Specialization and finiteness (a)symmetry in imperative negation: with a comparison to standard negation

3.1.1 Characterization

(Negative) imperatives have been the subject of much cross-linguistic research (e.g., Mauri and Sansò 2012; Van Olmen 2021; Xrakovskij 2001) but, perhaps somewhat surprisingly, many such studies do not characterize their topic accurately (Van Olmen and Heinold 2017: 6–7). Van der Auwera and Lejeune (2013), for one, simply define the negative imperative as “the grammaticalization of a prohibition”. They do not discuss what counts as grammaticalized or as a prohibition. A possible issue that arises as a result is whether (6) should be considered as a negative imperative or not. The construction seems relatively conventionalized, as “the preventive particle moli is used only in combination with the Future affirmative second person forms in sentences intended to prevent unintended results of the hearer’s careless actions” (Maslova 2009: 498). However, the answer to the question whether the meaning ‘Make sure that you don’t’ qualifies as a “proper” prohibition is not so self-evident. Moreover, as a negative imperative, (6) would hardly be comparable to English Don’t shoot the dogs!, which need not have preventive or so-called admonitive overtones.

Likewise, Aikhenvald (2010: 1–2) defines the imperative only implicitly, as the construction in a language that is specialized for prototypically expressing a command. The term “command” is to be understood here as equivalent to Searle’s (1975: 355) concept of directive – the speaker wants and attempts to get the addressee to do something – and not as a more specific type of directive speech act involving a speaker with institutionalized power. A question that one can ask then, though, is whether (7) should be regarded as an imperative or not.

The construction is included by Aikhenvald (2010: 223) but it clearly does not convey directivity in the classical sense of speech act theory. The speaker is not actually trying to make anyone do anything here. Rather, they are giving their consent to something that the addressee wishes to do. In addition, as an imperative, (7) would hardly be equivalent to English Go!, which can fulfill functions other than permission.

A fairly recent attempt to address this lack of precision in earlier work comes from Jary and Kissine (2016). Their aim is explicitly to give a definition of, in particular, the imperative that enables accurate cross-linguistic analysis. The comparative concept – in Haspelmath’s (2010) sense – that they propose goes as follows:

[The imperative is] a sentence-type whose only prototypical function is to provide the addressee(s) with a reason to act [and] that is suitable for the performance of the full range of directive speech acts. (Jary and Kissine 2016: 132)^[3]

“Provid[ing] the addressee(s) with a reason to act” is how they interpret directivity. They argue that imperatives across languages typically serve not only to command but also to give advice and grant permission and that the latter uses are not captured by the Searlean notion of directivity. This observation has already been made for permission but, in advice too, the speaker does not truly try to get the addressee to do anything. Take an imperative like Read the instructions if you want to know more. It does not really matter to the advisor whether the advisee looks at the instructions or not. The speaker just mentions a course of action that could be useful in their view and that the addressee is free to take or not. Both permission and advice can, however, be said to give someone a motive to act, i.e., the speaker’s approval and counsel respectively.

Given the compatibility of the two uses with this novel view on directivity, one presumes that they are part of what the comparative concept calls “the full range of directive speech acts”. Interestingly, though, Jary and Kissine (2016) are not very clear about what (else) this range is meant to include. Their lack of specificity is probably partly due to the fact that classifications of directive speech acts vary greatly in the number and types of distinctions that they make and the terms that they use for them (cf. Aikhenvald 2010: 198–203, 223–228; Siemund 2018: 231–235; does one, for instance, need to differentiate between advice and recommendations and, if so, how, and what distinguishes them from suggestions?). The closest that Jary and Kissine (2016: 121) come to spelling out the intended range is an unstructured list containing “commands and orders”, “requests, suggestions and pleas”, “warnings and advice” and “permission”. Some of these speech acts could be described as directive in the classical sense (e.g., commands, orders, pleas, requests). Others appear to be acts in which the speaker proposes something to the addressee but has no particular desire to see their proposal adopted (e.g., advice, warnings, suggestions). A last speech act (i.e., permissions) could be characterized more broadly as one where the speaker commits themself to a course of action (i.e., allowing the addressee to act on their desires), a general type that might be taken to encompass offers and invitations too (cf. Jary and Kissine 2016: 125).

The ability to convey this entire array of acts is a necessary feature of the imperative for Jary and Kissine (2016). Accordingly, they do not consider as imperatives constructions that are dedicated to specific directive speech acts, like the Nivkh permissive in (7) (and, by extension, they would presumably not analyze the Kolyma Yukaghir admonitive in (6) as a negative imperative either). As this section’s initial paragraph already suggests, I share Jary and Kissine’s (2016) assessment that such constructions are best excluded from a study comparing imperatives across languages. They cannot be regarded as equivalent to, say, the Plains Cree imperative – which is described as suitable for performing “orders, commands, invitations or requests” (Okimāsis 2018: 60) – or the Albanian one – which Breu (2010: 462–463) writes may be used for “commands, requests, instigation, … advice … [and] permission”. It is probably also important to add here that, for no language in the sample, ruling these more specialized constructions out leads to an analysis under which it has no imperative (or negative imperative, for that matter). Nivkh, for instance, still has the construction in (8).

Two comments are in order, however. First, establishing whether the imperative in a language is suitable for Jary and Kissine’s (2016) whole range of directive speech acts is difficult, if not impossible, for a typological investigation. For some languages, the descriptions do specify at least a certain range of potential uses. For others, it can be inferred from the examples – which may seem more or less forceful, more beneficial to the speaker or the addressee and so on – that the imperative can convey some distinct directive speech acts anyway but, for yet other languages, the available information is insufficient to determine whether what is identified as their imperative indeed allows a variety of uses. Many sources just state that imperatives are directives in the traditional sense, a usage type that is in fact mentioned for every sample language’s imperative in one way or another. Second, Jary and Kissine (2016: 123) offer little evidence for their claim that “in the vast majority of languages these same functions [i.e., the full range] are happily performed with the imperative”. They even argue, with no real proof, that, for languages with constructions like (7), “it is not always clear that these forms genuinely stand in complementary distribution with imperative sentences” (Jary and Kissine 2016: 123) and thus that the imperative in (8), for instance, cannot express permission.

To avoid these assumptions, and in view of the practical concerns about what sources can tell us, an adapted version of Jary and Kissine’s (2016) comparative concept is adopted here. While acknowledging the likelihood of imperatives conveying advice, permission and the like, it puts Searlean directivity center stage again (cf. Aikhenvald 2010: 1–2 implicitly; van der Auwera et al. 2013):

The imperative is a sentence type whose only prototypical function is to provide the addressee with a reason to act and that is at least suitable for the performance of directive speech acts where the speaker wants and attempts to get the addressee(s) to do something [my emphasis].

A similar definition of the negative imperative that guarantees comparability with the positive constructions under investigation would be (cf. Van Olmen 2021: 526–528):

The negative imperative is a sentence type whose only prototypical function is to provide the addressee with a reason not to act and that is at least suitable for the performance of negative directive speech acts where the speakers wants and attempts to get the addressee(s) not to do something.

The term “negative directive speech act” covers any situation where the speaker presents the addressee(s) with a reason not to act. Such uses include not only what may generally be referred to as prohibitions (e.g., an order like ‘Don’t pick your nose, son!’) but also, for example, advice (e.g., ‘Don’t use social media if you want to stay sane’). So, in my view (Van Olmen 2011: 45), positive and negative directive speech acts can fulfill similar specific functions but contrast in their fundamental orientation of action and non-action.

Let us now turn to the notion of “sentence type” in the above definitions. The operationalization of it which is adopted here requires constructions to be morphologically and/or syntactically distinct from other constructions to count as (negative) imperatives. This distinctiveness may arise as some form of dedicated marking by affixes, clitics or separate words. The Lezgian prohibitive suffix -mir in (9) is a case in point and so is the Pacoh imperative particle Ɂaw in (10).

(Negative) imperatives may differ from other constructions in (purely) syntactic terms as well. The omission of subject markers is a common phenomenon. Maung in (11) can serve as another example. In this language, the verb’s irrealis form combines with da ‘if’ to express conditions and with marig ‘not’ to convey negative past indicatives. Its second person instantiations in particular can also be employed on their own, to issue directives (Capell and Hinch 1970: 67–68). The absence of da or marig makes such uses syntactically distinct and (11) therefore counts as an imperative.

Not included as (negative) imperatives are (negatively) directive constructions that do not differ from other constructions grammatically. English Stop walking!, for instance, offers the addressee a reason not to act but does not constitute a negative imperative. Morphologically and syntactically, it is identical to a positive imperative like Keep walking!. Its negative directivity simply results from the lexical meaning of Stop.^[4] Furthermore, I take sentence type to refer to a construction that exhibits productivity or, in other words, can feature a wide range of different verbs (all verbs would not be a satisfactory requirement since imperatives are known to be incompatible with, for instance, stative verbs in some languages; e.g., Aikhenvald 2010: 150–153). Thus, constructions limited to particular verbs fall outside the present article’s scope. In Chalcatongo Mixtec, for example, the imperative is characterized by the usage of the potential stem of the verb and the absence of pronominal subject clitics. The directive forms of the verbs ‘go’ and ‘come’, however, are idiosyncratic: the former lacks clitics too but takes the progressive verb stem whereas the latter may (but need not) appear suppletively as ñã́Ɂã ‘come!’ (cf. potential kii ‘come!’; Macaulay 1996: 134–135). Neither construction is taken into consideration here.^[5] I do believe, however, that future research should look at (a)symmetry in suppletive (negative) imperatives too.

A second component of the above definitions that needs to be explained further is that, for sentence types/constructions to be (negative) imperatives, (negative) directivity should be their only prototypical function. This requirement excludes cases like (12) and (13). The construction in (12) is the primary way that speakers of Ghomara issue negative directives but it is not dedicated to the expression of negative directivity. It is a negative declarative about the future, which, in context, can be used to provide the addressee with a reason not to act. The language does not actually possess a specialized construction for this function and is therefore analyzed here as lacking a negative imperative. In the same vein, the construction in (13) is the main directive strategy in Lokono. It does not constitute an imperative, however, since offering the addressee(s) a reason to act is not the prototypical function of this present declarative. It is said to only acquire a directive interpretation with a particular intonation. The language has no construction that meets the present criteria for an imperative and, hence, it is analyzed as lacking one.

(Negative) present or future declaratives are the constructions that most languages in the sample with no (negative) imperative rely on to express (negative) directivity. Other recurrent non-specialized constructions, which need not be directive, are: (i) declaratives with a verb of volition, meaning ‘I (don’t) want you to’ or ‘You (don’t) want to’ (e.g., Katukína-Kanamarí; Groth 1988: 55–56); (ii) declaratives featuring some modal (e.g., Kannada; Steever 2020: 205); (iii) irrealis constructions vague, for instance, between directive, optative and dubitative readings (e.g., Fur; Waag 2010: 268). Languages having to turn to such constructions to express (negative) directivity because they do not possess more specialized means (like those between brackets) are regarded here as having no (negative) imperative. Note, though, that these types of constructions may develop into (negative) imperatives over time. In Makalero, for instance, a declarative with a modal verb meaning ‘not allowed to’ gets a dedicated negative directive interpretation when the subject pronoun is left out (Huber 2011: 446).

3.1.2 Problems

The following list summarizes, for convenience’s sake, the criteria that, as discussed in Section 3.1.1, a (negative) imperative needs to satisfy:

1. a morphologically and/or syntactically distinct sentence type

2. that has (negative) directivity as its only prototypical function

3. and can at least be used for acts where the speaker wants and tries to get the addressee(s) (not) to do something.

Applying them to the information provided in grammatical descriptions is not always straightforward, however.

A case in point is Kanuri in (14), which presents what Hutchinson (1981: 130) calls the “imperative/subjunctive” paradigm of lè ‘go’. All forms are said to be specialized in some way: the second person and first person plural ones are exclusive to this paradigm; the other person-number forms are morphologically identical to the conjunctive but syntactically distinct because of their main clause usage here.

Hutchinson (1981: 130) only characterizes the meaning of the complete paradigm, as “expressing orders, commands, wishes, desires, blessings, etc.” This description is suggestive of optative interpretations alongside directive ones, which is borne out by some third person examples translated as ‘may (s)he/they …!’. It is, however, not clear from the grammar whether or not the second person forms permit optative readings too and do not therefore satisfy criterion (ii). It is nonetheless not unreasonable, from a typological perspective, to assume that they are imperatives. Earlier research (e.g., Aikhenvald 2010: 76–77; van der Auwera et al. 2013) has revealed a distinctive trait across languages of paradigms dedicated to expressing directivity: if they go beyond the second person, the first other person-number value that they always include is the first person plural (see French in footnote 3), with further extensions to the third person and then the first person singular being increasingly infrequent. Looking at Kanuri in this light, the fact that only the second person and first person plural forms are really specific to (14) suggests that they make up an imperative paradigm like many attested cross-linguistically and thus that lèné ‘go!’ and lènówó ‘you all go!’ are indeed imperatives. The other, conjunctive, forms can be analyzed as a hortative construction that fills in the “gaps” in the imperative paradigm (like let in English).

For another example, take Lao in (15). The basic way of issuing a directive in this language is given in (15a). This construction has no overt subject but, as zero anaphora are very widespread in Lao, it can function not only as a directive but also as another type of speech act/sentence type, with a second person or non-second person subject. As Enfield (2017: 191) writes, “imperatives share their general characteristics – i.e., basic constituent order – with other sentence types, including interrogatives and declaratives.” In other words, example (15a) does not satisfy criterion (i) (or, for that matter, the other two criteria) and cannot be regarded as an imperative here.

Lao has a highly grammaticalized class of sentence-final particles, though. Some of these are specialized for directivity and they are in complementary distribution with a range of declarative and interrogative ones (Enfield 2017: 190). Dèè1 in (15b) is one such particle, serving to mitigate the imposition of a request. Others are: mèè4, used to urge the addressee to do something to which no impediments exist; vaj2, serving to ask the addressee to hurry up and act; and duu2, used to implore the addressee to do something for the speaker (Enfield 2007: 63–68). In my view, sentences such as (15b) can be argued to be syntactically distinct, with the absence of a subject and the presence of a particle like dèè1 as their characteristics. Directivity is, moreover, their only prototypical function and they all express acts in which the speaker wants and tries to get the addressee(s) to do something. For that reason, Lao, as well as some other South East Asian languages with similar phenomena, is still analyzed here as possessing an imperative, in spite of (15a).

4.1.1 Characterization

4.1.2 Problems

A first, ostensibly fundamental, issue concerns the (non-)finite status of (negative) imperatives. The verb forms of second person singular imperatives in particular are often considered to be intrinsically non-finite. As Nikolaeva (2007: 139–141) and Heine (2016: 252–257) among others note, imperative verbs tend to be very restricted, for instance, in the tense-aspect-modality distinctions that they can make – compared to other parts of the verbal paradigm but not unlike “ordinary” non-finite forms. There have been examples of this phenomenon in previous sections: what characterizes the imperative in languages like Vitu in (25a) and Chinantec Lealao in (26a) is the absence of marking of (other) verbal categories. Accordingly, (second person singular) imperatives “are often the simplest and most straightforward forms in a language” (Aikhenvald 2010: 89). Yoruba in (16a) and Yuhup in (24a) are cases in point. Nikolaeva (2007: 140) also observes that, like many non-finite forms, imperatives “have limited co-occurrence with subjects” (see Aikhenvald 2010: 92; Heine 2016: 252–252 too). In some languages, such as Adang (Haan 2001: 76), they do not appear to be able to combine with subject pronouns at all. In other languages, such as Pima Bajo (Estrada Fernández 1996: 33), subject pronouns are optional with the imperative and retaining them may produce particular pragmatic effects (e.g., emphasis in Pima Bajo).

If (negative) imperatives are indeed non-finite, it does not seem sensible to analyze them in terms of finiteness (a)symmetry. None of the above tendencies are exceptionless, though, as Nikolaeva (2007) and Heine (2016) themselves point out. In numerous languages, verbs display imperative-specific morphology and agreement with their subject. Itelmen in (28) is one of them: the q- prefix marks the verb as a second person imperative while the -xç/-sx suffixes mark it as having a singular/plural subject.

The situation in Klamath is especially noteworthy. In this language, indicative verbs do not convey person-number subject agreement but imperative ones do (Stegnij 2001: 81). It is also fairly common for specialized imperative constructions to feature finite forms from other parts of the verbal paradigm. The plural imperative verb form in Western Farsi, for one, is identical to the second person plural subjunctive, which normally occurs only in subordinate clauses but is dedicated to expressing directivity in main clauses (Mahootian 1997: 247–251). Moreover, imperatives do frequently make aspect distinctions as well as construction-specific tense distinctions. An example of the former is Kiowa: it distinguishes perfective (‘Do!’) from imperfective (‘Keep doing!’) imperatives (Watkins 1984: 169–170). The latter usually involves a contrast between immediate and delayed compliance, like in Menggwa (de Sousa 2006: 380–385). Finally, there are also languages where subject pronouns are actually required in the imperative. Teribe (which is not part of the sample) is a case in point. In this language, intransitive verbs take the imperative suffix -zong but need to be preceded by a second person singular or plural pronoun to indicate the addressee (Quesada 2000: 79–80).

Even if one nevertheless wishes to maintain that imperatives are generally non-finite, one should at least recognize that, as Heine (2016: 262) argues, they “have little in common with nominalization and other forms of verbal non-finiteness: the latter operate within the morphosyntax of a sentence”. In other words, verbs normally appear nominalized or as infinitives or participles because they fulfill a particular function or occupy a particular slot (e.g., a nominal one) in a clause. Imperatives, by contrast, “do not seem to serve primarily the structuring of a sentence; rather, they belong to a different space of speech processing, [i.e.] … the component of speaker-hearer interaction” (Heine 2016: 262). This distinction is adopted here: the characteristics mentioned in the first paragraph of the present section are not regarded as sufficient – for present purposes – to call (negative) imperatives less finite or non-finite. I do analyze them as such if the LV has any of the features discussed in Section 4.1.1.1.

Accordingly, I do not consider English Keep quiet! as non-finite (e.g., keep has no overt nominal traits and is not syntactically dependent on anything else). Its Wagiman counterpart in (29a), however, does count as non-finite in my view. ‘Be quiet’ is part of an open, non-finite class of roots called co-verbs or participles. They may express aspect (-na is non-perfective) but, to form a clause, they ordinarily require the presence of a member of a closed class of roots able to take inflections (Krauße and Harvey 2021: 97). Ya ‘go’ in (29b) is one of them: it carries the tense and agreement marking and combines with the non-perfective participle of durdut ‘run’ to yield a complete sentence. These participles can even be case-marked, as (29c) shows – as well as, of course, occur on their own with a directive interpretation, like in (29a). Since, in this imperative construction, the verb is in a form that is usually found in syntactically dependent environments, it is regarded as non-finite.

There is, by the way, no A/Fin in Wagiman. As (29d) demonstrates, the participles also occur in the negative imperative, where the abessive suffix -ne’en serves as the negator (Cook 1987: 150, 256).

A more concrete problem than the finite or non-finite nature of (negative) imperatives is that it is unclear at times which subtype of A/Fin a specific case of negation belongs to (see also Miestamo 2005: 87–96). An earlier example is Vitu IN in (25), which is ambiguous between a Neg-FE and a Neg-Cl analysis. A case that does not fit any subtype is IN in Wari’. Imperatives like (30a) are marked by the use of what Everett and Kern (1997: 34) call “second person realis future VICs [verbal inflectional clitics]”. They can also occur at the end of “leading questions” (Everett and Kern 1997: 12) but are otherwise essentially limited to the imperative. In fact, Apontes (2015: 166) analyzes them as imperative pronouns. Negative imperatives like (30b) resemble complementizer clauses. The latter are characterized by the use of tenseless VICs and the presence of one of a set of “INFL[ectional]” morphemes that only appear in such clauses or in noun and relative clauses (Everett and Kern 1997: 41–42). In (30c), for instance, note the tenseless VIC caca ‘they’ (cf. realis future ra-in ‘you-him’ in 30a) and the realis future INFL morpheme ta (others include realis non-future iri and irrealis xi). Crucially here, the negative imperative in (30b) also features a tenseless VIC (ma-on ‘you-him’) and an INFL morpheme, suggesting that it has a complement(-like) structure. In fact, Everett and Kern (1997: 169) point out that it differs from the SN complement construction in (30d) (see Miestamo 2005: 359–360 too) only in the absence of the negative verb ’om ‘not exist’. Compared to the imperative (30a), (30b) can thus be argued to exhibit a loss of finiteness.

Still, unlike (30c) with ma’ and (30d) with ’om, (30b) does not include any overt element that the construction clearly complements. There is actually no negator at all either in this negative imperative. Wari’ IN is, in other words, simply unclassifiable as Neg-LV, Neg-FE, Neg-Cl or NegVerb. It is therefore grouped together with cases like Vitu as miscellaneous instances of A/Fin.

A last issue concerns languages with multiple negative constructions that vary as regards A/Fin. This phenomenon is mainly found in SN, where specific tenses and/or aspects, for instance, may be negated in different ways. Kannada is a case in point, as briefly discussed in Section 4.1.1.1. An even more diverse example is Awa Pit (see Miestamo 2005: 138–139, 258). This language’s SN includes a construction with no A/Fin, a NegVerb construction and a construction ambiguous between NegVerb and Neg-LV. Such languages will generally be analyzed simply as displaying A/Fin. The details will be considered when relevant, though. The comparison of the subtypes of A/Fin occurring in SN versus IN, for instance, will take into account the two subtypes found in Awa Pit (see Section 4.2).

4.3.1 A/Fin in both domains and economy?

The 19 languages exhibiting A/Fin in both SN and IN could be taken to point to economy as a motivation. Miestamo (2005) sees it as the force behind the language-internal analogy in symmetric SN: the negative copies the structure of the positive “to maintain a formally coherent system [, which is] … economic – easier to store and to process” (Miestamo 2005: 205). This “desire” for coherence/economy might be argued to apply to negation across domains as well. The Warao negatives in (31b) and (31d), for instance, are both asymmetric compared to their positive equivalents in (31a) and (31c). They have the exact same Neg-LV construction, however. In each case, the LV takes the negative suffix -naka and the inflection appears on the FE ta. One could therefore say that Warao maintains a coherent system of negation, with A/Fin in both domains.

Still, “full” coherence, like in Warao, is only attested in five of the 19 languages with A/Fin in SN and IN. In eleven of them, there is still some difference between the two domains. In Yuhup, for instance, SN involves a Neg-LV structure where the LV occurs in a form normally used to convey simultaneity and all tense-aspect distinctions are neutralized (Ospina Bozzi 2002: 171). Its IN exhibits the same A/Fin but requires the presence of imperative ‘be’ as the FE, possibly to mark/distinguish the clause type (see Section 4.1.1.2). Most of the eleven languages are like Itelmen, though. Its SN in (32a) and (32b) displays Neg-LV asymmetry. The LV no longer has its tense-aspect and agreement affixes, which may appear on an optional copula, and it is marked as negative by the suffix -kaq and the particle qaɁm, which are negative existential in origin (and also serve to express ‘without’ with nouns). IN in Itelmen exhibits the same subtype of A/Fin asymmetry, as (32c) and (32d) show, but the negative particle is not qaɁm but the dedicated prohibitive marker zaq. There are thus clearly factors other than just economy at work in languages like Yuhup and Itelmen.

More generally, it remains rather speculative to assume language-internal analogy between two domains of negation (from SN to IN or vice versa) without extensive historical or comparative evidence. However, the former type of data does not exist or is very limited for most languages under investigation and the latter would require a more in-depth look at specific sets of closely related languages than the present typological study can realistically take. Moreover, the impact of economy here is probably quite small anyway – given that, across the entire sample, A/Fin in SN and A/Fin in IN seem to be fairly independent phenomena (see Section 4.2) and that, in most languages with A/Fin in both domains, there are still differences between them.

4.3.2 A/Fin in IN only and diachrony

Of the 18 languages fitting this description, nine have NegVerb asymmetry and, for seven of them, there are data about the negative FE’s source. In one case, it derives from a modal verb meaning ‘not allowed to’ (see Makalero in Section 3.1.1) and, in all other cases, it comes from an LV meaning ‘leave’ or something similar. Kresh is a case in point. Its SN, like in (33a), exhibits no A/Fin (Miestamo 2005: 304–305). The negative counterpart of its imperative in (33b) may be symmetric (Santandrea 1976: 160) but it may be the construction in (33c) too. It features the LV as an infinitive, the adverb a’da ‘down’ and the negative FE a’bi, which has a ‘cease, leave’ origin (and can even combine to say ‘Don’t stop!’).

It is likely that at least some of the other, diachronically obscure, prohibitive auxiliaries in the languages discussed thus far derive from a verb with a meaning like ‘leave’ as well. It is a cross-linguistically common source for prohibitive markers after all (Aikhenvald 2010: 364). It may also explain why NegVerb asymmetry occurs comparatively more frequently in IN than in SN (see Section 4.2). Such languages are interesting in yet another respect. A/Fin in their IN essentially results from the grammaticalization of a form that is initially an explicit directive to desist from something. This original form can therefore be said to an overt manifestation of the illocutionary dynamicity of negative imperatives, i.e., the fact that they still expect (the) action (of desisting) from the addressee. In other words, contra Miestamo and van der Auwera’s (2007) appeal to this dynamicity as a motivation for less A/Fin in IN, not only does it not appear to limit the overall rate of A/Fin in IN (see Section 4.2), but explicit expressions of it may even give rise to the asymmetry.

It is important to point out here that the character of the FE is actually part of Miestamo’s (2005) argument for stativity as a motivation for A/Fin in SN. It is not just the LV’s loss of finiteness that is seen as reflecting stativity. Miestamo (2005: 207, 221–224) also considers it indicative of the role of stativity that the FE tends to be or can typically be traced back to a (negative) existential or (negative) stative copula. The languages in the sample confirm this trend: for every language where the FE in SN is recognizably more dynamic, such as fe ‘do’ in Imonda in (34), there are three languages where it is clearly stative, like dì ‘be’ in Yuhup in (35).

The FE’s status in IN is quite different. In half of the languages for which there is enough information, it is dynamic. Kresh a’bi in (33c), coming from a verb with the meaning ‘leave’, is a case in point. In the other half, the FE is stative in origin. Yet, with one exception (i.e., a negative existential in the Adang polite negative imperative; Haan 2001: 78), it is always marked as an imperative. This form can be said to coerce a more agentive reading and to reflect illocutionary dynamicity in a sense. In the Yuhup negative imperative mentioned in Section 4.1.1.2, for one, the FE ‘be’ appears as an imperative and arguably urges the addressee to make it so that they are in a certain state. This contrast shows, again, that A/Fin is not the same in IN as in SN, notwithstanding its equally frequent occurrence.

A/Fin in IN may result, as pointed out earlier, from the grammaticalization of ‘leave’ into a prohibitive marker. Another mechanism that produces the asymmetry, in both languages that have it just in IN and languages that exhibit it in the two domains of negation, is insubordination. Evans (2007: 367) describes insubordination as “the main clause use of (prima facie) subordinate constructions”. Likely cases already discussed are Chinantec Lealao (see Section 4.1.1.2) and Wari’ (see Section 4.1.2). Another possible example comes from Copainalá Zoque. Its imperative in (36a) is specialized through the suffix -ʌ and the missing pronominal prefixes. In the negative imperative in (36b), the LV takes the suffix -i, which marks it as dependent on a negative auxiliary and also occurs in SN (Miestamo 2005: 367–368), and uy ‘don’t!’ appears. Interestingly, uy can introduce subordinate clauses of negative purpose too: uy chʌjcu ‘lest you don’t do it’ (Harrison et al. 1981: 442). A plausible connection between the two uses is the insubordinated use of the ‘lest’-clause to indicate apprehension (‘(Be careful) lest you fall!’). Such an expression verges upon and can become a warning (‘Make sure that you don’t fall!’), which may then in turn develop into a more general negative directive/imperative (‘Don’t fall!’). This pathway has been documented for many languages (e.g., Aikhenvald 2010: 278, 355–356).^[8]

For Evans (2007: 391), insubordination includes the free-standing use of non-finite verb forms. In three of the sample languages, this phenomenon has resulted in A/Fin specific to IN. Korowai in (37) is one of them.

The imperative involves a set of dedicated suffixes, like in (37a). Negative imperatives can be formed in two ways. The finite one in (37b) features the prohibitive adverb belén and a negative irrealis verb form that lacks the first part of the normal negative circumfix be- … -da. In its non-finite alternative, the LV always appears as an infinitive and, as (37c) and (37d) show, belén may be present or -da may be used instead alongside the desiderative clitic =xup.

Such independent occurrences of non-finite verb forms do not seem to count as insubordination for Aikhenvald (2010: 275–284). She does indicate that they share a sense of incompleteness with insubordination. At least as originally non-conventionalized directive strategies rather than negative imperatives, both phenomena do not explicitly encode the offer of a reason not to act to the addressee and leave it to be inferred. Another phenomenon in the data that fits this description is the abessive (or privative) construction. It has been pointed out before (e.g., Phillips 2020) that a construction with the meaning ‘without’ is able to turn into SN. In three of the sample languages, however, a negative construction of abessive origin is attested solely in IN and SN displays no A/Fin. Murui in (38) is a case in point.

SN in this language has no A/Fin, as (38a) makes clear. IN does, however. Imperatives can be expressed by a range of suffixes (e.g., -kai for immediate compliance; Wojtylak 2017: 462). The one for a “basic” imperative of a disyllabic verb is -ø, like in (38b). The negative imperative (38c) has the LV taking the negative suffix -ñe and the abessive marker -no. The comparison with (38d) shows that this construction is normally reserved for syntactically dependent contexts. It also suggests a literal translation of (38c) as ‘without walking’ – which, crucially here, contains no overt expression of negative directivity.

An important question to answer now is why these processes take place. The motivations for the developments are clearly varied. Verbs originally meaning ‘leave’ or something similar occur in constructions that are, initially at least, analytical/compositional and can be argued to highlight the negative character of the directive act. A free-standing non-finite verb form often – though not necessarily – serves at first to indicate non-negotiable and instant compliance. Its presentation of the event as a “given” contributes to the effect and so does the likely brevity of the utterance, iconically echoing immediacy (Aikhenvald 2010: 283). Insubordinated constructions, by contrast, are frequently – though, again, not always – used originally to express a less forceful directive.^[9] Leaving the speech act implicit, they can be regarded as implementations of Brown and Levinson’s (1987: 227) politeness strategy “be incomplete, use ellipsis” (Aikhenvald 2010: 277). A good illustration of the phenomenon is from Chinantec Lealao.

The subordinate construction found in the negative imperative in (38c) actually has a positive equivalent – in (39b) – and it functions as a polite alternative to the imperative construction in (39a).

A perhaps even more important question is why such constructions seem to conventionalize more often into negative imperatives than into imperatives, giving rise to A/Fin. Languages do employ independent non-finite verb forms in positive contexts as well, after all (e.g., Dutch infinitival Zwijgen! ‘Shut up!’; Van Olmen 2010). As (39b) shows, insubordination too can produce positive directive strategies (e.g., requestive If you could open the door in English). A possible explanation is that negative imperatives are generally much less stable diachronically than imperatives. The latter are often archaic forms “inherited, unchanged, from the protolanguage” (Aikhenvald 2010: 339). The verb stem plus the final vowel -a in Bantu is a case in point: it can still be found in 97 % of the languages in the family (Devos and Van Olmen 2013: 10). The former appear to undergo changes and/or be replaced much more quickly. The result in Bantu is substantial synchronic variation within the family (Devos and Van Olmen 2013: 24–43). Similarly, in their diachronic study of negative imperatives in Egyptian, Grossman and Polis (2014: 42) contrast “the frequent innovation of prohibitive … constructions” with “the relative stability of the affirmative imperative system” (see also Van Olmen 2010: 495–496 on Latin).

The reasons for this instability are probably complex. It has been suggested, for instance, that languages may develop a novel negative imperative with a preverbal negator to satisfy the need for negative first. This principle would be even more powerful in directives than in assertives. Horn (2001: 450) formulates it as follows: “While a violation of Neg First by the postverbal negation in [declaratives] … might result in temporary confusion, a similar transgression in the context of [imperatives] … would literally constitute a matter of life and death (Kill him – oops – not!).” Another potential reason is described in Grossman and Polis (2014). For some of the changes in Egyptian, they present the idea of a process reminiscent of a Jespersen Cycle: an existing negative imperative construction is felt to be insufficiently strong, a more emphatic one is introduced and this innovation may replace the older construction (and weaken in turn). A partially opposing view is expressed by Devos and Van Olmen (2013). They agree with Evans (2007: 393) that “the face-threatening nature of requests and commands places strong pressures on the language system to come up with new variants whose pragmatic force is freed from the history of existing formulas”. Additionally, they hypothesize that negative imperatives may be under more such pressure than imperatives. They are both normally a threat to the addressee’s negative face, since trying to get them (not) to do something restricts their freedom of action. Unlike imperatives, though, negative imperatives also tend to threaten the addressee’s positive face, i.e., their desire to be appreciated or approved of: telling someone to refrain from doing something “amounts to an unfavorable evaluation of the hearer’s expected behavior” and telling someone to stop doing something “comes down to a direct rejection of his or her behavior” (Devos and Van Olmen 2013: 44).

To replace a negative imperative, for whichever of the reasons mentioned in the previous paragraph, a language is bound to look at the directive strategies that it has at its disposal. The strategy that meets its needs the best is then likely to increase in frequency and may eventually conventionalize as a new dedicated construction. If this strategy happens to involve a less finite or syntactically dependent verb, the outcome will be an A/Fin relative to the more stable imperative. In other words, it is perhaps better to avoid proposing any comprehensive explanation for A/Fin in IN – like Miestamo (2005) with stativity as a motivation for its occurrence in SN – and to regard it simply as a possible by-product of the diachronic differences between imperatives and negative imperatives. This view is, in fact, in line with the conclusion of an earlier study of (ir)realis asymmetry in IN. Van der Auwera and Devos (2012: 182) argue that, rather than some overall motivation, “the explanation [for the presence/absence of irrealis marking in imperatives and/or negative imperatives] … must be that imperatives and prohibitives are the results of partially independent diachronies” (see also footnote 1 on irrealis in SN).

4.3.3 A/Fin in SN only

As mentioned before, A/Fin in SN is often the result of diachronic processes that see existential negation (Veselinova 2014, 2016) or nominal negation (e.g., Phillips 2020) being used for verbal negation. There are various ways in which such an extension may occur. Some could be said to be motivated by SN’s stativity (Miestamo 2005: 221–224). If a negated event like ‘She didn’t sing’ is presented with a negative existential and a nominalized verb form, i.e., as ‘There was no her singing’, or with an abessive marker, i.e., as ‘She was without singing’, the event’s state of non-existence or absence is arguably highlighted. However, other mechanisms that still produce A/Fin in SN may have little to do with its stative character or could be explained in a different way too.^[10] Veselinova (2014: 1336), for example, shows that the Bulgarian SN of the future tense, which involves negative existential njama ‘not have’ and the LV in a subordinate clause, is simply inherited from a ‘have’-future construction in Old Church Slavonic that was particularly frequent in the negative. Veselinova (2014: 1357) also argues that negative existentials regularly expand into SN as “higher verbs in a complex clause structure”, where they essentially mean ‘it is not the case that’ and the LV occurs in the subordinate clause. One might wonder whether such A/Fin constructions really manifest SN’s stativity or whether they just reflect the fact that the negation has scope over the whole proposition. Another common feature of such extensions, irrespective of origin or motivation, is that they remain limited to specific SN contexts (e.g., the future in Bulgarian) and/or alternate with an earlier SN construction in some way that stays remarkably stable over time (Veselinova 2016: 158). As a result, there is often variation with regard to A/Fin in SN that has no counterpart in IN.

In ten of the 17 languages displaying A/Fin solely in SN, its absence from IN lies in the latter domain’s symmetry, in two non-mutually exclusive ways. First, a language’s IN may be symmetric with an SN construction that does not exhibit A/Fin. Nivkh in (40) is a case in point. It has a number of SN constructions. One of them is a NegVerb one (Miestamo 2005: 86) and a negative existential cycle involving the verb q ^h aw- ‘not exist’ is responsible for it (Gruzdeva and Fedotov forthc.). Another, less frequent one features -̓gavr. This marker is negative existential in origin too but, synchronically, it functions as a negative suffix and the construction is symmetric, as (40a) and (40b) make clear (Gruzdeva and Fedotov forthc.). It is the latter pattern – more or less restricted to the future in declaratives – that is followed in IN, as (40c) and (40d) show.

Second, IN may not have A/Fin because negative imperatives are symmetric with imperatives. Epena Pedee can serve as an example. As argued by Miestamo (2005: 274–275), SN of the past in this language involves a Neg-LV construction and the past and future are not negated symmetrically either, since the tense suffixes are suppressed in the negative (Miestamo 2005: 274–275). Consider its IN in (41), though: (negative) imperatives carry the same imperative suffixes and differ only in the presence of the prohibitive marker -náa.

It is again possible (see Section 4.3.1) that language-internal analogy and economy are factors in the coherence that IN exhibits with part of SN in a language like Nivkh or the coherence that a language like Epena Pedee displays within IN. However, more data about -̓gavr’s expansion and -náa’s diachrony is needed to determine whether they indeed played a role.

Another reason why languages may have A/Fin in SN only is the manifestation of a different form of asymmetry in IN. In this type, negative imperatives are asymmetric with imperatives because their respective markers belong to the same mood paradigm (cf. Miestamo 2011: 2, 12–13 on polar interrogative marking as part of mood). Awa Pit is one of five languages in the sample where such an asymmetry is attested. It has a range of SN constructions, some of which exhibit A/Fin (see Section 4.1.1.1) and might be the result of a negative existential cycle (see Krasnoukhova and van der Auwera 2019: 455 on shi). Its (negative) imperative suffixes, including first person object imperative -zha in (42a) and second person plural negative imperative -man in (42b), form a set of markers that all occupy the same slot on the verb. The full second person paradigm is given in (42c) and it can perhaps even be extended to first person imperative suffixes such as -pay ‘let’s both …’ (Curnow 1997: 248–250) and other “mutually exclusive” (Curnow 1997: 239) markers for moods like the potential and the obligative.

In languages like Awa Pit, SN involves a separate exponent of negation whereas, in IN, the negative is arguably encoded as simply one of the options/variants within a mood paradigm dedicated to expressing directivity, on a par with the positive.

In the remaining languages with A/Fin in SN only, the negative imperative shares little to no resemblance to the imperative or any SN construction. The explanation probably lies in other analogies or in its independent diachrony. For most of the sample languages, the origin of the negative imperative is hard to determine without checking historical evidence or related languages. There is information, for instance, for Japanese. It has some A/Fin in SN (Miestamo 2005: 294). Its plain imperative is expressed through the suffix -e for consonant-root verb, as in (43a). Its negative imperative consists of the prohibitive marker na and the verb in its non-past form, as in (43b).

According to Jarkey (2017: 175), the idiosyncratic nature of (43b) results from the grammaticalization of a construction with third-person-imperative-like semantics: “This particle [na] is a reduced form of the Classical Japanese na-kar-e, itself a reduction from na-ku ar-e (neg-infin be-imp) ‘let there not be’”. IN in Japanese may, in other words, have involved A/Fin originally but it has been obscured by the further development of na into a particle.

4.3.4 Reverse A/Fin

As discussed in Section 4.3.2, processes like insubordination and the free-standing use of non-finite forms are not unique to negative directives. For Northern Paiute, for one, Thornes (2017: 155–156) observes that nominalized clauses such as (44) can be employed as positive directives and makes no mention of the option for negative ones. Their degree of specialization is unclear, however, since “nominalized clauses pervade Northern Paiute grammar as a means of packaging information in running discourse” and are said to fulfill the specific speech act of “offering advice” (Thornes 2017: 55).

Despite the respective diachronic (in)stability of (negative) imperatives, it is not inconceivable that a strategy like (44) grammaticalizes into a proper imperative. If the negative imperative in such a language did not exhibit any loss of finiteness, this would be a case of “reverse” A/Fin. There is little actual evidence of this pattern in the sample, though. The only concrete instance of reverse A/Fin comes from Japanese. Unlike its negative imperative in (43b), the imperative in (43a) has a polite alternative in which, as (45) shows, the LV appears as an infinitive.

According to Jarkey (2017: 175), the construction is “modelled on the productive subject honorific pattern – o-V- infin -ni.nar-u (hon-V-infin-s[ubj.]hon-npast) … – which has a variant form that utilizes the subject honorific verb nasur-u ‘do’ when it occurs with verbal nouns of Sino-Japanese origin”. The suffix -nasai in (45) derives from the imperative of this verb.