they constitute marked uses that are intuitively rare 19 and require contextual cues to arise. By having put forward a novel definition of MLN that rests on two properties, I hope to have contributed to defining the metalinguistic usage of negation.
This article has raised a number of topics for future work. 20 In particular, more needs to be known about the competition between connectives in view of MLN and denial uses of negation, and whether the equivalents of but distribute between the two readings in a language-specific or in a cross-linguistically stable way. Also, the relationship between MLN and denial uses seem to overlap in some cases. There are cases where it is the whole of the clause that is rejected: One might therefore assume these to be simple denials. However, the target interpretation is available only when ranging over discourse-old material, and exclusively with a correction. The complementary distribution of a MLN that relates to a focused segment, and denial that relates to the clause as a whole seems to collapse. A further example (provided by Laurence R. Horn) is from an ad which against the background image of bread dough being kneaded stakes the following claim: 19 This remains to be established, as pointed out by a reviewer. One difficulty, that proves the point, is that MLN is so rare that looking for it in a corpus can be a long and frustrating affair. A preliminary look at a corpus of French political speeches where MLN should be prevalent point to a rate well under 0.5% of MLN usage among clausal negatives. The same conclusion is reached independently by Silvennoinen (2017), who, looking at contrastive constructions where MLN should be common, finds that they are less than 1% in written material, and even rarer in spoken material. 20 Other topics for future research raised by reviewers are corpus investigations of MLN, an assessment of their contribution to discursive interaction, and an extensive cross-linguistic survey of morpho-syntactic reflexes.
(41) German Ich backe keine Brötchen. 1-sg-nom bake-pr-1sg no bread Ich arbeite am perfekten Morgen. 1-sg-nom work-pr-1sg on perfect morning 'I do not bake buns. I work on the perfect morning.' The target interpretation requires the correction, that bears on (almost) all of the clause, that is discourse-old by accommodation, and transformed into discourse-new information because of its focal status. While this does not affect the definition of MLN as a corrective act that turns the corrected, old information into new, it shows that the border between MLN and denial should be revisited in further work. The ability to frame these questions is hopefully an indication of the fruitfulness of the claims made in this paper that MLN is a corrective speech-act that has a paradoxical Information Structure.