Selectives (“topic markers”) on subordinate clauses

2.1 Toward a set of hypotheses about the functions of selectives

Authors of grammars and dictionaries make highly different choices as to whether or not calling the elements considered in this article “topic markers”, “topicalizers” or the like, which is due to entirely different traditions in the literature about what topics are considered or not considered to be.

Topic and focus (formerly also termed “theme” and “rheme” or in Czech základ ‘basis’ and ohnisko ‘focal point’; Mathesius 1939) are traditional notions of linguistic information structure, also called functional sentence perspective. Non-topic parts of sentences are called “comment” and non-focus parts “background”. A classical definition of topics is the aboutness-definition: “the speaker announces a topic and then says something about it” (Hockett 1958: 201). Due to the close association of the understanding of subject and topic going back to Aristotle’s hypokeimenon, translated to Latin as subiectum, aboutness is often understood in such a way that the topic is a sort of semantic or psychological subject and occurs as a variable in the valence of the semantic predicate (Jacobs 2001: 647). This makes it difficult to accommodate adjunct topics, often called “hanging” or “dangling” or “Chinese-style” topics (Chafe 1976; Li and Thompson 1976), which is why Haiman (1978: 585) entirely rejects the aboutness-definition. Haiman (1978) has been very influential in the descriptive tradition but has largely been ignored in theoretical approaches to topics, where various versions of aboutness continue to thrive. Jacobs (2001: 650) interprets aboutness as addressation (Jacobs 2001: 650) and Reinhart (1982) has suggested that topics function like file cards as units under which information is stored (see also Krifka 2008: 265).

Given that selectives tend to be extremely frequent, we must expect them to have a considerable range of different functions in all or at least most languages where they occur but notice that “different” does not mean “entirely different” here, but rather “similar”. Massive crosslinguistic comparison is useful for investigating which functions are more generally relevant and which are entirely parochial.

I will now introduce some likely candidates for functions of selectives in form of hypotheses that the empirical part of this study can relate to. The hypotheses are summarized in Table 1.

Selectives are likely to have something to do with the flow of information in speech and since the flow of information on the most superficial level of language structure is identical with linear order, a straightforward assumption is that selectives will strongly correlate with word order. Reasonable hypotheses are that selectives tend to occur early in the sentence (H[ypothesis] 2), but late within the constituents they are associated with (H3) (recall Ameka’s [1991] term “terminal marker” for Ewe).

H2 actually contains an even more basic hypothesis H1 which is that the domain of selectives is not the clause and not discourse units larger than the sentence, but rather the sentence (or, more precisely, a unit containing one main clause with its associated subordinate clauses). This entails that selectives will be largely indifferent to units of the clause such as the grammatical relation subject. Thus, I do not mean the same thing as Hockett (1958: 201) when he speaks of “the topic of the sentence” (this is rather topic of the clause when he discusses subject) – or Maslova and Bernini (2006) when they speak of “sentence topics” in a broader sense than what is meant here. The specific domain sentence is also one reason why “European-style” topic markers such as English as to, concerning, when it comes to, and French en ce qui concerne, quant à are excluded here, because these markers and constructions often have wider discourse scope than just the sentence that contains them.

H3 (“Selectives tend to occur after or at the very end of their constituent”) presupposes a connection to some kind of syntactic constituent. Selectives can be expected to be usually associated with a constituent of some kind (H4). If H4 is correct, this means that selectives are of high interest for empirical syntactic studies, since they are a kind of contrast agent for constituency. It follows from H4 and H2 that if selectives do not come particularly early, this can be because they mark long initial constituents, which is most efficiently achieved if the marker is positioned at the end of that constituent (H3). Selectives can have scope over more than just initial words. This is illustrated in (4) from Nalca with two selectives following two different nominalizers that trigger two different genders, the first one at the end of a headless relative clause (double underline), the second one at the end of the complement clause (underline).

It can further be expected that selectives mark constituents of very different length (H5) – compare Examples (1) and (4) – which will entail a high degree of psycholinguistic complexity.

If selectives, as this study intends to show, occur on constituents or “hosts” of very different kinds, ranging from noun or determiner phrases including pronouns to subordinate clauses and including arguments and adjuncts, it can be expected that selectives will have a high degree of freedom of host selection (H6), which is a typical property of clitics. Selectives can hence be expected to exhibit clitic-like behavior (but not necessarily phonologically). Since high degree of freedom of host selection is rather rare, it can be expected that the increase of freedom of host selection will characterize the grammaticalization processes of selectives whatever their sources (H7).

Marking constituency entails that selectives will have such effects as grouping (H8) (indicating which words form a constituent together), delimiting (H9) (indicating where a constituent ends or starts) and linking (H10) (indicating that something else is expected to follow). Linking is related to the notion of projection in interactional linguistics (see Schegloff 1996, and Tanaka 2015 for Japanese wa), but – due to H1 – largely restricted to the realm of the sentence. Delimiting and linking may seem to be contradictory at first glance but are in fact not. If there is nothing that follows there is no need for delimiting what precedes.

The hypotheses above are very general and highly underspecifying. They apply to many more environments than those where selectives are actually attested. It must therefore be expected that the use of selectives is constrained semantically or pragmatically. This article has no ambition to flesh out a comprehensive semantics of selectives, which is actually impossible given its focus on subordinate clause domains. The expected outcome of the empirical study is that different types of subordinate clauses differ in their propensity to take selectives will result in some sort of hierarchy that can be interpreted semantically (scales are simple semantic maps, see Croft 2003, Ch. 5).

Selectives are usually absent from units that bear illocutionary force (assertion, interrogation, command) (H11). This restriction is most important for clauses. Many complement clauses, for instance, have illocutionary force of their own, so complement clauses may be expected to be less prone to display selectives than other kinds of subordinate clauses. (However, as will be discussed in Section 3, selectives can anticipate the illocution associated with the main predicate – question or command – which testifies to the linking function of selectives.)

As already suggested in the introduction, selectives are expected to express selectivity (H12). As far as subordinate clauses are concerned, selectivity is akin to condition rather than concession. The apodosis of a conditional constructions applies only partially: under the condition that the protasis is true. The protasis has thus a selective function. Concessive clauses, however, do not restrict the occurrence of events reported in main clauses. Hence, if “selectives” express selectivity (H12) they can be expected to occur more often with conditional than with concessive clauses.

2.2 Contrast

In the vast literature on contrast, a frequent distinction made is the one between (i) semantic opposition (John is tall, but Bill is short) and (ii) denial of expectation (John is tall, but he’s no good at basketball) (Lakoff 1971: 133).^[4] The examples also differ between “double contrast” (two pairs of alternatives) and “single contrast” (one set of alternatives; Umbach 2005: 218). Unified approaches to (i) and (ii) include those of Umbach (2005), who suggests the confirm+deny condition (one part of the contrast is explicitly or implicitly affirmed and one part is explicitly or implicitly denied) and Winterstein (2012: 1883), who argues that two elements are contrastive if they are counter-orientated toward an argumentative goal. For our purposes, the distinction between semantic opposition and denial of expectation is relevant, because pairs of selective-marked constituents in contrast sequences can be semantically opposed, as in (1) “ You are from below; I am from above”, but need not, as in (5), where it is not “my house” and “you” that are contrasted.

In (5), the selective constituents (underlined) are semantically independent of each other. The first one introduces the topic of the whole contrast sequence, the second one narrows down the range of application of the counter-orientated statement.^[5]

Mann and Thompson’s (1988) Rhetorical Structure Theory operates with the two units “nucleus”^[6] (roughly corresponds to a semantically non-hypotactic clause) and “satellite” (a semantically hypotactic clause). According to Mann and Thompson (1988: 248), the contrast relation has two nuclei; it is thus not a hypotactic relation. For our purpose it is important to notice that the two nuclei can have different properties, which is why it is useful to label them distinctively. I will call the first contrast nucleus “non-adversative” and the second one “adversative”; thus, “contrastive” [both contrast nuclei] and “adversative” [the second contrast nucleus] do not mean the same thing. The distinction allows us to discuss such questions as to whether Dimasa distinguishes non-adversative (de) and adversative (la) selectives, as Example (5) suggests (Jacquesson 2008 speaks of non-adversative and adversative topic markers). Example (6) illustrates that the notions non-adversative and adversative apply to clausal points of departure in the same way as to nominal ones.

A further frequently posited use of but and other adversative connectives is concessive. According to Hermodsson (1978: 59–60) and König (1991), concessives are negatively causal or “incausal”. Umbach (2005: 227) argues that “there is no ‘concessive but’ just as there is no ‘causal and.’ The meaning of but does not include incausality, just as the meaning of and does not include causality.” This is a problem for Rhetorical Structure Theory where a discourse relation cannot be contrast (two nuclei) and concessive (satellite and nucleus) at the same time. As we will see, it is very common across the languages of the sample to use adversative connectives as translation equivalent for European-style concessive subordinators. The concessive use of adversative connectives, however it should be interpreted, is thus highly relevant for the study of interaction of selectives and contrast across subordinate clause domains.

Many languages have several different adversative connectives, but according to Winterstein (2012: 1866) this is not an argument against a uniform analysis of English but; rather other languages “sport connectives with meanings that are more specified”. Contrast is so far rather poorly investigated crosslinguistically, hence general statements seem to be premature. However, there does not seem to be any clear borderline between contrast and lack of contrast. For instance, in the N.T. but is often used in an inflationary manner at the beginning of new episodes (But when Herod was dead… 40002019 eng-x-bible-kingjames vs. Now after Herod had died eng-x-bible-lexham) which is probably due to translation effects from Koine Greek de and Latin autem (Defuncto autem Herode…). In fact, Classical Greek de has similar functions as selectives. However, as will be argued in Section 3, what makes it differ from selectives is that it does not occur in the non-adversative part of contrast sequences. It is similar to the adversative selective la in Dimasa in (5) and (6), but there is no non-adversative selective in Koine Greek.

2.3 Subordination

There are many conceivable types of subordinate clauses. Kortmann (1997: 81) distinguishes 32 interclausal relations only for adverbial clauses. No typological investigation of subordinate clauses is comprehensive, and this study is no exception. Since we know at least since Haiman’s (1978) seminal article that selectives are expected in conditionals, this study focuses first of all on various kinds of conditional clauses. Table 2 lists the types of subordinate clauses considered in this paper – the order anticipates the one of Figure 1 in Section 4. The full list of clauses is given in Appendix E.

Figure 1:

Selectives-across-subordinate-domains tendency scale.

By subordinate clauses of a certain type I mean constructions that display translation-equivalence with such clause types irrespective of their syntactic status. As far as adverbial clauses are concerned, this is well in line with Matthiessen and Thompson’s (1988: 301) hypothesis that “enhancing hypotactic clause combining” [which is their term for adverbial subordination] “has evolved as a grammaticalization of rhetorical relations in text of the enhancing Nucleus-Satellite kind”. In many languages, constructions are used that exhibit the hallmarks of what is called “cosubordination” in Role and Reference Grammar: shared operators and switch reference (Van Valin 2005: 187). Coordination cannot be excluded either. In Cuicatec and Sierra de Juárez Zapotec, the selectives have the same form as the coordinator ‘and’.

My approach to subordination is well in line with the one adopted by Cristofaro (2003: 34–48), who has launched the “Asymmetry Assumption” according to which the state of affairs expressed by the subordinate clause is pragmatically not-asserted irrespective of the structural properties of any clause type. Non-assertion of subordinate clauses also fits H11 “Selectives tend not to mark clausal units carrying illocutionary force”. Note also that according to Maslova and Bernini (2006: 68) topics are kept outside the scope of assertion.

Functional equivalence of subordinate clauses across languages makes Cristofaro (2003) rely on translation, which is well in line with the parallel text approach used here. However, Reesink (2014) has argued against Cristofaro that idiomatic translation cannot be relied upon. I agree with Reesink (2014: 259) that “different translation possibilities for various examples” is a relevant issue, but it can be addressed by using the cumulative evidence of multiple examples rather than just isolated ones. Hetterle (2015: 36) argues that Cristofaro underestimates the continuous character of the subordination–coordination distinction. For practical reasons, I use “subordination” as a convenient cover term for the clause types that are traditionally subsumed under subordination. This does not necessarily imply syntactic subordination or a cognitive construal of non-assertion. Finally, I fully endorse Hetterle’s (2015: 146) point that linguistic variation in clause-linkage phenomena should be investigated in a bottom-up way, this is exactly what the parallel text approach is good for.

2.4 Conditionals and other types of subordinate clauses

Conditionals (consequence relations) differ from many other interclausal relations in that they are not veridical (Asher and Lascarides 2003: 157). Veridical means ‘and’-like in the sense that the relation is true if and only if both members of the relation are true. Concessives, most temporal clause constructions, purpose and result and cause (explanation) are all veridical, which makes them more amenable to coordination-like construals than conditionals. Non-hypotactic construal of conditionals is possible, however, if the protasis is framed as a question (which also entails non-veridicality).

4.1 Introduction

This section will introduce the selectives-across-subordinate-domains tendency scale and discuss the relationship of selectives and preposed versus postposed order of subordinate clauses (4.1) before addressing some minor exceptions of the tendency scale in Section 4.2. Section 4.3 deals with the preferred position of selectives after the constituent they have scope over and proposes a functional explanation in terms of ambiguity avoidance. Section 4.4 addresses the question as to why selectives, which are frequent on conditional clauses, tend not to be conditional connectives. While 4.1–4.5 emphasize general trends that can be observed for selectives across subordinate clause domais, the summarizing remark 4.6 also highlights their crosslinguistic heterogeneity.

4.2 The selectives-across-subordinate-domains tendency scale emerging from the visualization of the database

A major result of surveying selectives in 81 languages across subordinate clause domains is that there is a tendency scale (no strict hierarchy, has minor, but no blatant exceptions):

The scale emerges from Figure 1, which visualizes the database underlying this study. The underlying set of codings is available at https://doi.org/10.5281/zenodo.5825998. The 81 translations of the N.T. reflecting 81 doculects are given as rows (start by three-letter ISO 639-3 codes) ordered according to their frequency of selectives in the 200 clauses contexts surveyed (x-axis). You can see the tendency scale as a triangle in the left upper corner when looking at the figure from a distance and when disregarding different colors. Colors indicate the linguistic area (see legend). For instance, purple rows reflect the five Oto-Manguean languages all from the Cuicatlán area and it can directly be seen that, for instance, Chiquihuitlán Mazatec (maq) and Teutila Cuicatec (cut) exhibit very similar patterns. Each column stands for a clause and columns are grouped according to rough semantic domains as explained earlier in Table 2 in 2.3. Domains (separated by black vertical lines) are ordered according to a compromise of frequency of selectives and semantic similarity of domain. Within each domain, the clauses with most selectives across doculects are placed to the left.

The generalization in (18) formulates the tendency scale on a rather abstract level. As one zooms in to more fine-grained domains and exemplars, there are many special circumstances which complicate matters. For instance, concessive conditionals (ConcCond) are intermediate between conditionals and concessives, generalized temporal clauses (GenTime) are intermediate between conditional and temporal, and ‘until’ clauses (Termin) behave in many ways like purpose clauses rather than other temporal clauses.

Small black crosses indicate postposed subordinate clauses (the main verb corresponding to the verb in the English main clause precedes the main verb corresponding to the verb in the English subordinate clause). Postposed clauses exhibit a largely complementary distribution to selectives (almost all small crosses have white background). In no language of the sample do postposed clauses have a preference for selectives. This finding supports H2 “Selectives tend to occur early in the sentence.”

Figure 2 displays the percentage of postposed subordinate clauses (x-axis) as compared to percentage of selectives across the doculects of the sample, which reflects an implicational relationship as the space on one side of the diagonal is empty. However, word order determines occurrence of selectives only partly. Concessive clauses, for instance, are almost always preposed in the dataset, but lack selectives almost completely. Put differently, no postposed clauses attract selectives, but not all preposed subordinate clauses attract selectives.

Figure 2:

Relationship between selectives and word order (Each dot is a context from the corpus, as listed in Appendix E. Numbers indicate percentages of languages).

Turning back to Figure 1, “Symbols” in the legend distinguish different kinds of marking. Notice that there is a difference between “None” (small dot: no selective) and “Not coded” (empty: the passage is lacking in the translation, the translation is too literal, or it was not possible to make a decision). There are special symbols for selectives following after a comma explicitly marking a break at the end of the subordinate clause (“TOP after comma”, see 4.4 vi) and selectives following an anaphor picking up the subordinate clause in the main clause (“TOP on anaphor”, see 4.4 v). These two kinds of special cases have been counted as selectives. Other symbols refer to selectives occurring with main rather than subordinate clauses, lexicalized selectives with scope only over connectives, and “Similar markers” that are not identical with selectives. All these special cases will be illustrated and discussed below.

4.3 Apparent counterexamples to the scale (but no blatant violations of it)

4.4 The position of selectives with respect to subordinate clauses

There is an astonishing uniformity in the positioning of selectives. Usually, selectives simply follow the subordinate clause: S top or S=top, which supports H3: selectives tend to occur after or at the very end of their constituent. This section discusses the few deviations from this general trend, summarized in Table 4.

(i) Clause-final selectives: In most languages of the sample, selectives occur at the very end of subordinate clauses – often as clause-final clitics (see, for instance, Curnow 1997: 387 for Awa-Cuaiquer). This preferred order best averts ambiguity between clausal (wide) and phrasal (narrow) scope of the selective. According to H5 “Selectives mark constituents of very different length”, which may cause scope ambiguities. H4 “Selectives are associated with a syntactic constituent” presupposes that a selective can be unambiguously attributed to a constituent it scopes over. Ambiguity-avoidance thus testifies to the relevance of scope of selectives. Since clause-final constituents are usually not topic-like, S=top usually prevents all risk of ambiguity between selective clause-internal constituents and selective-marked clauses. Yanesha’ (ame) (27) – where subjects are sometimes final in subordinate clauses – can also disambiguate by the choice between simple -pa’ [top] and a complex -ña-pa’ [sequence.marker-top]. Only -pa’ can occur at the end of subordinate clauses (Duff-Tripp 1997: 190).

(ii) Initial selectives: Only three languages of the sample – Gude (gde), Botolan Sambal (sbl) and Bolinao (smk) – have initial selectives. In the two Sambalic languages, selectives are restricted to NPs and headless relative clauses and the Botolan Sambal inversion marker ay preceding the verb often follows the selective constituent, hereby indicating where it ends. In Gude, nominal selective constituents have an initial particle ma and can be followed by prepause -a (Hoskison 1983: 127). According to Hoskison (1983: 129), ma also introduces conditional clauses. However, the N.T. has in fact maɗa and counterfactual maci (Hoskison 1983: 130), which contain the initial selective ma, but cannot mark NPs. Thus, in (28) ma-ci is the clausal selective and the nominal selective constituent ‘this person’ is clearly singled out by final -a.

(iii) Clause-internal selectives: In Parecis (pab), the selective atyo often follows the first element – a position with potential scope conflict. However, in subordinate clauses, it typically follows the conditional clitic -(i)ya, which in turn follows the first or the “thematic” element according to Rowan and Burgess (2008: 26), which prevents scope ambiguity, as shown in (29). [In hypothetical contexts -(i)ya also occurs in main clauses.]

In contrast to the closely related language Amanab, Waris (wrs) exhibits such a considerable inflation of selectives that it is perhaps easiest to describe which words do not have selectives. They are mostly lacking with finite verbs in main clauses, some function words and closely integrated attributes in continuous NPs. Selectives on subordinate clauses follow the verb, which in conditional clauses bears an irrealis markers (-ta ‘unanticipated/undesired’ or -ra ‘anticipated/desired’; Brown 1990: 64). The verb need not be clause-final, but often is, as in (30).

The Parecis and Waris examples also show that there is no fundamental difference between clause-final and clause-internal order. In (29) and (30), the selectives are actually also final. Also positioning of selectives following the verb is not exclusive to Waris, in many languages with clause-final selectives the verb is always clause-final.

(iv) Selectives on connectives: Clause-final positioning is not the only scope-disambiguating position. If the selective immediately follows the connective, it cannot have NP-scope. In Iyo (nca), the combination of relative qu and selective -ko is lexicalized as adversative connective quko ‘but’ (Minter 2009: 68). This is a case of a selective in a lexicalized unit indicated as “TOP on connective” in Figure 1. As in many other languages, adversative connectives in Iyo are used in concessive constructions (although p, q = p, but q), illustrated in (31).

A parallel can be found in Muyang ay ti [thus top] ‘but’, where the source for lexicalization, is a dangling adverb.

It has been pointed out that selectives often develop from definite articles or demonstratives. However, interestingly, most languages where such a grammaticalization occurs, place determiners in absolutely final position or the demonstrative or article shifts to final position during the grammaticalization process. This suggests that the grammaticalization of determiners to selectives is conditioned by word order properties (implying absolutely phrase-final position as attractor position). The question thus arises as to whether there can be any non-final selective determiners. In the sample, there is only one candidate. Zarma (dje) has the demonstrative /wô/ ‘this’ that follows selective personal pronouns (araŋ wo [2pl this/top]), hereby meeting the criteria for selectives. If we consider the definite article -o to be related to this demonstrative, it is instructive to look at its word order, which is N-o S. The unusual position N- marker S for a selective goes hand-in-hand with an unusual distribution of markers in Zarma on the selectives-across-subordinate-domains scale, not including conditional clauses (see Figure 1: dje). Many temporal clauses consist of relative clauses with a temporal head noun such as waat ‘time’. Example (32) shows that occasionally even generalizing (“indefinite”) temporal clauses can have such definite marking.

(v) The selective on anaphor strategy: An alternative to selectives within or on the clause is to place the marker after an anaphor following the subordinate clause: S, anaph-top M. The anaphoric strategy – to the extent that it does not involve clausal anaphors – is best suited for general relative clauses and for conditional clauses with the same subject as the main clause. Example (33) from Burum-Mindik (bmu) indicates that the anaphoric strategy may occur side-by-side with clause-final selectives, here with a nominalized subordinate clause. Burum-Mindik exhibits a further unexpected parallelism in that the anaphoric pronoun with the selective bears the same nominalizing marker as the selective-marked clause (underlined in (33)). [Borong (ksr), closely related to Burum-Mindik, only has anaphoric selectives.]

Among the languages of the sample, the anaphoric strategy is most pervasive in Warlpiri (wbp), which has a clausal anaphor ngula and where the anaphoric strategy is the only possibility to mark clauses with selectives (see also Bittner 2001). Warlpiri is well-known for its many second position clitics and the selective is one of them. Restricting selectives of clauses to the anaphoric strategy has the effect of minimizing the potential for scope ambiguity. Whereas the selective with general relative and conditional clauses usually follows the bare anaphor ngula, temporal clauses can have more extended forms such as ngula=jangka-ju [then=after.that-top] or ngula=kurra-ju [then=all-top]. Posterior clauses such as (34) exhibit the most baroque forms with both negative (“privative”) and temporal adverbial suffixes.

(vi) The selective “after comma” strategy: A more laconic, but similar, strategy is to skip the anaphor as a separate element, but to place the selective still outside of the subordinate clause together with the main clause following a clause break indicated by comma, as illustrated in (35) from Suena (sue).

As indicated in glossing, Suena awa is not only the selective but also the object form of the distal demonstrative pronoun, and (35) illustrates that it is not only used as a selective following conditional clauses, but also as complementizer in object complement clauses. It is not unlikely that the selective use after conditional clauses has originated from object complement clauses. The “subject” or perhaps rather ergative forms of the distal demonstrative Suena ami and Zia amimene in general relative clauses are coded as “similar markers” in Figure 1.

In Waskia (wsk), the selective mu – identical with the determiner mu ‘the, that’ – can occur on either side of the comma in conditional constructions. The occurrence of agi ‘or’ in conditional clauses (36) testifies to the close relationship with complement questions, which also can bear the selective (37):

Given the definition of selective in Section 3 as forming a unit together with a constituent (the personal pronoun), it may be surprising that there is such a thing at all as a selective “after comma” strategy but note that the definition only applies to selectives following a noun-phrase-like constituent. Selectives cling “to the wrong side” only with subordinate clauses, not with NPs. However, the question arises as to whether such selectives following commas are in fact spurious selectives at least diachronically, as the case of Suena and the closely related language Zia (zia) suggests, where it is likely that the markers are an instance of demonstrative-to-complementizer grammaticalization, which is well-attested even in languages where no selectives are involved, as in English that.

While selectives in Suena, Zia and Waskia have pronominal origin, I do not know what their origin in Doromu-Koki (kqc) and Cuicatec (cux, cut) is, where the selective “after comma” strategy is attested as well, but in all these languages, forms used as selectives also occur sentence-initially, which suggests that they all have some sort of anaphoric or conjunctional properties. According to Bradshaw (2012: 50), the Doromu-Koki selective bi can follow a zero anaphor. Cuicatec na is also a coordinator ‘and’ (Anderson and Roque 1983: 630).

(vii) Selectives on (and in) main clauses: Given that selectives are not prototypical subordinators, it is not surprising that they occasionally follow preposed main clauses (M=top S), which is common in some Chadic languages, as illustrated in (38) from Matal (mfh).

In Matal, the selective kà also introduces postposed purpose clauses and, in collocation with the relativizer, kà uwana [top rel] ‘because’ introduces causal clauses (Verdizade 2018: 34). This appears in Figure 1 with the symbol “TOP on main clause”.

A further phenomenon I have so far only encountered in Chadic languages is illustrated here from Merey (meq), a language with rather extended use of selectives. In (39), the selective na follows the background (single underline) with the question word in focus sentence-finally (see also Gravina 2007: 6). It is interesting to notice here that the selective-marked background (underlined) is not a straightforward simple phrase (NP, adverb or clause) as in all other examples, even though it probably does not violate H4 “Selectives are associated with (have scope over) a syntactic constituent”.

To summarize, despite much diversity in how selectives are used in complex sentences, selectives almost always occur after the selective constituent, largely confirming H3 “Selectives tend to occur after or at the very end of their constituent”. This is a position with maximally little potential for scope ambiguities, and in the few cases where the position is not absolutely terminal, there are usually other, more language-specific, strategies to avert ambiguity.

4.5 The relationship between conditionals and selectives

According to Comrie (1986: 88), conditionals differ in degree of hypotheticality, without there being any neat divisions between real, hypothetical and counterfactual. Degree of hypotheticality plays a role for selectives only in a few languages of the sample, and there is no uniform trend. In Abau (aau), non-hypothetical conditionals lack selectives, because they take the same connective ankin as temporal clauses not followed by the selective (Lock 2011: 367). Non-hypothetical conditionals are more akin to temporal clauses than hypothetical ones, and Abau testifies to the fact that temporal clauses can occasionally trigger the lack of selectives in non-hypothetical conditionals. Siane (snp) shows the opposite trend. The hypothetical-conditional connective neko (perhaps from ne-ko [exist-since/only]) is compatible with the demonstrative and selective ya ‘there’ (Potts and James 1983: 12) only if ya follows after the comma. In Ga (gaa), the counterfactual is marked initially by àjí and finally by kṹ!lɛ́ (or kṹ!ɛ̃́), with the definite article, which serves as selective, as a second component (Campbell 2017: 619).

If word order is disregarded, conditionals generally behave rather uniformly when it comes to selectives, which suggests that there is considerable potential for lexicalization. Indeed, several languages have markers in conditional clauses that look suspiciously similar to selectives, even though they are not selectives strictly speaking (in Figure 1 indicated with the symbol for “Similar marker”): Meyah (bera//beda), Moskona (era//eda), Yawa (-amo//weamo), Una (-ra//ura), Mian (-le//mole), Iyo (-ko//-qo), Korean (-nun//-myen) and Japanese (wa//ba). This is hardly a coincidence. In some of these, selectives and the markers attested in conditional clause will ultimately derive from the same source. For Korean (kor), there is actually diachronic evidence that this is the case (Koo 1999). In Japanese, ba and wa share some morphophonological properties (Shibatani 1990: 164). Also, in casual speech, the conditional and topic particles drop their consonants and become homophonous (Shibatani 1990: 176). In Moskona (mtj), top erá and the apodosis marker edá ‘then’ are minimal pairs (Gravelle 2010: 24). The same holds for the closely related language Meyah (mej) with top berá and the apodosis marker bedá ‘then’ (Gravelle 2004: 33). However, in generalized relative clauses both markers occur side-by-side, as in (40):

In Yawa (yva), Una (mtg), and Mian (mpt), the only problem is that the initial parts of the markers used in conditional constructions have no etymology, while the final parts are identical with selectives.

There is also evidence for on-going lexicalization. Chipaya (cap) nïki ‘if’ contains a fossilized masculine demonstrative form *ni(i)-ki [that/the.m-top] (Cerrón-Palomino 2006: 264). The Chipaya form may have originated in general relative clauses, where it occurs side-by-side with selective-marked relative forms, as shown in (41):

Ama (amm) has both optional and non-optional uses of selectives in conditional clauses. There is a lexicalized future conditional V-aki-person-mo [top], with the future conditional morpheme -aki- occurring only in this construction, where selectives are obligatory (Årsjö 1993: 14). A token of lexicalization is that the construction with -aki- can occur in postposed conditional clauses as illustrated in (42):

Also, Chipaya nïki ‘if’ occurs in postposed conditional clauses which further testifies to its lexicalization.

Comrie (1986: 86) argues that, if conditionals are topics, there must be some degree of grammaticalization (which is the same thing as lexicalization of the conditional marker), since conditionals can occur in non-topical uses, notably in responses to questions, such as Under what circumstances will you leave? with answers with conditional clauses in focus, such as I will leave, if you pay me. Comrie’s argument is similar to the one adduced by Chappell and Creissels (2019) against topic possessives as a possible basic type in the typology of predicative possession (Stassen 2009). However, clear cases of focal conditional clauses in the sense of answers to information questions are rare, in the N.T. only attested in some cases of free translation. An example is (43) from Huallaga Quechua (qub), where conditionals are adverbial clauses with adverbial markers distinguishing switch reference (-r ss in (43)) and usually followed by the selective <-ga> /-qa/ (Weber 1989: 415). However, in (43), the conditional clause, which is an answer to a question, takes the direct evidential marker that also indicates focus.

The example suggests that -qa is no lexicalized expression of conditional in Huallaga Quechua.

The question thus arises as to what extent selectives in conditional clauses can be considered conditional connectives. Markers with similar meanings have similar distributions in parallel texts and this can be easily measured with collocation measures (see, e.g., Dahl and Wälchli 2016 and Wälchli 2019). When searching by automated means for candidates for conditional markers in the 81 N.T.s of the sample with English if as a seed gram, in most languages not selectives, but entirely different conditional markers, are extracted. This is because selectives generally tend to have very high text frequency. Put differently, even though selectives often occur on conditional clauses, they are not dedicated markers of conditional clauses. The selectives that turn out to match best in distribution with English if (technically speaking “collocate best”) fall in three groups:

1. selectives originating in subordinate clauses (see 4.3.2): Gulay ɓaa;

2. a less frequent selective or a less frequent allomorph: Fon hǔn ‘top.imp/thus’, Kenga jaay, Korafe a-mo that-top, Aymara -xa; and

3. The extracted marker is a combination of the selective with some other marker: Gude maɗa and maci (see 4.3), Toura laá top.foc, Haka Chin ah-cun loc-top.obl, Waris -ta-va irr-top, Konai de-ba-be aux/do-pfv.irr-top and ba-be pfv.irr-top, Chipaya nï-ki dem.m-top.

This suggests that selectives do not tend to grammaticalize (or lexicalize) to conditional connectives, but occasionally combinations of selectives with another morpheme can lexicalize to conditional markers, as in Haka Chin locative plus the oblique case form of the selective (see Barnes 1998: 68).

Further evidence that selectives are not conditional connectives comes from word order. In a considerable number of languages of the sample (Chadic, Kwa, East Bird’s Head and Oto-Manguean languages), selectives and conditional connectives end up at the two different edges of the conditional clause, which results in a kind of bracketing marking of conditional clauses. Table 5 lists some examples for verse 43014007 ‘If you have known me, [you will know my Father also.]’ (CondHyp8).

Almost all of these languages are VO languages, where adverbial subordinators are expected to be initial (Dryer 1992: 108).

4.6 Summarizing remark

In Section 4 I have argued that the range of possible variation of selectives across languages in subordinate languages is strongly constrained. The distribution of selectives across subordinate clause domains follows a scale without blatant exceptions (4.2 and 4.3). Word order is strongly biased toward initial, rather than final, selective-marked clauses (4.2), which supports H2 “Selectives tend to occur early in the sentence” and toward occurring at the end or after the clause they scope over, which supports H3 “Selectives tend to occur after or at the very end of their constituent” (4.4). Despite the frequent occurrence of selectives in conditional clauses, selectives tend not to be conditional subordinators (4.5), if they do not originate from subordinators (4.3.2).

However, the regularities go hand-in-hand with a very high degree of language-specific heterogeneity. Here is a just one further example. In Folopa (ppo), the selective -ta almost exclusively follows nouns and pronouns and there happens to be a lexical preference to express events related to speech with relative clauses with the head noun fo ‘talk’: ya̧-lo yale fo [1sg-erg do.rel.pst talk] ‘what I said’, lit. “talk I did” (Anderson 2010: 50), which triggers the use of the selective, as in (44):

This idiosyncrasy accounts for the specific pattern of Folopa (ppo), different from all other languages in Figure 1 with sporadic occurrences of the selective -ta across different subordinate clause domains.

5.1 Are selectives a fruitful conceptual framework?

If – as argued in Sections 1 and 3 – selectives are an arbitrary conceptual framework for crosslinguistic comparison, how fruitful has the concept been so far?

First, it can be pointed out that the notion is easily applicable from a crosslinguistic point of view and is especially well-suited for a massive parallel text corpus where even languages that so far are poorly described can be considered.

Second, it can be observed that there are micro- and macro-areal patterns that testify to the usefulness of the concept, both such that already have been associated with “topic markers” (e.g., Ga-Ewe-Gen-Fon; New Guinea) and such that do not yet figure prominently in the literature (notably Cuicatlán).

Third, the concept has proven to be relevant beyond its definitional domain. Markers that were identified on the basis of their use with personal pronouns in contrastive use also occur in non-contrastive uses and in many languages also across various subordinate clause domains (see Section 4), very much as has been suggested already by Haiman (1978).

Fourth, and most importantly, selectives have turned out to be sensitive to a number of scalar phenomena such as summarized in Table 6. Among those, (iii) Contrast is most trivial, since it at least partly follows from the definitional domain, but (i) and (ii) testify to very strong correlations with word order (irrespective of other word order typology of the languages involved), confirming or at least partly confirming Hypotheses 1 and 2 in Table 1 (see Section 2.1) and the selectives-across-subordinate-domains tendency scale (iv) suggests that there is a hierarchy of more and less prototypical contexts for selectives across subordinate clauses.

However, it cannot be denied that there are still very many open questions connected to the suggested concept, some of which will be addressed below.

5.2 Are selectives a gram type?

Dahl (2016: 435) argues that clusters of language-specific category types that he calls crosslinguistic gram types are not only comparative concepts constructed by the researcher, but “objects that are ‘out there’ to be discovered”. Members of a gram type are not the same, but just “similar enough to be put in the same bucket”. They are expected to “share a set of prototypical uses, where the probability of their use comes close to 1” (Dahl 2016: 435). The underlying idea is that prototypical uses define the semantic core of the gram type which is crosslinguistically uniform whereas grams belonging to a gram type can be rather diverse in their peripheries. Gram types can be “polycentric”, which can mean that members of a gram type can have different preferred sources of grammaticalization or that there can sub-clusters within the more general larger cluster.

Selectives are a challenge to the gram approach for various reasons. As pointed out in Section 3, selectives tend to be communicatively optional, despite their high text frequency, except sometimes when they fuse with other markers for specific functions such as conditional as discussed in 4.5. This suggests that selectives are perhaps more of a pragmatic rather than a strictly grammatical category type. While this study has identified more prototypical contexts of use for selectives with personal pronouns, on the one hand, and with subordinate clauses, on the other hand, it is not yet clear whether there is a semantic prototype for the set of markers irrespective of particular uses. As pointed out, a major difficulty is that selectives used with nouns are difficult to distinguish from other phenomena (all associated with definiteness). Another difficulty is that selectives tend to have an extremely wide range of use in many languages where they occur, which makes it rather difficult to make out the most prototypical contexts of use. However, it can be taken for granted that the crosslinguistic set of selectives have prototype-like (scalar) behavior at least when associated with personal pronouns and with subordinate clauses, which makes them at least similar in this respect to a gram type.

One thing that is certain is that selectives are not a monocentric gram type. In 4.3.2, we have seen that a number of selectives in different languages originate in conditional or in temporal clauses. At least in Dii and Gulay, the clausal uses clearly prevail in frequency. Thus, there is at least a distinct subtype with clausal origin (including Dii máa, Gulay ɓaa, Khalkha Mongolian bol, Kalmyk bol-xla and Sougb kaba).

5.3 Further dimensions of variability

Throughout this paper I have emphasized the crosslinguistic heterogeneity of selectives, but this does not entail that a selective in one language is totally different from selectives in all other languages. If this were the case, the concept would obviously not be fruitful. Rather there are a number of dimensions of variability, whose identification actually may contribute to the fruitfulness of the concept of selectives. Dimensions of variability are also useful for understanding how selectives relate to similar categories that do not meet the definitional criteria for selectives. Let us consider here a few important dimensions.

(i) Some selectives can be sensitive to adversative meaning: In 4.3.1, we have seen that Sochiapan Chinantec and Toura selectives prefer adversative contexts in subordinate clauses. This groups them, on the one hand, with adversative selectives, such as Dimasa la, as opposed to non-adversative selectives (see Sections 2.2 and 3), but also with adversative markers that have been excluded from consideration (such as Koine Greek de, see Section 3).

(ii) Some selectives can be separated from subordinate clauses: These selectives follow anaphora (see 4.4v), have anaphoric properties themselves (4.4vi) or are at the same times conjunctions (4.4vi). Among the phenomena excluded, these are similar to apodosis-introducers, such as English then or clausal conjunctions meaning ‘and then’.

(iii) Some selectives are wholly or mostly restricted to NPs: This holds for most markers at the bottom of Figure 1, which do not combine with subordinate clauses or only to subordinate clauses with nominal heads, such as generalized relative clauses. There is reason to believe that a smaller proportion of markers would have been attested in subordinate clauses if nouns rather than personal pronouns had been chosen as a definitional domain. This means that the fact that this study seems to confirm H7 (high degree of freedom of host selection of selective) is largely a consequence of how selectives have been defined.

In many Central Eurasian languages, possessive suffixes have developed topic-like uses (Yurayong 2018; for Colloquial Persian, see; Etebari et al. 2020) and these elements seem to have difficulties to spread both to personal pronouns and to subordinate clauses (i.e., they do not easily increase in freedom of host selection). Eastern Mari is a language where third person singular possessive affixes can occur as selective on personal pronouns of speech act participants. However, not in all styles and registers to the same extent. Forms, such as me-že [1pl-top/poss.3sg], are common especially in folk poetry (see, for instance, Saarinen 1994: 220, 390). In the Eastern Mari N.T. translation, there is just a single occurrence, Example (45) with second person plural:

The Mari example reveals that the test for selectives suggested in Section 3 is not always equally easy to apply. What to do if the definitional criterion holds, but is extremely rare? In this case I opted for not including Mari, because the use with personal pronouns was too marginally attested. It follows from this discussion that markers which do not have a high degree of freedom of host selection simply did not make it into the sample.

(iv) Multiple selectives within a sentence: Especially some languages of New Guinea allow for multiple selectives within the same sentence, a phenomenon that Heeschen (1994: 57, 1998: 162) and de Vries (2006: 814) call “thematizing” (a discourse preference), here illustrated in (46) from Abau with the beginning of the Parable of the Prodigal Son where selectives follow two NPs containing new (off-stage) information.

The most extreme case in the sample is Waris (see Section 4.3). Indeed, the Waris selective has a very different overall distribution, for instance, when compared to Gulay ɓaa that mainly follows subordinate clauses and has clausal origin. However, both Waris and Gulay selectives have similar uses in person pronoun contrast and on subordinate clauses.

These are just some examples of how various kinds of selectives may differ from one another, but this does not mean that the markers differ in completely unrestricted ways, otherwise there would not be any scalar effects.

5.4 Are there any semantic invariants?

The multiplicity of scales (Table 6) clearly testifies to the semantic and functional complexity of selectives. Let us now consider some possible semantic explanations of the selectives-across-subordinate-domains tendency scale.

Why are concessives so much less prone to have selectives than conditionals? There is no major difference in word order. In 2.4, I have suggested that selectivity (H12) is part of the explanation. Furthermore, concessives are veridical, which allows for non-hypotactic construals. In many languages, concessives are expressed mostly by adversative coordination. Adversative coordination is a natural paraphrase of concessives: although p, q = p, but q. Put differently, concessives are often expressed with adversative connectives that also occur in contrast constructions. Now, contrast constructions actually favor selectives, but not in the position where the concessive relationship is expressed (double underline), as illustrated in (47) from Sougb.

In Section 4 it has also been shown that complement clauses are not a characteristic environment for selectives. This may seem strange since there is some evidence that indirect question complements may have contributed to the origin of selectives in conditional clauses in such languages as Suena and Waskia (see 4.4). However, given that complement clauses have illocutionary force, it is no surprise that selectives hardly ever become fully entrenched in complement clauses, which is consistent with H11 “Selectives tend not to mark clausal units carrying illocutionary force”. Such marginal contexts as ‘Whether he is a sinner I do not know’ (see (10) from Hills Karbi in 2.4) are most inclined to bear selectives. These are examples where the question in the complement is actually not to be answered and rather a point of departure than an illocutionary act (indicated in English by the marked initial word order). This renders such complement clauses similar to alternative concessive conditionals, such as Whether we get any financial support or not, we will go ahead with our project (Haspelmath and König 1998: 653).

As far as temporal clauses are concerned, veridicality may also be a factor why conditionals outperform temporal clauses. Another important factor is the higher degree of delimiting of conditional clauses. Temporal sequence is one of the most characteristic contexts for what Foley and Van Valin (1984) and Van Valin (2005) call co-subordination, where operators, such as tense, are shared. In Figure 1, it can be seen that especially many languages of New Guinea, where co-subordination is widespread, have selectives in conditionals, but not usually in temporal clauses. Foster (1981: 15) points out for Timbe that one of the functions of âmâ is that it limits the scope of tense/mood/aspect and negation to that clause. In (48), where the subordinate clause has its own TAM and polarity values, here in the only example from the concessive domain where âmâ/amâ is used:

Similarly, Årsjö (2016: 255) reports that the selective in Konai may “be used as a barrier, to certain features spreading”; see also Reesink (2014: 256) for Usan. All this supports H9 “Selectives often have the function of delimiting”. However, note that delimiting is not an automatic consequence of selectives; Bickel (1993: 33, 38) discusses instances of ambiguity involving illocution and negation in Belhare. Delimiting is not restricted to clauses. Several grammar writers report that selectives preferably follow zero-marked subjects (Poppe 1951: 99 for Khalkha bol, van Gijn 2006: 114 for Yuracaré, and Cerrón-Palomino 2006: 123 for Chipaya).

However, selectives do not only delimit constituents from the rest of the sentence or clause, they also do what appears to be the opposite: linking (H[ypothesis]10) the selective constituent with the predicate or nucleus that follows. Linking is most apparent in instances where selectives anticipate properties of the nucleus predicate. In Section 3, we have seen that some languages have interrogative and imperative selectives. There are also examples for less grammaticalized instances of the same tendency. Usan uses the form eng originating from a demonstrative (Reesink 2014) as a selective. In many instances with imperative-like main predicates, this marker is preceded by /gâb/, the same-subject-form of the verb ‘see’, as in (49). The range of use is broader than the imperative selective in Fon (see Section 3), most important is that the nucleus predicate somehow engages an addressee.

This construction suggests that conditioned commands are primarily persuasion by directing perception to crucial circumstances.

However, there are also uses of selectives where there is nothing to link to, which are apparent counterexamples to linking (H10) and to H11 “Selectives tend not to mark clausal units carrying illocutionary force”. In many languages selectives can occur sentence-finally, often with some modal connotation. Sentence-final selectives are akin to insubordination, “the conventionalized main clause use of what, on prima facie grounds, appear to be formally subordinate clauses” (Evans 2007: 367). Insubordination has the functions of “indirectionalizing, modalizing, and presuppositionalizing” (Evans 2007: 423). Object-topicalized clauses in Kayardild are actually one of Evans’ (2007: 425) examples. In Sochiapan Chinantec, sentence-final né³ is used as a request for further clarification (Foris 2000: 358). In Aymara, sentence-final -xa is an attenuator “I guess, OK?” (Hardman 2001: 171). In Usan, “[c]onstructions with sentence-final e(ng) convey the speaker’s attitude of assertiveness, indignation or surprise” (Reesink 2014: 249). Also, at least in some languages, there are referent-plus-selective-questions, such as Japanese Jun kun wa? ‘What about Jun-kun?’ (Tanaka 2015) or Tena Lowland Quichua Jenny=ga? ‘What about Jenny? Where is Jenny?’ (Grzech 2016: 40), whose crosslinguistic distribution unfortunately cannot be assessed in the N.T., where there are no such examples. These are an interesting type of turns without overt predicates. They provide counterevidence to H1 that selectives are units of the sentence. However, it might be argued that in such “what-about”-questions the predicate is simply omitted, and that if it is not, as in (50) from Nalca, the illocution follows the selective constituent, as expected.

The questions as to how selectives relate to the domain of the sentence or any related unit and the extent to which selective constituents can bear illocutionary force needs further investigation.