The article addresses empirical and methodological issues that are of central concern for an assessment of uninflected function words doing services in clause-combining and/or in indicating the speaker's stance toward illocutionary force or propositional content. Such units have been variably treated: either just as 'particles', as subordinating conjunctions or complementizers, or as auxiliaries of 'analytic moods' (marking directive or optative illocutionary force). Whatever they are called, all these units scope over clauses and manipulate their reality status. A discrimination of these types of units is difficult or hardly possible, first of all, because core notions (especially '(analytic) mood' and 'complementizer') are ill-defined and their consequent cross-linguistic application suggests an almost arbitrary exchangeability: since the notional contrasts behind them are basically identical, clear criteria based on form and paradigmatic organization are warranted. Jointly, one needs to specify the format of the relevant units in terms of clines between morphemes and words, and between words and constructions, first of all for North Slavic by and South Slavic da. Concomitantly, the delimitation of discourse coherence from syntactic subordination poses notorious problems. First, embedding is a property on a gradient, mainly because symptomatic shifts of egocentricals need not (and often do not) occur simultaneously. Second, there is an enormous grey zone of clausal complements vs adjuncts leaving ample space for indeterminacy. Both intensional and extensional approaches to determining clausal complements have their inherent and empirical weaknesses, and one wonders whether these might be recompensated by combining both types of approaches. The article gives a complex account of general theoretical and empirical pitfalls, with illustrations from a comprehensive body of data across Slavic on a typological background. The article also shows a principled divide between volition- and cognition-based clause connectives (and of their constructions), for which it points out inner-Slavic areal clines.
Keywords: Complementizers; (analytical) mood; clause-initial particles; subordination; discourse syntax; Slavic
"границата меѓу помошните елементи и комплементизаторите и
помошните глаголи и глаголите не е строгa"
(Čašule 1989: 101)
Clausal complementation has been defined as
[
In abiding by this definition we have to be aware of its problems. Some of them will be examined in this article (for a more systematic discussion cf. [
[
This definition is very vague and, as we will see in the following, insufficient. The main problem is that it does not restrict the format of the units in question on a word—affix cline. By contrast, [
Therefore, complementizers, clause-initial particles and auxiliaries of assumed analytic moods form a triangular relationship, with complementizers appearing somewhat "parasitically" on the latter two. This article addresses problems in the analysis of elements that either occur as clause-initial connectives or as parts of such connectives. Such elements have variably been treated as 'particles' or as subordinators, in particular as complementizers, or as kind of auxiliaries in analytical mood, in particular as 'subjunctive markers'. See the highlighted elements in the following examples.
Russian
(
1sg.nom want [ipfv]-prs.1sg
čto. by my osta-l-i-s' vs-e vmeste.
comp.irr 1pl.nom remain [pfv]-pst-pl-rm all- pl.nom together
'I want us all to remain together.' ('... that we all remain together.')
(RNC; Daša, 2004)
Polish
(
chc-e, by inn-i pozna-l-i ją
want [ipfv]-prs.3sg comp other- pl.vir.nom acquaint [pfv]-pst-pl.vir 3sg.f.acc
również.
'He makes a good face for a bad game, he tries really hard, because he knows his worth and wants others to know it, too.' ('... that others know it, too.')
(PNC; M. Olszewski: Chwalcie łąki umajone. 2005)
Bulgarian
(
think[ ipfv].prs.1sg con go_out[ pfv].prs.1sg a_bit on air
'I think I'll go a little bit out into fresh air.' (V. Kampf, p.c.)
Czech
(
wish [ipfv]-prs.1sg 2sg.dat opt rm happy- adv return [pfv]-fut.2sg
'I wish you that you return safely' (or '... may you return safely!')
(from Meyer 2010: 373)
Slovak
(
neg -can -prs.1pl 3sg.m.dat say [pfv]-inf
nech si kupi materiál sám.
opt rfl.dat buy [pfv]-fut.3sg material- (acc) self- (sg.m.nom)
'We can't tell him to buy the material himself.' (lit. '... may he buy the material himself')
(from Meyer 2010: 373)
Slovene
(
afternoon aux.prs.3sg boss- (nom.sg) finally order [pfv]-pst-(sg.m)
naj me pokličej-o k nj-emu.
opt 1sg.acc call [pfv].prs-3pl to 3-dat.sg.m
'In the afternoon the boss finally ordered that they call me to (come to) him.'
(lit. '... May they call me to him')
(from Uhlik 2018: 412)
Polish
(
jakoby -m odby-ł spotkani-a z wojewod-ą.
comp.rep-1sg perform [pfv]-pst-(sg.m) meeting- pl.acc with voivode- ins
'There appeared some information that (as though) I had meetings with the voivode.'
(PNC; Wieści podwarszawskie 2007)
In all of these examples the clause-initial unit can be analyzed as a complementizer, i. e. a unit which marks a clause as an argument of a higher predicate. Obviously, the justification for such an analysis arises from that higher predicate, not as such from the clause with the element in question (see Section 3.3). All these units can also be categorized in at least one alternative way:
(i) as subordinators marking clausal adjuncts, e. g. purpose clauses (Russ. čtoby, Pol. by, Srb./Cro. da, Sln. naj), see (
(ii) as 'particles', i. e. units whose position in the clause is flexible (to a larger or smaller extent), which do not form constituents of their own, but operate on propositional and/or illocutionary content, as they connect this content with the propositional and/or illocutionary meaning of the preceding utterance and/or the actual discourse context; see (10–11);
(iii) as connectives somehow in-between (i) and (ii), namely: as clause-initial units linking "their" clause with a preceding one as its consequence; see (
(iv) as auxiliaries, i. e. morphemes which need a lexical verb to supply the argument structure for a complex predicate. The auxiliary contributes a modal, causative or illocutionary component to the meaning of the entire complex, and it often conveys grammatical information like tense and/or person-number agreement. To the extent that such auxiliaries modify the illocutionary force of their clause, in particular as a directive (hortative, jussive, permissive, etc.) or optative, or that they suspend the propositional content of their clause, such auxiliaries might be, and have been, considered as markers of 'analytical mood', which somehow mark the utterance content as 'irreal'. See (13–14), which illustrate syntactically independent clauses (as does ex. 11).
Polish
(
pochylić,
by słysze-ć jego głos.
conj.irr hear [ipfv]-inf his voice- (acc)
'Mick looked up and, staring at the ceiling, began to speak so softly that Phil hat to bend down again to hear his voice.' ('... in order to hear his voice.')
(PNC; A. Barczyński: Ślepy los. 1999)
Slovene
(
opt 2sg.acc somebody. nom hear [(i)pfv]-(prs.3sg) emph be. fut.3sg bad_mood- nom
'Should somebody hear you, bad mood will arise.'
(from Sonnenhauser 2021: 449)
Macedonian
(
ptc enough pn lest 1sg.acc call [pfv]-(prs.3sg) dad
'Boško, I must go, lest my father call for me.'
Czech
(
opt rm soon recover [pfv]-fut.2sg
'(May you) get well soon!' (from Meyer 2010: 373, translation adapted)
Polish
(
by leniwi-e osią-ść na dni-e więzienn-ej cembrowin-y.
conj.irr lazy- adv sit.down [pfv]-inf on bottom- loc prison lining- gen
'The tongue of the window light licked the rusty bars Mikołaj Rej's cell from Nagłowice to lazily settle down at the bottom of the prison lining.'
(PNC; Z. Smektała: Chcica czyli Billie Holiday to kurwa: poemat romantyczny. 2006)
Slovak
(
opt leave [pfv]-fut.3pl everybody- pl.nom
'May they / Let them all go away!' (from Meyer 2010: 373)
Macedonian
(
con leave [pfv]-prs. 1 sg q
'Should I go?'
The following Serbian example illustrates how one and the same surface unit can fulfil different syntactic functions, in fact, all of the aforementioned ones: da is employed as [a] complementizer, [b] connective within complex predicates ('mood marker'), and [c] adverbial subordinator (purpose clause):
Serbian
(
pn.nom think [ipfv].prs-(3sg) aux.prs.1sg say [pfv]-pst-sg.f
[a] daću [b] danapiše-mknjig-u
fut.1sg write [pfv].prs-1sg book- acc
[c] dapostane-m slavn-a.
become [pfv].prs-1sg famous- nom.sg.f
'Mary thinks that I said that I will write a book in order to become famous.'
(from Todorović 2015: 25)
Correspondingly, da has been given a different status in South Slavic languages (see Section 2.3).
The categorization of mood has also been made an issue in connection with clause types linked to some (presumably) superordinate clause with some of the units exemplified in (1–7), namely with clause-initial connectives marked with –by (in North Slavic, see 1–2), with da (in South Slavic, see 3) and recently also with Sln. naj (see 6). In Polish, many clause-initial units with the segment –by (as well as by alone, see 2) are characterized by their role as hosts of enclitic person-number desinences (see 7). Yet other units, like Russ. pust'/puskaj, have been considered as markers of analytical mood, but not as clause-initial connectives with the potential to mark subordinate clauses (see Section 2.5), in contrast to units like Slk. nech (in 5), which have sometimes figured as (adverbial) subordinators, but not as mood markers. In turn, Cz. ať (see 4) has been considered as 'analytic imperative' and subordinator at once (cf. MČ/III 1987: 494, 512–514).
What do all elements, as were exemplified in (1–15), have in common? Obviously, almost all of them somehow manipulate the reality status of the clause in which they occur; that is, all of them are clausal operators. They either modify illocutionary force, as some of them indicate directive or optative speech acts, or they signal some kind of epistemic distance toward the proposition coded in the clause. This binary division implies that we distinguish between clauses that code a proposition – i. e. the denotation of a state of affairs (SoA) which has reference by being anchored to some specific time and space and which can therefore be judged as true or false – and clauses devoid of a proposition, because they express specific non-representative speech acts. From a slightly different perspective, this divide amounts to a distinction between knowledge/belief-based utterances and volition/emotion-based utterances. Almost all of the examples (1–15) show a decrease in reality (or factuality) status, but for different reasons; (
Therefore, provided difficulties in determining whether we are dealing with mood markers, subordinators or particles is not just the consequence of some negligent façons de parler or of terminological confusion, we should become aware of the theoretical premises of these terms and (if we do not want to abandon some of them) we should clarify where their "neuralgic" areas of interference are. Such zones are most interesting since they create playgrounds of diachronic change, provided, again, we understand their relationship and analyze data consistently. Slavic languages feature whole bunches of phenomena that call out for such a clarification, the examples presented in this introduction are only a representative tip of an iceberg; the discourse conditions leading to the rise of patterns that may get tighter (or not) are basically located beneath the surface. However, in this contribution diachronic processes will only be accounted for as a backdrop to elucidate some of the observations, the main objective of this contribution is a methodological one. I therefore also do not aim at systematic stock-taking (nor at very systematic references), but will concentrate on pinpointing the functional domains in which complementizers arise most likely and recurrently (certainly not only in Slavic languages). The paper assesses the applicability of criteria that are required when it comes to analyzing linguistic data from a usage-based perspective. Attention is paid to properties of complementizers which make them a gradient category organized in a potentially open class with core and periphery. In all likelihood, the "semi-open" character of the complementizer class, as it is described in this paper, has its counterpart in the presumed perception of laypersons (so-called 'language users') for whom the categorial status of the auxiliary or connective probably in most cases does not matter – if only it contributes to some particular function in discriminating the factuality (or reality) status of the entire utterance. This aspect, which is treated as 'oscillation' of syntactic functions in Mendoza and Sonnenhauser (this volume), complements this paper, so much as the present contribution aims at spelling out the gradients and criteria along which discrimination between the categories can (but need not) become possible and how a consistent analytic treatment may (hopefully) be reached.
Thus, the key concepts of this paper are: Complementation, complementizer, mood (auxiliary), subordination. In the next section (Section 2) I will first examine under which conditions clause-initial connectives and (purported or acknowledged) mood markers cause troubles of analysis. This section ends with an attempt at pinpointing the neuralgic zones where these concepts potentially interlace (Section 2.9). On this basis, I will dwell upon issues connected to subordination and complementation and evaluate criteria proposed as diagnostics of complementizers on Slavic data (Section 3). The final section summarizes the findings and formulates some conclusions (Section 4).
The following "technical" remarks seem appropriate. First, if needed, scope is indicated by square brackets. Second, 'conjunct' refers to any clausal unit occurring in a sequence with other clausal units (as for 'clause' see Section 2), 'connective' is used as an umbrella term for any kind of function word indicating some meaningful relation with the content of the immediately preceding conjunct(s). Third, if no specific status is to be assigned to a clause connective it is glossed by using con , irr , assigned to a morpheme participating in any sort of irrealis marking (jointly with a specific verb form). Morphological segmentation will account not only for clitics (marked with =), but also for fused segments whose morpheme status may become debatable (marked with. in the glosses or otherwise with |). The reasons for this approach will become evident in Section 2.
Let us start with complementizers and mood auxiliaries. Both are sensitive to the relation of the denotational content with some illocutionary purpose and/or speaker's stance toward some kind of reality (judged on some conversational background, in Kratzer's 1981 terms). However, while complementizers are markers of subordination, since they "link" clauses which betray traces of structural asymmetry (see Section 3), mood auxiliaries are not necessarily indicative of subordination. Although some "traditional" moods are typically associated with subordinated clauses (compare, e. g., the very term 'subjunctive'), other 'moods' defy subordination because of the illocutionary force they are associated with; this pertains, in particular, to moods that mark directive speech acts. At least in this domain, complementizers and moods appear to be antipodes.
Briefly, I opt for understanding complementizers as heads of clauses which in toto occupy an argument slot of the predicate in an adjacent clause; this clause is often called 'matrix clause', its predicate 'complement-taking predicate' (CTP). This relation between clauses entails embedding (on which see Section 3). Simultaneously, complementizers are heads of embedded clauses; they, as it were, flag the dependency relation with the respective CTP (just like case endings or adpositions serve as flags on dependent constituents in simple clauses). In this respect, a complementizer functions as an explicit sign that transmits the head relation of the CTP to its dependent clause. This is compatible with Noonan's (2007: 55) definition of a complementizer as given in [
As concerning moods, due to the most conservative (European) traditions they have been narrowly captured as morphological categories marked on the verb by sets of endings (cf. [
In typological literature, mood (and modality) has been discussed in connection with the marking of ir/realis distinctions. In their critical survey [
In turn, an influential paper from language philosophy is [
What indifference for the locus of marking leads to becomes evident, for instance, in formal syntax research, where 'subjunctive' is often used as a label given to clause patterns that are associated to irrealis (in the aforementioned sense), marked by various types of expressions, among others by clitics and clause-initial elements (often called 'subjunctive particles' or similarly), and which typically occur in embedded (mono- or biclausal) structures and act like replacements of the infinitive. This practice is particularly common in generative syntax research into South Slavic, where it concerns, first of all, the ubiquitous morpheme da. Actually, da behaves very differently in the particular South Slavic languages (see Section 2.3), but the problem to be pointed out here becomes clearer if we first turn to Romanian. Contrary to Balkan Slavic (and Slavic languages in general), Romanian has specific desinences on the verb which are treated as a morphological subjunctive. However, simultaneously clauses with some irrealis meaning can be embedded after semantically suitable verbs, and this embedding may (or must) be indicated by clause-initial să. This connective stands in an opposition to că, which is considered a default factuality complementizer since it does not imply anything specific about the reality status of the embedded clause ([
Romanian
(
hope. prs.1pl comp.irr win. 1pl. sbjv a prize
" sbjv "
'We hope to win a prize. / We hope (that) we win a prize.'
(Irimia 2010: 324, adapted, her translation)
One can of course argue, that within clause boundaries, 'subjunctive' is marked twice; alternatively, one may say that 'irrealis' is marked twice. This would raise an impression of terminological arbitrariness and leave us with the question whether 'mood' should not, after all, be treated as a form-related notion, while 'ir/realis' might better be maintained as a conceptual distinction that is less dependent on form and on tighter paradigmatic relations. Regardless, either of these notions pertains to the clause level, and both are highly abstract, though for different reasons.
The problem can be approached with less circularity if we examine cases in which there are no specific desinences on the verb (other than for the indicative), and the only proper signal of ir/realis marking is the clause-initial element, or some other marker that is not a verbal affix. This situation can be observed in Balkan Slavic. In the Bulgarian example (
Bulgarian
(
hear[ ipfv].prs-1sg con / how / comp beat[ ipfv].prs-3sg clock
'I (can) hear the clock beating.' (V. Kampf, p.c.)
Moreover, (
Bulgarian
(
think[ ipfv]-prs.1sg comp solve[ pfv]-aor.1sg all task. pl on test. def
'I think I (have) solved all tasks from the test.'
Here different tenses are used (aorist in 18, present tense in 3), but the crucial difference only becomes evident from distributional restrictions: while če allows for any tense available in Bulgarian, da cannot combine with the aorist or the future (marked with šte), being practically restricted to the present tense and the perfect (with l-participle), while imperfect is possible only in narrated past. This distributional restriction does not allow us to deduce whether both če and da can count as complementizers, but it shows that in Bulgarian (and Macedonian), the ir/realis distinction is not really a matter of the verb phrase. Instead, it hinges on morphemes that somehow manipulate the available set of forms on the predicate. These forms are not specialized for mood marking (unless one treats mood as an abstract dimension that simply coincides with ir/realis distinctions).
Certain properties of da in Balkan Slavic can be used as arguments counter its complementizer status, but in favor of treating it like a mood marker (for a comprehensive discussion cf. [
Macedonian
(
con neg aux.prs.2sg 3m.dat 3m.acc give [pfv]-lpt-sg.f more
'Don't give it to him anymore, or else!'
(from Bužarovska and Mitkovska 2015: Ch. 1; cf. also Spencer and Luís 2012: 125)
This is why only other clitics can separate da from the following verb marked for tense:
Macedonian
(
order [ipfv].prs-1sg pn
da dojd-e vednaš. (vs *da Marija dojde)
con come [pfv].prs-3sg immediately
'I order Maria to come immediately.' (more lit. ...that Maria comes...)
This example illustrates complementation, inasmuch as nareduvam requires a clausal argument (see Section 3.3). Concomitantly, da frequently occurs clause-initially both in (assumed) dependent clauses (see 3, 17, 20) and in independent clauses (see 14, 19). However, although this effect is frequent, it is incidental in that da's clause-initial appearance entirely depends on its leftmost position in verb-oriented proclitic clusters.
The tendency toward proclitic treatment is a property that has increased in Balkan Slavic together with other phenomena of internal differentiation within South Slavic. In the (north)western part of South Slavic, i. e. in SerBoCroatian and Slovene, da has gone another way. It has lost its irrealis restrictions and thereby turned into a default complementizer that is widely used in realis contexts. Concomitantly, da no longer restricts the choice of admissible tenses. It can occur with the future (as ex. 15 shows), although it occurs predominantly with the present tense, if the clause which it introduces marks some irrealis-meaning. That is, da can be considered a complementizer that is indifferent for ir/realis contrasts, but the former restriction to irrealis (and present tense) still surfaces in a split of present tense interpretation and in different restrictions to the array of admissible tense-aspect forms in the lexical verb.
Moreover, da keeps its clause-initial position; it is easily separated from the verb by any kind of clitic, word or constituent, and it can itself serve as prosodic host of 2P-enclitics (cf. [
SerBoCroatian
(
doubt [ipfv]-prs.3sg comp irr 3-pl.nom stop [pfv]-lpt-pl
'S/He doubts that they would stop.'
(from Szucsich 2010: 405)
The contiguity between da and bi- can be disrupted by negation. Admittedly, the frequent collocation of da and negation is a common feature of entire South Slavic, which has yielded connectives (or particles) like Mac. da ne, Bulg. da ne bi (see Section 2.6). However, the frequency and functional load of the bi-subjunctive (also dubbed 'conditional') has decreased in Balkan Slavic, in particular in Macedonian, where we mostly find it in frozen phrases with an optative meaning, e. g. in curses and blessings (cf. [
Slovene
(
fear [ipfv].prs-1sg rm comp neg irr forget [pfv] - lpt-(sg.m) purse- acc
'I fear that I might / will forget my purse.'
(M. Uhlik, p.c.)
In summary, whereas in Balkan Slavic (= eastern South Slavic) da has become an integral part of verb-oriented proclitic clusters and first position in the clause is thus possible only if the verb occurs as the first stressable word (= prosodic host), western South Slavic employs da as a morpheme which can serve as host of 2P-enclitics, it is thus a real clause-initial connective. This behavior goes hand in hand with the loss of irrealis restrictions, and as such da still freely collocates with the bi-subjunctive. In contexts associated to irrealis functions, da basically betrays a behavior resembling Balkan Slavic da, but such contexts are found primarily in complementation patterns with a high degree of semantic integration, e. g. after desiderative and manipulative verbs. Of course, as is common for isoglosses, the different patterns in eastern and western South Slavic do not constitute a clear-cut boundary, but a gradient (in many respects). However, the point is that the severe consequences for an inner-South Slavic differentiation in morphosyntax that are associated to this cline have also created different conditions for complementizers (and other clausal subordinators) and their delimitation from, and interaction with, verbal mood – provided we distinguish the one from the other. Note, furthermore, that none of the properties pointed out for da in either "half" of South Slavic are eo ipso helpful as "diagnostics" of complementizers; these properties just apply to a distinction between clause-initial connective and verb-oriented proclitic. For a clause-initial connective to be qualified as a complementizer, other criteria must be considered (see Section 3).
The situation is entirely different in North Slavic. Da has never played any considerable role, but by- (+ l-form or infinitive) has been retained as periphrastic subjunctive marker, regardless of the embedded or independent status of the respective clause. Moreover, we encounter clause-initial connectives and particles to which by attached and eventually got incorporated; compare Russ. čtoby, Pol. żeby, jakoby. Pol./Cz./Slk. aby, etc.: the morpheme boundary can still be recognized, but –by cannot be separated. This stage represents almost the endpoint of a morphologization process (cliticization > agglutination > fusion) and can be explained as the consequence of by's behavior as an original 2P-enclitic which clusters with person-number enclitics in a fixed order. This order has remained stable in West Slavic (while with the l-form East Slavic lost person-number enclitics altogether), which explains why pronominal clitics of first and second person almost always inseparably adhere to by. This applies even if the cluster [by+person-number] no longer occurs in 2P, but can take virtually any position between the first constituent and the verb or immediately after the verb; compare the following examples:
Polish
(23a) Kiedyby-śsięwreszciezabra-łdorzecz-y?!
when irr-2sg rm finally get.at [pfv]-lf-(sg.m) to thing- gen
(23b) Kiedy zabrał=by-ś się wreszcie do rzeczy?!
(23c) Kiedy wreszcie zabrał=by-ś się do rzeczy?!
(23d) Kiedy wreszcie do rzeczyby-śsię zabrał?!
(23e) *Kiedy zabrał do rzeczyby-śsię wreszcie?!
'When will [lit. would] you finally get at your stuff?!'
Remarkably, the clitic cluster [by+person-number] has remained stable even after by became inseparable from its host; compare (
(
want [ipfv]-pst-f-1sg comp.irr=1pl go [pfv]-lf-pl.nvir together on walk- (acc)
ale usłysza-ł-a-m, że jest umówion-a.
but hear [pfv]-pst-f-1sg comp be. prs.3sg appointed- sg.f.nom
'I wanted that we go for a walk together, but I heard that she had an appointment.'
(PNC; Dziennik Zachodni, 19.09.2006) (vs. *... żeby poszły-śmy)
In connectives like Russ. čtoby or kak by (...ne), Pol. żeby the segment by has turned into a "morphological hermaphrodite". In the first place, by cannot be separated from these connectives, it is thus a segment of these lexical units (see the notation in 23–24). These connectives contrast with etymologically related connectives that are void of by, but can combine with the "ordinary" subjunctive, i. e. in combination with the l-form or the infinitive. A conundrum arises from the fact that, all over North Slavic, despite its inseparability from the relevant connective units, by shows the same requirements to the form of the predicate in its clause as does by in the "free" subjunctive: In either case by consistently requires the l-form or the infinitive; compare (
Polish
(
very= irr.1pl rfl-dat wish [ipfv]-lf-pl.vir
że.by nasz-e aut-a by-ł-y sprawn-e.
comp.irr our car- pl.nom be- lf-pl.nvir operational- pl.nom.nvir
'We would very much wish our cars were operational.'
(PNC; Gazeta Krakowska, 13.12.2006)
(
think [ipfv]-prs.1sg comp exactly here can- lf-pl.vir=irr.1pl
rozwiną-ć nowy nurt.
develop [pfv]-inf (new trend)- acc.sg
'I think that this is where we could develop a new trend.'
(PNC; Dziennik Zachodni, 15.11.2002)
Russian
(
wh- (acc) irr 1pl.nom man- pl.nom without 2pl.gen do [ipfv]-lf-pl
'What would we men do without you?'
(RNC; kollektivnyj. Forum: Mužčina v škole. 2011)
(
čto on-a by i dv-ux dn-ej ne potjanu-l-a.
comp 3-f.sg irr ptc two- gen day- gen.pl neg draw [pfv]-lf-sg.f
'I look at my 17 year old daughter and think that she wouldn't make it even for two days.'
(RNC; Naši deti: Podrostki. 2004)
(
how irr 1pl.dat neg want [ipfv]-lf-n-rm
no oni est', i my s nimi stalkivaemsja.
'As much as we would like, but they exist, and we face them.'
('However we would like...')
(RNC; kollektivnyj. Forum: 12 časov v den'?. 2010–2011)
This grammatical requirement is analogue to Balkan Slavic da (see Section 2.3): both by-connectives and Balkan Slavic da restrict the admissible range of verb forms, but for –by the degree of fusion is greater than for da. Importantly, the by-segment in the North Slavic connectives shows all the aforementioned properties practically only if these connectives occur clause-initially, i. e. in a position where they have the "best chance" to serve as complementizers. As particles (whose position is not fixed) these units do not require any specific verb forms (see 30, 32), nor do they (in West Slavic) practically ever serve as hosts of enclitics (see 31).
Polish
(
peasant- pl.nom gossip [ipfv]-prs.3pl comp there
jakoby
rep wander [ipfv]-(prs.3sg) rm on court- loc spirit- (sg.nom) heir- gen
'The peasants rumor that the heir's spirit is allegedly wandering around the court.'
(vs ...plotą, jakoby błąka-ł / *błąka się duch)
(PNC; T. Bojarska: Świtanie, przemijanie. 1996)
(
but dem.n what. acc rep find [pfv]-pst-pl.vir=2pl
'But what you allegedly found is just mine.' (vs
(PNC; A. Sapkowski: Wieża Jaskółki. 2001)
Russian
(
čto
gosudarstv-o
kak by
comp state- sg.nom as_though neg react [ipfv].prs-3sg
ili ne v sostojanii reagirovat' na sobytija podobnogo roda.
'One gets the impression that the state does not seem to react [does not as though react] or is not able to react to events of this kind.'
(RNC; Vremja MN, 5.08.2003)
On the one hand, such differences in morphosyntactic behavior create distributional contrasts, i. e. patterns which justify postulating different minor word classes ('function words'), here particles as distinct from subordinators. In particular, such contrasts are decisive for Pol. jakoby to become a complementizer, since person-number enclitics attach to it almost only if it occurs clause-initially, i. e. in complementizer position (see above and fn. 18). Jointly, only as a complementizer does jakoby require the l-form (which, in turn, is a precondition for the person-number enclitics to occur). On the other hand, the morphologization of –by eventually occurring as a fused final segment of connectives creates new lexical units. This applies even if its diachronically motivated relation to the l-form on verbs remains transparent and continues to mark irrealis functions (incl. any kind of 'epistemic distance'). What we thus observe is the univerbation of an irrealis element with clause-initial connectives, this process was favored by frequent co-occurrence due to the 2P-rule ('Wackernagel's law') before this rule began to faint.
In addition, we observe that Pol. by itself has become a subordinator (on a par with aby and żeby) able to introduce purpose clauses (see 33) and clausal complements (see 34). It requires the l-form or the infinitive and serves as host of person-number enclitics, but, as a subordinator, it must have lost its enclitic properties.
(
Chc-emy tylko, że.by zdrowi-e
want [ipfv]-prs.1pl only comp.irr health [n]-sg.nom
dopisa-ł-o ,
turn.out.adequate [pfv]-lf-sg.n
by=śmy jak najdłużej mog-l-i się sob-ą nawzajem cieszy-ć.
irr=1pl as.long.as can- lf-pl-pl.vir rm rfl-ins mutually rejoice -inf
'At our age, you don't dream anymore. We just want our health to be good, so that we can enjoy each other for as long as possible.' ('... that our health be good...')
(PNC; Gazeta Krakowska, 30.11.2002)
(
pan profesor chcia-ł=by
Mr professor [m]-(sg.nom) want [ipfv]-lf-(sg.m)=irr
by=ście z radości-ą świętowa-l-i t-en dzień.
irr=2pl with joy- ins celebrate [ipfv]-lf-pl.vir this day- acc
'I think that the professor would like you to celebrate this day with joy.'
('...that you celebrate this day with joy')
(PNC; Dziennik Bałtycki, 22.06.2007) (vs *chciałby by z radością świętowa-l-i=ście)
Moods are normally regarded as grammatical categories that manipulate the reality status of "their" clause or, respectively, "their" utterance. In practice, this means that this manipulation relies on patterns which are discernible by a sufficiently regular relation between forms and their functions. The pertinent expressions are expectable and discernable enough in contexts for which reality status is relevant. Such form:function relations may be strengthened by paradigmatic replacement conditions with other forms (or constructions) that are comparable functionally, but which – in these functions – are constrained by conditions caused by additional functions and/or categorial oppositions of the given language. For instance, Russ. pust'/puskaj occurs predominantly with the 3. prs -indicative in a jussive or permissive function and seems to paradigmatically complement the "synthetic" imperative related (in its primary directive function) to the immediate addressee (second person), in which pust'+ prs.ind occurs at best exceptionally (Dobrushina 2019: Section 3.1). For this reason, the periphrasis pust'/puskaj+ 3prs.ind has been suggested as a member of an imperative paradigm in Russian ([
Nonetheless, Dobrushina (2019) denies it this status, because (among other reasons) in a minority of cases pust'/puskaj also combines with the subjunctive (by+l-form). She argues that grams representing categories of the same type (here: mood) usually do not combine with each other, but show complementary distribution. A similar point could be made for Sln. naj, which can combine with the imperative (see Section 3.4) or the bi-subjunctive (cf. [
First, the combination of pust'/puskaj and the by-subjunctive yields a transparent counterfactual meaning composed of two components: pust' adds its directive (or optative) meaning to the meaning of an unrealized event contributed by the by-subjunctive; the latter conveys an epistemic (i. e. belief/knowledge-oriented) function, whereas the former is based on volition. See one of the examples in Dobrushina (2019), who herself admits the transparent additive meaning:
Russian
(
оpt perish [pfv]-lf-sg.n irr all humanity- sg.nom.n
(i) 'Let all humanity perish, and the boy would remain / would have remained alive!'
(ii) 'Even if all humanity perished, the boy would have remained alive!'
(F. Iskander. Son o Boge i d'javole. Znamja. 2002)
An analogous point could be made for combinations of the by-subjunctive with equivalent units in other Slavic languages. Thus compare Sln. naj (see above) and the following examples in which a permissive (Slk. nech) respectively optative (Pol. niech) meaning of the 'let'-derived unit scopes over the subjunctive, which itself contributes a hypothetical reading:
Slovak
(
nech by sa ocit-l-i kdekoľvek,
opt irr rm find [pfv]-lf-pl wherever
mohli by očakávať len horšie zaobchádzanie.
'It didn't take long and a decree was issued to leave the city, which frightened the Jews, for they knew that wherever they were, they could only expect worse treatment.'
(SNC; Krížne cesty. Bratislava, 2000)
Polish
(
Tylko niech by- m dosta-ł choć materiał.
only opt irr-1sg get [pfv]-lf-(sg.m) at.least material- (acc)
'I don't see a problem. I would have made myself. Just let me get some material.'
('... If only I got some material.')
(PNC; Gazeta Krakowska, 07.08.2002)
Second, there are many cases in which categories of the same "dimension" allow for combinations, or even have to combine. For instance, Bulgarian systematically combines values of two different aspect oppositions, aorist:imperfect ´ perfective:imperfective (cf. [
Regardless of such objections, strict replacement conditions can also obtain between clause-initial units that mark some meaningful relation to the preceding clause. That is, replacement conditions hold prior to any specific status which we might want to assign to the units (or constructions) involved in paradigmatic replacements. Pertinent examples can be easily construed with probably all of the units illustrated as potential complementizers in (1–7) and Section 2.6. Compare the corpus example (38a) with its modifications in (38b-c):
Polish
(38a) Powiedzmu,niechjutro
say [pfv]-(imp.sg) 3sg.m.dat opt tomorrow come [pfv]-(fut.3sg) to cantor- gen
'Tell him, may he come to the cantor tomorrow.'
(PNC; Wł. St. Reymont: Ziemia Obiecana. 1898)
(38b) Powiedzmu,a.by / byjutro
say [pfv]-(imp.sg) 3sg.m.dat comp.irr tomorrow come [pfv]-lf-(sg.m)
do kantor-u.
to cantor- gen
'Tell him that he come to the cantor tomorrow.'
(38c) Powiedzmu,żejutro
say [pfv]-(imp.sg) 3sg.m.dat comp tomorrow
aux.prs.3sg come [pfv]-inf to cantor- gen
'Tell him that he is supposed to come to the cantor tomorrow.'
The clause introduced by niech in (38a) marks a directive speech act. Although directive illocutions are considered to be lost in subordination, it is tempting to count this as an example of complementation, since powiedzieć requires an argument coding the content of speech and the niech-clause exactly fits this purpose (see Section 3.4). Irrespective of this, niech can be replaced by aby or by in (38b). This clause refers to the same directive speech act as does (38a), although aby triggers the l-form (see Section 2.4), whereas niech predominantly occurs with the present indicative (as does Russ. pust'/puskaj); the respective other verb forms are unusual (or downright inacceptable) in combination with aby and niech. Finally, (38c) demonstrates yet another way to (roughly) "say the same" in relation to the preceding conjunct: The standard complementizer że is employed together with a modal auxiliary (mieć + infinitive), which refers to the same directive speech act. Simultaneously, the information concerning reality status has been "distributed" over the second conjunct in another manner: the complementizer does not specify anybody's stance, instead this function is fulfilled by the auxiliary.
The comparison of (38a–b) illustrates not only that two clause-initial units may replace each other to denote (roughly) the same situation, but also that their co-occurrence restrictions with grammatical forms on the predicate tend to be complementary. In fact, many connectives restrict the set of grammatical forms on the verbal predicate which follow them in the same clause (from among the set of verb forms that are generally available in the given language), just as we noticed this above for da and –by. Such restrictions have been considered to be characteristic of non-indicative moods as well, and they are related to deranked (vs balanced) marking in clause linkage (cf. [
Simultaneously, a comparison of (38a–c) shows that purported auxiliaries of analytical moods happen to occur as initial element in a clause that serves as a complement to a verb (or noun) in the preceding clause, and this is what makes this element suitable as a "replacement" of another element which is considered a complementizer. We are facing the same problem as with, e. g., the Balkan Slavic verbal proclitic da, with Cz. ať (cf. MČ/III 1987: 512–514), or with by-connectives in North Slavic (see Sections 2.3–2.4). The would-be auxiliary need not be adjacent to "its" verb (see 38a), but it can, as e. g. in (
Graph: Figure 1 Paradigmatic relations between clause-initial connectives
Now, if we were to apply Noonan's definition of 'complementizer' cited in [
Figure 1 admits for more semantic relations to be illustrated as possible instances of clausal complementation, together with admissible clausal "linkers" occurring at the left edge of a subsequent conjunct. Here are some additional examples. The languages are chosen for convenience, but with the purpose to show the broadest possible array of semantic relations (at least for Slavic); the domains covered are named before the language headings.
Bulgarian
(
old_man. sg-def.sg.m see. aor-(3sg) how pn go_down[IPFV]- (prs.3sg)
'The old man saw how Elka quickly comes down the road.'
(from Petkova-Schick 1973: 279)
Similarity (comparison), with different degrees of conformity with reality
Croatian
(
this -nom.pl airplane- nom.pl look- prs.3pl
kao da će upravo poletje-ti.
like comp fut.3pl just fly [pfv]-inf
'These airplanes look as if they are about to fly off.'
(T. Sočanac, p.c.)
Macedonian
(
rm behave [ipfv].prs-(3sg) as_if be- lpt-sg.f in America
'She behaves as if she had been to America.'
Russian
(
seem [ipfv]-pst-n-rm as.if burn [ipfv]-prs.3sg house- (sg.nom)
'The fire was dying out, and it was thrown on the dark logs of the hut. It seemedas if the house was on fire.'
(RNC; O. Radzinskij: Proščanie. 1985)
Polish
(
sprawia wrażeni-e jak.by słucha-ł-a Łukasz-a
cause [ipfv]-prs.3sg impression- acc as_if. irr listen [ipfv]-lf-sg.f pn-gen
bardzo uważni-e.
very attentive- adv
'A close-up of focused Kasia who seems to be listening to Łukasz very carefully.'
('...creates the impression as though she is listening...')
(PNC; Samo życie, odcinek 214; TV script)
Reportive (see also ex. 7, 41, 97 for Pol. jakoby)
Russian
(
3sg.m.nom say [ipfv]-pst-(sg.m) as. if work [ipfv]-prs.3sg on factory -loc
'He told (me), as though/that he worked at a factory.'
(Letuchiy 2023)
Striking is that – as far as I know – none of these domains has ever been associated to mood; however, all the units (and many more) in their function as clause-initial "linkers" in the second conjunct might be considered potential complementizers. One cannot but admit that the second conjunct fills an argument slot of the verb (or alternatively, of an event-related nominal part of speech) in the preceding conjunct (see Sections 3.2–3), and since those clause-initial connectives in the second conjunct serve as cohesive devices sensitive to the meaning of both conjuncts, they can easily be considered complementizers. In many cases, they cannot even be omitted as "linkers" (e. g., Mac.
The domains illustrated in (42–46) are all related to perception or knowledge and belief, i. e. to epistemic attitudes or, as with the reportive complement in (
The matter is slightly different for apprehension; conceptually, this is probably the most complex domain. Apprehension relates to events judged as likely and undesirable at once, it thus combines an epistemic assessment that a situation S has occurred, is occurring or will occur, with a wish that S better be (or: had been) avoided. In this sense, apprehensional clauses unite features of negative optatives with epistemic evaluation. In Slavic, only few units show a "specialization" as markers of apprehensional complements, and there seems to be only one such unit (or construction) which is restricted to this use, namely Russ. kak by... ne:
Apprehensional (see also ex. 61)
Russian
(
fear [ipfv]-inf-rm how irr his uncertainty- gen
ne zameti-l-i ostal'n-ye.
neg notice [pfv]-lf-pl other- pl.nom
'And the guy began to fearthat the others would notice his uncertainty.'
(RNC; V. Bykov: Boloto. 2001)
There are some more clause connectives that are able to function as complementizers of apprehensional clauses jointly with negation; compare Pol. żeby... nie, Cz. aby... ne, Mac. da ne, Bulg. da ne bi (da...). In contrast to Russ. kak by... ne, all these connectives are polyfunctional, and the degree to which the negation remains semantically transparent differs (cf. [
Optative marking has rarely been considered as another mood in a Slavic language; if anything, such a 'mood' was understood not as a verbal category, but defined as a clausal construction whose formal characteristics were considered in a paradigmatic relation with other clausal constructions bearing different illocutionary force (cf., e. g., [
In all, disputes about 'analytical moods' as full-fledged paradigms have arisen practically only for discontinuous combinations of some form of a lexical verb (indicative present, l-form, infinitive) with auxiliary-like elements that add some volition-based illocutionary force (optative or directive: permissive, jussive, hortative). They have hardly ever arisen for combinations from the epistemic/evidential domain (based on cognition), the latter being more tightly associated with perception. The only exceptions to this are the by/bi-subjunctive (or conditional) inherited from Common Slavic and da-clauses with some irrealis semantics in (much more recent) Balkan Slavic. These modifications of clausal semantics can be either volition- or cognition-based.
The list of meaning relations discussed for the cognition-based domain in Section 2.6 lacks two types of connection between clauses: statements about the reality of a proposition (a.k.a. assertions) and statements about propositions that are logically presupposed (according to [
Czech
(
2sg.nom rfl.dat think [ipfv]-prs.2sg comp neg.know [ipfv].prs-1sg
že jsi ho očarova-l-a?
comp aux.prs.2sg 3sg.m.acc charm [pfv]-lpt-f.sg
'You think that I don't know that you charmed him?'
(ČNC; D. Fo: Marcolfa; translated in 2009)
The next example illustrates a factive complement:
Macedonian
(
good- n be. prs.3sg comp be_silent [ipfv]-impf.2sg
'It is goodthat you kept silent.'
Not all Slavic languages have factive complementizers; indeed, such units are prominent only in South Slavic; compare Mac. što, SerBoCroatian što, jer, Bulg. deto. But even in these languages, the default complementizer is often used with factive complements. Factive complementizers show an affinity to causal conjunctions, from which they often derive diachronically; this is reflected in functional overlap (compare SerBoCroatian jer, Bulg. če). This affinity is most plausible for (some) emotive factive predicates like, e. g., 'be sad', 'be happy'. Providing a reason (or motivation) for some state or event entails that this state/event is taken for granted (thus, presupposed). Emotive predicates, thus, can function as door-openers for factive complementizers.
The crucial point to be made is that statements about propositions to which an attitude holder gives full epistemic support (cf. [
In principle, the same considerations can be extended to interrogative complementizers. Propositions need not be asserted (or denied), as in declarative sentences, but may become the target of yes/no-questions. These are conventionally identified with if -units; they acquire complementizer function from embedding yes/no-questions, but usually they also mark syntactically independent yes/no-interrogative clauses (Russ. li, Bulg. dali, Pol. czy, etc.). Thus, the only difference from that -units consists in weaker epistemic support, which is just a consequence of yes/no-questions.
These things may seem trivial (or well-established), but, remarkably, units which are employed as default or as interrogative complementizers have not been considered as mood markers; instead, there are extensive discussions concerning the problem whether the relevant units mark subordination (and thus are able to serve as complementizers) or simply set off clausal units that may have their own illocution and/or epistemic vantage point. This problem is not only a consequence of the usual person-deictic shifts known as a diagnostics of direct (quoted) vs indirect (reported) speech, but it is ubiquitous with any sort of egocentric expressions. These either altogether defy shifts (and thus subordination), or they change their referential anchorage under subordination (if they are not reinterpreted). Thus, [
Trivially, egocentricals are defined via attitude holders. Attitudes may not only be related to cognition (e. g., epistemic stance), but also to intention, and these are the basis of illocutionary force. Both have to be assigned to, respectively, a judging (epistemic) or a desiring (intending) subject, and this subject may differ from the speaker and be transferred to the subject of a clause linked by a clause-initial unit that generally is considered a factual or interrogative complementizer. Thus, see the following examples, in which the illocution (indicated by square brackets) is not that of the speaker, but belongs to another attitude holder. (
Polish
(
że [niech go lepiej prokurator uwolni]...
comp opt 3sg.m.acc better prosecutor- (sg.nom) free [pfv]-(prs.3sg)
'But, rabbi, you said a moment ago that [may the prosecutor better free him]...'(PNC; J. Dobraczyński: Święty miecz. 1949)
(
ask [ipfv]-prs.1sg q can- lf-(sg.m)-irr-2sg rm fuck.off [pfv]-inf
'I'm asking, could you please fuck off?!' (lit.... whether you could fuck off)
(PNC; Samo życie, odcinek 294. 2002–2010)
Russian
(
čto [pust'kul'tur-apoterp-it].
comp opt culture -sg.nom be.patient [pfv]-fut.3sg
'The president said at the supreme council that [may culture be patient].'
(RNC; V. Davydov: Teatr moej mečty, 2004)
(
čto [budto [opjat' k nam v sel-o karatel-i ed-ut]].
comp as.if again to 1pl.dat in village- acc punisher- pl.nom ride [ipfv]-prs.3pl
'And there is also such a rumor that (as though) punishers are coming to our village again.'
(RNC; A. I. Panteleev. Nočnye gosti. 1944)
Notably, if the illocution is volition-based (see 51, 53) the clause can in toto be assigned to the other attitude holder, i. e. including the marker which is indicative of that illocution (niech, pust'), and can, jointly with that marker, be read off as direct speech. If, however, the illocution is cognition-based (see 54) the marker indicative of the other attitude holder (budto) is not part of such an original speech act. Finally, Pol. czy (in 52) could be part of that speech act, but this is because as an interrogative marker it can be used equally well with clauses coding a proposition (and a representative speech act) and with directive speech acts.
After all, we may say that factual complementizers are compatible with all kinds of shifts to changing attitude holders, because of their versatibility among various attitudes. Alternatively, we could say that in cases like (51–54) no complementation applies and, accordingly, that here Pol. że, czy and Russ. čto are not complementizers, but simply markers of clausal boundaries which here happen to coincide with the boundaries of utterances assigned to different attitude holders (which may be the same person, as in 52, or not).
For factive complementizers, this issue does not arise, because they presuppose propositional content as true, so that they do not create any "leeway" for epistemic attitudes, and they are insensitive to illocutionary distinctions, since such presuppositions can only arise with representative speech acts. Thus, any purported moods related to directive or optative speech acts are out of issue.
Once we admit this, another problem arises. Many conjunct pairs as in (51–54) lack such versatile factuality markers; see (3–7) and (38a). Are these eo ipso to be considered complementizers by themselves? Of course, this question should be examined taking into consideration more general issues of subordination, to which we come in Section 3.4. Let us first summarize.
We have seen that a delimitation of units that can be considered either as (emergent) auxiliaries of analytical moods or as clause-initial connectives ('particles') cannot be successful on functional, or conceptual, grounds alone, since all relevant units scope over the propositional and/or illocutionary content of the clause they appear in and thereby manipulate the reality status of the utterance. This implies restrictions on the grammatical forms of the predicate (in comparison to their generally available array of forms in the given language). Unless we want to discard distinctions between mood markers (auxiliaries) and (clause-initial) connectives altogether (or use these terms arbitrarily), what seems to be called for are form-related criteria and/or criteria based on grammatical distribution. This, apart from restricting the form of the predicate, leads to a tighter paradigmatic integration of sufficiently regular forms correlated with ir/realis contrasts. The less such form:function correspondences are concentrated in the verb phrase and, conversely, the more they occur scattered over different places of the clause, the more it seems problematic to relate them to 'mood' – otherwise this term easily becomes just an equivalent of ir/realis distinctions defined on conceptual grounds. Defining 'mood' by considering occurrence restrictions for various kinds of dependent and main clauses – or after particular matrix predicates – is not fruitful either. First, this already implies that we can clearly tell apart main and dependent clauses (for this problem see Section 3) and, second, clausal connectives which contribute to marking ir/realis distinctions occur in the same environments and may, under specific circumstances, be analyzed as complementizers. This is to say that we cannot distinguish moods and clause connectives (resp. complementizers) sensitive to ir/realis contrasts simply on the clause level, because all of them operate on the clause level regardless of how we might take stance to the subordination issue. The same applies to the possible interlacing of (epistemic or volitional) attitude holders (see Section 2.7).
However, we can treat moods and ir/realis-sensitive clause connectives as categories with core members and peripheries, so that we have gradients which, as it were, run into one another (see the boxes in the right half of Figure 2 below). Provided we understand mood as a primarily verb-oriented category and clause connectives primarily as word units which tend to occur at the left edge of their clause (at least in European languages), we can capture the fact that peripheral members of one category (or better: expression class) become similar to more typical members of the respective other expression class. This goes hand in hand with expectable definitorial and analytic problems in the middle part of the conceptual space created by these opposed clines as well as for units that incorporate morphemes otherwise associated with (analytic) mood. What does this mean for Slavic languages?
In Slavic, the middle part is dominated by word units (or uninflected morphemes) derived from let -verbs (Russ. pust'/puskaj, *nehati > Pol. niech, Cz./Slk. nech, USorb. njech, Ukr. (ne)xaj, Sln. naj, other South Slavic neka) and usually vaguely classified as 'particles'. They can be either interpreted as analytic markers of mood-like distinctions of the volition-based domain (directives, optatives, but also causatives) or as left-edge clause connectives. The former can be motivated the better, the tighter the morphosyntactic connection with the clausal predicate and the higher the paradigmatic integration with already "established" moods (with a high degree of morphological fusion) can be shown to be. The latter becomes the more likely the more consistently these 'particles' are associated with clause-initial position, because this renders them suitable for paradigmatic replacements of otherwise well-established clausal subordinators (among them complementizers); see Figure 1.
In Slavic languages, this "lingering" status is well-attested. However, since uninflected morphemes deriving from 'let' do not show any signs of morphologization, but can easily occur clause-initially, their association with clausal connectives can become much stronger than with verbal moods. Balkan Slavic da is an exception, which, apart from being of an entirely different provenance, has acquired weak affix-like properties in verb-oriented clitic clusters, so that it occurs clause-initially rather "by accident". This development closely correlates with the preservation of da's irrealis function. In western South Slavic this functional restriction is lost; concomitantly, da has become the default factual complementizer. Thus, across South Slavic, da covers almost the entire range of "guises" regarding its format and status displayed in the right-most box of Figure 2. As yet it has not become a verbal desinence (even in Balkan Slavic), but it has entered etymologically complex function words (at the other end of the cline) even in Balkan Slavic, which probably derives from a process that must have begun prior to da's integration into verb-oriented proclitic clusters.
Tight morphological integration, close to fusion, can be observed with da's "irrealis counterpart" in North Slavic, namely with by, otherwise involved as kind of auxiliary in the "free" subjunctive/conditional: -by has become an inseparable part of whole series of 'particles' and other clause connectives in the particular languages. It becomes associated to complementizers (or other clausal subordinators) under the same conditions of linear placement in subsequent clausal conjuncts as have the auxiliary-like units derived from 'let' discussed above, however only as integrative part of word units. An exception is Pol. by since it can also serve as subordinator on its own (see ex. 2, 8, 12). Regardless, the subordinator function of by-connectives in contemporary Polish, and their delimitation from positionally flexible particles, is supported by the fact that person-number enclitics (associated to the l-form) almost exclusively attach to these connectives only if they occur clause-initially (not as particles). In this respect, a discrimination among minor word classes does not apply to niech, which can serve as host of by+person-number enclitics in any position within the clause (compare 35–37).
Graph: Figure 2 Mood and clause-initial connectives as expression poles of an 'irrealis space'
The proposal forwarded here would work at least in European languages, since in these complementizers usually take the form of 'function words' (including 'particles') that, as a rule, occur at the left edge of the clause ([
The clines shown in Figure 2 also allow to capture the diachronic processes which lead to the aforementioned analytical problems on synchronic levels. We use a model that allows us to compare products of these processes within single and across different (Slavic and other) languages without being apodictic about any presumed categories. Moreover, since all units (or constructions) on this united cline bear on notional contrasts from largely the same dimension, we can capture combinations of them (as in Romanian, see Section 2.2) as well as the fact that in some cases (languages, historical stages) one mode of expression is preferred over the other, and that this proportion may change.
It should be remembered that all of these interpretations (with their caveats) are drawn by linguists, they may be (and probably mostly are) unimportant for ordinary "language users". Notably, this relates also to the last big issue which, as it were, builds on top of the points made so far. The previous discussion has focused on the relation between mood markers, 'particles' and clause-initial connectives. This relation can be considered without asking whether the respective clauses in which these units occur can, or should, be treated as complements of predicates in a superordinate (a.k.a. matrix) clause. Noonan's definition in [
From a structural perspective, subordination implies embedding, which means that a syntagm X is dependent on a syntagm Y. Compare Lehmann's (1988: Ch. 1) definition:
[
This definition presupposes constituency (the constituent X is part of a larger constituent Y). It is at the foundation of practically all tests of subordination that have been proposed (see Section 3.2). Both constituency and embedding have been put into question as guiding principles, first of all, by cognitive linguists and by some functionally oriented typologists, mainly for two reasons. First, many criteria based on them cannot be applied cross-linguistically, or at least not universally, as they are based on rather specific structures of particular languages. Correspondingly, sets of criteria resemble checklists from which linguists choose particular features (because they apply to some construction), whereas they neglect or downplay others (because they cannot be applied or provide evidence contradictory with other features). It may not be obvious which criteria are more important (or necessary and sufficient) than others, and why (usually, this question is not even asked). Second, the structural criteria might prove to be only symptoms of more deeply rooted conceptual or pragmatic principles (including information structure); cf., for instance, [
As concerns the first objection, "checklist approaches" are indeed common in different strands of linguistics. In particular, they are applied to clausal complementation and complementizers as well (see below), but also to broader phenomena like 'subject', or 'privileged syntactic argument', and finally to 'grammaticalization' or the notion of 'word'. However, the actual problem seems to lie in determining feature hierarchies (in order to handle feature conflicts) and in understanding underlying principles for which the features chosen may well turn out only as symptoms. As concerns the second issue, it shouldn't be doubted nor ruled out that syntactic structures, in particular techniques and devices for clause linkage, have their roots in discourse pragmatics. Even after tighter constraints establish themselves, discourse-pragmatic factors continue to play a role in the exploitation of syntactically tighter constructions. For instance, extraction, or other marked focus techniques, can be accounted for as constituting pragmatic strategies to elicit attention. Likewise, the linear sequence of purpose, conditional, temporal or various complement clauses with respect to another clause in relation to which they are assumed to be subordinate often shows preferences (for pre- or for postposition), and insertion may be generally rarer because of communicative motives. Thus, in narration many temporal clauses tend to precede "their" main clause because there is a strong tendency toward an iconic correspondence between the sequence of clauses and the sequence of the related events. Likewise, in conditional sentences, the protasis naturally precedes the apodosis because it seems to be cognitively preferable to first state a condition and then the consequence. By a similar token, a reason why clausal complements of predicates related to cognition and speech usually follow "their" matrix clause appears to lie in a preference for first stating the act (and the subject) of cognitive judgment and/or the speech act before specifying its content.
Furthermore, there is no need to assume specific syntactic configurations as parts of speakers' mental representations in the grammar of their language. Yet the denial of such configurations does not entail that we cannot (and should not) assume different layers of structure that may sometimes get into conflict, but which also allow for contrasts (possibly useful in communication). These would not be available if tighter ("syntacticized") structures could not "go against the grain" of merely discourse-oriented expectations. In fact, there is some (certainly unintended) irony in Cristofaro's (2014) argument that "[m]any phenomena that are usually regarded as distinctive for subordination (...) do not actually provide evidence for such a category" (2014: 73). She discusses a couple of cases in which criteria like extraction, clause-internal word order, backward anaphora, or relative order of conjuncts and the range of interpretations (i. e. of semantic relations) of their combinations yield different results for comparable clause types in different discourse contexts. However, when she motivates the variable order of a purpose clause in relation to another ("superordinate") clause on the basis of the function which this clause has on the backdrop of the preceding discourse, Cristofaro cannot but presuppose some notion of embedding marked by some connective (usually at the beginning of the 'embedded' clause). Thus, following [
Another case discussed by Cristofaro is the complementizer taki in Sranan (an English-based creole). It derives from a say -verb (< Engl. talk), but has extended into contexts that are not directly related to speech reports; among other things, taki can be used with factive complements of emotive predicates. Since the evaluation of a fact can provide the main communicative purpose, and the propositional content of factive complements is logically presupposed (and thereby strongly associated to topic status), it seems natural that the complement with taki precedes the clause which contains the evaluative predicate, as we see in (
Sranan
(
comp pn neg kill pn make 1pl happy
'That Kofi didn't kill Amba made us happy.'
(from Heine and Kuteva 2007: 292; glossing adapted)
We may skip the question whether factive complements really tend to generally precede their matrix clauses (some observations run counter to this assumption, see Section 3.2). The crucial point to be made here (and glossed over by Cristofaro) is that these sentences contain a connective which "keeps" its place with the conjunct, regardless of which discourse motivation there may be for the linear sequence of conjuncts. What, then, is the reason for which we nonetheless may argue for a tighter structural bond of this clause with the preceding or the subsequent clause?
The answer lies in information structure: subordinate clauses do not have an independent topic-focus structure. In other words: a high degree of syntactic integration manifests itself in a unitary topic-focus structure embracing both clauses; cf. [
[
If the connective "migrates" with its conjunct, it becomes a conventionalized sign indicating this asymmetric relation, and can therefore be considered a subordinator. If there is no such sign, i. e. the connection is asyndetic, the same principle of a unitary topic-focus structure applies, and intonation is left to organize this structure. I skip over problems that arise for asyndesis since here we are interested in the analysis of explicit, primarily clause-initial, means marking clause linkage.
Definitions of subordination that are based solely on semantic-pragmatic terms suffer from a principled disadvantage: they cannot sufficiently discriminate synchronic variation and diachronic change. In fact, if subordination is defined as "a cognitive situation corresponding to the non-assertion of one of the linked SoAs" ([
Assertiveness is an important concept, as it unveils asymmetries between adjacent clauses (which may be prosodically separated) in terms of communicative salience. This asymmetry can be disclosed by targeting the asserted part, e. g. by question or negation tests. However, asymmetries in terms of assertiveness also characterize parenthetical comments, i. e. utterance parts that are neither asserted nor integrated syntactically, as well as introductions to (pseudo-)quotes. Compare the verb that is used parenthetically in (56c) with the same verb as a CTP in (56a) and as an introduction to imagined direct speech in (56b). The brackets mark the clause on which dumat' 'think' comments on:
Russian
(56а) V Rossi-ičastodumaj-ut,
in Russia- loc often think [ipfv]-prs.3pl
[čto [politik-a – del-o nesložn-oe]].
comp politics- nom matter- sg.nom uncomplicated- sg.nom
'In Russia, people often think that politics is not so difficult.'
(RNC; «Izestija». 2003.02.19)
(56b) On ot nee [pravdy o bolezni] prjačetsja,
duma-et : [drug-ie umiraj-ut, no ne ja].
think [ipfv]-prs.3sg other- pl.nom die [ipfv]-prs.3pl but neg 1sg.nom
'He hides from it [the truth about his illness], thinks: others are dying, but not me.'
(RNC; I. Grekova. Perelom. 1987)
(56c) Ladno, budem govorit' za ženščin. Xot' segodnja i ne 8 marta,
no, dumaj-u , [ocenj-at].
but think [ipfv]-prs.1sg appreciate [pfv]-fut.3pl
'Okay, let's speak for women. Although today is not March 8, but, I think, [they will appreciate it].'
(RNC; kollektivnyj. Forum: Mužčina v škole. 2011)
A question like Razve (tak)? 'Really (like this)?' targets the asserted part. With the CTP (in 56a) it can address either the matrix clause (dumajut) or only its complement. In the quotative construction (56b), a focus on dumaet is less easily available, the question rather addresses the imagined quote (in brackets); finally, in (56c) the question cannot address the parenthetical verb (dumaju), but only the clause which it comments on (ocenjat). That is, with a CTP and its complement clause (56a) the communicative fore-/background may switch, while with parentheticals (56c) this is excluded and they are always in the background, and quotative-like constructions (56b) seem to occupy an intermediary stage.
Consequently, assertiveness as a diagnostic of subordination becomes problematic when the conceptual level (to which assertiveness belongs) and the structural level are confused or used interchangeably. If it is events (situations) that are viewed as subordinate or non-subordinate ([
Russian
(
1sg.nom neg agree- (sg.m) as.if our people- (sg.nom) with year- ins.pl
v čem-to tam prozreva-et,
in something- loc there see.clearly [ipfv]-prs.3sg
čto-to v nem nazreva-et...
something- nom in 3sg.m.loc mature [ipfv] - prs.3sg
'I do not agree that over the years our people see somewhat more clearly, something in them matures...'
(RNC; A. I. Solženicyn. V kruge pervom)
On the other hand, the diagnostics of subordination based on information structure within chains of conjuncts appears to be crucial for more specific properties of structural dependency, in particular in complementation, such as limited possibilities of island extraction, or of the preposition or insertion of clauses with less established complementizers (see Section 3.2). For an elaborate discussion of tests on subordination cf. [
The points just made hold true for clausal complementation, too, as it is "included" in subordination. When it comes to determining whether a clause-initial connective can be considered a complementizer, different tests are variably suited. For some units, tests are successful only with a handful of verbs, and only with verbal attachment sites (e. g., Russ. točno with kazat'sja, slovno with kazat'sja and snit'sja; [
(a) Letuchiy considers obligatoriness not to be a very useful criterion. If understood in semantic terms (presence in a predicate's frame of government) it often cannot be tested; if understood in syntactic terms, it turns out unreliable for many syntactic actants (e. g., with bojat'sja 'fear'). In fact, both problems are opposite sides of the same coin (see Section 3.3). Notably, later Letuchiy points out that complement clauses allow for the omission of an object less freely than adjunct clauses. This applies even more pronouncedly for infinitival than for finite complements. This criterion is a facet of the obligatoriness issue, which is thereby admitted access via the backstage.
(b) If complement clauses are inserted in their matrix clause (center embedding), they anyway prefer the position immediately after their CTP (2020: Ch. 0.6.2). This observation applies to marginal (or emergent) complementizers as well, in fact it might point to a factor which favors their interpretation as complementizers. Although suitable examples are usually difficult to find in spontaneous discourse, constructed examples like the following with Sln. naj do not appear weird:
Slovene
(
message [n]-nom.sg opt rm the_quicker announce [pfv].prs-1pl
je bilo objavljeno na oglasni deski.
'The message that [we should show up as quickly as possible] appeared at the bulletin board.'
(by courtesy of M. Uhlik)
In fact, if the naj-clause were non-contiguous to its assumed attachment site, this would evoke Free Indirect Speech (FID).
Remarkably, emergent complementizers can pass the insertion test, but they are not used if the entire complement clause is preposed ([
Russian
(
say [ipfv] - pst-pl as.though as.though pn-sg.nom leave [ipfv]-prs.3sg
tol'ko dv-a čelovek-a.
only two -nom person -gen.sg
'Only two people were saying that Petja was going to leave.'
(
as.if 3-(sg.m.nom) suffocate [ipfv]-prs.3sg-rm
emu ne kaza-l-o-s'.
3sg.m.dat neg seem [ipfv]-pst-n-rm
intended: 'It did not seem to him that he suffocated.'
This also applies to Sln. naj ([
However, the avoidance of preposition does not seem to be conditioned by the specific function of the complementizer (e. g., evidential as in ex. 59–60), rather we may suspect general properties of information structure to be at work. Complement clauses with established complementizers that precede "their" matrix clause often turn out weird as well, e. g.
Russian
(
comp.irr come [pfv]-pst- (sg.m) my -nom.sg.m brother [m]-(sg.nom)
ja xoč-u.
1sg. nom want [ipfv]-prs.1sg
lit. 'To come my brother I want' (intended: 'That my brother comes I want').
See Wiemer (2021a: 115–119) for discussion.
(c) Restrictions on nonverbal attachment sites. These are not a very reliable indicator of complementizerhood. Some emergent complementizers appear unacceptable (or at least marked) with nominal attachment sites, while they sound normal with verbal attachment sites (e. g., Russ. vrode by, vrode kak with speech act or seem -CTPs; cf. Wiemer and Letuchiy 2021: 430–432). However, other emergent complementizers, e. g. Sln. naj, can introduce complement clauses to nouns (see ex. 58 above), for contemporary Pol. jakoby nominal attachment sites even constitute the majority in discourse (Wiemer 2015 b: 227 f.).
(d) Ability to introduce clauses that occupy the subject position in the embedding clause. Emergent complementizers may show restrictions here (or sound awkward/marked); for instance, Russ. kak by... ne only very rarely introduces a subject clause (denoting the reason of the apprehension), as in the following example:
Russian
(
'But our "patriots" are more concerned about how foreign firms might get in there and, God forbid, restore, say, the façade to their liking.' ('... are concernedlest...')
(RNC; N. V. Koževnikova. Sosed po Lavruxe. 2003; from Letuchiy 2021 b: 208)
By contrast, complement clauses in subject positions of factive predicates are much more common (e. g., Menja udivilo/udivljaet, čto 'It astonished/astonishes me that P'). Moreover, other marginal (or emergent) complementizers seem to introduce complement clauses in subject position quite freely; compare, for instance, Russ. budto (by):
Russian
(
'It was allegedthat he penetrated the African American management.'
(RNC; Žizn' nacional'nostej, 06/05/2002)
(
'In the same evening, Olga sat down in the wardrobe, in the morning it was announced to the whole apartment that she left the city in an unknown direction.'
(RNC; V. P'ecux: Škaf. 1997)
This point probably applies to clause connectives related to speech (and thus to propositional arguments) in general – which remains to be investigated. Moreover, a rehearsal of corpus examples gives the clear impression that subject complement clauses with emergent complementizers practically never occur before "their" matrix clause (see (b)); this also concerns factive clauses.
(e) Island extraction appears to be more easily available for complement than for adjunct clauses, but its use is often marked/or it is altogether unavailable, particularly in Slavic languages; compare:
Russian
(
what. acc 2sg.nom want [ipfv].prs-2sg comp.irr 1sg.nom 2sg.dat say [pfv]-lf-sg.f
'What do you want me to tell you?' ('what do you want me that I tell you?')
[RNC; M. Traub. ploxaja mat'. 2010; from Letuchiy 2021 b: 41)
(f) Complement clauses are more liable to employing tense as taxis (understood sensu largo); compare the temporal overlap in (
Russian
(
'I was surprised [lit. it surprised me] that I was treated like that.'
All in all, these standard tests and symptoms of clausal complementation often yield ambiguous results, and they constitute another example of a checklist approach to a gradable phenomenon. Nonetheless, we are left with two issues that, to some extent, are independent from information structure and which are particularly crucial for clausal complementation. First, the distinction between complementation and adverbial subordination hinges on the argument—adjunct contrast (or cline); second, subordinate, in particular complement, clauses have been claimed to be deprived of independent illocutionary force. I am going to finish my discussion with these two issues in this order.
In Noonan's widely cited definition, clausal complementation is conceived of as "the syntactic situation that arises when a notional sentence or predication is an argument of a predicate" ([
[
Roughly, approaches to determining argument relations divide into two groups, which I propose to call intensional vs extensional. Intensional approaches try to define argument relations via semantic paraphrases of predicates (i. e. of expressions able to function as predicates); from this perspective, arguments constitute variables that have to be specified in order to yield an exhaustive lexicographic description of the respective predicate. This method heavily relies on in-depth (or informed native speaker) knowledge of the respective language and is, thus, truly applicable only for language stages (and varieties) that are contemporaneous to (and actively used by) the linguist. Moreover, we know of many cases in which syntactic valency is smaller than the number of semantic valency slots established by the semantic paraphrase ([
Extensional approaches, in turn, basically rely on observed distribution, i. e. collocations, in real discourse. Their reliability is limited for another, as it were complementary, reason. Since [
Concomitantly, we face a problem which is shared with intensional approaches: "Phrases and clauses whose omission leads to ungrammatical sentences can be considered arguments, but the converse does not hold: the optionality of an element in a sentence does not automatically rule out argument status" (i. e. syntactic valency may be smaller than semantic valency). However, "sometimes the picture is complicated by the fact that, in the absence of any clear morphosyntactic criteria for argumenthood, obligatoriness is virtually the only clue that one could turn to" ([
For practical purposes, and to avoid vicious circles as well as the influence of prescriptive norms, Schmidtke-Bode, faute de mieux, eventually adopted a liberal baseline:
[
Actually, this "policy" opens the field for a broad grey zone of potential complementizers (or other complementation devices, e. g. non-finite verb forms), but it also mirrors the fact that only a handful of occurrences of an assumed, or potential, argument relation may not be particularly informative (as argued above).
What these opposite sides of the same coin lead to in practice can be seen in the following: certain types of clauses that start with specific connectives ('particles') can be observed quite regularly, though not very frequently, after particular verbs (or other predicative units) with which they are compatible. Compatibility may either arise from "semantic harmony" (or concord) with these predicates, or because the connectives in question add some reason, or motivation, to the situation described by these predicates (which is typical of factive complements of predicates denoting emotive states; see ex. 66–67). In such cases, could it be argued that these clauses fill out an argument position of these predicates, even if they occur with them only occasionally, and that, correspondingly, the clause-initial particles have turned into complementizers?
Let us look at some examples. Consider first Mac. da ne:
(
most_of_all rm fear [ipfv]-impf.1sg
da ne ja razočara-m.
irr neg 3f. acc disappoint [pfv].prs-1sg
'Most of all I was afraidnot to let her down.' (lit. '... lest I disappoint her.')
(from [
Plašam se 'be afraid' (as other fear -verbs) narrows down the function of the subsequent clause, which would supply a reason for the emotional state, but we need not assume that such a clause is required.
The next example presents us with an analogical situation:
Russian
(
a
ja
and 1sg.nom fear [ipfv]-prs.1sg-rm
kak by on ne sta-l pryga-t'.
comp.irr 3-(sg.m.nom) neg begin [pfv]-lf-(sg.m) jump [ipfv]-inf
'He shouts back at me and waves his arms, and I'm afraid he might jump.'
('...afraidlest he jump')
(RNC; A. Gerasimov. Žanna. 2001; from Letuchiy 2021 b: 115)
Rather the opposite situation applies in this case:
Russian
(
and 1sg.nom think [ipfv]-pst-(sg.m) about this- loc.sg.m
kak by ne povtori-t' žalkost'...
comp.irr neg repeat [pfv]-inf pity- (acc.sg)
'And I thought abouthow not to repeat the pity...'
(RNC; S. Šargunov. Priključenija černi. 2009)
Dumat' 'think' does not narrow down the function of the subsequent clause, but this clause undoubtedly fills an argument slot of dumat' o tom.
Notably, this holds true regardless of a possible ambiguity of the meaning relation between both clauses (if taken out of context): the second conjunct can either verbalize apprehension ('lest I repeat the pity') or a deliberative question ('how to avoid repeating pity?'). In the former case the combination [kak + by + negation] has lost transparency, in the latter case this combination remains transparent. Concomitantly, in case we are dealing with a deliberative question we wonder whether embedding occurs, since the kak by ne-clause retains its own illocutionary force. This becomes more apparent when the subject of the (purported) matrix clause and the speaker do not coincide, i. e. when we replace it with third person: Ona dumala o tom, kak by ne povtorit' žalkost' 'She thought about how not to repeat the pity' – the deliberative question-reading would then favor FID, and this has to do with the behavior of egocentricals (see further Section 3.4).
Domains associated to apprehension are particularly well-suited to show how meaning relations between clauses can arise – and again fade away; this corresponds to a possibly transient nature of complement relations with narrow classes of predicates. Consider (69–70): although usually kak by... ne has not been associated to purpose (Wiemer, forthcoming: Ch. 4.1.1), this sense can arise occasionally after verbs that denote attentive activity, since these readily evoke a goal (= purpose). Since negation is involved, the "goal" is meant to be avoided (or prevented). This leads to a meaning of apprehension or of negative purpose. Both meanings may overlap because shared semantic components condition each other, so that in particular discourse tokens these meanings can be difficult to distinguish.
(
look [ipfv]-pst-pl comp.irr neg allow [pfv]-inf death- gen
ėt-ogo čudak-a...
(this eccentric)- gen.sg
'Or maybe they look athow to prevent the death of this eccentric...'
('..how not to cause...')
(RNC; G. N. Vladimov. General i ego armija. 1994)
(
while driver [m]-(nom.sg) watch [ipfv]-pst-(sg.m)
kak by ne vreza-t'-sja v odn-o prepjatstvi-e,
comp.irr neg drive_into [pfv]-inf-rm in one obstacle- acc.sg
na puti vozniklo vtoroe.
'While the driver was watchingnot to crash into one obstacle, a second one appeared on the way...'
(RNC; «Vstreča» (Dubna), 23/04/2003; from Letuchiy 2021 b: 207)
Concomitantly, if there were some prosodic break after pogljadyvali or sledil, respectively, an effect of FID would become likely again, even if the subsequent conjunct were not interpreted as a deliberative question. The latter would imply kak by... ne as a transparent composition of the manner WH-word kak 'how', the irrealis marker by and the negation. This reading seems to be excluded (and one wonders why), but regardless of the different discourse functions, the question remains whether pogljadyvat' and sledit' require a goal/purpose complement (with negation then being transparent).
Alternatively, verbs may imply an epistemic (and emotional) attitude toward a posterior event. Consider Russ. ždat' or ožidat' 'wait': connectives such as kogda, poka are "harmonic" with this semantic element, as they allow to specify the moment when this event was/is expected to occur ([
(
1sg.nom simply wait [ipfv]-pst-(sg.m)
kogda vs-ё ėt-o konč-it-sja.
when all this- nom.sg.n end [pfv]-fut.3sg-rm
'I just waited for this to end.' ('...waitedwhen/until all this will end.')
(RNC; A. Gelasimov. Ty možeš'. 2001)
Compatibility (as between ždat' and kogda) should become obvious in shared meaning components. However, we notice that accepted authoritative semantic paraphrases, as in NOSS, account for the epistemic attitude, but they do not explicitly mention this time interval, or its closure. We might say that this closure is imposed by a subsequent clause which adds appropriate new information with a suitable clausal connective, all the more if this connective (here kogda) is already indicative of subordination. As a result, commonplace assumptions about discourse coherence (which cooperative interlocutors presume by default) force a meaning relation between the two adjacent conjuncts which only arises from implicatures.
Moreover, kogda (but not poka) can introduce factive complements if the clause denotes an episodic situation (e. g., Russ. Ona ne zametila / propustila, kogda... 'She didn't notice when...'). Furthermore, kogda "fits" after verbs which imply the potential (or habitual) occurrence of a situation (e. g., Russ. On ljubit, kogda... 'He likes when / if...'). Potential situations can also be marked with esli. The difference between kogda and esli becomes evident with factive CTPs: contrary to kogda, esli can induce the potentiality of the entire complex sentence ([
(
in Hollywood- loc nobody- nom neg surprise [pfv]-fut.3sg-rm
esli uslyš-it vopros:
if hear [pfv]-fut.3sg question- (acc.sg)
"Kakov vaš goroskop?"
'In Hollywood, no one will be surprisedif they hear the question: "What is your horoscope?"'
(RNC; «Sovetskij ėkran». 1960)
Esli can introduce complement clauses also of a few verbs implying emotions and the possibility of an alternative ('S can happen/be chosen or not'): Russ. vozražat' 'object', byt' protiv 'be against' (usually in rhetoric polar questions) and, again, bojat'sja 'fear' ([
Are observations like these to be taken as illustrations of insufficient lexicographic accounts (of ždat', zametit', etc.)? In all likelihood they are not, since such observations are too recurrent to reflect accidental inconsistencies. It is more likely that we are observing manifestations of an intricate mechanism on the semantics-pragmatics interface, i. e. of an interplay between coded and inferred meaning. Such issues are calling for intensional approaches; from an extensional perspective the issue is what would frequency counts of occurrences of such clause pairs tell us. In particular, is it reasonable to assume any thresholds for the occurrence of clause pairs in which the second conjunct provides a meaningful continuation to the preceding conjunct, more precisely: to a predicative expression that does not seem to require (or imply) that continuation by all means? The meaning relation can be evoked by context, but how many such occurrences would be "sufficient" to assume that a tighter bond (with the given connective in the second conjunct) has arisen, or otherwise: That an invited implicature has become sufficiently conventionalized, so that the connective can be considered a complementizer? Expectability after particular semantic types of predicates is an issue – and it is connected to frequency – but it causes problems: first, frequencies differ for different types of predicates (and for each particular predicative lexeme), and, second, predictions about the occurrence of particular clause connectives have to be calculated on the backdrop of some 100 %-expectability of occurrence provided all conditions are known (see fn. 41). Obviously, such a calculation seems utopic, even if good corpora are available.
Apart from that, connectives like those considered above do not lose their more specific (temporal, conditional, etc.) semantics under any condition ([
Subordination is considered to bar independent illocutionary force of the embedded clause ([
Here I will consider only cases related directly to complementation, i. e. mainly units regarded as standard factual complementizers (e. g., Russ. čto, Pol. że, Sln. da) and units derived from 'let'-verbs which indicate directive or optative speech acts. Both types of units can follow each other. See examples (
Polish
(
'The old man replied that [may he rot even in prison].'
(PNC; T. Dołęga Mostowicz: Znachor. 1988 [1937])
Slovak
(
'I gave him the green paper and the guy wanted to kiss my hand. I told him that [may he not dare do such a thing].'
(SNC; Listy do zahraničia. 2000)
Upper Sorbian
(
'Today the teacher said before the prayer that [God the Lord may warn the country about
disturbances].'
(from Scholze 2010: 388)
Russian
(
'She lived as a hermit, believing that [let everyone remember her as elegant, beautiful and young.'
(RNC; I.Ė. Kijo. Illjuzii bez illjuzij. 1995–1999)
Slovene
(
'They concluded that work should continue [lit. maythe work continue] everywhere in all commissions and subcommissions.'
(Gigafida; Delo, 1998) (cf. also Sonnenhauser 2021 for ample evidence of such examples)
The parts in brackets can be ascribed their own illocution, which can correspond either to a quote (
Polish
(
'‒ Please take it. Only here they have such cookies. I nodded that [let it be].'
(PNC; W. Myśliwski: Traktat o łuskaniu fasoli. 2007)
Hardly can kiwnąć be assigned a propositional argument, but że can even mark off clauses that do not serve as clausal arguments of any assumable CTP, as is demonstrated in (
(
'Mother tried to stop him, that [may he wait, she will make his bed for him] (...).'
(PNC; W. Myśliwski: Traktat o łuskaniu fasoli. 2007)
Here we are dealing with interpretive use connected to the quotative function of że, as analyzed meticulously by [
However, in actual occurrences it is often not easy to distinguish the clause-linking and the (pseudo-)quotative function of że. The same applies to its functional equivalents in other languages, although some such units, like Russ. čto, have not been attested as quotative (or interpretive) markers on their own. Thus, in examples like (
The same holds true for the Polish niech-clause in (
This gradience can give rise to oscillation (as presented in Mendoza and Sonnenhauser, this volume), but it also shows, again, which conditions can serve as a door-opener to complementation and complementizers: the immediate linear precedence of predicative expressions which can (or must) have arguments of clausal format. If this format and the preceding predicative expressions denote content that has to do with speech acts (or their semiotic equivalents) and their evaluation, such a combination can easily be interpreted as complementation with a propositional argument, and the unit at the beginning of the second conjunct as a factual complementizer. Other clause-initial units, like those deriving from 'let'-verbs, have to abandon their directive/optative function, or else they remain restricted to CTPs that denote such speech acts (like 'beg', 'ask', 'demand') and keep a non-factual function related to reportive evidentiality. This is what has been demonstrated by [
If both complementizer and (pseudo-)quotative use of such a unit have the same diachronic origin (see Pol. że), we naturally ask which one was first attested diachronically, or whether both developed in parallel as the result of a bifurcation from that common source, or otherwise: from a stage in which the predecessor(s) of że in modern Polish (or its equivalents) served a diffuse (resp. oscillating) function of marking off new discourse units, each of which had its own illocution (and possibly its independent attitude holder). We are not concerned here with the directionality of change, but possibly uses as in (74–78) are manifestations of such an earlier diffuse usage. More important from a synchronic (or rather: panchronic) perspective are some other issues.
To begin with, even when there is an established default complementizer which can be distinguished from a (pseudo-)quotative marker (of whatever origin), this complementizer can introduce quotes. This concerns factual and interrogative complementizers likewise. Examples for factual complementizers (Pol. że, Russ. čto, Sln. da) were provided above. An illustration of an interrogative complementizer was given for Pol. czy in example (
Polish
(
(= 52) 'I'm asking, could you please fuck off?!' (lit.... whether you could fuck off)
(PNC; Samo życie, odcinek 294. 2002–2010)
Pytać only denotes a request for information, while the czy-clause codes a directive speech act (request for action), its presumable intonation is suggested by interpunction. The speaker obviously refers to an earlier directive speech act of their own. This is why the whole sentence resembles a performative utterance: Directive illocutionary force is in what is encoded in the complement of first-person pytam 'I am asking', which at once describes that speech act.
This may be a very specific example, but there are better known cases in which clauses introduced by acknowledged complementizers contain an imperative. Such cases occur consistently in Slovene; cf. [
Slovene
(
say [pfv]-pst-pl aux.prs.3pl 2sg.dat con bring [pfv]- imp .sg beer- acc
'They told you that you bring the beer.' (lit. '... (You) bring the beer!')
(
say [ipfv]-(prs.3sg) con rm if 2sg.dat aux.prs.3sg hot
slec-i .
undress [pfv]-imp.2sg
'S/He says that, if it's too hot for you, you may undress.'
(lit. undress!, with the če-clause separating the reflexive clitic from the verb which it belongs to)
Many examples discussed by Dvořák serve as reminders of previous speech acts (on singular occasions), which brings them close to quotatives (see above). This relation to previous speech acts also applies to imperatives that occur, as it were, embedded in WH-questions (see 84) or in relative clauses. This, however, is by no means a rule, as examples like (
(
ptc aux.prs.1sg 2sg.dat say [pfv]-lpt-(sg.m)
kam se skri-j !
where.to rm hide [pfv]-imp.2sg
'I did tell you where you have to hide (, didn't I?).' (lit... where hide!)
(
dem-nom.sg.m rel 3sg.m.dat send [pfv]-imp.2sg help [f]-(acc)
naj je bo tudi potrében.
opt 3sg.f.gen be. fut.3sg also in.need.of- (nom.sg.m)
'That one whom you provide help should also need it.' (lit....whom provide help!)
Which conclusions are we to draw from facts like these? The imperative is the grammatical form that denotes directive speech acts par excellence. If they appear subordinated – in particular in complement clauses, which are said to exclude independent illocutions – are we to conclude that the imperative changes its function, with a concomitant loss of directive illocutionary force? What would this function be: an interpretive one, which primarily serves to recall earlier directive speech acts? We saw that this does not apply in all cases (see 85). Or are we to say that we are not dealing with embedding, but then where would we go with the argument requirements of speech verbs to which in their majority these imperative-clauses seem to attach (see 82–84)? Or should we say that subordination does not (or not always) "absorb" independent illocutions? This most radical conclusion would be tantamount to giving up one of the hitherto accepted hallmarks of subordination.
To my mind, the solution of this conundrum lies in the way we treat quotation, which – via associated domains of interpretive use – constitutes another gradient with cognition-based domains, such as reported speech and reportive evidentiality, and the development of specialized reportive complementizers. Approached from this angle, the use of Pol. że as a factual complementizer appears as a special case within a broader functional domain: że just serves as a device to mark off clauses. The illocutionary content of these clauses can vary, and że simply serves to organize information structure into blocks. This is in line with Topolińska's (2008a) original claim about że as an "all powerful introducer of new clauses", which also takes into account that że serves as a prosodic host of 2P-enclitics with which it may coalesce. However, an analogous line of thinking should be expanded to da in the western part of South Slavic, where it has become the default factual complementizer, but still acts as clitic host, among others for enclitic bi- (see Section 2.3). In addition, it should apply also to default factual complementizers of yet other provenance, basically in East Slavic (compare Russ. čto and cognate units). Moreover, this line of reasoning should be expanded to more recent, or still emergent, complementizers like Sln. naj: their complementizer function obviously has been evolving from directive speech acts; these can be "attached" to suitable speech act verbs, and this attachment may remain loose (→ discourse dependence), either with or without the "mediation" of an established factual complementizer (see 51, 53, 74–78), or it may develop a tighter bond with that CTP. After all, the latter point is difficult to analyze, because it hinges on the distributional characteristics of expectability which were discussed in Section 3.3.
This leads us to the next issue. Uninflected units originating from 'let'-verbs (pustiti/puskati > Russ. pust'/puskaj, *nehati > Pol. niech, Slk. nech, Ukr. (ne)xaj, Mac./Bulg./SCr. neka, Sln. naj) have been qualified as complementizers (or as adverbial subordinators) only exceptionally. In fact, only Sln. naj seems to have been analyzed as an emergent complementizer; cf. [
(
'He just came up to him one day and said – let her stay.'
(RNC; A. Gelasimov: Žanna. 2001)
Pol. niech (lack of interpunction as in the original)
(
marynarz to kapelan powiedział by niech szczęści mu się na niebieskich oceanach pełnych
okrętów podniebnych...
'Please explain what it is about because I do not understand. If a sailor died tragically, the chaplain would say, let him be happy in the heavenly oceans full of ships under the sky...'
(PNC; Usenet – pl.soc.religia, 27.08.1998)
(
'They were three, the seven of us. We immediately jumped up to them and told them may they give the girls peace, and then we escorted them so that nothing would happen to them.'
(SNC; Slovo. Bratislava, 2003)
One can try to do away with such examples by saying that they just indicate quotation (see above). However, sometimes we observe person-deictic shifts that are characteristic of reported speech; compare (
Slovene
(
say [pfv]-prs.1sg 2sg.dat good- adv think.over [pfv]-imp-pl
'I tell you, think it over well.'
(Uhlik and Žele 2018: 225)
Zero complementizers are not a very illuminating concept; in fact, such constructs will remain unverified for as long as we cannot formulate water-tight criteria on when "omission" applies (so that zero realization acquires a meaning). Alternatively, one can try to show that the clause-initial particle was reanalyzed as a complementizer because of its expectability. This is probably what happened to South Slavic da in Old Serbian, following the analysis in [
Old Serbian (13th c.)
(90a) Tavisterek-l-idasestane-mo
this 2pl.nom aux.prs.2pl say [pfv]-lpt-pl irr rm meet [pfv].prs-1pl
(90b) Ta vi ste rekli │da se stanemo. juxtaposition
'Well, you told us. Let's meet.' (two monoclausal sentences, see [a])
(90c) > Ta vi ste rekli da se stanemo. complement clause
'Well, you told us to meet.' (one biclausal sentence, see [b])
(Stare srpske povelje i pisma; from Grković-Major 2004: 198)
The generalized schema for this reanalysis is given in Figure 3:
[a] [... verb / event noun]
[b] > [[... CTP]
Graph: Figure 3 Reanalysis: clause-initial particle > complementizer
It should be emphasized that the initial structure can still be maintained and, again, the distinction between [a] and [b] poses a problem for linguists rather than for "naïve" speakers (cf. [
This brings us to a third issue. An analogous argument can be raised for units that follow on standard complementizers with an epistemic and/or evidential function, e. g. Pol. jakoby, Russ. budto (by), vrode 'as though'. Notably, practically all such units derive from similative markers (comparison); for the semantic connection see Section 2.6.
Russian
(
'It seemed as if they saw each other yesterday.' ('... that as though they saw...')
(RNC; «Zvezda». 2003)
(
(= 54) 'And there is also such a rumor that (as though) punishers are coming to our village again.'
(RNC; A. I. Panteleev. Nočnye gosti. 1944)
(
'They say that Osip Emilievich Mandelstam once brought down the stairs a colleague who came to complain that his poems are not being published.'
(RNC; Otečestvennye zapiski. 2003)
(
'The media wants to hammer into people's brains that the people are afraid of the coming of the communists to power.'
(RNC; Sovetskaja, Rossija. 23.08.2003)
(
'Isn't it about you printed in the newspaper that it seems like you rescued the girl, Marusja Loginova, from under the car.'
(RNC; M. Sergeev: Volšebnaja galoša. 1971)
(
'And what is this Chonkin doing there? – He's standing there, – Paxomov shrugged his shoulders. – They say that/as if he even got married.'
(RNC; V. Vojnovič. Žizn' i neobyčajnye priključenija soldata Ivana Čonkina. 1969–1975)
Polish
(
'I entered the enormous building of the Warsaw Headquarters, which was rumored to be [that as though it was] so tall that one could see Siberia from the cellars.'
(PNC; M. Sokołowski: Gady. 2007)
(
'Ambassador Grzybowski did not accept the note. He also rejected the thesis contained in it as though the Polish state ceased to exist.'
(PNC; Eu. Duraczyński: Rząd polski na uchodźstwie 1939–1945. 1993)
We can either say that the respective units found themselves in clause-initial position because a default factual complementizer was "dropped", or simply because they could be reanalyzed as complementizers after suitable predicative expressions in the immediately preceding clause. This reanalysis might have been supported by paradigmatic replacement conditions with formerly established complementizers (see Figure 1). It will be a future task for corpus-based diachronic research to clarify what really happened.
After all, complementizers that emerge from similative expressions might be less conspicuous than expressions from the directive-optative domain because no initially volition-based illocutions are involved that might be "absorbed" by embedding. In addition, similative expressions are able to scope over propositions prior to becoming complementizers, as they can introduce clauses with representative illocutionary force in isolated clauses, i. e. as a particle (99a-b), and in adjunct clauses (100a-b) as well. See the following Polish examples with jakoby, whose similative function has become obsolete, and its cognate jakby, which is widely employed in this function in modern Polish:
Particle (no subordination):
(99a) Jesteśjakobychory 'Allegedly, you are ill.' reportive (personal knowledge)
(99b) Las wyglądał na gęstszy, bardziej skudlony. Imadło nieba i ziemijakbypoluzowałoszczęki.
'The forest looked denser, fuzzier. The vise of heaven and earth, as it were, loosened its jaws.'
(PNC; M. Olszewski: Chwalcie łąki umajone) similative
Introduction of adjunct clause, only similative:
(100a) Old Polish
Ach myloscz, czosz my vczinyla, eszesz me tak oslepila (...), yakoby ch nykogo na swecze znal.
'Oh love, what have you done to me, that you have blinded me (...) as if I didn't know anybody (else) on earth.'
(SłStar; MacDod 43. 1408; from Wiemer 2015 b: 249)
(100b) Modern Polish
Napis zabłysł nagle, jakby biała dłoń Rysia oświetlała go przelotnie.
'The inscription flashed suddenly, as if Rysio's white hand was illuminating it fleetingly.'
(PNC; J. Iwaszkiewicz: Brzezina i inne opowiadania. 2006)
That is, contrary to connectives derived from 'let'-verbs, similative sources of reportive complementizers do not cause any "clash of illocutions" between different attitude holders, since similative connectives have always been related to the cognition (knowledge/belief) domain based on perception. As a consequence, when it comes to complementation, no shift of illocutions (forced, as it were, by embedding) occurs, different attitudes affect only epistemic or evidential functions and are, thus, restricted to propositional content. Probably for this reason particles based on perception-cognition have never been considered as markers of mood. In all other respects, i. e. as for syntax, the conditions for a reanalysis (clause-initial particle > complementizer) are identical to those of 'particles' connected to the directive-optative domain.
I have attempted to disentangle the relation between units (or constructions) to be regarded as operators on the reality status of utterances: complementizers, mood auxiliaries and related connectives with more flexible syntactic behavior ('particles'). All of them modify the illocutionary and/or propositional content of clauses, but they differ in terms of their integration into syntactic, partially also morphological contexts, i. e. as for the format of the construction in which they partake. Standard definitions and approaches toward these types of units (or constructions) are often too vague, and thus indiscriminative, in particular with respect to these formats. This is a central reason of their unanimous, often even contradictory or incommensurable treatment in the linguistic literature. Another important point is that only the notion of 'complementizer' is inherently related to subordination. Therefore criteria that are useful to distinguish auxiliaries from particles, including those at the left edge of clauses (see Section 2), are of another nature than those that prove essential for assigning complementizer status to "left-edge particles" (see Section 3); see further below under Figure 4.
The aim of this paper has not been to present a universal solution for the treatment of mood auxiliaries and complementizers, and how they relate to other connectives. Since probably the only feature that unites them from a crosslinguistic perspective is a notional one – the modification of the reality status of the utterance – any attempt at discriminating them without criteria based on form and morphosyntactic distribution is deemed to fail. I am agnostic as for whether concepts of mood (auxiliaries) and complementizers can be defined that are at once comparative and contrastive with regard to each other. Instead, the task has been to disclose (a) the conditions under which representatives of these expression classes, if contrasted in principle, are difficult to distinguish and (b) the conditions which prepare the ground for diachronic changes. Slavic languages supply ample material to illustrate such conditions. These have been pointed out from a usage-based perspective here, which is to mean that the analysis, though, of course, not theory-neutral, does not make any specific formal assumptions about projections, specifier positions, and the like. It also does not make any claims as for how speakers (subconsciously) interpret the relevant constructions in which reality status is manipulated.
Thus, my proposal assumes that, unless we want to drop at least one of the aforementioned notions (because they turn out as superfluous or are explained in a vicious circle), we can discriminate them in Slavic and, more generally, European languages by treating auxiliaries of 'analytical moods' and "left-edge connectives" (including complementizers) as expression classes on opposite poles of syntactic integration, each with its typical representatives and a large transitional grey zone in between. Both particles and complementizers, on the one hand, and auxiliaries, on the other, are typically word units, but while the latter are understood in their relation to the verbal predicate (without fixed word order), the former tend toward the initial position of the clause; in addition, complementizers are considered to head these clauses, although for unrelated reasons (see below). Anyway, regardless of their (morpho)syntactic relation to the verbal predicate, non-factual moods (i. e. anything except the indicative) and non-factual complementizers, as a rule, restrict the choice of tense-aspect (da in Balkan Slavic) or otherwise of formal characteristics (the l-form or infinitive in North Slavic) on this predicate. These restrictions remain salient wherever a unit on the gradient between mood auxiliary and left-edge connective "moves", unless this unit loses its non-factual (= irrealis) meaning, in which case it tends to occur in a fixed clause-initial position (see da in the western half of South Slavic).
After all, both complementizers and auxiliaries usually derive from various kinds of uninflected connectives of different provenance, among them trunks, contaminations or otherwise petrified verb forms, most prominently from 'let'-verbs (Russ. pust'/puskaj, *nehati > Sln. naj, Pol. niech, etc.). In this respect, complementizers and auxiliary-like units of purported analytic moods are parasitic on such particles, which are themselves products of fossilization. Thus, the crucial question is to which extent such fossilized units interact in morphosyntax, including their prosodic treatment. This becomes even more apparent when we realize that the gradient between auxiliaries and left-edge connectives interferes with another gradient, namely: the degree to which a morpheme with irrealis function undergoes morphologization either with an already existing clause-initial connective or with the verb (see Figure 2). The latter has to some extent occurred with da in Balkan Slavic, as it has become an integral part of verb-oriented proclitic clusters, in which it occupies the leftmost slot (Section 2.3). The former is true of the morpheme by in North Slavic (inherited, in combination with the l-form, from the Common Slavic subjunctive/conditional), which, as originally a strict 2P-enclitic, has univerbalized with uninflected connectives (as its former prosodic hosts). However, even incorporated in left-edge connectives, -by keeps its requirement of the l-form (or the infinitive) in North Slavic and it continues behaving as a host for person-number markers in West Slavic. These properties are almost lost for heterosemic cognates that do not function as subordinators, but as particles with flexible position in the clause (compare, e. g., Pol. jakoby, Russ. kak by), so that we also observe a tendency toward the discrimination of minor word classes on the basis of their morphosyntactic behavior (Section 2.4).
For these reasons, a discrimination of auxiliary-like units and complementizers or, more broadly, irrealis-sensitive particles is problematic when such units occur clause-initially (see Figure 1 and ex. 19 for Macedonian). In practice, discussions about analytic moods and their discrimination from clause connectives have concerned their possible status as complementizers almost only for South Slavic (first of all, Balkan Slavic) da and clause-initial connectives with morphologically incorporated –by in North Slavic. These two units function in both the volition- and the cognition-oriented domain. Otherwise analytic mood has predominantly been viewed in contrast to irrealis-sensitive connectives that are only loosely integrated into clausal syntax ('particles'), but not in contrast to subordination (an exception is clause-initial and non-incorporated Pol. by, another, more recent one is Sln. naj). This restriction concerns notional distinctions which are, first, marked by units younger than by and da and which, second, are based on volition and intention. That is, analytic moods have been an issue practically only if directive and optative illocutions are concerned; representative illocutions (together with their epistemic or evidential modifications) and interrogatives, all of which imply propositional content, have hardly ever raised discussions on mood. Nonetheless, the general problem of discerning complementizers from other clausal connectives applies to units related to the cognition domain as well. In this domain, units with a provenance related to perception and comparison (similatives) figure most prominently (e. g., Russ. (kak) budto, vrode, Pol. jakoby, Mac. kako da, SerBoCroatian kao da); see Sections 2.6–2.7.
Figure 4 provides a graphic illustration of the conceptual relation between complementizers and mood auxiliaries jointly with other notions relevant for the rise of complementizers. What still needs to be explained is their relation to subordination.
Graph: Figure 4 Connectives, subordination and 'analytic moods'
The relation between uninflected morphemes that can potentially turn into auxiliaries of analytical moods, on the one hand, or into left-edge particles, on the other, can be considered in the confines of simple clauses. As explained above, frequent clause-initial occurrence, possibly discontinuous with the predicate in the same clause, may turn into a factor that favors the reinterpretation of such units as complementizers, and purported mood auxiliaries also have ample chances to occupy this favorable position in their clause (see Section 2.5). However, whether deriving from such emergent auxiliaries or from left-edge particles, such units can count as complementizers only if sufficiently clear symptoms indicate that the clause with such a clause-initial connective is embedded in another clause. In addition, the left-edge connective must be in semantic concord with a predicative expression in that other clause (= CTP) and be sufficiently expectable from the valency requirements of that expression. Since however these requirements are difficult to test in a methodologically impeccable way and the majority of properties ascribed to subordination in general and to complementation in particular cannot always be tested, and they derive from distributional properties anyway, clausal complementation more often than not remains a matter of degree. In addition, many clause-linkage devices behave as complementizers only with a very narrow, although predictable class of predicative expressions; apart from particles (e. g., Cz. ať, Russ. budto, vrode), this includes acknowledged subordinators of clausal adjuncts, namely temporal and conditional conjunctions. Practically all left-edge particles (which we have focused on here) that show properties of complementizers under specific conditions, are also used in independent clauses (Section 3.3).
Subordination as such is seriously challenged when it comes to contiguous combinations of units acknowledged as default factual complementizers with markers of directive or optative illocutionary force (e. g., Pol. że niech, Sln. da naj, Russ. čto pust') or with the imperative (e. g., in Slovene), or of interrogative complementizers (e. g., Pol. czy) in clauses with an obviously independent directive illocution (Section 3.4). This issue cannot be solved by analyzing empirical facts and distribution, instead a principled clarification is required concerning one's approach toward quotation and interpretive use (or echoic deontics) and toward the decisive role of illocutions. Concomitantly, illocutionary force should be considered a stronger egocentrical than person-deictic expression, with the consequence that illocutions that are apparently embedded (since they occur with a unit considered to be a default factual complementizer) easily produce effects of Free Indirect Speech. This is tantamount to a perspective clash of different attitude holders. This effect is less conspicuous if the clash does not affect illocutions based on volition, but only epistemic attitudes. Such a clash can occur in clauses that contain 'particles' operating on propositional content, and restricted to representative speech acts, in combination with default factual complementizers (Pol. że jakoby, Russ. čto budto, etc.). However, such clause connectives have never been taken into account as potential auxiliaries of analytical mood (see above). This again shows that the mood issue (and its delimitation from clause-initial particles) has largely been restricted to the volition-based domain (directives, optatives), which interferes with causation.
Now, in case these insights are convincing enough, which consequences do they have for linguistic analysis? First and foremost, we have to do justice to several gradients and the fact that many properties typical for expressions at either end of the respective gradient can be detected and described only on the backdrop of larger distributions. This is important especially for units (constructions) situated in large grey zones between the respective poles of a gradient. Distributional properties should somehow be measured against each other and with an account of paradigmatic replacement conditions and syntagmatic combinability of the relevant units within their respective larger frame (morphemes or just segments in words, words in clauses, clauses in pairs or series of conjuncts). On the one hand, gradients concern morphosyntactic and prosodic relationships; they are at work in morphologization clines (word/free morpheme > clitic > affix: agglutinated > fused) which create new 'function words', such as clausal connectives discussed here. On the other hand, gradients apply for syntactic dependency and expectability from discourse in linking clauses into larger units. Syntactic dependency (and embedding) appears as an extreme case of expectability from discourse (jointly with unified topic-focus structures of adjacent clauses). Grey zones pop up from time to time because meaningful relations between clauses may be only partially, or vaguely, based on semantic requirements of some predicative expression while context-triggered implicatures do the rest of work for the interlocutors to achieve a relevant, thus coherent, contribution to the illocutionary purpose (or, in addition, the propositional content) of adjacent clauses. If, then, a clause making such a contribution contains a unit (particle or conjunction) at its left edge, this unit can be interpreted as a complementizer provided its function is "in harmony" with the meaning relation at issue and some predicative expression in the adjacent clause.
Now, only frequent recurrence of such combinations of clauses can make this complementing relation more salient and expectable, but it is not clear how expectability should be quantified. A main reason is that we cannot assume that different meaning relations arise, and different predicative expressions are employed, with comparable frequency; it is therefore not clear how to define a baseline for an assessment (Section 3.3). A viable road out of this methodological impasse might be to start with conditional frequencies, which account for the occurrence of some element (e. g., a left-edge particle) under specific contextual conditions; cf. [
Another conclusion arises from the observation that complementizers based on left-edge particles clearly divide into two groups concerning their "illocutionary provenance". In one group we find left-edge particles related to perception, epistemic stance and information source (= cognition domain). Consequently, these are tightly associated with propositional content and representative (declarative or interrogative) speech acts. In the other group we find left-edge particles that are indicative of directive or optative speech acts. Such illocutions are based on volition, or intention, and the respective clauses do not code propositions. This division is very clear-cut, even Sln. naj has not (yet) expanded into the domain of propositions and representative speech acts (let alone its much less "advanced" cognates all over Slavic or Russ. pust'/puskaj), nor do cognition-related units expand into the domain of directive or optative speech acts, and practically all apprehensional markers in complementizer function keep their connection with negative purpose (thus, volition).
However, regardless of this division between cognition and volition domain and their respective associations with illocutions, units of either group can follow on the connective which, for the particular language, is considered the default factual complementizer (Russ. čto, Cz. že, etc.). Apart from the principled question whether non-representative illocutions can be embedded – an issue that should be clarified prior to the testing of empirical facts (see above) – this recurrent phenomenon might raise the impression that such default complementizers function as door-openers for more specific complementizers (considering all the other conditions discussed throughout this paper). Nonetheless, it would be premature to assume "complementizer ellipsis" (e. g., Russ. Æ
Finally, it is quite possible that in many concrete discourse tokens the categorial status of the units in question (complementizer vs auxiliary or particle) cannot be determined unanimously. This would demonstrate that their employment has become, or remained, diffuse, and this favors oscillation – provided these categorial distinctions correspond to different syntactic structures (cf. [
I want to thank Daniel Weiss and a very constructive anonymous reviewer for their useful comments. I am also obliged to Anca Găţă for consultations on Romanian. Of course, the usual disclaimers apply.
1, 2, 3 first, second, third person; acc – accusative, adv – adverb, aor – aorist, aux – auxiliary, comp – complementizer, con – connective, conj – conjunction (adverbial subordinator), dat – dative, def – definite article, emph – emphatic marker, f – feminine, fut – future, gen – genitive, imp – imperative, impf – imperfect, inf – infinitive, ins – instrumental, ipfv – imperfective, irr – irrealis, lf – l-form, loc – locative, lpt – l-participle, m – masculine, n – neuter, neg – negation, nom – nominative, nvir – nonvirile, opt – optative, pfv – perfective, pl – plural, pn – proper noun, prs – present, pst – past, ptc – particle, q – question marker, rep – reportive, rfl – reflexive pronoun, rm – reflexive (light) marker, sbjv – subjunctive, sg – singular, vir – virile
By Björn Wiemer
Reported by Author