Ouk ´Ismen Oudén: Negative Concord and Negative Polarity in the History of Greek

in Journal of Greek Linguistics

In Ancient Greek a single set of indefinite enclitic pronouns was used indifferently in both negative/affective environments (i.e. like negative polarity items (NPI)) and in positive ones (i.e. like positive polarity items (PPI)). At the same time the negative pronouns used as negative quantifiers (NQ) were also employed as emphatic NPIs, with negative concord. The two functions of each class (i.e. PPI-like vs NPI-like, NQ vs NPI) were determined by syntactic distribution. In the specific case of negative sentences, an indefinite before a sentential negative marker (NM) functioned like a PPI but after a NM like an NPI, while a negative pronoun before a NM was an NQ but after an NM an NPI. This pattern was at odds with the canonical VSO clause structure that evolved in later antiquity, in which focal constituents were contrastively stressed and fronted to the left periphery: neither indefinite nor negative pronouns could be focalised because of the prosodic and/or semantic restrictions on their distribution. This deficiency was eventually remedied by formal/prosodic recharacterisation, the loss of NQs and the generalisation of NPIs to all syntactic positions available to DPs, including the focus position, a process that triggered their reinterpretation as involving universal quantification over negation rather than, as before, existential quantification under negation. The Modern Greek PPI kápjos and NPI kanís are traced from their origins in Ancient Greek and their role in the evolution of the system is explored. The final outcome is typologically to be expected in so far as NQs are redundant in a system in which NPIs appear freely both before and after NMs.


In Ancient Greek a single set of indefinite enclitic pronouns was used indifferently in both negative/affective environments (i.e. like negative polarity items (NPI)) and in positive ones (i.e. like positive polarity items (PPI)). At the same time the negative pronouns used as negative quantifiers (NQ) were also employed as emphatic NPIs, with negative concord. The two functions of each class (i.e. PPI-like vs NPI-like, NQ vs NPI) were determined by syntactic distribution. In the specific case of negative sentences, an indefinite before a sentential negative marker (NM) functioned like a PPI but after a NM like an NPI, while a negative pronoun before a NM was an NQ but after an NM an NPI. This pattern was at odds with the canonical VSO clause structure that evolved in later antiquity, in which focal constituents were contrastively stressed and fronted to the left periphery: neither indefinite nor negative pronouns could be focalised because of the prosodic and/or semantic restrictions on their distribution. This deficiency was eventually remedied by formal/prosodic recharacterisation, the loss of NQs and the generalisation of NPIs to all syntactic positions available to DPs, including the focus position, a process that triggered their reinterpretation as involving universal quantification over negation rather than, as before, existential quantification under negation. The Modern Greek PPI kápjos and NPI kanís are traced from their origins in Ancient Greek and their role in the evolution of the system is explored. The final outcome is typologically to be expected in so far as NQs are redundant in a system in which NPIs appear freely both before and after NMs.

1. Introduction

1.1. The Scope and Purpose of the Article

When the negative pronoun oudeís ‘no one’1 appears preverbally in Ancient Greek (AG, c. 8c bcc. 7c ad), the sentential negative marker (NM) ou(k) ‘not’ cannot be used simultaneously unless a double negative reading (e.g. ‘no one didn’t see Socrates’) is intended, cf. (1a).2 By contrast, when a form of oudeís appears postverbally, ou(k) is all but obligatorily present, at least with finite verb forms (see 2.2 below, and cf. Chatzopoulou 2012 for a full discussion). In this case no double negation is involved, cf. (1b):


‘No one saw Socrates.’

article image

‘Socrates saw no one/didn’t see anyone.’

In Modern Greek (MG, c. 17c ad–present), by contrast, there is no negative pronoun corresponding to oudeís,4 and the NM ðen ‘not’ (‪<‬ AG oudén ‘nothing’ used adverbially = ‘not at all’) appears obligatorily in combination with kanís/kanénas ‘anyone’5 in the translation equivalents of both (1a) and (1b): i.e. forms of kanís/ kanénas appear in both pre-verbal (2a) and post-verbal positions (2b):


‘No one saw Socrates’.

article image

‘Socrates saw no one/didn’t see anyone’.

Neither sentence has a double negative reading. Indeed, the fact that kanís/kanénas can also appear in certain non-negative contexts = ‘anyone’ shows that it cannot be inherently negative, cf: íðes kanéna? [saw-2sg anyone], = ‘did you see anyone?’ The use of these items with an apparently negative meaning in isolation from a NM (e.g. íðes kanéna?kanéna ‘did you see anyone?’—‘no one’) is therefore assumed here to be a matter of ellipsis,6 i.e. kanéna (ðen íða) [anyone (not saw-1sg)], cf. Giannakidou 2000a: 485–7.

In the case of indefinite pronouns, AG made no formal distinction between ‘someone’ and ‘anyone’, using the enclitic pronoun tis for both.7 After negatives, therefore, tis overlaps with oudeís, tis being neutral, oudeís more emphatic (cf. (3d) with (1b)):


‘Someone saw Socrates’.

article image

‘Socrates saw someone’.

article image

‘Someone did not see Socrates’.

article image

‘Socrates saw no one/didn’t see anyone’.

MG, however, has two formally contrasting items corresponding to English some(one) and any(one), namely kápjos and kanís/kanénas:8


‘Someone saw Socrates’.

article image

‘Socrates saw someone’.

article image

‘Someone didn’t see Socrates’.

article image

‘No one saw Socrates’.

article image

‘Socrates saw no one/didn’t see anyone’.

The purpose of this article is to trace the development from a pronominal/ specifier system that contrasted ‘no X’ with a formally undistinguished ‘some/any X’ into a system that contrasts ‘some X’ with ‘any X’ and has (virtually) dispensed with ‘no X’. The analysis of the evolution of negation in Greek will, however, also be used to advance a claim that languages typically lack items meaning ‘no X’ when those meaning ‘any X’ may appear both before and after the negative that licenses them (as in MG, cf. (4c.ii) and (4d), but not in English, cf. *anyone didn’t see Socrates).9 It is also argued that the availability of pre-negative ‘any’ depends on which of two possible semantic interpretations is assigned to the items in question (on which see 1.3).

1.2. Some Key Concepts

Many languages, including English, make a formal distinction between ‘positive polarity items’ (PPIs = ‘some X’) and ‘negative polarity items’ (NPIs = ‘any X’). The former are used in positive assertions, as in (5a), the latter in conjunction with a negative element, such as the enclitic NM -n’t in (5b):

(5) a. John saw someone.

b. John didn’t see anyone.

Since the negative that licenses English NPIs always precedes them (cf. *anyone didn’t see John), it seems that NPIs in English must always fall within the semantic scope of negation: ‘[it is not the case that [any X …]]’ We should note, however, that many NPIs, including any in English, may also be used in a range of ‘affective’ environments, including interrogative, modal, habitual, conditional, future-referring, and imperative sentences, where they might more precisely be called ‘affective polarity items’ or APIs. These all involve what Giannakidou (1998, 2000a, 2000b, 2005, 2011) has called non-veridicality, i.e. semantic functions that do not ensure truth. Unlike in the negative case, which is specifically anti-veridical, non-veridical contexts usually offer a choice between the use of PPIs and NPIs associated with differences of specificity and/or existential commitment: cf. if someone comes … /if anyone comes … etc. Recall that wider usage of this kind tends to preclude the possibility that the NPIs in question are themselves inherently negative (cf. the discussion of (2) above), though it is clear that routine association with a NM may lead over time to a situation in which the NM becomes subsidiary (and may even be lost), and the NPIs themselves evolve towards NQ status (see fn. 13 for the case of colloquial French). This may then entail a situation in which the former NPIs retreat from affective environments leaving PPIs as the sole survivors in this domain.

The term ‘negative concord’ (NC) is standardly used to describe the use of more than one negative item in a construction that carries only a single negative reading, as in the Italian example in (6), where nessuno ‘no one’ is necessarily accompanied by the NM non ‘not’ but the meaning is simply ‘Gianni saw no one/didn’t see anyone’, involving just one instance of negation semantically:


‘Gianni didn’t see any one.’

In such cases the pronoun appears to ‘agree’ with the NM in negativity without contributing a negative meaning of its own (though see 1.3 below for further discussion). The situation in AG, as illustrated in (1b), is very similar.10

By contrast, when no one is combined with not/-n’t in (standard) English, the two negatives are interpreted separately to give a double negative reading equivalent to an emphatic positive, as in (7a)).11 When a negative meaning is intended, no one appears without not, as in (7b)), which is semantically equivalent to (5b):

(7) a. John didn’t see no one. (= ‘John did see someone’)

b. John saw no one.

Negative items like no one, which retain a negative reading of their own when in combination with another negative, are called ‘negative quantifiers’ (NQ).

1.3. Some Important Issues

Consider now the Italian sentences in (8). When nessuno ‘no one’ appears preverbally as a subject, as in (8a), it functions as a NQ and no further negation is required; indeed the presence of non forces a double negative reading. But in (8b) (= (6)), where nessuno is a postverbal object, non is obligatory, i.e. postverbal nessuno must fall within the scope of a licensing negative, leading to a case of NC:12


‘No one saw Gianni.’

article image

‘Gianni didn’t see anyone.’

In (standard/formal) French, by contrast, there are no NQs, and NPIs like personne ‘anyone’ appear both pre- and post-verbally in conjunction with obligatory sentence negation, as in (9a) and (9b), neither of which has a double negative reading:13


‘No one saw Jean.’

article image

‘Jean didn’t see anyone.’

Furthermore, since French NPIs, unlike their English counterparts (cf. the discussion of (5)), may precede as well as follow the licensing negative, it seems that they are not required to fall within the scope of the licensing negation (cf. the MG example in (4c)(ii)): ‘for any X [it is not the case …’ It is again assumed that the use of such items in isolation from a NM (e.g. qui est là? (‘who’s there?’)—personne (‘no one’)) is a matter of ellipsis, personne (n’est là) [anyone (not be-3sg there)].

To summarise, the following pre-/post-verbal pairs of single-negation structures involving NQs and NPIs have been illustrated from Greek, English, Italian and French:

article image

Since the postverbal case of (10c) involves items that may otherwise appear unambiguously as NQs (cf. oudeís in (1b) with preverbal oudeís in (1a), or nessuno in (8b) with preverbal nessuno in (8a)), the question arises as to whether these should be treated as ‘real’ NQs or, despite their negative form, as NPIs homophonous with the corresponding NQs (on which see immediately below). For the moment, however, note that only unambiguously negative NQs, i.e. those that contribute negation in the absence of an overt NM, allow double negative readings in combination with a further negative, and that there is a significant semantic difference between the pre- and post-verbal cases, cf. (11a) and (11b) respectively:

article image

Given the differing forms and distributions of NQs and NPIs across languages, there has been a great deal of debate about the semantics of these items, e.g. whether NPIs are themselves semantically negative,14 whether NQs and NPIs should be treated as distinct sets of elements,15 etc. There is further disagreement about how best to deal with NC, a matter which depends largely on whether NQs/NPIs are treated as negative or not. See Giannakidou 2000a, 2000b, 2005, 2011, Horn and Kato 2000, Werle 2002, and Penka 2010 for a range of views and their consequences, and Horn 2010 for a recent general bibliography. The detailed investigation of such issues is beyond the scope of this article, but the positions adopted here (more or less following Giannakidou 1998, 2000a, 2000b, 2005, 2011) must first be briefly stated.

It is assumed, hopefully uncontroversially, that NQs express, without the presence of any further negation, the non-existence of items with the specified attribute(s) and as such contribute to double negative readings when used with other negatives. Accordingly, oudeís in (1a), no one in (7a/b) and nessuno in (8a) are all treated as NQs.

It is also assumed that formally ‘non-negative’ NPIs such as MG kanís/kanénas, anyone and (standard) French personne (cf. the discussion of (2), (5) and (9)), though licensed by negation, are not themselves negative in meaning and therefore do not contribute to double negative readings. Rather more controversially, oudeís in (1b) and nessuno in (8b) are also treated here as NPIs despite their ‘negative’ form, primarily on the grounds that they too co-occur with a NM without contributing negation of their own.16

The choice of what can serve as an NPI is therefore between items that are already associated with non-veridical contexts (viz. APIs) and items that are already associated with negation (viz. NQs). The appearance of APIs as NPIs in specifically anti-veridical environments presumably reflects a fairly simple and natural process of extension, but the latter case is clearly more difficult to explain. We saw in (11), with regard to sentences involving double negation, that there is a significant difference between NQ subjects, which take wide scope over negation (‘there is no X such that it is not the case that …’), and NQ objects, which have narrow scope under negation (‘it is not the case that there is no X …’). This relates directly to the absence of NC readings of sentences like no one didn’t take a turn: (apparent) NQs in subject position cannot be read as NPIs because they do not lie within the scope of the negation (cf. also the discussion of anyone in (5)). The AG data in (1) confirm this conclusion and further suggest that the reanalysis leads to the (eventual) demise of ‘bare’ NQs in positions in which they could in principle fall within the scope of negation, i.e. if an NM were introduced. The key to this reanalysis is clearly the specific scopal sequencing ‘not [no (X)]’, which becomes ‘not [any (X)]’ through ‘deletion’ of the negation lexically incorporated within no. What this process actually amounts to, and what exactly drives it, are again complex issues beyond the scope of this article.

Whatever the proper analysis of negative-form NPIs, we shall henceforth treat NC as a rather more general phenomenon than example (6) suggested, namely as the selectional relationship between negation, prototypically a NM, and whatever set of NPIs (if any) it licenses in a given language, whether these are of negative or non-negative appearance. Languages like (standard) French and MG would then exhibit ‘strict NC’, i.e. a NM licenses NPIs in all syntactic environments,17 while AG, Italian and English have ‘non-strict NC’, i.e. a NM licenses NPIs only in those syntactic environments in which the NPI falls within the scope of the negation.

When NPIs of whatever form appear syntactically only to the right of the negative element that licenses them, i.e. when the language in question has non-strict NC, as with oudeís in AG (cf. (1b)), any in English (cf. (5b)) and nessuno in Italian (cf. (8b)), they are interpreted here as involving narrow-scope existential quantification under negation:

(12) ‘it is not the case that there is some/any X such that [X has role R in event E]’

This crucially links the obligatory pre-NPI position of the licensing negative18 with a reading in which negation necessarily has wide scope.19 In other words, it accounts for the impossibility of placing an NPI before a licensing negative and motivates the simultaneous presence of NQs in the relevant languages, assuming we understand these as items that combine negation and existential quantification lexically as an alternative means of expressing (12) where NPIs (+ NM) are either unavailable or disallowed. Though in principle stable (cf. Italian), this form of complementary distribution is potentially vulnerable to levelling, e.g. through the generalisation of NPIs to pre-NM position via the adoption of a different, though synonymous, reading (thereby fatally undermining the role of NQs).

Thus when NPIs occur syntactically before their licensing negatives (i.e. when a language has strict NC), as with kanís/kanénas in MG ((4c.ii) and (4d)) or personne in standard French ((9a) and (9b)), they can only be understood to involve wide-scope universal quantification over negation:

(13) ‘for any X, it is not the case that [X has role R in event E]’

‘Any’ so interpreted does not bind the relevant set of entities collectively (like ‘all’), or individually and specifically (like ‘each/every’), but on the basis that a random, potentially hypothetical,20 selection of a member of the set will in every case identify individuals of whom the associated negative proposition is true. But the crucial thing here is that this interpretation of NPIs works satisfactorily regardless of syntactic position, given that the universal quantifier invariably has scope over the negation.21 It is therefore reasonable, and certainly more economical, to assume that, in languages where (13) applies at all, it applies by default to NPIs in all environments. Languages that treat NPIs in this way have no need of NQs, and any NQs that may survive from an earlier period in which interpretation (12) was in play are likely to be abandoned or reinterpreted. The highly relevant example of Greek is considered in detail in Sections 2 and 3.

In the light of this discussion non-strict NC seems to be inherently associated with the reading (12) for NPIs, strict NC with reading (13), i.e. the two readings distinguish languages like Italian and AG that do not allow [… NPI … NM-V …] from those like (standard) French and MG that do. Further implications of the two different readings of NPIs will be explored below. Here we simply note that Greek has shifted from an Italian-like position in its ancient form to a (standard) French-like position in its modern one, or equivalently from exhibiting non-strict NC to having strict NC. See Zeijlstra 2006 for the different contribution of NMs in languages with strict NC and non-strict NC, Haspelmath 1997 and Israel 2011 for recent discussion of the full range of issues surrounding indefinite pronouns, negative polarity items and their relations.

2. Ancient Greek to Medieval Greek I: The Data22

2.1. Indefinite tis in AG vs Indefinite tis/tinás in MedG23

A number of issues relating to negation and negative polarity in Greek, both ancient and modern, have been examined in the recent literature (e.g. Giannakidou 1998, 2000a, 2000b, 2005, 2011), Klein 2011, Tsimpli and Roussou 1996, Willmott (2013), and, in part, Chatzopoulou 2012); Kiparsky and Condoravdi 2006 deals specifically with Jespersen’s Cycle in relation to Greek. But the focus of this article is different in that it deals specifically with the transition from the AG to the MG system of negative indefinites (i.e. NQs and NPIs), as outlined in 1.1. The key to an understanding of this process is the treatment of negation and negative indefinites in Medieval Greek (MedG, c. 8c–c.16c ad).

In AG and in written forms of MedG that reflect the contemporary vernacular in some degree24 forms of tis/tinás correspond to English some(one) in both pre- and post-verbal positions in positive sentences, ie. they assert the existence of one or more people/things. In MedG, however, and unlike in AG where it was enclitic (see fn. 7), tis/tinás may routinely appear clause-initially and/or be emphatically stressed (e.g. as in (14a)(ii), where the focal status of tiná is indicated by italicisation), indicating it had lost its clitic status and taken on the role of an indefinite quantifier:25

(14)a. preverbal(14)
article image

‘Cebes is always on the look-out for some discussion …’

Plato (428–347bc), Phaedo 63a2
article image

‘Of her many (complaints) I have set before you (just) some’

Ptochoprodromica (12c ad), 1.114
b. postverbalarticle image
article image

‘But why do some take pleasure in spending time with me?’

Plato (428–347bc), Apology 33b9
article image
article image

‘For they contain (some) expressions full of bitterness’

Ptochoprodromica (12c ad), 1.122

Forms of tis/tinás occurring after a negative element, however, correspond to English any(one) in both AG and vernacular MedG, i.e. in conjunction with a negative they assert the non-existence of people/things with the relevant attributes. Again MedG tis/tinás in this sense may freely appear in clause-initial position and/or receive emphatic stress (e.g. as in (15b)):

article image

‘He neither buried any of the dead nor ransomed any of the prisoners’

Hyperides (c. 390–322bc), Fragment 76.16
article image
article image

‘She neither kissed me nor spoke to me at all’

Digenes Acrites G (13/14c ad MS), 1.308

The parallelism in (15a) between the neither-clause and the nor-clause, together with the fact that the latter contains an unambiguous use of the NPI oudéna = ‘ANYone’ (i.e. the emphatic equivalent of tiná, cf. (1b) and (3d)), argues strongly that tinás in the former should also be taken as falling within the scope of the negation and that it an NPI-like rather than PPI-like role (the latter = ‘he didn’t bury some of the dead’, with the existential quantifier having wide scope). This conclusion is supported by the fact that positive readings in negative sentences in AG involve preposing of the indefinite before the NM, as in (16), where syntactic position and semantic function correspond, i.e. with tis standing to the left of the NM just as the existential quantification lies outside the scope of the negation:26

article image

‘Some (inlets) empty with the tides, but others the water never leaves at all’

Strabo (1c bc – 1c ad), Geographica
article image
article image

‘And as Aratos did not allow the Achaeans to cross a certain deep ravine …’

Plutarch (c. 46–120ad), Agis and Cleomenes 27.4.1

In the medieval period, however, a major change took place in this pre-negative use of tis/tinás in the spoken language and in written styles that reflected the vernacular most closely. Contrast the examples in (16) with the MedG data in (17):

article image

‘I act according to my own principles, (namely) that I fear no one’

Vitae et Miracula Sancti Anastasii Persae: Miracula Romana (?10c ad), B. Flusin, Saint Anastase le Perse et l’histoire de la Palestine au début du VIIe siècle, 1. Paris, CNRS 1992: 165–187, section 7,13
article image

‘Let no one boast of that’

Digenes Acrites E (?11c ad, 15c MS), 36
article image
article image

‘Prison makes a bad enemy …, … it has no friend’

Michael Glycas (12c ad), Verses written while imprisoned, 239–40
article image

‘The noble girl had never loved anyone’

War of Troy (13/14c ad), 310

At least from the 10th/11th centuries ad onwards, but most probably earlier, forms of tis/tinás in this position can only mean ‘any(one)’ (i.e. ‘no(one)’ in combination with the following NM), and the PPI reading of AG is excluded. Given the difference between the ancient and (vernacular) medieval senses of, for example, (16a) (viz. ‘there are some inlets that empty out with the tide and others that do not’ vs ?? ‘there are some inlets that empty out with the tide and none that do’), many uneducated people must at times have felt extremely puzzled when listening to older forms of Greek, e.g. in biblical readings etc.

Evidently, the shift to a ‘modern’ distribution and interpretation of indefinites, as illustrated with corresponding modern forms in (4c) and (4d), had occurred by around the end of the first millennium ad. We should note here that only written registers above the most vernacular still retain plural forms of tis/tinás and a significant use of this item with PPI readings. Otherwise, it exhibits mostly singular forms (with innovative nominative tinás) and is increasingly used in negative/affective environments to the exclusion of positive ones. Relevant numbers are given in (18) for the Escorial (E) Digenes Acrites (?12c ad, 15c MS), the Ptochoprodromica (12c), the Grottaferrata (G) Digenes Acrites (13/14c MS), the War of Troy (13/14c, first 5000 lines), and both the Copenhagen (H) and Paris (P) manuscripts of the Chronicle of the Morea (14c). These are all to some degree ‘vernacular’ texts but are divided here into ‘less’ vs ‘more’ vernacular according to the level(s) of language employed. The numbers of nominative and accusative forms of tis/tinás with PPI and NPI readings (the latter in both negative and affective environments) are then given, together with an indication of whether plural forms occur:

article image

Given that the innovative PPI kátis/kápjos ‘some(one)’ and NPI kanís/kanénas ‘any(one)’ were already in competition with tis/tinás in this period (see 3.1 for details), it seems that the appearance of these formally contrasting pronouns/specifiers in the spoken vernacular was intimately bound up with:

(19) a. The progressive loss (other than in written styles retaining aspects of traditional practice) of tis/tinás with PPI readings in favour of the true PPI kápjos.

b. The convergence (other than in written styles retaining aspects of traditional practice) of the use of tis/tinás with an NPI reading with that of the true NPI kanís/kanénas—which has only a singular paradigm and is used both before and after licensing negatives.

To pursue this investigation of the transition from the ancient to the modern system of negative/indefinite pronouns and specifiers we must therefore consider the origins and development of a formal ‘some/any’ contrast in MedG, and the associated issue of the loss of the NQ oudeís ‘no(one)’ in favour of the generalisation of the NPIs kanís/kanénas and tinás to pre-NM environments (for which see 3.1). But this requires that we first examine the use and distribution of oudeís/uðís29 ‘no(one)’ in AG and MedG.

2.2. Oudeís/uðís: NQ vs NPI in AG and MedG

We begin with the distribution of oudeís as a NQ in AG, as illustrated in (20) and (21) for pre-verbal and (apparent) post-verbal positions respectively:

(20)preverbal (double negation possible before NM ou(k))(20)

‘But when no one came out to fight’

Thucydides (c. 460-c. 395bc) 1.6
article image

‘And I never yet robbed anyone of their charm’

Plato (428–347bc), Hippias minor 372c5
c. (with double negation):30article image
article image

‘Then none of those watching failed to suffer’

Xenophon (c. 430–354bc), Symposium 1.9.4
(21)postverbal (no double negation possible after NM ou(k))31(21)

‘No one plans and puts into practice on a parallel basis’

Thucydides (c. 460-c.395bc), 1.120.5
article image
article image

‘Taking (them) away they put them to death with no exceptions’

Thucydides (c. 460-c. 395bc), 3.68.2
article image

‘They never wronged anyone’

Demosthenes (384–322bc), Epitaphios 7.4

It is important to note, however, that while the preverbal use of oudeís as a NQ is normal in positive sentences and (just about) possible in negative ones containing ou(k) ‘not’ (see fn. 30), the corresponding postverbal use is very rare in positive sentences and impossible in negative ones after NM ou(k). Furthermore, when oudeís does occur postverbally there is usually contextual evidence that the verb has been fronted over it for reasons of emphasis or contrast, as in (21a) and (21c).32 Since sentences of this kind do not have pragmatically neutral word orders they do not constitute evidence for the ‘normal’ use of NQs in postverbal position. Even (21b) might be read as involving topicalisation of eksaíreton, with an associated focalisation of the verb (‘as for exceptions, they made none’).33 Evidence for the regular postverbal use of oudeís as a NQ is therefore vanishingly rare in Greek of the 5th-century bc and onwards, and we find instead oudeís used as an emphatic NPI after a preceding negative in parallel distribution with tis in its NPI-like uses (cf. (15) above), and even this combination becomes steadily less frequent in later antiquity:

article image

‘And no one either appeared on the wall …’

Thucydides (c. 460-c.395bc), 5.7.5
article image
article image

‘I cannot tell you of any of today’s orators’

Plato (428–347bc), Gorgias 503b5
article image

‘Nor does any of our friends talk to us’

Demosthenes (384–322bc), Philippics 4.54.1
article image
article image

‘I do not think anyone will lay this charge against us in a moderate way’

Polybius (c. 200–118bc), Historiae 9.20.5

Turning now to MedG, and using the same corpus of vernacular texts as for tis/tinás (cf. (18)), the numbers of nominative/accusative forms of uðís (including rare examples of innovative nominative uðénas), in pre- and post-verbal positions, both without and with a NM, are as follows:34

article image

This distribution shows that the postverbal use of uðís was by now in its death throes, even as an NPI following a NM. The evidence from the War of Troy for an extension of NPI uðís (i.e. without double negation) to pre-NM/preverbal position, thus replicating the MedG distribution of tinás with NPI-like function (cf. (17)), is entirely out of keeping with the general trend and probably best explained as due to editorial intervention (see fn. 34):

article image

‘There is none of us in all the army …’

War of Troy (13/14c), 7967

Uðís thus survives principally as a preverbal NQ, mostly as a subject (nominative) and without a following NM, i.e. more or less in continuation of the AG use in this position. But even here the numbers are small, and the almost complete absence of uðís from the more vernacular of these texts, ignoring the War of Troy but including the relevant parts of the Ptochoprodromica, is striking. It is tempting to speculate that it was no longer current in everyday spoken Greek or in poetry with a strong oral/popular background like the Escorial Digenes, and that its already ‘literary/archaic’ quality might also have rendered it inappropriate for the ‘of-the-moment’ urban satire of the Ptochoprodromica.

3. Ancient Greek to Medieval Greek II: Analysis of the Developments

3.1. AG ‪>‬ MedG: The Loss of NQs and the Reinterpretation of NPIs

The uses of tis and oudeís in AG may be summarised as follows:

article image

The fundamental contrast is between tis = ‘someone’ (PPI-like) and oudeís = ‘no one’ (NQ). The former appears in positive sentences pre- and post-verbally and the latter in negative sentences pre-verbally without an NM (post-verbal oudeís as a NQ is marginal at best, as noted above). These same meanings are retained when a NM is present, but only when the two items precede it (though the double negative use of oudeís is very rare, as noted). This contrast between tis and oudeís is neutralised, however, in negative sentences after a NM, where both function as NPIs = ‘any(one)’ (albeit with a difference of emphasis, again as noted).

Since NPI-like tis cannot be generalised to pre-negative positions (pre-negative tis is always PPI-like), an alternative means of expressing the non-existence of people/things had to be employed in sentences in which a negative would otherwise follow an indefinite, viz the NQ oudeís. The latter, however, was (all but) impossible postverbally and was therefore in (virtual) complementary distribution with ou(k) + NPI tis/oudeís. Thus of the two possible readings of NQs/NPIs (cf. (12) and (13)) only that involving narrow-scope existential quantification under negation can account for the distribution of the relevant pronouns/specifiers in AG, as noted. The complementary (26a) and (26b) therefore both mean (26c):

(26) a. oudeís [NQ] … V …

b. ou(k) V … tis/oudeís [NPI] …

c. it is not the case that [there is some/any X such that [X has property P]]

where the order of the negative and indefinite elements (compounded in (26a), separate in (26b)) directly reflects the order of the semantic operators in (26c).

In vernacular MedG, however, PPI-like tis has largely been replaced by PPI kátis/kápjos and tis/tinás is now employed before as well as after a NM as a true NPI. Correspondingly, the emphatic NPI oudeís/uðís has mostly been replaced by kanís/ kanénas, which, like NPI tinás, has also been extended to pre-NM environments. Both NPIs are now stressed normally and may also receive heightened stress, e.g. for emphasis/contrast. In association with a NM, therefore, kanís/kanénas and tinás more or less replace both the preverbal NQ oudeís/uðís (in the order NPI + NM) and the post-negative NPI oudeís/uðís (in the order NM + NPI). PPI tis and NQ oudeís thus serve chiefly as markers of more conservative registers, or at least of efforts to appropriate something of their prestige as ‘ancient’ forms. It should be noted further that the loss of NQs in the vernacular entailed the automatic loss of double negation.

The key changes are summarised in (27), where it can be seen that the principal contrast is now between the PPIs kátis/kápjos on the one hand and the NPIs tinás and kanís/kanénas on the other. The most striking differences between AG and MedG are marked in bold. Note in particular that pre-NM tis (always PPI-like in AG) is formally replaced in MedG in its PPI function and continues here only as an NPI (a role unavailable in AG):

article image

Before examining the two newcomers in detail, however, we should first emphasise that the across-the-board generalisation of NPIs to pre-NM positions, resulting in the elimination of NQs, crucially presupposes their reinterpretation as involving universal quantification over negation (cf. again (12) and (13)), a shift associated in Greek with the need to allow the relevant items to be focalised (for which see 3.3). Once this shift has taken place, the universal quantifier has the wider scope, and the indefinite pronouns/specifiers may now freely precede or follow the licensing negative:

(28) a. tinás/kanísou(k) V … tinás/kanís

b. for any X [it is not the case that [X has property P]]

This decisive interpretative change means that, unlike in AG, a uniform meaning is carried by a uniform construction in all syntactic positions, viz. NPIs in association with a (preceding or following) NM. It also means that PPIs and NPIs now interact with negation in the same way, i.e. both involve a form of wide-scope quantification over negation, the former existential ((29a)), the latter universal ((29b)):

(29) a. there is some X such that [it is not the case that [X has property P]]

b. for any X [it is not the case that [X has property P]]

Following discussion of the possible origins of kátis/kápjos and kanís/kanénas in 3.2 we turn in 3.3 to the underlying motivation for this redistribution and reinterpretation of NPIs.

3.2. The New Indefinites: PPI kátis/kápjos and NPI kanís/kanénas

For well-known cultural and historical reasons, relatively little material composed in more ‘natural’ forms of Greek has survived from the period between the 7th and 11th/12th centuries ad. One of many unfortunate consequences of this state of affairs for the historian of Greek is that the origin and spread of the innovative indefinite pronouns characteristic of MedG cannot be traced in the documentary record. By the time the relevant forms are securely attested, the medieval system outlined above is already in place. The most one can hope to do, therefore, is to link the late antique to the medieval system through reconstruction of the key interim developments.

Fortunately, there is AG evidence that is potentially indicative of the origins of both kanís and kápjos to help point the way. Beginning with the former, heîs ‘one’ (later eîs, then is) already overlapped with tis as an indefinite pronoun/specifier in more vernacular registers of AG, sometimes in combination with the latter as heîs tis, sometimes alone as in the following example, where heîs clearly means ‘a certain’ or ‘some’:

article image

‘One/some lame innkeeper was called ‘partridge’’

Aristophanes (c. 448–380bc), Birds 1292–3

This usage became increasingly normal with the passage of time and was well-established by late antiquity, by which time (h)eîs/is had acquired a clear role as an indefinite article, exactly as in MedG and MG:


‘And some/a farmer hired him’

Joannes Moschus (c. 550–619ad), Spiritual meadow 183 (Migne p. 3056, l. 5)

The AG emphatic particle ká:n ‘even’ was the product of the fusion of intensifying kaí ‘even’ with the conditional conjunction á:n ‘if ever’, ‪<‬ eá:n ‪<‬ *ei án.35 In classical Greek (5th/4th c. bc) the modal particle án in its generic function combined only with subjunctives in subordinate clauses. Ká:n was therefore originally used in future-referring/generic conditionals with subjunctive verb forms and the expected meaning ‘even if (ever) …’. But it also came to be used more generally as an intensifier by abstraction from its use in elliptical expressions such as that in (32), where the first ká:n appears in a complete conditional clause and the second in a conditional clause where the verb pése:i [fall-3sg-SUBJ] ‘he fall’ has been omitted, thus opening the way for reinterpretation simply as an intensifier of the following PP:

article image
article image

‘But a man, even if he grows his body great, must expect to fall, even (if he falls) from a small misfortune’

Sophocles (c. 496–406 B.C), Ajax 1077–8

Ká:n thus starts to appear as a variant of intensifying kaí in a range of non-veridical environments including not only conditionals but also modal verb forms, imperatives, futures, habituals, etc (cf. the discussion of (2) and (5) above). The examples in (33) all contain conditionals or imperatives, while ká:n in (33a) actually modifies a conditional conjunction (ei ‘if’ + optative marking a ‘remote’ possibility), confirming that, at least in colloquial varieties partly reflected in comedic dialogue, extension of the range of ká:n had begun by the later 5th century bc (Frogs was first performed in 405bc):


‘Even if you were to beat me I wouldn’t reply’

Aristophanes (c. 446bc-c. 386bc), Frogs 585
article image

‘But if once he makes even a small improvement …’

Aristotle (384–322bc), Categories 13a.25
article image

‘Come in even now (if you must)’

Menander (c. 341/42–c. 290bc), Fr 342, 1
article image
article image

‘If I touch even his clothes, I shall be saved’

St Mark (1st c. ad), 5. 28

Unsurprisingly, we eventually start to find examples in specifically negative environments too:36


‘… who had never yet owned even an ass’

Lucian (c. 125-post 180ad), Timon 20
article image
article image

‘With proper names there is no harm done even without their own articles’

Apollonius Dyscolus (2c. ad), On Syntax (Grammatici Graeci 2.2, p 114.7)
article image
article image

‘And he didn’t soften his own statement even to some degree’

Origen (184/5–253/4ad), Commentary on the gospel of St John 20.24.216

Ká:n was quite frequently combined syntagmatically with (h)eîs (= ‘even one’), as in (35), where preceding negative elements are also marked in bold:

article image

‘It has become impossible to recruit even one of them’

Lucian (c. 125- post 180ad), The fugitives 21.4
article image
article image

‘… (books) even one of which he only needed to have read’

Galen (129-c. 200ad), Against Lycus (Opera Omnia ed Kühn, 18a, p 219.6)
article image
article image

‘If even one (of the particulars) should appear opposed to the others …’

Sextus Empiricus (c. 160–210ad), Pyrrhoniae Hypotyposes 2.195
article image
article image

‘Let him show not many nor even a few but even one of such a calibre’

Origen (184/5–253/4ad), Against Celsus 2.8.3
article image

‘though there was not even one person to console them’

Origen, On Lamentations Fr 10.12 (Origenes Werke 3, ed. Klostermann)

As recognised at least since Jannaris (1897:163–4), such examples illustrate the kinds of context in which ká:n + (h)eîs/is might first have developed into a compound. In the specific case of examples falling within the scope of negation this coalescence seems to have led to a natural semantic development (i.e. one appropriate to the pronominal status of the compound) whereby ‘not … even one’ ‪>‬ ‘not … anyone.’ It is easy to imagine how, during the early middle ages, a vernacular pronoun kanís/kanénas might first have replaced the emphatic NPI uðís in positions after a NM and then, once generalised to pre-NM positions and reinterpreted as in (13)/(28), replaced the now redundant preverbal NQ uðís as well.37 Tis/tinás in NPI function was naturally drawn into the same distributional pattern, thus losing its enclitic status, before finally being replaced by its rival in the early modern period.

The resulting uniformity of role and consistency of interpretation of kanís/kanénas across all syntactic positions (i.e. as an NPI licensed by a NM) clearly represented a significant overall simplification vis-à-vis the functional and distributional complexities of ancient oudeís. In the following extracts, dating from c. 11th – c. 14th centuries ad, forms of kanís appear before and after a NM both as subjects and as direct objects:

article image

‘And if the mares follow you, no one will catch you’

Digenes Akrites E (?11c ad, 15c MS), 281
article image
article image

‘No one was found to get the better of me’

Digenes Akrites E (?11c ad, 15c MS), 155
article image

‘They don’t allow anyone to leave, (anyone) to go’

War of Troy (13/14c), 1065
article image

‘For they didn’t find anyone whatsoever there’

War of Troy (13/14c), 10005

But inherently negative/affective indefinites require inherently positive indefinites to complement them. It is interesting to note that MG NPIs other than kanís/kanénas are simply ‘strengthened’ forms of the enclitics that in AG appeared freely in both positive and affective/negative environments: thus accented tinás (AG tis) ‘anyone’, típota (AG ti) ‘anything’, accented poté (AG enclitic poté) ‘ever’, puϑená (AG pou) ‘anywhere’. On the other hand the corresponding PPIs all contain the accented prefix ká(n)-: thus ká-tis ‘someone’, ká-ti ‘something’, kám-posos (AG enclitic posós) ‘some amount/number (of)’, ká(m)-pote ‘sometimes’, ká-pu ‘somewhere’. Some tentative thoughts on the probable sequence of events in the development of the PPI/NPI opposition are offered below, once the likely origin of the PPIs has also been discussed.

There are no examples of PPI-like tis/tinás (cf. (18)) or any ‘modern’ PPI replacement in the Escorial Digenes Acrites, nor are there innovative PPIs in the more learned Grottaferrata version. But there are examples in other vernacular texts from c. 12c onwards, where the forms okátis/okápjos, okáti and related adverbs are common, ukátis/ukápjos, ukáti and related adverbs less so. Examples of okátis/ukátis etc are given in (37), and of okápjos/ukápjos etc in (38):38

article image

‘and now someone was eating a melon in his sleep’

Michael Glycas (12c ad), Verses written while imprisoned 266
article image
article image

‘If some men or some women take or steal my birds …’

Assizes B (Cyprus, 13c ad/ms 15c) 450.23
article image

‘… the sides somewhat rounded’

Ermoniakos (fl. first half 14c), Iliad 3.170
article image
article image

‘… even when they came to some water to cross’

Chronicle of the Tocco (15c) 2316
article image

‘even some neighbour’s dress might fall apart’

Ptochoprodromica (12c), 3.162
article image
article image

‘… that some king had done many things’

Chronicle of the Morea H (14c), 882
article image
article image

‘And if it happens that you find it some time in the future …’

Spaneas V (ms c. 1200ad) 42
article image

‘They were singing some songs’

Alexander Romance F (16c), 80.3

It is important to note that both the o-forms and the rarer u-forms are first attested from c. 12c onwards (cf. (37a) and (38c)), and that both are sometimes attested in a single text, e.g. the 13c Assizes from Cyprus (cf. (37b) with … afíni ukátinos ape ton víon [leave-3sg.PRES someone-GEN from the-ACC life-ACC] ‘… leaves to someone part of his/her estate’ in Assizes B 388.6), and the 14c Chronicle of the Tocco from the Ionian islands/Epirus (cf. (37d) with examples like (próloɣon) ton élejen okátis [(prologue-ACC) which-ACC spoke someone-NOM] ‘(introduction) that someone spoke’ at 1702). Though both the Chronicle of the Tocco and Ermoniakos’ Iliad (cf. (37c)) are associated with a ‘northern’ region where raising of unstressed mid-vowels took place in the middle ages (i.e. /e/ ‪>‬ /i/ and /o/ ‪>‬ /u/, see e.g. Newton (1972), Horrocks (2010: 404–6)), the u-forms also appear in texts from places where no such changes occurred, e.g. Cyprus, as above. Furthermore, u káti in the Chronicle of the Tocco is unique (o-forms are the norm), and in Ermoniakos’ poem (where u-forms are standard) there is no general notation of unstressed mid-vowel raising. This evidence taken together therefore suggests that the u-prefix may well be an original form rather than simply a regional variant of o-, albeit a residual one with sporadic attestation by the time of our earliest vernacular texts. This possibility is explored immediately below.

Forms without the u-/o- prefix begin to be attested somewhat later (kápjos is the standard MG form, as noted), though forms with and without a prefix co-existed for some considerable time:

article image

‘Some nobleman wants, asks to harvest it (a vineyard)’

Achilleid N (?14/15c, ms 15/16c),1050
article image

Some find them in the deep’

Assizes B (Cyprus, 13c ad/ms 15c), 299.12
article image

‘She was watching over some sheep’

Voskopoula (Crete, 16/17c), 11
article image

‘He sent him on some service’

Life of Aesop K (16/17c), 167.4

The variation between the root elements -tis and -pjos is readily explained: the former is the ancient indefinite, the latter an early medieval innovation bringing the masculine/feminine interrogative and indefinite pronouns into line with the majority of other interrogative/indefinite pronouns and adverbs beginning with /p/ (though, curiously, the neuters ‘what?’ and (ká)ti ‘something’ were retained). Thus pjós ‘who?’ (‪<‬ AG poîos ‘what sort of?’) replaced the interrogative tís ‘who?’ just as the indefinite (ká)pjos ‘someone’ (‪<‬ AG poiós ‘some sort of’) replaced the indefinite tis ‘someone’, cf. pósos ‘how much?’ and (kám)posos ‘some amount/number’, póte ‘when?’ and (ká)pote ‘sometime(s)’, ‘where?’ and (ká)pu ‘somewhere’, pós ‘how?’ and (ká)pos ‘somehow’.

The element o/uka(n)- is more problematical. One standard account (advocated, for example, in the online Dictionary of Standard Modern Greek40) traces ka(n)- directly to the intensifier ká:n in composition with AG indefinite enclitics, and accounts for the general, though not total, loss of /n/ as due to the influence of a mis-segmentation of kanís in line with its syllabification as /kaˈnis/.41 The additional o-prefix is then explained as a generalisation of the o-element seen in indefinite relative pronouns and adverbs such as ópjos ‘whoever’, ópote ‘whenever’, etc. One problem with this approach is that AG ká:n, as we have seen, was associated with affective and later, negative, environments and it is therefore difficult to understand why the apparently near-synonymous tis and (h)eîs should have evolved in such different ways when they eventually came to be compounded with this element. Even more damaging is the fact it is difficult, perhaps ultimately impossible, to find convincing ancient examples of any relevant combination other than kà:n (h)eîs, whether in affective or positive environments (the common sequence ká:n tis means ‘even if anyone/ someone …’). This suggests that ká:n did not originally combine with enclitics at all. Furthermore, we might reasonably ask how convincing a model indefinite-generic relatives would have provided for the analogical extension of o- to indefinite but specific PPIs. In any case, this version of events does not deal with the distribution of u-, which, as we have seen, cannot simply be a ‘northern variant’. These issues must be addressed if a satisfactory account of kátis/kápjos is to be provided.

The syntactic string ouk án tis, comprising NM + modal particle + indefinite pronoun, together with an optative or past indicative verb form was common in pre-verbal position in main clauses from the earliest texts through to the classicising writers of later antiquity. Properly, án here modified the following verb to create a ‘potential’ meaning, with the whole expressing what one ‘could not do/could not have done’:

article image

‘… and all the other things that one cannot foresee at present’

Thucydides (c. 460-c. 395bc) 1.122.1
article image
article image

‘And no one young or old could/would say …’

Plato (428–347bc), Symposium 182b3
article image

‘And no one … could/would have recognised the place’

Josephus (37-c. 100ad), Jewish war 6.8

In the later post-classical period (c. 2nd/3rd c ad onwards), however, when contrastive vowel length had been lost in spoken Greek, along with the optative mood and potential án (now markers of ‘learned/written’ registers, albeit often used unclassically, cf. Horrocks forthcoming), many speakers may have regarded this string as a kind of compound and resegmented it as ou-kán-tis [NM-intensifier-indefinite] = ‘not-even-one’ ‪>‬ ‘not-any(one)’ (cf. the proposed origin and development of ka:neís/kanís above), thus creating a novel preverbal NQ to complement [ou(k)ka:neís/ kanís] just as preverbal oudeís/uðís complemented [ou(k) … tis] (cf. (26)).42 The different position of the accent compared with ka:neís/kanís now follows automatically: án followed by an enclitic was already accented in AG, as in (40). At this stage past indicatives, originally counterfactual, cf. (40c), would have been reinterpreted as simple negations of past events, and ou-kán-tis extended to the full range of affective environments associated with intensifying ká(:)n. The sentences in (41), for example, contain present and future indicatives, subjunctives and modal verbs, none of which originally combined with potential án:43


‘No one may/will understand the mind of the Lord’

Cyril of Alexandria (c. 376–444ad), Commentary on Isaiah, PG 70, p85.10
article image
article image

‘And none of the right thinking will deny this’

Cyril of Alexandria (c. 376–444ad), Thesaurus, PG 75, p64.36
article image
article image

‘… (husks) which no one can hold onto because of their smallness’

Orion (5c ad), Etymology: Alpha, p 7.12 (Orionis Etymologicon ed. Sturz)
article image
article image

‘No one audits the past unless he lacks interest in what he has in hand’

Photius (c. 810-c. 893ad), Epistle 281.22

If we suppose that oukántis/uká(n)tis had evolved into a vernacular rival to oudeís/uðís by the early middle ages, its apparently miraculous metamorphosis into a PPI in fact has a natural explanation in the context of the reinterpretation of NPIs proposed in 3.1 above. For as long as NPIs such as tinás and kanís were restricted to post-NM positions and interpreted as involving existential quantification under negation, NQs were understood as their synonymous pre-verbal counterparts = ‘not-any (X)’, i.e. ‘no X at all’ (cf. (26)). This treatment would at first have applied equally to the innovative pair uká(n)tis and u(k)/(u)ðen … kanís. But when NPIs were generalised to pre-NM positions and reinterpreted as involving wide scope universal quantification = ‘for any (random) X’, we might have expected uká(n)tis to become redundant in exactly the same way as uðís: a morphological compound comprising [NM + NPI] cannot be reanalysed to give a reading in which the NPI has scope over the negation because a lexical meaning ‘any-not’ (as opposed to ‘not-any’) is manifestly a nonsense. But while uðís was indeed dropped on this basis, ukátis was recycled to conform with the ‘universal’ interpretation of NPIs but crucially, still with narrow scope under the negation as required in a compound, so that uká(n)tis, originally = ‘not-any X (at all)’, came to mean ‘not-any (random) X’, i.e. entailing ‘some (particular) X’ (cf. fn. 21).

In this way NQ uká(n)tis could easily have become a PPI partner to NPI kanís, thus producing the split seen in MedG and MG and the final loss of double negation. If so, the positive use of quondam NQs as PPIs would have instigated generalisation to all positive environments and induced a rapid dissociation of u- from the notion of negation. This in turn might have led to a shift of u- to o-, perhaps on the analogy of other indefinites beginning with o- (though the potential difficulty noted above remains). But in either guise this element was now to all intents and purposes meaningless and it eventually disappeared, through a combination of aphaeresis (many unstressed initial vowels were lost in the middle ages) and the influence of the complementary but prefix-less kanís/kanénas (influence which might also explain the widespread loss of -n- in PPIs, as outlined above).

A probable chronology of events may conveniently be summarised here. The surviving AG data and the apparent impossibility of combining intensifying ká:n with an enclitic suggest that the first indefinite compound to emerge, already in post-classical antiquity, was ka(:)neís, which originally had the distribution of an API but eventually took on that of an NPI as well, finally being generalised in this role to pre-NM (and therefore preverbal) position in the early Middle Ages. But when the inherently preverbal AG sequence ouk án tis was first reanalysed in late antiquity as the NQ ou-kán-tis/u-ká(n-)tis, and then reinterpreted as a PPI in the early middle ages, the two items, originally synonymous and in complementary distribution, immediately became contrastive. A full set of PPIs was then modelled on uká(n)-tis, using the ancient indefinite enclitics as a base, and the same items were strengthened, through (normal) accentuation and/or suffixation, to provide a full set of NPIs to complement kanís. Tis/tinás, lacking the o/uka(n)- prefix, naturally fell in with the NPIs and adopted their distribution, eventually giving way in the early modern period to its rival kanís.

3.3. The Motivation for the Generalisation of NPIs to Pre-NM Position

The principal outstanding issue is why NPIs were generalised to pre-NM positions in the early medieval vernacular, thus forcing the reinterpretation discussed above. It is well known that in the post-classical period there were significant changes in the ordering of constituents. In particular, the possibility of informationally neutral verb-final order within VP was lost and verb-initial order became the rule, not only within VP but as a regular option within clauses, where VSO and SVO were both potentially neutral orders in informational terms.44

With the exception of subjects, therefore, late antique Greek, followed by MedG and MG, no longer allowed preverbal constituents that were neither topics (i.e. peripheral constituents with a scene-setting/resumptive role) nor foci (i.e. contrastive or emphatic constituents marked by heavy stress). We are concerned here only with focalisation. One option was simply to stress items in situ, but another was to combine emphatic/contrastive stress with displacement from their grammatically defined positions within VP to the left periphery of the clause containing them (see, among many others, Alexiadou and Anagnostopoulou 1998, Horrocks 1983, 1994, 2010: 108–9, 277–80, and Roussou and Tsimpli 2006 for various instantiations of this basic idea):45

article image

Even in classical Greek clausal constituents could be fronted for emphasis or contrast (see Dover 1960, and for general considerations of focus and word order in Greek and Latin, Devine and Stephens 1999, 2006), but for a combination of prosodic and semantic reasons both indefinite tis and NQ/NPI oudeís fell outside this pattern and could not be assimilated to it (cf. 2.1 and 2.2). Thus enclitic tis could not stand initially in a clause or be emphatically/contrastively accented; and in negative sentences its meaning changed according to whether it appeared before the NM (= ‘there is some X that did not …’) or after it (= ‘there is no X that did …’). Correspondingly, preverbal oudeís was a NQ but postverbal oudeís (after a NM) an emphatic NPI; and fronting over an NM again changed the meaning (‘there is no X that did …’ ‪>‬ ‘there is no X that did not …’).

The function of tis (i.e. PPI-like vs NPI-like) and oudeís (i.e. NQ vs NPI) in any given sentence was therefore determined by the presence or absence of a NM and, in the presence of an NM, by structural position with reference to it. But in a post-classical world in which the majority of native speakers were no longer from ‘Greece’ and Greek was widely learned as a second language, the fact that neither of these elements could be assimilated to the regular focalisation rule seems increasingly to have been perceived as confusing and problematical. By the early middle ages, therefore, steady pressure towards greater constructional and interpretational uniformity had led to the emergence of a system in which NQs were eliminated, the clitic status of certain indefinites was lost, and the entire class of indefinites was divided into contrasting sets of PPIs and NPIs (the latter licensed by NMs in all environments), both of which could now be focalised in the regular way.46 It was precisely this distributional assimilation of NPIs to the norm, including the possibility of focalisation, that first necessitated the semantic reinterpretation discussed in 3.2.

4. Conclusions

It has been argued that the complexities and anomalies of the AG system of negative indefinites (NQs and NPIs) eventually led, in the Early Middle Ages, to a wholesale systemic reconfiguration in which, inter alia, the distribution of properly licensed NPIs was extended to all syntactic positions available to DPs (i.e. that Greek shifted from being a language with non-strict NC to become one with strict NC). The ‘price’ paid for this simplification included the loss of NQs and double negation, and the forced reinterpretation of NPIs as elements involving universal quantification over negation (a property of strict NC languages generally).

The likely origin of the formal distinction between NPIs and PPIs has also been explored, and the innovative items kátis/kápjos and kanís have each been given a full, if partly reconstructed, history, with their beginnings located in recurring sequences of elements in AG that were eventually lexicalised. Interestingly, kátis/kapjos seems in origin to have been an NQ that was subsequently reinterpreted as a PPI at the time when NPIs were reinterpreted ‘universally’.

Finally, since the occurrence of NPIs in both pre-and post-NM positions makes NQs redundant, it is suggested that languages in which NPIs are understood as involving wide scope universal quantification over negation (i.e. languages with strict NC) are languages that also lack NQs. The proper testing of this claim is, of course, a matter for further research.


Alexiadou, Artemis and Elena Anagnostopoulou. 1998. Parametrizing AGR: Word Order, V-movement and EPP Checking. Natural Language and Linguistic Theory 16: 491–539.

Chatzopoulou, Katerina. 2012. Negation in Greek: a Diachronic Study. Unpublished PhD dissertation. University of Chicago.

Devine, Andrew and Laurence Stephens. 1999. Discontinuous Syntax: Hyperbaton in Greek. Oxford: Oxford University Press.

Devine, Andrew and Laurence Stephens. 2006. Latin Word Order: Structured Meaning and Information. Oxford: Oxford University Press.

Dover, Kenneth (1960), Greek Word Order. London: Cambridge University Press.

Giannakidou, Anastasia. 1998. Polarity Sensitivity as (Non)Veridical Dependency, Amsterdam× John Benjamins.

Giannakidou, Anastasia. 2000a. Negative … Concord? Natural Language and Linguistic Theory 18: 457–523.

Giannakidou, Anastasia. 2000b. Negative Concord and the Scope of Universals. Transactions of the Philological Society 98.1: 87–120.

Giannakidou, Anastasia, 2005. N-words and Negative Concord. In Martin Everaert, Henk van Riemsdijk, Rob Goedemans and Bart Hoolebrandse (eds.) The Blackwell Companion to Syntax, Vol III, ch. 45: 327–391. Chichester: Wiley-Blackwell.

Giannakidou, Anastasia, 2007. The landscape of EVEN. Natural Language and Linguistic Theory 25: 39–81.

Giannakidou, Anastasia. 2011. Negative and Positive Polarity Items: Variation, Licensing and Compositionality. In Klaus von Heusinger, Claudia Meienborn and Paul Portner (eds.) Semantics: an International Handbook of Natural Language Meaning Vol. II, ch. 64: 1660–1712. Berlin: De Gruyter Mouton.

Haspelmath, Martin. 1997. Indefinite Pronouns. Oxford: Oxford University Press.

Horn, Laurence. 2010. Negation and Polarity in the New Millennium: a Bibliography. In Laurence Horn (ed.) The Expression of Negation, 287–329. Berlin: Mouton de Gruyter.

Horn, Laurence and Yasuhiko Kato (eds.). 2000. Negation and Polarity: Syntactic and Semantic Perspectives. Oxford: Oxford University Press.

Horrocks, Geoffrey. 1983. The Order of Constituents in Modern Greek. In Gerald Gazdar, Ewan Klein and Geoffrey Pullum (eds.) Order, Concord and Constituency. Dordrecht: Foris. 95–112

Horrocks, Geoffrey. 1994. Subjects and Configurationality: Modern Greek Clause Structure. Journal of Linguistics 30.1: 81–109.

Horrocks, Geoffrey. 2010. Greek: a History of the Language and its Speakers (2nd edn.). Chichester: Wiley-Blackwell.

Horrocks, Geoffrey. Forthcoming. High Register Medieval Greek: ‘Diglossia’ and What Lay behind it. To appear in Caterina Carpinato and Olga Tribulato (eds.) Storia e Storie della Lingua Greca. Venice: Università Ca’ Foscari.

Israel, Michael. 2011. The Grammar of Polarity. Cambridge: Cambridge University Press.

Jannaris, Antonius. 1897. An Historical Greek Grammar Chiefly of the Attic Dialect. London: MacMillan. (Reprinted 1987, Hildesheim: Georg Olms).

Kiparsky, Paul and Cleo Condoravdi. 2006. Tracking Jespersen’s Cycle. In Mark Janse, Brian D. Joseph and Angela Ralli (eds.), Proceedings of the 2nd International Conference of Modern Greek Dialects and Linguistic Theory, 172–197. Patras: University of Patras. [http://www.stanford.edu/~kiparsky/Papers/lesvosnegation.pdf]

Klein, Jared. 2011. Negation and Polarity in the Greek, Gothic, Classical Armenian and Old Church Slavonic Gospels: a Preliminary Study. Oslo Studies in Language 3(3): 131–54. [= Eirik Welo (ed.) Indo-European Syntax and Pragmatics: a Contrastive Approach. Oslo: University of Oslo].

Newton, Brian. 1972. The Generative Interpretation of Dialect. Cambridge: Cambridge University Press.

Penka, Doris. 2010. Negative Indefinites (Oxford Studies in Theoretical Linguistics 32). Oxford: Oxford University Press.

Roussou, Anna and Ianthi-Maria Tsimpli. 2006. ‘On Greek VSO Again!’ Journal of Linguistics 42: 317–54.

Tsimpli, Ianthi-Maria and Anna Roussou. 1996. Negation and Polarity Items in Modern Greek. The Linguistic Review 13: 49–81.

Werle, Adam. 2002. A Typology of Negative Indefinites. In Mary Andronis, Erin Debenport, Anne Pycha, and Keiko Yoshimura (eds.) Papers from the 38th Meeting of the Chicago Linguistic Society Vol. 2: The Panels, 127–143. Chicago: Chicago Linguistic Society.

Willmott, Joanna. 2013. Negation in the History of Greek. In David Willis, Anne Breitbarth and Christopher Lucas (eds.), The History of Negation in the Languages of Europe and the Mediterranean, Vol. 1, 299–339. Oxford: Oxford University Press.

Zeijlstra, Hedde. 2006. The Ban on True Negative Imperatives. In Olivier Bonami and Patricia Cabredo Hofherr (eds.), Empirical Issues in Syntax and Semantics 6. http://www.cssp.cnrs.fr/eiss6/. 405–424.

Ancient Greek ouk ísmen oudén = [not know-1pl nothing], lit. ‘we don’t know nothing’.

My grateful thanks to Julián Méndez Dosuna and Marjolijne Jansen, both of whom read this article in draft and saved me from myself on numerous occasions. For better or worse inherent stubbornness has stopped me taking their advice in one or two places; any residual errors and deficiencies are, of course, my own responsibility. Thanks are also due to two anonymous readers for JGL, who made invaluable suggestions for improvements of both content and layout.

Oudeís ‘no one’ is a compound of oudé ‘not even’ + heîs ‘one’ (masculine), and is almost exclusively singular.

Most negative sentences below contain the NMs ou(k) (AG)/u(k) (MedG) or (u)ðen (MedG/MG) [= NEG-1], though a few have the alternative NM : (AG)/mi(n) (MedG/MG) [= NEG-2], which is characteristic of ‘non-veridical’ contexts (see the discussion of (5), and Chatzopoulou (2012)). For the purposes of this article, the choice of NEG-1 or NEG-2 is immaterial.

As noted, this sentence with ou(k) added is in fact grammatical, but only on a double negative reading (though see 2.2 below for some qualification).

Other than as a residue from AG, with limited uses (for Medieval Greek see 2.2).

These are distinct only in the nominative (the forms are partly interchangeable), and are treated here as a single item. The component -is/-énas is again the numeral ‘one’ (is is the modern pronunciation of the AG masculine form heís, énas a medieval innovation): kanís/kanénas has only a singular paradigm. Note that when used with a NM, as here, these and other related items may be emphatically stressed, while in non-negative environments they are always unstressed, cf. Giannakidou 1998 and subsequent work.

As also in formal/academic French, cf. (9) below.

Tis has both singular and plural paradigms. As an enclitic it forms a phonetic word with its host, does not appear clause-initially or after a pause and is not normally accented (though disyllabic forms receive a secondary accent on their final syllable to accommodate cases that would otherwise break the rule that an accent must fall on one of the last three syllables). There are, however, a few examples used contrastively at the beginning of a clause, presumably, via a natural semantic extension, as an indefinite quantifier, a role in which it is accented in its own right (cf. (16a) below).

Kápjos has both singular and plural forms, whereas kanís/kanénas, as noted, has a singular paradigm only.

It is also worth noting that many languages with inherently negative pronouns/specifiers fail to distinguish between ‘some’ and ‘any’, and use the negative elements without a supporting ‘not’ both pre- and post-verbally, e.g. Germanic other than English.

A distinction is commonly drawn in the literature (e.g. Giannakidou 1998 and subsequent work) between strict and non-strict NC, the former requiring the co-presence of a NM with negative pronouns and adverbs in all environments, the latter with such items only in a subset of environments.

An NC reading as opposed to a double negative reading of (7a) is acceptable in many colloquial varieties (= ‘John didn’t see anything’), though this is impossible in standard English.

Italian treats post-verbal subjects in the same way as objects: contrast nessuno (*non) è venuto [no-one (*not) be-3sg come-pple] with *(non) è venuto nessuno [*(not) be-3sg come-pple no-one].

In colloquial French, by contrast, the NM ne has been largely abandoned, with the result that its formal partner pas has developed into the regular/dominant NM (cf. je ne sais pas ‪>‬ je sais pas ‘I don’t know’). In very informal varieties former NPIs such as personne have become NQs (cf. j’ai vu personne ‘I saw no one’), which has in turn led to double negative readings alongside NC ones in cases such as personne (n’) a vu rien = ‘no one saw nothing’/‘no one saw anything’.

(Standard) French personne, for example, appears to have a non-negative form while Italian nessuno a negative one, but both have functions as NPIs.

NQs might, for example, be treated as NPIs licensed by ‘covert’ as opposed to overt negation, as argued in Penka 2010.

Note that (unstressed) nessuno may sometimes appear in affective environments where a negative meaning is not a prerequisite: ha arrivato nessuno? [have-3sg arrived anyone], = ‘has anyone arrived?’ This is not the case for oudeís, however.

(Standard) French and MG happen to have ‘non-negative’ NPIs but this is not necessarily the case in strict NC languages. Russian, for example, has clearly ‘negative’ NPIs in all contexts: nichevo ne proizoshlo [nothing not happened], on nichevo ne zdelal [he nothing not did], vs on ne zdélal nichevó [he not did nothing]. Correspondingly, English and AG show that languages with non-strict NC may also have NPIs of either type.

The discussion here relates to NPIs only. The ‘free-choice’ use of any (cf. any man will do/won’t do, etc) is a different matter, usually treated as involving a form of universal quantification.

By contrast PPIs in English involve wide-scope existential quantification over negation (if present) and the meaning is quite different, viz. ‘there is some X such that it is not the case that [X has role R in event E]’, i.e. at least one X did not have role R in event P, but other X’s (probably) did. Contrast I didn’t read any of the books with I didn’t read some of the books.

Hence the more general association of NPIs with non-veridical/affective environments, cf. also the discussion of (5) above.

By contrast, when negation has wide scope, ‘not … any …’ is normally understood to involve existential quantification, as described above. But this combination may also involve universal quantification, and be taken to mean that the relevant proposition is true not of a random individual of the specified kind but (by implication) of some specific one. Thus, subject to the necessary intonational requirements, she didn’t meet anyone may mean either ‘she met no one’ or ‘she didn’t meet any random person’ (implying that she did meet someone specific). The importance of this for Greek will become apparent in 3.2.

I am grateful to Marjoijine Jansen for her wide-ranging help with the Medieval Greek data.

Tis and tinás are both nominative singular forms: the former reflects the ancient language directly, the latter is a medieval innovation. As we have already seen, texts from the earlier periods of Greek attest a single indefinite enclitic tis with an ‘existential’ reading independently of the presence or absence of negation. We may, however, reasonably say that its role is ‘PPI-like’ or ‘NPI-like’ according to whether or not there is a licensing negative and if so, whether the existential quantification falls inside (= NPI-like) or outside (= PPI-like) the scope of the negation. In what follows terms such as ‘PPI reading’ and ‘NPI reading’ are used on this understanding, and there is no associated implication that there were two distinct but homophonous lexical items, one a PPI the other an NPI.

The registers used for medieval Greek writing were largely genre-determined: belletristic and official (imperial and ecclesiastical) texts required styles that affected a classical appearance in lexicon, morphology and (to some extent) syntax, while popular forms of Christian writing, low-level documentation and certain poetic genres (satire periodically and romance more generally) allowed the use of more vernacular-looking varieties. The focus here is necessarily on the latter since these texts offer a more realistic, though still partial, view of the spoken realities ‘on the ground’.

Extrapolating from the corresponding MG practice, and assuming that the new (non-clitic) distribution was associated with a normal range of stress options.

Since AG has only the single indefinite tis, and since there is no semantic difference between PPIs and NPIs when the latter are given the ‘existential’ reading (12), it seems reasonable to argue that AG was rather like modern Germanic (other than English), with ‘positive’ and ‘negative’ readings of tis associated automatically with its position (pre- or post-NM position, cf. (15) and (16)).

This is one of the rare examples where tis, being both clause-initial and contrastive, is accented normally (cf. fn 8).

The first is l. 5348: Enéas tiná eskótose [Aeneas-NOM some-ACC killed-3sg], ‘Aeneas killed someone’.

Uðís is the MedG/MG pronunciation of AG oudeís.

The construction illustrated in (20c) is in fact very rare and the usual expression of double negation is oudeís hóstis ou(k) [no-one who not], ‘(there is) no one who (does) not …’

Interestingly, when oudeís is associated with following negatives other than the NM ou(k) (or :), the latter are always treated as NPIs, e.g. Plato, Philebus 65e5: phróne:sin kaì noûn …oudeìs pó:pote oúthhúpar oút’ ónar aiskhròn oúte eîden oúte epenóe:sen oudamê:i oudamô:s oúte gignómenon oúte ónta oúte esómenon [wisdom-ACC and mind-ACC no-one-NOM ever neither awake nor asleep disgraceful-ACC neither saw nor thought-of nowhere in-no-way neither becoming-ACC nor being-ACC nor about-to-be-ACC], ‘No one ever, either asleep or awake, either saw or thought of wisdom and mind as in any way (or) by any means becoming or being or about to be unseemly.’

Thus in (21a) ‘planning in good faith (tê:i pístei)’ is contrasted with ‘delivering in practice (tô:i érgo:i)’ by means of a chiastic order created by the fronting of enthumeîtai: [plan in-good-faith] X [in-practice deliver]; in (21c) the particle mén is inherently focalising, its purpose being to contrast what precedes it with something in the next clause (also focalised and immediately followed by the particle ‘but’).

If this and similar cases really are cases of postverbal oudeís used as a NQ, the option may have been deliberately chosen for stylistic effect as reflecting an older phase of the language in which oudeís etc were invariably NQs.

The reader should be warned that the numbers for the War of Troy, based on the edition of Papathomopoulos as used by the TLG, are highly questionable, since the editor sometimes prints oudeís where the MSS have tinás. This observation does not detract from the overall argument, and in fact enhances it, as noted in the text.

I am indebted to Julián Méndez Dosuna for pointing out the best approach to understanding ká:n, and for some of the supporting data below.

The use of kan ‘even’ in negative environments survives into MG, see e.g. Giannakidou 2007 for discussion.

Unlike NPI oudeís, however, kanís was never restricted to negative contexts and so continued to be used in the full range of affective environments once occupied by tis.

No satisfactory explanation of this ‘prothetic’ o- (or u-) has yet been proposed. In the case of o- appeals are sometimes made to the analogy of ‘indefinite’ relative pronouns and adverbs which also begin with o- (cf. ópjos ‘whoever’, ópote ‘whenever’, etc), but these are generic (= ‘any X that …’) and so rather unlike the PPIs under discussion, which typically mean ‘a certain/some particular X’, etc.

The two elements are graphically separated in the ms.


Only (u-/o-)kámposos ‘some amount/number’ regularly shows kam- (with assimilation of the nasal), though (u-/o-)kámpote ‘sometime(s)’ is also occasionally attested.

The fact that no compelling contemporary written evidence survives for the assumed compound status or the postulated resegmentation is unsurprising. The formal education presupposed by literacy reinforced both the traditional spelling and traditional usage involving optatives etc.

In all these examples the postulated (o)u-kán-tis is actually written ouk án tis, see fn 33.

In terms of the diagram in (42), assuming that V raises to Infl(ection), these two orders might be accounted for on the basis that, under specified conditions, subject DPs either remain in Spec VP or raise to Spec IP.

There are many different interpretations of the data and (42) is intended merely as a non-committal sketch of the properties of Greek sentences from late antiquity onwards.

Though the persistence of non-uniform systems cross-linguistically, as in Italian, shows that complexity per se is not enough to guarantee the onset of regularising change.


The first is l. 5348: Enéastináeskótose [Aeneas-NOM some-ACC killed-3sg], ‘Aeneas killed someone’.