Diachronic evolution of the subordinator kak in Russian

3.1 Kako in independent clauses

Historical dictionaries and grammars give the following list of functions for kakъ/kako in OR (the original form of kak; both phonetic variants are referred to below as kako) in independent sentences: a question word of manner (13), more rarely – a question word with ablative, temporal and reason meaning, also used in polar questions of surprise (14) and exclamations, in the complex e(g)da kako meaning ‘suppose’ (15) (Slovar’ 1975; Sreznevskij 1893).^[1]

Questions of indignation and surprise may be introduced by kako with the interrogative particle li.

This meaning can also be introduced by constructions da ne kako (lit. ‘let neg how’), ci kako (lit. ‘whether how’). Another specific construction is the use of kako in the beginning of an independent sentence (often with the particle da) in the meaning ‘let it be as [God will/Saint Virgin will etc.]’:

Note that in (13) and (16) kako preserves the manner meaning, while in constructions like (14) and (15) it does not: the focus is on the truth value/epistemic value of the sentence, and not the manner of never seeing his son/repentance. Thus, the constructions in (14) and (15) semantically involve propositions (see Section 3.2.2 for details). Both of these uses are lost in MR.

3.2 Complement clauses with kako

3.2.3 Distribution of the functions of kako in complement clauses

As shown in Sections 3.2.1 and 3.2.2, the list of functions of kako in complement clauses is much larger than in MR, including not only eventive (18) and manner complements (17), but also propositions (25), irrealis complements with the conjunctive mood and the infinitive, see (20)–(21) (this is summarized in Scheme 1 below). We aimed to identify the stages of evolution of these meanings and uses of kako and identify which meaning or use is basic at a given period, and which meaning or use is diminishing or increasing. To attain this goal, we counted the frequencies of various uses of kako in a limited sample of texts, following the methodology proposed by Hilpert and Gries (2009) and Diessel and Hilpert (2016). We compiled a sample of several large manuscripts starting from the earliest available documents of the 11th century up to the 16th century and retrieved all the examples with kako (omitting the use of kako as an indefinite pronoun with the meaning ‘somehow’). The sample was constructed in such a way as to include documents from various periods. To make our sample consistent from the point of view of language varieties, we limited our research, first, to documents created in the north or center of Russia and, second, to original documents (translations were excluded). The list of the documents is given in Appendix 1. For the 18th century, we used a randomized query from the RNC.

Scheme 1:

Functions of kako.

It must be specified that even though OR is sufficiently represented in corpora (the historical subcorpus of the RNC and the OR part of the TOROT Treebank accessible at https://torottreebank.github.io/), the corpus queries are hardly useful when trying to conduct semantic analysis of OR. Most often, subordinate clauses and subordinators allow for more than one interpretation, given that most of the subordinators and particles had a much larger distribution among the types of subordination than in the modern language (Bulaxovskij 1958: 398–399). Moreover, it has been argued that in OR texts, like in other languages of this period, the notion of a sentence, or a complex sentence, is itself problematic. The texts are not constructed as sequences of (complex) sentences but as paragraphs containing narrative chains (Keršiene and Preobraženskaja 1983). Each member of the chain was introduced by a specific particle or subordinator (often both a particle and a subordinator), and sometimes it is impossible to decide which clause is the embedding clause, the one on the left or the one on the right. Thus, the semantic interpretation of each example requires an analysis of the larger context, and the context of one or two sentences does not offer enough information to decide their exact meaning. This is why we did not base our analysis on the material of corpus queries for the 11th–16th centuries, and only used it for the material of the 18th century (texts from the 17th century have not been studied). For other periods we have studied the original manuscripts. Solitary ambiguous examples have been rejected. All this explains why our sample is rather small: it includes 68 examples for the earliest period (11th–12th centuries), 108 examples for the second period (12th–14th centuries) and 73 examples for the 15th–16th centuries. For the 18th century, we made use of a randomized corpus query, taking the first 396 sentences. The results on the distribution of complement, adverbial and independent clauses in the whole sample are represented in Figure 2 in Section 4.1.

Let us now focus on complement clauses, which are examined in detail in Table 1. The whole number of complement clauses comprises 35 examples from the 11th–12th centuries, 57 examples from the 12th–14th centuries, 28 examples from the 14th–16th centuries and 19 examples from the 18th century. The distribution of complement clause types is represented in Table 1 and Figure 1. This table is a fragment of the larger Table A1 given in Appendix 2. Here we have omitted the complement types that are represented by a small number of examples and are irrelevant to the present discussion: first, the indirect question of degree (“I did not imagine how big the project could be”) and irrealis complements (31); both emerged in the 18th century. We have also omitted ambiguous examples allowing both eventive and propositional reading. As was discussed in Section 3.2.2, these two meanings cannot always be distinguished, and many corpus examples are ambiguous between the two readings. Since our article is focused on the distribution of events and propositions, we omit the ambiguous examples in the discussion below Table 1. The results are visualized in Figure 1.

Table 1:

Distribution of major complement types among complement clauses with kako.

Simple/complex complementizer	The single complementizer kako			Complex complementizers with kako	Total number of examples in columns to the left
Morphosyntactic properties	Complements with the INF/CONJ	Indicative complements		kako li, e(g)da kako, kako ti
Semantics	(%)	Propositional (%)	Event and manner (%)	(%)
11th–12th centuries	3 (9.4)	2 (6.3)	22 (68.8)	5 (15.6)	32
12th–14th centuries	17 (32.7)	12 (23.0)	21 (40.4)	2 (3.8)	52
15th–16th centuries	7 (25.9)	2 (7.4)	15 (55.6)	3 (11.1)	27
18th century	0	2 (12.5)	14 (87.5)	0	16

Figure 1:

Proportions of complement types in the four periods of OR (the plot shows the observed proportions with a 95 % confidence interval according to a multinomial logistic regression model, with ‘period’ as an ordered quadratic predictor; see the Supplementary Materials for details).

Figure 2:

Proportions of independent sentences, complement, adverbial clauses and other types of clauses with kak(o) (the plot shows the observed proportions with a 95 % confidence interval according to a multinomial logistic regression model, with ‘period’ as an ordered cubic predictor; see the Supplementary Materials for details).

It must be taken into account that the resulting sample of complement clauses with kako is rather small hindering any solid statistically-based conclusions. However, our main goals are, first, to show that the discussed uses are attested in the discussed periods, and that they are not marginal. Second, even if small nuances in figures are not telling, we may at least observe basic tendencies in the development of functions, i.e., their expansion or loss.

Complement clauses with kako have been attested since the 11th century (Shevelёva 2009), and our results show that they could not only encode events and manner complements (the reason for grouping these two meanings is given in Section 3.2.2), but, in principle, all the major types of complements. More specifically, these include propositions, purposive-like complements with the conjunctive mood or with the infinitive, infinitival complements of modal verbs, complexes with the particles ti and li and with the subordinator e(g)da. These complexes were used to encode irrealis prospective complements (events and propositions), polar indirect questions and complements of the verb ‘fear’, respectively.

The statistical results visualized in Figure 1 reveal the expansion of infinitival (conjunctive) complements in the 12th–14th centuries and their gradual loss in the following periods; in the 18th century they are completely absent from our sample. Along with the expansion of the infinitival complements the event/manner complements decreased in frequency in the 12th–14th centuries and constantly increased afterward. In the modern language, these two types of complement clauses are the only ones preserved from the OR period.

As for propositions and complements with complex complementizers, the change in frequency is too small to arrive at robust conclusions and does not exceed the expected values (the values expected under the null hypothesis, that is, the hypothesis of random frequency). It is well-known that the analyzed complex complementizers were completely lost by the 18th century. However, the data on propositions offers an interesting result, which has not been yet mentioned in the literature: namely, that kak could introduce propositions in the 18th century. Given that in the modern language it is not possible (Arutjunova 1988), the complete loss of this meaning must have happened in the 19th century.

In sum, in complementation we do not observe a spread of functions of kako, but a narrowing of functions, see Scheme 1. It must be specified that MR has developed a complex complementizer kak budto (by) encoding irrealis complements of perception verbs (Letuchiy 2015):

(31)

…mne	po-kaza-l-os’,	[ kak	budto	ja	sejčas	tol’ko	rodi-l-sja].
I.dat	pv-seem-pst-n.sg-refl	how	as.if	I	now	only	be.born-pst(m.sg)-refl
‘It seemed to me as if I was just born.’
(RNC: M. Saltykov-Ščedrin. Nezaveršennye zamysly i nabroski. 1869–1872)

However, this is the only context in complementation that arose in Middle Russian. The process of grammaticalization of kako as a complementizer involves a narrowing of its functions, as shown in Scheme 1. This does not fit well with the scenario proposed in (i): this scenario predicts that the first function to develop from manner complements is that of eventive complementizer. Thus, we would expect to find examples of kako as a manner complementizer and eventive complementizer at some stage of language development, while other uses would be found in more recent documents. Instead, it seems that kako first propagated to all types of complements and then narrowed its functions to eventive and manner complements as shown in Scheme 1.

Thus, in the 11th century kako functions as a complementizer with a general meaning, which is specified in the larger context; note, however, that even at this point the manner/event use is the dominating one (as it covers 67 % of its use).

3.3 Adverbial clauses with kako in OR

For kako in OR, Sreznevskij (1893), Istrina (1923), Lomtev (1956), Bulakhovskij (1958), Slovar’ (1975), and Morozova (1983) list the functions of a “comparison marker” (a standard marker in similative and implicit equative clauses, see Kobozeva and Serdobolskaya 2021), an adverbial subordinator of time, condition and manner, a marker of parenthetical clauses with the meaning ‘source of information’ and as forming part of a complex subordinator with the particle by (which occurs in irrealis complements as in (21) and purposive adverbial clauses). The adverbial functions of kako that we have identified in our sample follow the list of Slovar’ (1975) and Sreznevskij (1893). Temporal clauses with kako have the meaning of immediate sequence (‘as soon as’) as in (32) or specification of time. This latter can denote the beginning of the situation (‘since’) as in (33) or the overlap of two situations in time (34); both include clauses making a more precise indication of time used with time-denoting nouns and expressions.

These functions are also observed in MR; however, with restrictions concerning the ordering of clauses. Conditional clauses with kako, not attested in MR, are only found in OR.

Clauses of reason (36) are marginal and stylistically restricted in MR, see the comment to (12) above.

Adverbial manner clauses are widely attested both in MR and in OR:

Even though historical dictionaries claim that kako is a comparison marker, Kobozeva and Serdobolskaya (2021) show that in the 11th–14th centuries the use of kako in comparison is restricted to equative clauses involving an implicit parameter:

The subordinate clause in (38) refers to the exact conditions of selling the wine, which is explicitly indicated in the context. It must be emphasized that it is not about similarity; rather, the speaker aims at identifying the extent of a parameter he did not mention explicitly (conditions and price). In Kobozeva and Serdobolskaya (2021) this meaning is introduced as “an implicit parameter equative”. They show that other comparison contexts are chiefly introduced by aky in the 11th–14th centuries. Starting from the 14th century, kako expanded into the contexts of aky, and in Middle Russian it is largely used as a similative and equative standard marker (in terms of Treis 2017).

Purposive clauses may be introduced by the construction with kako and the conjunctive mood or the infinitive with the irrealis particle by (Istrina 1923: 196; Lomtev 1956: 526–532; see also above):

Our sample also includes several instances of kako in complex adverbial subordinators such as ašte ti kako ‘if’ (lit. ‘if EMPH how’).

In sum, adverbial subordinate clauses with kako involve both propositions (conditional and reason clauses) and events (manner, similative, purpose and temporal clauses).

The exact figures for the above-described constructions are given in Table 2. This Table is a part of the larger Table A2 given in Appendix 2. Here we omit a number of adverbial clause types: first, all the examples with a potentially ambiguous interpretation (e.g., a well-expected ambiguity between condition and temporal clauses, reason and temporal clauses); second, complex subordinators including kak(o) as a part (since they are independent subordinators, and their functional distribution is different from that of plain kak[o]). The percentage in parentheses is calculated from the number of examples in the next to last column. The last column in italics gives the whole number of adverbial clauses for each period (see the details in Table A2 in Appendix 2).

In OR kako has more functions in adverbial subordinate clauses, in comparison to MR (see also Bulaxovskij 1958), and Table 2 shows that its frequency drops in three types of clauses. Conditional clauses with kako are not attested in our sample after the 15th century, clauses of reason become stylistically restricted already in the modern period, and purpose clauses were lost by the 18th century. The other three types, marked by grey color, remain widely used in MR.

The loss of conditional and reason clauses can be explained as follows. Semantically, conditional and reason clauses involve propositions, while temporal, manner and similative/equative clauses are eventive (see Section 3.2.2). Thus, eventive subordinates are widely used in MR, while propositional subordinates (conditional and reason clauses) become severely constrained in use or lost. In this way, adverbial clauses demonstrate the same restrictions in MR as complement clauses: in OR kako is widely used to introduce both propositions and events, while in MR kak is restricted to events. Note that propositions are attested in both complementation and in adverbial clauses even in the 18th century.

Infinitival purpose clauses disappeared along with the loss of this construction in complementation (with manipulative, speech commands and modal verbs) by the 18th century.

There is one type of adverbial clause that developed after the 16th century and is widely used in MR – i.e., similative, equative and related meanings. By related meanings we refer to comparative clauses (inequality comparison), additive constructions (“X, just as Y”) and role meanings (“he applied as a student”). Kobozeva and Serdobolskaya (2021) show that these constructions were raised in the 17th century, where kak(o) expanded into the contexts previously covered by aki(aky), replacing the latter. In that function, kak spread quickly, which can be seen from the fact that in the 18th century, this type’s frequency was the same as the sum of all other types in Table 2 (48.3 %).

4.1 A specification of meaning of a complementizer kako

In Section 3 above we showed that in OR kako is widely used both in independent and in subordinate complement and adverbial clauses. The exact distribution among the three types of constructions is given in Table 3 and Figure 2. (The full version of this table including a detailed elaboration of the columns “Independent clauses” and “Other types” is given in Table A3 in Appendix 2.)

Figure 2 shows that complement clauses and independent sentences with kako are by far more frequent in OR texts than adverbial clauses. The increase of adverbial clauses is observed in the 18th century concurrently with the gradual decrease of complements and independent sentences in frequency. At this point, many old functions of kako in complementation and independent clauses were lost (conditional, complex kako li, e(g)da kako, kako ti) or diminished in use (propositional complements). By contrast, in the domain of adverbial clauses kako developed the use of a marker of a standard of similarity/equality and expanded into the contexts of aky (which completely disappeared towards the 19th century), see Kobozeva and Serdobolskaya (2021). Furthermore, in the 18th century, kak developed a large number of other functions hardly attested in older texts, such as a parenthetical use, a large number of new complex subordinators, specific constructions with verbs, and a use as a conjunction introducing NPs with role meaning, and with the meaning ‘example’, see Table A3 in Appendix 2. It seems plausible that concurrently with the decrease of kako in complementation, it underwent semantic specification as an eventive/manner complementizer.

We shall now focus on the specific complement types and especially on the two grammaticalization scenarios sketched in the introduction. As discussed above, the scenario associated with the semantic analysis in Arutjunova (1988) involves the development from manner complements to eventive complements (i), while the scheme suggested in Boye and Serdobolskaya (2018) involves the development of both eventive and propositional complements from manner complements (ii).

Synchronic polysemy of event and manner complements in MR seems to point to the grammaticalization scenario proposed in Arutjunova (1988), see (i). The basic explanation for this meaning shift is the expansion of internal characteristics of the carrying out the action (manner with ‘sing’ and ‘dance’, speed with motion verbs, etc.) to the description of the situation as a whole. However, assuming this development scheme, we would expect kako to be used in manner complements, then expand to eventive ones, while propositions may be the next step of development. Assuming this development pattern, we would not expect to find propositional complements in the earlier time periods. This does not conform to our data. By contrast, Boye and Serdobolskaya’s (2018) hypothesis predicts that we may simultaneously find eventive and propositional complements at the earliest time periods, which is our case: in OR kako introduces all the major semantic types of complement clauses, including events and propositions.

Thus, it is most likely that kako did not follow the scheme (i), but first propagated to complement clauses in general (as predicted by (ii)), covering all the major complement types in OR, and then specialized on the event and manner contexts in MR. This specification happened gradually, after the complete loss of several complex complementizers and the general decrease of the frequency of kako in complement clauses.

Another point is how exactly the complementizer use was grammaticalized in specific complement types. Both the schemes (i) and (ii) assume that eventive (propositional) complements grammaticalize from manner complements such as I know how he got to the metro (2) (which, presumably, developed from indirect questions such as I asked how he got to the metro based on a “from question to subordination” shift, cf. Heine and Kuteva 2006: 204–243). This explanation seems to be the most straightforward one for manner complements: manner is the core meaning of kako in OR, and it is likely to have grammaticalized from independent manner questions. However, for other complement constructions observed in OR we offer another explanation based on the fact that similar constructions are present in independent sentences.

For example, it is most likely that e(g)da kako ‘suppose’ with verbs of fear (24) grammaticalized from apprehensive contexts involving e(g)da kako in independent clauses (15), kako li in polar questions (22) developed from kako li in independent questions (14), and kako with infinitives/conjunctive mood (19) grammaticalized from purposive clauses hosting the same construction (39).

Manner is known to be the vaguest of all situation modifiers (such as time, place, condition, reason, etc.), and manner expressions are well-known to develop a range of meanings that are hard to interpret and largely depend on the semantics of the verb they modify (Švedova 1998: 149). For example, in Russian kak is not only used in questions about a parameter of a situation like speed in (40a), but also in questions about a situation description, termed by Filipenko (2003) as “situation structure” ((40b) and (40c)), and even about some obligatory participants of a situation (40d).

Questions similar to (40b) and (40c) usually inquire about a description of the situation as a whole, not just one of its aspects like manner in (40a) and route in (1) above. An appropriate answer to (40b) requires a whole story describing the weather, route and visited places, undertaken activities in (40c), while an appropriate answer to (40d) is a proper name.

Thus, the manner meaning of kak is expanded to the situation as a whole, including all its arguments and modifiers. This use arguably gives rise to the eventive reading, where kak does not denote manner, but encodes the event as a whole (see also Arutjunova 1988: 115–117). Therefore, the eventive reading of kako does not arise in complementation but is developed already at the stage of independent questions, contrary to scenarios (i) and (ii). The proposition reading is also observed in independent sentences with kako: namely, propositions are introduced by kako li in polar questions (14) and e(g)da kako in ‘suppose’-clauses (15). These uses could have reinforced the propagation of propositional complements with kako at the stage of its grammaticalization as a complementizer. In the same vein, we suppose that the multifunctionality of kako in adverbial subordinates is the result of the grammaticalization of independent questions with the same meaning, see Section 4.3.

As a consequence, eventive and propositional complements of kako are both attested in the earliest documented periods.^[2] Then several uses are gradually lost and the meaning of kako specializes on manner/event contexts. The specification of this meaning could have been reinforced by the specific type of context analyzed in Section 4.2.

It should be taken into account that the proposed development scheme describes the facts observed in the period covered by written sources, that is, since the 11th century. We are not aware of the earlier facts of the language and can only speculate about them. It might be speculated that the hypothesis (i) is still plausible and that before the 11th century complements with kako first followed the “manner > event” semantic shift, then the “event > proposition” semantic shift, thus leading to the polysemy observed in Section 3.2.3. In the course of further development around the 11th century, this process might have turned backward and the functions of kako might have gradually decreased.

4.2 A possible bridging context

Following Heine and Kuteva’s (2019) scenario, it can be hypothesized that kako got fossilized as a complementizer following the scheme “question word in independent sentences > question word in indirect questions (embedded clauses) > complementizer”. However, we suggest that the specification of kako as a complementizer may have happened through a specific bridging context (in terms of Evans and Wilkins 1998). Our sample has a substantial number of examples (e.g., in the 11th century there are four examples) involving the following syntactic structure, see (29) and the following example:

The syntax of these examples poses the following problem: it is unclear whether the stimulus of the mental verb is actually ubogyixъ ‘the wretched’ or the clause introduced by kako ‘how they are lying pierced by raindrops (similarly, ‘birds of the sky’ or ‘how they do not sow or harvest’ in (29)). Intuitively, it seems that both of them are, but that contradicts the idea of one-to-one correspondence between arguments and the slots in the verbal argument structure.^[3] The second argument could be an afterthought, but this is primarily expected in colloquial examples, and not in the written language, such as hagiographies and chronicles (note that these “double object constructions” are absent from birchbark letters which are generally believed to be stylistically close to colloquial speech).

Another solution is that the dependent clause in (29) and (41) is a relative clause modifying the NP ubogyx (the wretched who are lying etc.). However, there are (almost) no relative clause examples with kako in either our sample or historical dictionaries and grammars (except for manner relativization as in (35)).^[4]

In MR the written translation of (29) and (41) would require a change in the syntax of the construction, most probably a relative clause (a word-by-word translation would yield a colloquial sentence, unacceptable in the written language). However, similar examples are possible with nouns taking a clausal argument (with a role of content) and denoting a document or an oral message: “a tale that…”, “a story that” etc.:

We suppose that these “double-object” constructions could serve as a bridging context consolidating kako as a complementizer:^[5]

It can be hypothesized that at first, the double object constructions involved two independent sentences with an anaphor in the second one (“Look at the wretched. How they are lying pierced by raindrops!”), then they were reanalyzed as constructions involving a double object in the same sentence. Next, the noun replaced the anaphor in the second clause and the clausal argument remained as the only stimulus of the complement-taking verb (“Look COMP the wretched are lying, etc.”) following Lightfoot’s (1979) principle of avoiding structural complexity.

It is noteworthy that similar constructions, in which a how-clause contains a pronoun referring back to the PP in the previous context, are documented in Old English; they are termed as CLAN-constructions in López-Couso and Méndez-Naya (1996):

Note that in Modern English the question manner word how is also used both in manner (46a) and propositional complements (46b) (“declarative” how in Huddleston 1971):

4.3 Development of adverbial clauses with kako

The considered data raises the question of why a manner question word could be used as an adverbial subordinator covering a rather large number of adverbial clause types: note that kako could introduce conditional and purpose clauses, which is impossible in MR, and clauses of reason, which are stylistically and syntactically restricted in MR. We explain this by the multifunctionality of kako in independent questions: in OR interrogative kako could be used not only in questions about manner (13), but, as Slovar’ (1975) states, also in questions about time (47), cause (48), and ablative; with the particle li it is used in rhetorical polar questions of indignation (14). Such a multifunctionality of manner words is well-expected, since among other ontological categories manner is the vaguest one and is susceptible to semantic shift.

There is no doubt, however, that manner is the core meaning of kako, since manner questions prevail in number in all kinds of texts (e.g., 10 examples among the whole 38 examples in PVL) and it is the first meaning in all historical dictionaries, while other interpretations are much rarer. However, it is important that kako could also be used in other types of questions. These contexts gave rise to indirect questions of various semantic types, which were then grammaticalized in adverbial clauses, following the scheme “from question to subordination” proposed in Heine and Kuteva (2006).

The remnants of this polyfunctionality are observed in MR: questions about reason and condition may sometimes be introduced by kak, as well as rhetorical “indignation” questions:

In (49) the answer includes a description of the motivation; in (50) the condition of success. Of course, both examples have a nuance of manner and do not constitute the core of the reason and condition questions; however, the answer to both of them presupposes an explanation of the reason/condition of situations under discussion.

Thus, the functions of kak in MR are beyond the internal characteristics of the situation or proper manner. However, word-by-word translations of (47) and (48) would be ungrammatical in MR. Thus, we claim that kako has undergone a gradual specification of meaning both in independent questions and in adverbial clauses, similar to complement clauses.