We begin with the most straightforward difference between the two types of particles, already highlighted in Section 1: one type follows and the other precedes its host. Hungarian orthography mandates a comma between the members of the iterations. The commas will help structure the examples for the reader, so they are retained.

When they occur in subject position, all of these trigger singular agreement on the verb, which is generally the case even with simple conjunctions in Hungarian:

The bluntest sign of a significant syntactic difference is that reiteration is optional in one type but mandatory in the other. Compare is in (9a) with mind in (10a). Kati is ‘Kati too’ by itself is perfect in both preverbal and postverbal position (and similarly for Kati sem ‘Kati nor’), whereas mind Kati ‘all Kati’ by itself is entirely unacceptable in both preverbal and postverbal position. It must come in tuples (pairs, triples, etc.).

The same is true of vagy, sem, and akár (10b,c,d), although confounds from what I believe represent different constructions may make the strings acceptable when particle+host occurs in the preverbal position (the postverbal versions are inescapably bad). See the footnotes; note 2 pertains to both (10b) and (10c). For the purposes of this paper, all the examples in (10) are ungrammatical.

These contrasts unambiguously point to a global difference between the two constructions, postulated in (1) and (3). Even though the particles that follow their focused hosts (is, sem) participate in “reiterated” constructions, those arise from the coordination of independent and self-sufficient propositions. In contrast, in the case of the particles that precede their hosts (mind, vagy, sem, and akár), grammar must ensure that the hosts form a tuple.

Reiterated constructions of both types optionally contain what I will call “connectives” – but different connectives, és vs. pedig. És is the default Junction head corresponding to and (Szabolcsi 2015).4 It optionally occurs in iterations with particles that follow the host, confirming that those iterations are vanilla conjunctions. The optionality of és is due to the fact that Hungarian generally allows asyndetic (connectiveless) conjunctions. Pedig requires a more elaborate introduction, see below.

We first demonstrate that és and pedig are not interchangeable and do not combine. (11)–(12) illustrate the lack of interchangeability and combinability with particle sem, whose two versions offer minimal pairs. Afterwards only the grammatical examples are given, see (13)–(14). Notice that unlike és, pedig intervenes between the particle and the host.

To my knowledge, the connective pedig has not been discussed in the modern literature on Hungarian (although data of the type of (14) are cited in Lipták 2001).

Pedig has what seem to be two distinct versions.5 One version, which may be called adversative, occurs in clause-initial position and can be translated as ‘even though’ or ‘despite the fact’. The present paper will not be concerned with this version.

The pedig that interests us follows the topic in the last member of a propositional coordination or list, and its contribution can be approximated as ‘on the other hand’ or ‘however’ or ‘lastly’. It marks the last member of a set of two or more partial answers to an overt or covert question, and thus indicates that the answer is now complete. Pedig is an entirely opaque element. It will be glossed as PEDIG, for lack of a non-misleading option.

In the reiterated particle construction, pedig occurs in the second (or last) member of the iteration, intervening (as noted above) between the particle and its host. Its interpretation is very much in the same spirit as in its sentential use illustrated in (16)–(17).6 For example,

The interpretation of pedig jibes with the fact that it occurs in that construction which always comes in tuples. The tuple as a whole provides an answer to an overt or implicit question under discussion, enumerating propositions that serve as partial answers. If pedig is present, the list is complete. The particle mind ‘all’ tells us that all the propositions in the tuple are true; vagy ‘or’ tells us that one of them is true. In the next subsection we will see this even more vividly.

Hungarian forms its quantifier words by combining particles with indeterminate (wh-) pronouns. They are built with the particles that are confined to tuples. Those particles precede their hosts both in the reiterated construction and in combination with indeterminate pronouns. Vala- ‘some’ is an allomorph of vagy ‘or’.

Tying these together with the findings of the previous subsection, compare:9

In contrast, the particle is that follows its host does not form quantifier words, in either order, see (21). The NC particle sem of course does participate in quantifier words but, given the double life it leads, there is every reason to believe that sen-ki, sem-mi, se-hol are simply instances of (19d).

Is participates in the composition of NPIs and FCIs, see (22), but in that case it combines with a full quantifier word, which is a different matter (see Szabolcsi 2017).

The data just reviewed converge with the observations above. Mind-type particles must be part of a reiterated construction that is semantically equivalent to a quantifier, and they also form actual quantifier words with the indeterminate pronouns that correspond to the same domains as their hosts. Is-type particles on the other hand do not have to be part of a reiterated construction; when they are, those are “accidental” coordinations. They do not build quantifier words.

At this point we have evidence for the overall structures of the two constructions and for the quantificational role of mind-type particles; less so for the nature of is and the sem that patterns with is (follows its host). Negative concord (NC) provides some fine-grained evidence about those syntactic details.

The negative concord particle sem (more colloquially: se) occurs in both constructions under investigation but, interestingly, builds different kinds of NCIs in the two cases.

Hungarian is a negative concord language that is usually thought to be of the strict type, where the sentential negation marker nem ‘not’ is always mandatory, as in Russian. But Surányi (2006) showed that Hungarian NC is a hybrid: there is a set of NC-expressions that do not co-occur with nem when they are in preverbal position, mimicking non-strict NC languages like Italian. Consider the n-expressions senki ‘n-one’ and senki sem ‘n-one nor’ in preverbal position (which distinguishes strict vs. non-strict negative concord) and in postverbal position (which does not distinguish them). We see that senki has the distribution of a strict NCI like Russian nikto, whereas senki sem has the distribution of a non-strict NCI like Italian nessuno. All three examples below mean ‘No one slept’.

I observe that the two types of reiterated sem construction differ in that the “particle precedes host” one builds a strict NC expression, and the “particle follows host” one builds a non-strict one.

In other words,

Notice that in the strict NC version, the particle sem precedes its lexical hosts as well as the indeterminate pronoun ki ‘who, viz. human’ that it combines with, whereas in the non-strict NC version sem follows its lexical hosts as well as the n-word senki that it combines with.

Szabolcsi’s (2016) proposes a unified analysis for the hybrid negative concord facts involving senki and senki sem. It will be briefly reviewed and extended to the new data in Section 5. At this point, we note that Szabolcsi treats the sem in the non-strict NC construction as an overt counterpart of Chierchia’s (2013) NEG, a functional head that invokes an abstract negation above its projection. The account of the contrast in (29)–(30) converges with the overall syntactic analyses for the two constructions.

The preceding subsections substantiated the claim that the host + particle and particle + host reiterations are syntactically distinct, over and beyond linear order. Reiterations of host+particle represent the coordination of two or more independent propositions, with or without ellipsis. Since each of the propositions can stand on its own, reiteration is not mandatory. This contrasts with the case of particle+host, where reiteration is mandatory, and the construction can be seen as providing a set of partial answers to a question. The quantifier particles indicate whether each, some, or none of those partial answers is true; addition of pedig to the last junct marks the set as constituting a complete answer. Finally, with the particle sem the first type constitutes non-strict and the second, strict negative concord expressions. We anticipated that this squared with the claim that the particles following their hosts are heads on the clausal spine, whereas the particles preceding their hosts spell out a (propositional) quantifier.

Having presented and compared the basic data of the two constructions, Sections 3 and 4 zoom in on each construction in turn. Negative concord is relegated to Section 5.

Section 2 enumerated descriptive arguments to the effect that the structure of mind_mind, vagy_vagy, akár_akár and (strict NC) sem_sem constructions is different from that of is_is and (non-strict NC) sem_sem ones. The juncts obligatorily come in tuples, each junct bears the same particle, and the connective és ‘and’ cannot be added. (Pedig might be a J head, with the initial quantifier particle adjoining to it, or it may be internal to the last junct. This paper uses pedig primarily as a diagnostic tool and does not attempt to settle its syntax.)

I argue that a JP structure is involved but, unlike in (33), it is the complement of an unpronounced but contentful quantifier that is overtly realized in the form of identical particles on each of the juncts. The syntactic relation between Q and the particles can be seen as feature-checking, as indicated in (35), which repeats (3).

Feature checking is a standard tool, and it captures the idea that Q is contentful, whereas the particles merely signal its presence. Feature checking ensures that each and every junct carries the appropriate particle if it is subject to an ATB-style condition.15

An alternative tool might be borrowed from Case concord, discussed in Bayırlı (2017), based on Pesetsky (2013). Case concord is specifically pertinent, because Case originates outside DP, not on the Num head (like plural) or on the N head (gender). The realization of the Q-feature can be seen to spread to the coordinates in the manner of Feature Assignment, subject to the intervention of phasal domains:

Imitating Bayırlı’s notation, we may instantiate mind_mind constructions as follows, with the arc indicating a domain that is not penetrated by the copied particles. Read S as a variable over appropriate categories.

This paper will simply work on the assumption that some well-attested syntactic mechanism can account for the realization of the Q feature in the form of particles on each of the juncts.

Den Dikken (2006) employs JP for either_or, neither_nor, and both_and structures. He does not specify what semantic contribution, if any, the J head makes, but his use of J in these varied cases indicates that it is meant to be semantically neutral. To make that idea precise, Szabolcsi (2015) couples JP with a proposal by Winter (1995). On that view, J merely forms an ordered tuple of the juncts. It is neither a conjunction, nor a disjunction. The tuple is converted to a set and, by default, an unpronounced operation interprets it as a conjunction, just as an unmarked sequence of sentences in a text is interpreted conjunctively. Additive (MO-)particles reinforce the conjunctive interpretation; in their presence, it is not merely a default. This is what happens in (33) above. But other particles on the juncts may override the default. Szabolcsi (2015) discusses KA-particles that invoke an unpronounced disjunctive operation. She argues that disjunctions are overtly marked, precisely because the conjunctive default needs to be overridden.

In what follows, I will likewise assume that J itself does not make JP a conjunction or a disjunction. It merely forms an ordered tuple, or set, of the juncts. What semantic operation is performed is determined by the particles on the juncts (i.e. by the unpronounced Q whose presence they signal).

Recall from Section 2.4 that the same particles that appear in (35) also form quantifier words with indeterminate pronouns, yielding the parallelisms below, and similarly for akár and sem:16

The parallelism recalls an insight from Kratzer & Shimoyama (2002) and Kratzer (2005), presented for quantifier words that are built in the same way as mindenki and its brothers.

The propositional quantifiers referred to above are defined as follows:

Kratzer and Shimoyama apply (41) in the interpretation of sentences with quantifier words. There, A is the set of propositional alternatives generated off of indeterminate pronouns (who) in Hamblin-style alternative semantics: the set of propositions such that, for any human there is, the proposition that this human sleeps is in the set.

In our case, A is the set of propositions that JP assembles from the juncts in it, as described in 4.2 above. Hamblin semantics can be used but does not play any particular role.17

For example, with particle mind ‘all’ attached to the juncts, the matching Q is ∀, and the result is the proposition that is true in all worlds where each of Kati sleeps, Mari sleeps, and Peti sleeps is true.

Note that at the end of the day the following two sentences have the same truth conditions:

The reason is that “every proposition in {p, q, r} is true” is logically equivalent to “p is true, and q is true, and r is true”. The overall meaning of these sentences leaves us in the dark as to how that meaning might be composed. Only detailed analysis can help. This is especially important to bear in mind if one investigates a language in which, for some reason or other, even linear order does not distinguish between the two types of particles.

We have arrived at propositional quantification from two directions. On the one hand, Kratzer & Shimoyama use it for the equivalents of Someone/Everyone sleeps, for reasons related to Hamblin semantics. (In this paper, we are not specifically concerned with quantifier words.) On the other hand, the ellipsis-free versions of particle+host reiterations are straightforward representatives of propositional quantification.

In other examples, the particles seemingly attach to smaller constituents (Mind Kati, mind Mari alszik, etc.). Starting with the introduction, I suggested that these are obtained by the optional ellipsis of shared material. The examples below provide further evidence that ellipsis or structure-sharing must be involved in some of the examples where the juncts are not complete propositions. I first illustrate the phenomenon with comparable examples from English:

These examples bring to mind Right-Node Raising, which is often handled by movement. But neither bottle(s) in (46), nor -handed child(ren) in (47) could be shared by the two juncts as a result of rightward across-the-board extraction. I thank A. Lipták (p.c.) for directing me to work by V. Valmala that solves the problem. Using Right Node Raising as a pre-theoretical cover term, Valmala (2012; 2013) distinguishes focal-pivot right-node raising and non-focal pivot right-node raising, and argues that the confusingly mixed properties of RNR can be neatly sorted out once this distinction is made. Valmala proposes a set of diagnostics; see his work for details. Below I give a bird’s eye review.

Focal-pivot RNR obtains when the shared string (the pivot) has focus accent. Focal pivots are always extractable expressions. For example:

Non-focal pivot RNR obtains when the last element before the pivot is focused and the pivot crucially is not. In this case the pivot may or may not be an extractable expression; non-extractable ones can only participate in this latter kind of RNR. (46) and (47) would be examples. Valmala argues that here the gap is not a result of movement; instead, interpretation is in-situ. He considers two possible mechanisms: ellipsis or structure-sharing (multidominance), without committing to a choice. I will not not address this theoretical issue, and simply use the label ellipsis.

As A. Lipták (p.c.) observes, Hungarian never moves constituents to the right to be assigned stress, even pretheoretically and non-anti-symmetrically speaking, and so it does not exhibit focal-pivot RNR, although it allows non-focal-pivot RNR of the same string:

Turning to the quantifier particle constructions, Hungarian exhibits a huge variety that exemplify the non-focal pivot RNR case. Verb-initial (50b), (51b) and (52) plausibly involve leftward across-the-board extraction of the verb, and (51b), the subsequent merger of negation.

These examples most likely derive from propositional coordination and quantification. The question arises whether the simpler ones involve DP-level (generalized quantifier) coordination and quantification, e.g.:

At least two works have recently called into question the availability of phrase-level coordination, arguing for conjunction reduction: Schein (2017) and Hirsch (2017). Schein’s arguments primarily come from plurals and event semantics. Hirsch investigates semantically simpler cases and proposes that syntactic well-formedness and the availability of certain scope relations are best understood if coordinate structures are derived from vP-coordination, coupled with well-attested mechanisms for ellipsis, e.g. gapping. Importantly, he notes that as soon as generalized quantifiers are coordinated, semantically speaking we have propositional (type t) coordination (though not necessarily full-clausal coordination). Therefore, in the domain of data that he investigates, the general unavailability of constituent coordination must be a syntactic matter.

The same argument extends to the question of (53). If the formation of the generalized- quantifier level JP and QP is not possible even for such an innocent-looking case, that probably has to do with what syntactic mechanisms are available. At this point I cannot pinpoint any syntactically disastrous effects that would emerge if (53b) were allowed to co-exist with the various possibilities for ellipsis. However, the data discussed in this paper will offer new grounds for hypothesis testing.

This situation calls for further research directed at the division of labor in this area of the syntax/semantics interface. What is the best way to strike the theoretical balance between the syntactic differences and the semantic similarities, within one language and across languages? Is it justified to generalize over the clausal head and the quantifier-internal versions in languages where the same particle morphemes show up in both roles? Are the reiterated constructions actually syntactically ambiguous in some of those languages, even if there is no difference in linear order that might draw attention to the possibility of a structural ambiguity? How to deal with this overarching question in languages that are largely similar to Malayalam, Japanese, and Hungarian in the pertinent respects, but quantifier words are generally not built from wh-pronouns and independently active particles?

Below I illustrate the interest of these questions with reference to Persian and Turkish on the one hand, and Telugu and Japanese on the other. This is followed in Section 6.2 by a survey of data from a broader range of languages that have both types of particle construction.

The possibility of syntactic ambiguities is highlighted by the sem data discussed in the foregoing sections. The right-hand column of (66) recaps how sem syntactically parallels both mind and is:

Now consider Persian in the right-hand column of (67) (A. Kahnemuyipour, p.c.):

(67b, c, d) make it plausible that Persian ham plays the syntactic roles of both Hungarian is and Hungarian mind, even though there is a gap in (67a): Hungarian mind builds mindenki ‘everyone’ (and serves as the floating quantifier mind ‘all’), whereas Persian ham does neither.

Turkish, in turn, borrows herkes and hem X hem Y from Persian (B. Kamali, p.c.), but has dA for a clausal head (Kamali & Karvovskaya 2013):

Again, the syntactic parallelisms in (68b, c, d) seem solid.

Since Persian ham X and Turkish hem X are unacceptable outside tuples (in contrast to X ham and X dA), the ham/hem that occur in ham/hem X ham/hem Y are probably not counterparts of Hungarian is ‘too’. Given the discrepancies in (67a)–(68a), it would be rash at this point to jump to the conclusion that they are synonymous with Hungarian mind ‘all’. This is why the solid parallelisms are qualified as syntactic. However, although Persian has no quantifier word *ham-ki ‘everyone’, the noun hame ‘all’ forms universal quantifier phrases with count and mass nouns and serves as a floating quantifier (Toosarvandani & Nasser 2017: 666, 683–4, 690).19 This is quite similar to the behavior of mind(en), illustrated in fn. 8. If ham and hame are at least historically related, then it is possible that the ham that precedes its host more generally parallels mind(en), in contrast to the ham that follows its host and parallels is. Naturally, the gaps with quantifier words require a systematic explanation, and the compositional interpretation needs to be developed.

Now consider the relation between clausal head particles and the universal quantifier particles from another angle. In Sections 2 and 3, it was observed that Hungarian is ‘too’ does not build quantifier words with wh-pronouns (*is-ki, *ki-is); this was taken to be one argument for its clausal head, as opposed to QP-internal, character. The same holds for counterparts in Russian, for example (see 6.2), but not in Japanese (Shimoyama 2011) or Telugu (Balusu 2017):

How can the Japanese and Telugu data be made sense of now, given our syntactic vigilance?

Balusu (2017) proposes that the Telugu particle -VV (harmonizing long vowel) primarily forms polarity sensitive expressions with wh-pronouns, and the availability of the universal reading is a result of subsequent exhaustification (limited to eppuD-uu in Telugu). If the same analysis can be extended to Japanese and the generally available universal reading for wh+mo is obtained by exhaustification, then it may be possible to maintain that both -VV and -mo are heads on the clausal spine that have the ability to accommodate wh-pronouns in their specifiers. That would in turn provide a link to the analysis of the Hungarian string senki sem, analyzed with senki in the specifier of the clausal head sem; see the discussion of (55) in Section 5:

Similarly, Hungarian is accommodates existential pronouns in its specifier, with the effect of creating weak negative polarity items, cf. (22), repeated below:

It remains an open question why Telugu -VV and Japanese -mo combine with bare wh-pronouns and Hungarian is/sem with ones that have (uninterpreted) existential particles. (Note that senki sem and valaki/akárki is are strings, not single “quantifier words”.)

Szabolcsi (2017) proposes a semantics for Hungarian is/sem that subsumes the additive, scalar, negative polarity and free choice readings. In brief, is/sem seeks out a set of alternatives induced by its host and activates them. These may be focus alternatives or subdomain/scalar alternatives. Activation in the sense of Chierchia (2013) entails that the alternatives must be incorporated into the meaning of the sentence by way of some strengthening (e.g. exhaustification) mechanism, typically with the assistance of further, overt or covert operators. According to Szabolcsi (2017), English too and either fall under the same generalizations, although they apparently specialize in working with focus-alternatives. This is in the same spirit as Balusu (2017), although the latter only investigated combinations with wh-pronouns.

With this last ingredient added and generalized cross-linguistically, it appears that the apparently contradictory properties of -mo and -VV can be reconciled.

The present paper cannot undertake a thorough descriptive and theoretical investigation of the questions raised above, but as a starting point, this subsection presents the results of a small cross-linguistic survey. The data establish the prevalence of two distinct constructions, even though cross-linguistically, the distinction does not track the relative order of the particles and their hosts.

The data were solicited from semanticists and syntacticians.20 The survey did not ask them to commit to syntactic or semantic analyses, although sometimes I was able to rely on their closely related publications; see especially Kobuchi-Philip (2009); Slade (2011); Kamali & Karvovskaya (2013); Balusu (2017); Esipova (2017); Jeretič (2017). The data typically came in the form of full sentences, but the summary below distills them into schemas. We did not try to track down all the pertinent constructions in the given languages, so the important distinction is between one or more representative, or no representative, of each type.

Each of Bosnian, Japanese, Mandarin, Persian, Russian, Modern Colloquial Sinhala, Telugu, and Turkish appears to have a counterpart of Hungarian is_is ‘too_too’. Some particles have restrictions on the size of the propositional (i.e. type t) hosts they combine with.21 Whether the particle precedes or follows its host is not cross-linguistically consistent and thus not diagnostic. Instead, the diagnostic properties are the following:

Consider the following schematic data in Table 1; comments follow below. Three juncts are displayed to bear out one critical difference between the two types of reiterated constructions. (Telugu -VV and Turkish -dA exhibit vowel harmony.)

In multiple languages, no optional connective is possible. This does not entirely correlate with whether ‘too’ is homophonous with ‘and’ in the table. Moreover, Mitrović & Sauerland (2016: 482) cite Macedonian [i Roska] i [i Ivan] ‘both Roska and Ivan’.

Whether the optional connective (if one exists) may appear between all the juncts is a somewhat delicate matter. Notice that in English, John and Mary and Bill is grammatical but dispreferred in comparison with John, Mary and Bill. For the same reason, the single and may indicate the completion of the list, but that is probably a Gricean inference. The Japanese and Bosnian data can be interpreted in the same way, but caution is in order.

Almost all the languages in the sample have one or more ‘either_or’ constructions that seem to exemplify quantifier-phrase internal particles. As above, whether the particle precedes or follows its host is not cross-linguistically consistent. The diagnostic properties are the following:

Consider the following schematic data in Table 2; comments follow below. The optional connective often intervenes between the particle and its host, indicated by a caret (^) in the column for junct Z.22

Usually, medial ‘or’ is inclusive and reiterated ‘or_or’ is exclusive/exhaustive, but there are two exceptions that show that the correlation is not necessary. Sinhala X-hari Y-hari and Telugu X-oo Y-oo are inclusive; Telugu X leeda Y ‘X if-not Y’ is exclusive/exhaustive.

Turkish and Persian in general do not build quantifier words from particles and wh-pronouns, as noted above; in Bosnian and Russian, only some particles fail to participate.

Note that Turkish dA leads a double life as an additive clausal head (Table 1) and as a “contrastive” connective (Table 2), calling for a semantic or pragmatic unification.

There are further examples of the “quantifier-phrase internal” construction, although in fewer languages in my sample. Syntactically, they come from the same mold as the disjunctions in Table 2: tuples are necessary; the optional connective is the same “contrastive” item that occurs in Table 2 and indicates the completion of the list; quantifier-word formation is possible.

At least Russian, Turkish, and Hungarian have two “alternation” constructions, whose semantics is analyzed in Esipova (2016; 2017).23 The first one involves particles, see Table 3. The second one involves wh-words. Lipták (2001; p.c.) suggests that it might be possible to bring them into the fold by postulating null particles; I will not pursue an analysis here. But notice the same general pattern. See Table 4.

Table 5 recaps the discussion of Persian ham_ham and Turkish hem_hem in Section 6.1.

French and other Romance languages have reiterating constructions that may represent the two types, but the analysis is not straightforward. See Table 6.

Finally, we turn to negative concord, so far as it involves specialized particles and/or optional connectives. Recall that Hungarian sem has both clausal head and quantifier-phrase internal versions, each of which solidly patterns with the other representatives of its type. Given this duality, it is not surprising that both types seem to be represented in the data, based on the optional particles, for example. But the properties are fairly mixed, and so it would be rash to attempt labels at this point. See Table 7.

Seeing how difficult ‘nor_nor’ type negative concord is to analyze in many languages, it is unsurprising but at the same time somewhat disappointing that our diagnostics do not straightforwardly point to one analysis or another. But they probably provide some useful starting points.