The interplay of contrast markers (‘but’), selectives (“topic markers”) and word order in the fuzzy oppositive contrast domain

2.1 The preselected domain of this study and what is not treated

Contrast entertains many relationships to other domains, not all of which can be treated here. The starting point will be contexts most often encoded by English but across different translations of the gospels, assuring all the three subdomains listed in (1)–(3) will be included. This choice excludes prototypical instances of concession (although), a domain often treated together with contrast (see, for instance, Rudolph 1996). Concession is typically encoded by subordinate clauses whereas contrast constructions are mostly coordinate. However, there are languages, such as Sanuma (Borgman 1990: 59), where contrast is generally expressed by subordinate concessive constructions when marked overtly and this investigation includes translation-equivalents of ‘but’ even if construed as subordination.

Contrast markers frequently grammaticalize from restrictive markers (Malchukov 2004: 194). However, words for ‘only’ are included here only to the extent they happen to be used in oppositive, counterexpectative and corrective contrast.

It is well-known that contrast is closely related to focus, see, e.g., Umbach (2005) – who also emphasizes the close relationship of contrast with denial (see also Spenader and Maier 2009). Dik (1997: 331–335) regards oppositive contrast (“parallel contrast”) and corrective contrast (“replacing contrast”) as types of foci. However, Matić and Wedgwood (2013) show that the notion of focus is problem-ridden from a cross-linguistic perspective both in parametrized approaches (where corrective focus is considered a focus type of its own) and in unified approaches. The consideration of focus is beyond the scope of the present study.

Contrast is also closely related to conjunction (‘and’-coordination; Haspelmath 2004: 5). Russian a, for instance, is often conceived of as intermediate between conjunction (‘and’) and contrast (‘but’). However, as shown by Krejdlin and Padučeva (1974), it also has specific syntactic properties in many of its uses, tending to be immediately followed by a topic (“theme”) irrespective of whether the use is contrast or conjunction as in (8), where one constituent from the first conjoined clause is picked up in a new role in the second conjoined clause (see also Jasinskaja and Zeevat 2008).

In the literature on information structure, “contrastive topic“, such as that marked by a specific rising intonation contour in English (“[t]he phrase denoting what the question being addressed is about”, Constant 2014: 17), is an important notion. However, it has to be pointed out that “contrastive topics” also occur in enumeration as in (9) with initial phrases in comparative sequences with more than two coordinands.

The present investigation is restricted to contrast constructions with only two coordinands, which are in-line with Mann and Thompson’s (1988: 248) definition of contrast as always having “exactly two nuclei”.

2.2 Saliency of the oppositive phrase

In Section 1, I have claimed that oppositive contrast, although a meaning, is also associated with a syntactic constituent type: the oppositive phrase. Grammatical markers are not seldom associated with phrase types; case, for instance, usually has a noun phrase host. However, oppositive phrases are special in that (i) they are not definable via dependency on predicates and (ii) there are few restrictions on what kind of constituent can qualify as oppositive phrase. Hence, if the oppositive phrase exists, there are only two alternatives: (a) either it is a pre-established category, such as Rizzi’s (1997) postulated Top(ic)P(hrase) – an option I do not seriously consider – or (b) it is a matter of saliency: it exists if there are properties that render it salient. This section deals with such properties.

A first step is to exclude what cannot be an oppositive phrase. The oppositive phrase is a subpart of the contrast sentence, not the entire contrast sentence or its main predicate or illocution. Put differently, the oppositive phrase cannot contain the main predicate of the sentence or any word directly associated with the main predication such as sentence negation (no verum-elements in the sense of Umbach 2005: 218).

Like conditional clauses, oppositive phrases play a role in stepwise establishing common ground (Lehmann 1974). As such, they will strongly tend to be sentence-initial or immediately follow a sentence-initial contrast marker. This yields Property 1 [P1]: sentence-initial position (note that a phrase following a contrast marker also counts as initial). P1 is most salient if the initial position goes against the expected dominant word order. However, P1 is particularly important in rendering all non-initial phrases unlikely candidates for oppositive phrases. It follows from P1 that contrast sentences with initial predication-level elements, such as finite verbs or sentence negation, are unlikely to contain oppositive phrases.

Saliency may be strengthened if an initial phrase is detached from the rest of the sentence by a marker, which yields P2a: oppositive-phrase-final marker, which can be achieved by a selective as in (5) or an oppositive-phrase-final contrast marker as in (6a). From P2a, we may derive the hypothesis in (10):

This is illustrated in (11) from Somali and (6) from Estonian.

P2a may suggest that oppositive contrast markers will most likely occur following oppositive phrases. However, oppositive contrast markers have to single out two constituents: (i) the oppositive phrase and (ii) the entire contrast sentence. Since the optimal position for contrast markers is between the anchor and the contrast sentences, (ii) is in conflict with (i), we may assume that (i) will often prevail over (ii). However, a sentence-initial contrast marker – even if not fully dedicated to the expression of oppositive contrast – can single out an oppositive phrase, if the contrast marker tends to collocate with an immediately following oppositive phrase. This yields P2b: specific contrast marker collocating with immediately adjacent oppositive phrase. If a language has two sentence-initial contrast markers sensitive to the oppositive–counterexpectative contrast distinction, such as Russian a opp and no cexp, it can be derived from P1 that the former will be immediately followed by an oppositive phrase more often than the latter. Krejdlin and Padučeva (1974) argue that Russian a in two of three major uses must be immediately followed by a “theme”, i.e., an oppositive phrase. Examples (12a) and (12b) differ in their construal in two respects that go hand-in-hand: (i) in the use of the contrast marker a versus no and (ii) in the presence (underlined) or absence of an initial oppositive phrase. Note that the English translation and (12b) lack oppositive phrases because the contrast marker is followed by predicate-associated elements.

It might be argued that sentence-initial position [P1] affecting the anchor phrase in the anchor sentence may have similar effects indirectly, as in (13). However, if me in (13) is an oppositive phrase, it is unlikely to interact in long distance with the contrast marker and contrast markers used in such examples are unlikely to be oppositive contrast markers (and, as expected, English but is a general contrast marker and not an oppositive contrast marker).

Turning to semantics, an oppositive phrase in the contrast sentence is typically opposed to some element in the anchor sentence. Sæbø (2002: 261) calls this element the alternative. Alternatives may be explicit or implicit. However, a referent is at least usually not opposed to itself. Hence, it is reasonable to postulate that the referent of the oppositive phrase cannot be the alternative [P3]. In (14) from Ukrainian, mı ‘we’ cannot qualify as oppositive referent, because it has the same role in the anchor sentence (same goes for you in (13)).

In (14), the semantically best candidate for an oppositive phase is the NP ‘this one [=Jesus, this man]’ with the explicit alternative ‘Moses’ in the anchor sentence and, indeed, many translations in many languages have ‘this (man)’ as initial oppositive phrase in the contrast sentence. However, in (14) the initial constituent in the contrast sentence is the entire underlined subordinate clause ‘where this one has come from’, marked with the second position clitic ž with oppositive function (follows the first word of the phrase, glossed as “(2)” in parentheses). It has a partly implicit alternative ‘Moses has come because God sent him to teach us’. Even though we may assume that [P4] explicit alternatives are better than implicit ones, (14) demonstrates that the initial position of the oppositive phrase is more important.

Example (15) from Lithuanian (o is cognate with Russian a) illustrates that what matters is alternative referents and not lexical semantics. The lexeme (aš ‘I’) can be the same if the referents are different (Note P2a indeed in the English translation).

As already pointed out in Section 1, emphasis [P5] is a very important factor. Oppositive phrases are emphatic. Emphasis implies that a meaning is explicitly expressed, which entails that emphasis always has a formal component. Given the same or similar meanings, forms that are more distinctly articulated (longer, prosodically more marked and more clearly detached) are more emphatic. Semantically, emphasis has various aspects. Particularly important for our purposes are (i) extreme position on quantity scales and unexpectedness/surprise (Trotzke 2017: 35 speaks of “scale of likelihood”).

The list of salient properties of oppositive phrases is summarized in Table 3:

Table 3 demonstrates that formal properties are as important as semantic ones. Oppositive contrast is a matter of family resemblance with more and less prototypical examples. This makes its investigation a challenge, because, as pointed out by Myhill and Xing (1996: 354), there is a risk of circularity. Myhill and Xing’s (1996) approach is to apply a rigid distinction based on two semantic criteria and to not include word order in the definition, which is a feature that their study correlates with oppositive contrast. What is done here instead is to start with the form of contrast markers and to have oppositive contrast emerge as a region of the similarity space of contrast markers. Put differently, I will rely almost entirely on the strength of P2b (and P2a will also play in, because different order of contrast markers, such as Estonian aga1/2, is coded differently). In Section 4.1 we will see that Dimension 2 of the similarity space of contrast markers corresponds to the semantic/pragmatic gradualness of oppositive contrast in Table 2, although all properties in Table 3 except P2a/b are initially ignored.

3.1 The design of this study and its theoretical and practical implications

The present study is (i) is corpus-based, (ii) considers markers extracted from texts, (iii) makes use of quantitative methods, (iv) uses parallel texts, (v) is massively cross-linguistic, (vi) uses translations of the New Testament (NT), and (vii) investigates contrast. This section discusses major practical advantages and disadvantages of these choices and their theoretical implications.

1. The encoding of a functional domain may differ in degree of language-internal variability ranging from entirely stable (no variation) to entirely emergent (not conventionalized). Hopper (1998) emphasizes the emergent nature of grammar where structure is always viewed as provisional. In order to make the notion fruitful for typological research it is better to speak of partially emergent grammar. There is a constant interplay between discourse and conventionalized structure. Using corpora has the advantage that emergent structure reflected as variation in language use is not precluded from consideration, but has the disadvantage that stable and emergent structure cannot easily be kept apart if only one text per language is available. This could be a major problem, if the aim of the study was to classify languages into types, but no attempt of classification of languages into types is made here. Instead, each text is considered a doculect (a documented language variety) of its own.

2. For the purposes of this study, very little annotation and analysis is needed. It is sufficient to identify markers in contrast constructions. No information about how these markers are integrated in language systems is required. The advantage is that even texts in languages with very restricted or no available grammars or dictionaries can be used.

3. In linguistics, there has long been a traditional tension between universals and diversity. This study makes use of quantitative methods for data mining that can detect major trends in fuzzy datasets that reflect the full range of attested diversity. I am using such methods because linguistic data are both highly diverse and highly constrained at the same time. The aim is to find strong generalizations in diverse, not strictly implicational, data. A disadvantage is that such methods always come in families of slightly different tools where it is not always easy to find the optimal tool for a certain set of data. Rather than striving at optimizing the choice of specific data mining tools, I will use simple tools yielding sufficiently good results.

4. The use of data from parallel texts comes at the cost of translated, rather than original, planned, rather than spontaneous and written, rather than spoken language. The motivation for the choice is that parallel texts allow us to compare languages at the maximally granular level of contextually embedded exemplars rather than on the level of preconceived abstract functional domains. It is often assumed that meanings can be taken for granted as extralinguistic comparative concepts (Haspelmath 2010: 681). However, nothing involved in language is truly extralinguistic. Functional domains relevant for cross-linguistic comparison are not given, but must be established as regions in similarity space from empirical cross-linguistic investigations. However, functional domains can only be established in cross-linguistic investigations if the unit of comparison is more fine-grained than the functional domain. Since there is no empirical extralinguistic way to describe the degree of similarity between pairs of meanings, similarity space can actually not be fully abstracted from form. This is also why I prefer similarity space over “semantic space” or “semantic map” here.

5. A massively cross-linguistic approach is adopted, because it is not easily possible to distinguish arbitrary from non-arbitrary structure in specific languages. This study relies on the assumption that cross-linguistically recurrent identity of form reflects similarity in meaning (Haiman 1985: 19). As a consequence, functional domains can be modeled as similarity spaces in which the distance between any pair of aligned exemplars in parallel texts reflects the probability that these are encoded by the same marker in any language (Wälchli and Cysouw 2012). However, it is well-known in typology that cross-linguistically recurrent patterns may also reflect large-scale arbitrary patterns due to language contact or genealogic relationships (see, e.g., Dryer 1992). Considering large and diverse sets of languages minimizes possible hidden effects of large-scale arbitrary patterns. Since it is difficult to remove bias entirely, it can also be useful to compare results in more and less biased datasets as will be shown in Section 4.1.

6. The choice of using translations of the New Testament follows from (iv) and (v). It is the only massively parallel text with many electronically available versions even in minor languages from all over the world with sufficiently variable content. One advantage of the New Testament for studying contrast is that there is a number of contexts with strong emphatic opposition of referents, which happen to represent the strongest kind of oppositive contrast (see Section 4.1). However, there is also reason to believe that the use of contrast markers in the Bible translations is not always fully representative of the languages of the translations. Koine Greek de ‘but’ is used extensively in narrative sequence, and translations strongly influenced by Greek or Latin reflect such use, as But when the Pharisees saw it (40012002) in the King James translation. Such contexts are excluded here, as only passages where but occurs in many translations are considered. It also has to be pointed out that the language of the Bible is prestigious in many languages, so that features of Bible translations can spread to other texts. For using Bible translations in typological investigations, see also de Vries (2007).

7. Contrast markers are known to be maximally unstable diachronically and are among the “most vulnerable items to contact-related linguistic change in grammar” (Matras 1998: 281). The wide spread of Spanish, Arabic and Russian contrast markers to unrelated and geographically distant languages makes clear that it is hardly possible in this domain to reconstruct the precolonial situation. However, it is important to notice here that there is not necessarily any correlation between (non-)arbitrariness and stability. In the Bible corpus used here, there is a single one among 27 German translations, the Grünewalder Bible, where aber is largely restricted to second position and used as oppositive contrast marker as opposed to counterexpectative (je)doch ‘but, however’ in initial position. This aber is used as an emergent oppositive contrast marker in a similar way as language-internally much more stable a in East Slavic languages. However, several Polish translations use both ale and lecz in non-oppositive (counterexpectative and corrective) contrast without any clear distinctive pattern. Andrason (2020: 17) shows that lecz is stylistically marked and occurs more often than ale in conservative Bible translations whereas it hardly occurs in informal Polish conversation. Polish lecz can therefore be quite arbitrarily distributed in Polish written texts and its use is not stable. To summarize, contrast is a highly interesting domain to study the interplay between emergent structures and cross-linguistically general trends.

3.2 Data and method

The material used is a corpus of translations of the New Testament originally compiled by Mayer and Cysouw (2014). For a list of 1,659 texts with 1,259 different ISO 639-3 language codes included in this study, see Appendix 3.2.A. However, all but one step are based on much smaller subsets.

This study makes use of several quantitative tools: (i) Principal Coordinate Analysis for constructing and visualizing the similarity space of contrast markers (Section 4.1), (ii) Principal Component Analysis for exploring the different loadings of markers (Sections 4.1–4.2), (iii) Partitioning-Around-Medoids for identifying clusters (Section 4.1) and (iv) t-value as a collocation measure for extracting sets of similar markers in the corpus with a certain distribution profile (Section 4.3).

1. Principal Coordinate Analysis (PCoA) [R: cmdscale()], also called classic metric Torgenson scaling, is a metric variant of multidimensional scaling (MDS). It is often somewhat sloppily called “multidimensional scaling” in the literature, but differs, for instance, from optimal classification MDS (Croft and Poole 2008; see van der Klis and Tellings 2022 for a survey of linguistic applications of various kinds of MDS). PCoA takes a dissimilarity matrix as input, here calculated with Hamming Distance (see Appendix 3.2.B for further explanation). A major reason for using PCoA is that it is most similar to Principal Component Analysis among different kinds of MDS.

2. Principal Component Analysis (PCA) [R: prcomp()] differs from PCoA most markedly by taking a set of variables rather than a distance matrix as input. In our application, each marker in each language is a variable (see Appendix 3.2.C). What is called “dimensions” in PCoA is called “components” in PCA. An advantage of PCA is that the contribution of each variable (i.e., marker) to each component can be quantified. The technical term for this is “loading”.

3. Partitioning-Around-Medoids (PAM) [R: pam()] is a clustering algorithm taking the same input as PCoA, a dissimilarity matrix. PAM is useful in typology because it reduces sets of exemplars to clusters roughly corresponding to functional domains that typologists using description-based comparison tend to posit as comparative concepts. As we will see, this works very well where meanings are strictly delineated, but not with fuzzy meanings.

4. t-Value (Manning and Schütze 1999: 163) is a collocation measure that can be used to rank the distribution in the parallel text corpus of all salient potential marker candidates (all wordforms and all character sequences within wordforms) in all languages according to how well they match a search distribution. If we can determine a search distribution for oppositive contrast (see Section 4.3 for how this is done), we can search all languages in the whole parallel text corpus for markers that best match this distribution profile. All markers with t-values above a certain threshold can then be considered good candidates for expressing oppositive contrast and the automatic output can be verified by using reference materials. This is an efficient procedure for identifying many markers belonging to a gram type (a cross-linguistic set of grammatical markers or “grams” sharing a core of prototypical contexts of use; see Dahl and Wälchli 2016).

Since advanced automatic processing is not possible right from the beginning, this investigation uses a stepwise procedure, where initial steps taken do not cover the entire diversity of the corpus, but whose results may then be a basis for later semi-automatic explorations. This procedure is summarized in Table 4, with steps feeding into each other consecutively numbered. The choice of number of languages in each step will be explained in Section 4.

In accordance with the stepwise procedure, Section 4 falls into various subsections, whose roadmap is anticipated here. For each subsection there are appendices providing additional data.

In Section 4.1, oppositive and counterexpectative contrast emerge as opposite poles of Dimension 2 in PCoA analyses of both a European and a world-wide sample. Europe is chosen first, because this allows for a comparison with Mauri’s (2008, ch. 4) investigation focusing on Europe. It is shown by means of partitioning that the oppositive–counterexpectative dimension does not lend itself easily for data reduction into types, unlike corrective contrast in Dimension 1.

In Section 4.2 it is shown, by means of a principal component analysis, that non-dominant non-verb-initial word order in VSO/VOS languages and selectives pattern together with oppositive contrast.

At this point, oppositive contrast is established as a region of the similarity space of contrast. However, we still have only very few candidates for oppositive contrast markers. These candidates are taken as “seeds” (see Dahl and Wälchli 2016 and, for a similar approach, Asgari and Schütze 2017) for identifying the most prototypical oppositive contrast contexts in the whole NT. With this oppositive contrast prototype domain we can then search the whole corpus for many more oppositive contrast markers. In Section 4.3, the oppositive contrast markers detected are then further coded manually for expectable differences between contrast markers and selectives (Table 1). In Section 4.4, oppositive contrast markers and selectives are arranged in a space defined by these properties, which allows us to determine to what extent oppositive contrast markers differ from selectives.

4.1 The similarity space of contrast markers

4.1.1 The similarity space of contrast markers in Europe

In this subsection, a similarity space is constructed with PCoA from data from 63 European languages (from eight families, 43 Indo-European, see Appendix 4.1.1.A).

There are many contrast contexts in the NT, English but ranges from occurrence in 706–2,082 verses depending on translation, and the necessity of manual coding requires sampling. The beginning was entirely Anglo-centric.^[5] English was chosen for two reasons. It has a general contrast marker but ^[6] and there are many different translations available, which allows for assessing prototypicality of contexts within a single language before cross-linguistic comparison is started. I first extracted 500 NT verses where but is most frequent across 32 English translations (Appendix 4.1.1.B). Next I selected, within those, verses where but usually only occurs once, thus – for reasons of convenience in annotation – minimizing the number of verses with multiple contrast contexts. Then I picked the 101 first verses, which were subsequently extracted in the set of 63 European languages that constitute the focus of this section. I later controlled for the effect of the Anglo-centric starting point, and the result is that many corrective contrast contexts would not have made it to the set of most prototypical examples with more diverse sampling (which is not surprising given that corrective contrast is often unmarked cross-linguistically, see Mauri 2008: 127).

Figure 2 displays the two first dimensions of the similarity space of contrast in Europe for six languages (for all languages, see Appendix 4.1.1.D; for other dimensions, see Appendix 4.1.1.E). Every symbol stands for a verse, and same color and shape of symbols indicate that the contrast marker is the same in the doculect visualized. Color and shape (and legend order) strictly follow frequency in the dataset in order not to suggest semantic patterns by coloring. Latvian has a general contrast marker bet ‘but’ as English but. On Dimension 1, corrective contrast (negative pole) – illustrated by Romanian ci ^[7] and Estonian vaid – is clearly set apart from non-corrective contrast (positive pole). Only a single context in the dataset is intermediate between corrective and non-corrective contrast (Do not swear falsely, but fulfill your oaths to the Lord). Koine Greek alla encodes corrective contrast, but also covers a part of counterexpectative contrast, and is opposed to de(2). Bulgarian and Russian both use a for oppositive contrast on the negative pole of Dimension 2, opposed to counterexpectative no on the positive pole of Dimension 2. However, in Russian corrective contrast goes with a; in Bulgarian it is split.

Figure 2:

Similarity space of the contrast domain in European languages.

A major result of the PCoA analysis is that Dimension 2 largely corresponds to the semantic/pragmatic gradualness of oppositive contrast as postulated in Table 2, which is shown schematically in Table 5 (see Appendix 4.1.1.C for more details).

Table 5:

Semantic/pragmatic gradualness of oppositive contrast in Dimension 2.

	Strong oppositive contrast >		Merely polar oppositive contrast >	Contexts such as non-responsive contrast >	No or dummy oppositive phrase
Example (shortened if given earlier)	(7) rich people give of abundance, this poor woman put in everything		(6a) poor always, me not always	(6b) they offer wine, he not take	… this must happen, but is not yet the end (40024006)
Value on Dim2	−0.301		−0.162	0.175	0.237
Catalan (see Figure 3)	en canvi2	en canvi1	però
Russian	a			no
Koine Greek	de(2)				alla

Figure 3:

Similarity space of the contrast domain in European languages (further doculects).

For assessing which markers express oppositive contrast we can use Principal Component Analysis. Principal Component Analysis [PCA] (prcomp() in R) provides us largely with the same result as PCoA (but “dimensions” are called “components” in PCA). However, PCA also assigns all variables (here contrast markers) negative or positive loading values for each component, which allow us to assess how strongly each contrast marker correlates with the component (see Appendix 4.1.1.G for the entire list). Table 6 lists selected doculects which all have oppositive contrast markers marked in boldface (they are all associated with the negative pole of Dimension/Component 2 when loadings in the Principal Component Analysis are considered).

Table 6:

Partitioning with three clusters (selected doculects, others in Appendix A.1.1.F).

	Cluster 2: corrective	Cluster 1: counterexpectative	Cluster 3: oppositive
Koine Greek (grc)	alla[45]	de(2)[26], alla[7]	de(2)[20]
Slovenian (slv)	ampak[29], temveč[15]	pa2[16], toda[6], vendar[4], a[3]	pa2[19]
Bulgarian (bul)	no[31], a[13]	no[28], a[6]	a[15], no[4]
Russian (rus)	a[26], no[6], a=liš[3], ZERO[3]	no[30]	a[10], no[6], že(2)[3]
Estonian (est)	vaid[45]	aga[19], kuid[7], aga2[4]	aga[9], aga2[7], kuid[3]
Basque (eus)	baizik3[19], ZERO[14], baizik[4]	baina[29]	baina[8], ordea2[7], berriz2[3]
Catalan (cat)	sinò[39]	però[34]	però[11], en=canvi2[5], en=canvi[4]
Albanian (aln)	por[40], madje[3]	por[33]	por[12], kurse[5], ndërsa[3]

1. Markers sensitive to oppositive contrast in boldface, fields blue when such contrast markers are most frequent in a cluster. Markers with one or two occurrences not listed.

Partitioning-Around-Medoids [R-function pam()] is the statistical equivalent to the common typological practice of assigning markers to rigid functional domains. Figure 3 bottom right shows how the R-function pam() clusters the distance matrix with three clusters.^[8] Three clusters are chosen for comparison with Mauri (2008), who uses the three comparative concepts oppositive, counterexpectative and corrective contrast. The three clusters obtained with Partitioning-Around-Medoids reflect the most optimal classification of contexts by means of rigid functional subdomains for oppositive, counterexpectative and corrective contrast. Blue color in Table 6 indicates all fields where an oppositive contrast marker (boldface) holds the majority within a subdomain. If oppositive contrast markers were equally distributed across doculects, blue color would be equally distributed across all rows of Table 6, which is not the case. Rather, oppositive contrast markers range from markers such as Koine Greek de(2) and Slovenian pa2 (top rows in Table 6), which are most frequent even in the counterexpectative cluster, to doculects such as Basque, where oppositive contrast markers are not even dominant in the oppositive contrast cluster.^[9]

We may conclude that Dimension 2 is gradual in the sense that markers that support Dimension 2 range from rare oppositive contrast markers (mostly restricted to strong oppositive contrast) such as Catalan en canvi to frequent markers such as Slovenian pa2. The large difference between Basque ordea2 and berriz2 and Catalan en canvi1/2, on the one hand, and Slovenian pa2, on the other hand, can also be seen in Figure 3 on Dimension 2 (y-axis) of the PCoA-plots for Basque, Catalan and Slovenian.

While corrective versus non-corrective contrast appears as a rather strict distinction in the similarity space, it is not entirely strict. There are minor expressions for ‘on the opposite’, as, for instance, Russian naprotiv [one occurrence]. But more importantly, there are more intermediate contexts than the PCoA plots suggest. They did not make it into the sample, because they happen to be non-prototypical ‘but’-contexts. Put differently, not only can they be encoded both by typical contrast and non-contrast markers, but they are often left unmarked or expressed by markers meaning ‘only’, ‘instead’ or ‘rather’, such as Spanish No teman a los que … Teman solo a Dios … (40010028) ‘And do not be afraid of those [who kill the body but are not able to kill the soul,] but instead be afraid of the one [who is able to destroy both soul and body in hell.]’ These examples often encode partial correction (hence ‘only’).

The difference in gradualness between the two different dimensions is of uttermost methodological importance. Corrective markers tend to be strictly delimited in prototypical contrast contexts, which is why they are detectable by means of more general comparative concepts, such as those used by Mauri (2008), whereas this is considerably more difficult for the fuzzier oppositive contrast markers. This is illustrated in Figures 4 and 5 where contrast markers with high opposite loadings are shown. The x-axis shows contexts in ranking order according to the single relevant dimension (Dimension 1 for corrective contrast and Dimension 2 for oppositive contrast). Figures 4 and 5 only display languages that reflect the relevant distinction.

Figure 4:

Corrective versus non-corrective contrast in Dimension 1 in selected European doculects.

Figure 5:

Contrast markers sensitive to Dimension 2 in selected European doculects.

Figure 4 shows that corrective (red) versus non-corrective contrast (black) is a rather strict distinction and Figure 5, where various kinds of oppositive contrast markers are plotted in different colors, shows that oppositive contrast markers are distributed in a much fuzzier manner.

4.1.2 Expanding to a world-wide sample

A set of 101 situations could not be handled practically with manual annotation in a larger world-wide sample of 193 languages (see Appendix 4.1.2.A) and was reduced to 52 situations (see Appendix 4.1.2.B). The sample is a diverse convenience sample and is biased towards languages with selectives and languages with verb-initial dominant word order, which will be needed in Section 4.2.

The first two dimensions of the similarity space are largely the same as for Europe irrespective of whether it is constructed with all 193 languages or only with the 129 non-European ones,^[10] which is shown in Figure 6 top left where arrows link the two similarity spaces built by PCoA.^[11] Each arrow stands for a contextually embedded situation and the arrows go from the space built without European data to the space including European data. Different colors help distinguish the arrows from each other. The two spaces are not completely identical, but very similar as shown by a congruence coefficient of 0.99 (see Mair et al. 2022: 39). Bulgarian (top right) is given as an example from the European dataset. This and further maps in this section are based on the 193-language sample.

Figure 6:

Similarity space of contrast markers based on a 193-language world-wide sample.

Let us now consider selected languages illustrating an oppositive–counterexpectative contrast distinction on Dimension 2 (for all languages, see Appendix 4.1.2.C; for evaluation of the dimensions, Appendix 4.1.2.D).

The following two examples demonstrate how PCoA-plots can contribute to description in particular languages.

Somali (bottom left) – the discussion is picked up here from Section 2.2 (example (10)) – has sentence-initial laakiin ‘but’, borrowed from Arabic, and second-position (Saeed 1999: 92) -se ‘but’ (rarely combined as laakiinse ‘but[… on the other hand]’). Mauri (2008: 152) classifies both markers as counterexpectative contrast, but -se is rather oppositive, which can be clearly seen in the similarity space.

In Cherokee (bottom right), =skinii ‘but’ (Montgomery-Anderson 2008: 158) is used for oppositive and corrective contrast. The clause-initial conjunctive adverb aseéhno ‘but/however’ links clauses (Montgomery-Anderson 2008: 535) and can be identified as counterexpectative.

Let us now turn to the discussion of a few languages where different markers interact with different kinds of word order.

Malagasy (Figure 7 top left) is particularly interesting because of its basic VOS order and because it has been argued to have clause-final topics (Pearson 2005). However, oppositive phrases are sentence-initial, combined with postposed kosa2 ‘on the contrary’ (Rahajarizafy 1960: 116) – illustrated in (16) – fa … kosa2 (both clearly associated with the oppositive contrast pole in Figure 7) or with bare fa with an initial NP. Bare fa ‘but; because; complementizer’ is not dedicated to any specific subdomain within contrast (since different word orders with fa are not indicated in Figure 7), whereas (ka)nefa ‘however’ and saingy ‘but aside from that’ are clearly counterexpectative.^[12]

Figure 7:

Further evidence for oppositive contrast markers from the world-wide sample.

(16)

Plateau Malagasy (plt-x-bible-interconfessional 42007046)
Ianao	tsy	n-anosotra	diloilo	ny	loha-ko;
2sg	not	pst-anoint	oil	def	head-poss.1sg
izy	kosa	n-anosotra	ilomanitra	ny	tongotro
3sg	but. opp	pst-anoint	perfume	def	foot.poss.1sg
‘You did not anoint my head with olive oil, but she anointed my feet with perfumed oil.’

Maori (Figure 7 top right) has a rich set of markers (Mauri 2008: 152 only lists engari). Of particular interest is Maori ko (called “topicalizing particle” in Bauer 1993: xvii, 398), which triggers subject-initial word order, so called “ko-fronting” (basic word order is VSO). However, atypically for a topic marker, it can mark contrast on its own without another contrast marker present. As can be seen in (17), ko may occur both in anchor sentences and contrast sentences (for further functions, see Pearce 1999). Maori ko has some properties of a topic marker and some properties of a contrast marker, without being a prototypical instance of any of them (and, as can be seen in Figure 7, it is not entirely restricted to oppositive contrast).

(17)

Maori (mri-x-bible 40024035)
Ko	te	rangi	me	te	whenua	e	pahemo,
“top”	def	sky	with	def	earth	tma	pass.away
ko	aku	kupu	ia	e	kore	e	pahemo.
“top”	1sg	word	3sg	tma	neg	tma	pass.away
‘Heaven and earth will pass away, but my words will never pass away.’

Ama is particularly interesting in two respects. The contrast clitic -su following NPs and ulai ‘but (signaling the unexpected)’ (Årsjö 1999: 84, 33) are oppositive and counterexpectative only as a tendency (see Figure 7 bottom left) and Ama can combine -su with a following selective mo top, as shown in (18). As in other languages where such a combination occurs, selectives tend to follow oppositive-phrase-final oppositive contrast markers.

(18)

Ama (amm-x-bible 40024035)
Yo- su	mo	nuwoi	muwoi.
1sg- but. opp	top	certain.day	neg
‘But I have no certain time.’

Tzotzil illustrates the complex interplay of different contrast markers with different word orders and a marker with properties of a selective. Tzotzil has VOS basic order, but contrasted phrases in oppositive contrast are placed before the verb, usually followed by a final particle =e (Aissen 1992: 49; glossed fin). The final particle behaves like a selective except that it also often occurs sentence-finally.^[13] The most frequent contrast marker is pero (from Spanish), used both with and without initial NPs. Since word order is not shown in Figure 7, pero extends over both oppositive and counterexpectative contrast. However, there is also the oppositive contrast marker yan ‘but’ illustrated in (19a). Example (19b) is a non-responsive example with pero, which, however, has an initial oppositive phrase marked with =e:

(19)

Tzotzil (tzo-x-bible-chamula 40026011, 41015023)

a.

Ti	buch’u-tic	abol	s-ba-iqu=e	li’
def	rel-pl	poor	3-refl-pl=fin	here
ono’ox	oy	ta	a-tojol-iqu=e.
always	ex	obl	2-in.presence-pl=fin
Yan	ti vu’un = e		mu	scotol-uc	c’ac’al
but. opp	def 1sg = fin		neg	all-every	day
li’	oy-un	ta	a-tojol-iqu=e.
here	ex-ptc	obl	2-in.presence-pl=fin
‘… the poor you always have with you, but you do not always have me.’

b.

Tey	laj	y-ac’-be-ic	y-uch’	ti	Jesus= e
there	aux.finish	pst.3-give-io-pl	pst.3.sg-drink	def	Jesus=fin
jun	ch’ail pox	ti	cap-al	ta	mirra=e.
one	liquor	def	mix-poss	of	myrrh=fin
but	def Jesus= fin		neg.ex	aux.finish	pst.3.sg-drink
Pero	ti Jesus = e		mu’yuc	laj	y-uch’.
‘… they attempted to give Jesus wine mixed with myrrh, but Jesus did not take it.’

In this section we found that the same two dimensions that characterize the contrast domain in Europe are equally manifest in a world-wide sample. As in Europe, the distinction between corrective and non-corrective contrast is more rigid (Dimension 1), and the oppositive–counterexpectative distinction is fuzzy (Dimension 2). This section has also illustrated a number of languages where contrast markers interact with different word orders and with selectives or selective-like markers. Such languages can also have dedicated oppositive contrast markers, such as Malagasy kosa2 and Tzotzil yan.

4.2 The oppositive-versus-counterexpectative contrast dimension in relation to selectives and sentence-initial oppositive phrases

This section explores the relationship of the oppositive–counterexpectative contrast dimension with word order and selectives in the world-wide sample introduced in Section 4.1.2. For this purpose we will zoom in on non-corrective contrast, which means that all examples with negative values on Dimension 1 in Section 4.1 are removed.^[14] As we have seen in Section 4.1, corrective contrast is a sharply delimited cluster whose examples are not particularly sensitive to the oppositive–counterexpectative distinction. This leaves us with 35 examples per doculect, which are now also coded for presence or absence of selectives in a set of languages with selectives (Appendix 4.2.A) and non-predicative initial word order in VSO/VOS-languages (Appendix 4.2.B).

Example (20) illustrates presence and absence of a selective (top) from Hills Karbi.

In non-corrective contrast, the oppositive–counterexpectative distinction is the dominant signal. Hence, the first component of a Principal Component Analysis will reflect it and it will also show how selectives and word order relate to it when this data is added as additional variables.

Appendix 4.2.C lists all PCA-loadings for Component 1, and it turns out that all but one selective [TOP = “topic marker”] values and all non-dominant non-verb-initial word-order-values are negative in the same way as oppositive contrast markers, such as Russian a, which means that the hypothesis of a correlation between oppositive contrast, selectives and initial oppositive phrases is fully confirmed. The loadings of all markers are plotted on the y-axis in Figure 8. Markers and constructions associated with oppositive contrast (selectives, non-initial phrases in dominant verb-initial word order and oppositive contrast connectives) – all grouped to a block in the legend – appear in the lower part of the figure as opposed to counterexpectative contrast connectives in the upper part.

Figure 8:

Markers and constructions sensitive to the oppositive–counterexpectative distinction.

There are also some differences between oppositive contrast markers, selectives and non-dominant initial word order. Clausal oppositive phrases are more prone to be initial than to host oppositive contrast markers. This is expected, since initial adverbial clauses are very common even in verb-initial languages. There are many infrequent oppositive contrast markers (few tokens, to the left in Figure 8), whereas selectives and initial oppositive phrases tend to be more frequent across the contrast domain (to the right in Figure 8). In languages with two types of marking strategies, the occurrence of an oppositive contrast marker is almost always combined with an initial oppositive phrase in VSO/VOS-languages, see, for instance, Tzotzil (19).

The only doculect in the sample lacking non-dominant non-verb-initial word order in oppositive contrast is Contemporary Welsh.^[15] The Welsh translation from 2013 has only one example with an initial oppositive phrase among contexts sampled, and this example has an initial adverbial subordinate clause, where initial order is expected. However, the translation from 1804 (a modified version the Morgan-translation from 1588, which is still close to Middle Welsh) has many cases of initial oppositive phrases, with the translation from 2004, with some conservative elements, being intermediate. The Irish, Waray and Car Nicobarese translations have low incidences of initial oppositive phrases. Car Nicobarese pöri ‘but’ (21a) usually follows the predicate (Braine 1970: 227). However, in two of the most prototypical oppositive contrast examples (see 21b), it follows an oppositive phrase rather than the predicate.

No systematic coding of initial oppositive phrases in OVS-languages was made, but “fronting” in contrast is common in Hixkaryana (OVS; Derbyshire 1979: 72), illustrated in (22). The adversative mak ‘contrary to expectation’ is placed after or before the verb sentence-medially, but oppositive contrast also tends to have an initial phrase followed by particles.

Hixkaryana maintains a flavor of OVS by often using initial cataphoric pronouns with the co-referential lexical item placed finally: Moson haxa ryhe […] wosà ‘this.one but. opp emphasis … woman’.

In word order typology, a distinction is sometimes made between languages with given–new order (as in Slavic) and languages placing “indefinite, new, or less expected information first”, such as (23) Tohono O’odham (Payne 1987: 802). Interestingly, this does not result in different word order preferences in oppositive contrast.

Both kinds of languages have initial oppositive phrases, which is fully compatible with the function to “instruct the hearer to mentally ‘tag’ the entity as something to be available for deployment” (Payne 1987: 798–799); this is tantamount to saying that common ground is established stepwise.

4.3 Extracting oppositive contrast markers with high cue validity

This section sets up an automatic search procedure for salient oppositive contrast markers across all doculects of the Bible corpus. Methodologically, this implies shifting from manual annotation to methods relying on automated discovery, which allow for larger samples of doculects. The procedure (inspired by Dahl and Wälchli 2016 and Wälchli 2019) consists of the following steps:

1. Collect “seed” markers for oppositive contrast (markers for which there is previous evidence that their distribution singles out oppositive contrast) from the items with top-ranked loadings in the Principal Component Analysis in Section 4.2 (Appendix 4.3.A).

2. Do the same thing for counterexpectative contrast, because we need a filter that removes all markers predominantly used in the counterexpectative contrast domain (Appendix 4.3.B).

3. Compile two sets of verses with most “seed” markers for (a) and (b), which define in which contexts oppositive contrast markers are expected to occur and not to occur (Appendix 4.3.C/D).

4. Extract all markers that collocate with the oppositive contrast domain above a certain threshold^[16] and that collocate more strongly with it than with the counterexpectative contrast domain, and let the counterexpectative contrast domain be stronger (i.e., larger) for eliminating any risk that markers not primarily restricted to oppositive contrast will be extracted by coincidence (implemented in a Python-program).

5. Check manually with available reference grammars and dictionaries whether the extracted markers indeed reflect oppositive contrast markers.

The search procedure is designed to have high accuracy (general and counterexpectative contrast markers are reliably removed), but limited coverage: only salient markers with high cue validity make it to the sample. Notably, affixes will be extracted to a much lesser extent than wordforms. Since the purpose of this study is to establish oppositive contrast markers as a category type rather than detecting all languages where they occur, it is sufficient to find many.

The procedure does not contain any filter to exclude selectives, which is why selective-like markers may occasionally be extracted. However, most selectives will not reach the collocation threshold, since they are not sufficiently dedicated to contrast (nearly all selectives are very frequent).

Only 36 of the 50 seed markers in (a) make it to the extraction (Appendix 4.3.F), but it is important to note that extracted markers can cover very different zones of Dimension 2 in Section 4.1, ranging from markers largely restricted to strong oppositive contrast, such as Basque berriz, to markers covering a broad zone such as Koine Greek de(2) (see Table 6). Collocation measuring applies on the level of verses and no filter prevents them from occurring in the anchor sentence (but most extracted markers are in practice restricted to contrast sentences).

Markers in 255 doculects from 199 different languages were extracted. In several languages, extraction is not constant across doculects, which testifies to the large amount of language-internal variability of oppositive contrast markers. For instance, French tandis is extracted only in one of 17 translations and Italian invece only in two of seven translations (see Appendix 4.3.G).

Extraction of counterexpectative contrast markers (checked also with reference grammars and dictionaries, wherever available) reveals that in all doculects with extracted oppositive contrast markers another or several other markers are used in counterexpectative contrast contexts (see Appendix 4.3.F). There is one minor type of wrongly extracted markers: emphatic personal pronouns of the third person singular, such as Xhosa (xho) yena, in only six languages (see Appendix 4.3.E). Emphatic personal pronoun errors are not unexpected since emphasis of oppositive phrases clearly goes hand-in-hand with oppositive contrast.

For further exploration, 20 verses where both oppositive phrases and anchor phrases can be quite easily identified (see Appendix 4.3.H) have been manually coded across all doculects with extracted markers for four properties ((b)–(d) are taken from Table 1). On the basis of the 20 occurrences in the 20 selected verses, the four following indexes are calculated (divided by the number of attested occurrences, hence ranging between 1.0 and 0.0) Figure 9 displays all four properties and shows that no property always assumes extreme values:

1. Frequency: whether the marker is found in that verse [0–20 tokens indicated by size of circle in Figure 9]

2. Word order: whether the marker occurs before (red in Figure 9, e.g., Russian a_) or after the oppositive phrase (blue in Figure 9, e.g., Georgian -ki; phrase-internal order as in Koine Greek de(2) is treated the same way as OP-final, since the two cannot be distinguished in phrases consisting only of one word);

3. Combinability with contrast markers: whether it combines with (left on x-axis in Figure 9) or does not combine with another contrast marker (right on x-axis in Figure 9),

4. Occurrence: whether it behaves selective-like in also occurring with the anchor phrase in the anchor sentence (bottom on y-axis in Figure 9) or not (top on y-axis in Figure 9). Examples with marking only on the anchor phrase, but not on the oppositive phrase are also counted as selective-like.

Figure 9:

Properties of extracted oppositive contrast markers.

All values are also given in Appendix 4.3.I.

The background of Figure 9 is a two-dimensional logarithmic histogram or heatmap where darkness on a grayscale reflects density of markers. The black square in the upper right corner shows that this area hosts the majority of extracted markers (almost all red, i.e., OP-initial). Table 7 shows that the dominant values are (i) not combined with other contrast marker (right), (ii) no marking on the anchor phrase (top) and (iii) initial position (red). However, if the construction contains another contrast marker (left), the oppositive contrast marker strongly tends to follow the oppositive phrase (blue dominates in Figure 9 except in the top right corner).

The Batakic markers Batak Dairi (btd) ukum and Toba Batak/Batak Angkola (bbc/akb) anggo stick out (red top left in Figure 9). They are OP-initial (red), mostly combine with another contrast marker and sometimes also occur before anchor phrases. The Batakic markers are also the only ones deviating from (24):

(24) is dominant even in Batakic (see also van der Tuuk 1864: 364), but, as illustrated in (25), an oppositive phrase can also, more rarely, occur sentence-finally in some kind of antitopic-position. This is a rare case where [P1] Sentence-initial position in 2.2 is violated.

If the extracted oppositive contrast marker and another contrast marker in the same language occur in different positions relative to the oppositive phrase, the oppositive contrast marker is usually OP-final or OP-internal (26):

(26) restates the hypothesis formulated in (9) in Section 2.2. An exception is the Western Arrarnta initial oppositive contrast marker kanha whereas another more general contrast marker pula occurs in second position.

In the 20 oppositive contrast contexts surveyed, oppositive phrases are almost always the initial phrase in contrast sentences if combined with one of the extracted oppositive contrast markers. This does not entail, however, that all oppositive contrast markers are immediately followed by an oppositive phrase if these markers are used in entirely different functions.

Very few examples go against the trend in (27). One is Bora (28), where áánetu (Thiesen and Weber 2012: 290) does not always have an immediately following oppositive phrase. The word order is similar to that of English … but you do not always have me , but English is no exception to (27), because but is no oppositive contrast marker. All exceptions, such as Bora (28), have initial contrast markers.

To summarize, many languages have salient oppositive contrast markers. The simple automatic search algorithm used here identifies them in more than a seventh of the languages of the corpus in languages from all continents.

4.4 The relationship between oppositive contrast markers and selectives

According to Wälchli (2022), contrast markers only occur once in the contrast construction (within or immediately preceding the contrast sentence), whereas selectives also occur in the anchor sentence. However, some oppositive contrast markers extracted in Section 4.3 sometimes occur even with anchor phrases in the anchor sentence, which suggests that there might be a continuum between oppositive contrast markers and selectives. Let us now look at how a set of 63 selectives (Appendix 4.4.A) behaves under exactly the same conditions. Figure 10 displays the same properties as Figure 9. Its left hand side maps selectives only, whereas its right hand side displays both oppositive contrast markers and selectives. Symbols for selectives are distinguished by light blue outline. The comparison reveals that selectives tend to occur following anchor phrases in anchor sentences in at least one of three cases (index value on y-axis below 0.66),^[17] in word order mostly follow the contrasted phrase (blue) and mostly occur in combination with a contrast marker (left).

Figure 10:

Ratio of occurrence with AP: selectives (left) and oppositive contrast markers with selectives (right).

While Figures 9 and 10 show that oppositive contrast markers and selectives generally tend to have different properties, there are a small number of extracted oppositive contrast markers that actually have properties similar to selectives.

There is only one selective also extracted as oppositive contrast marker in Section 4.3: Hills Karbi (mjw) ke. However, the extraction will have missed some intermediate markers (for instance, Maori ko, see (17)).

To summarize, although oppositive contrast constructions host both oppositive contrast markers and selectives, these two types of markers do not have the same distribution within oppositive contrast constructions. Whereas oppositive contrast markers tend to be strongly restricted to occurring with oppositive phrases in contrast sentences, selectives also occur with anchor phrases in anchor sentences. However, since markers differ in how often they occur with anchor phrases, there is a continuum between oppositive contrast markers and selectives, but the intermediate area is not particularly crowded. In most cases, oppositive contrast markers and selectives can be clearly distinguished.

5.1 The interplay of oppositive contrast markers, selectives and word order

In Section 4.2, it was shown that oppositive contrast markers, selectives and initial non-predicative constituents in VSO/VOS-languages all occur in the oppositive contrast domain. This overlap in use is evidence for some sort of interplay; it clearly demonstrates that the three phenomena types overlap in use. However, the categories neither strictly condition each other nor are they all conditioned by a single underlying syntactic structure. We have seen in Section 4 that there are a large number of markers and constructions that are somehow sensitive to oppositive contrast. Whether or not their opposition to other markers in the same language is stable or emergent, they do not cover exactly the same area of the oppositive contrast region of the similarity space, thus demonstrating by their cumulative cross-linguistic evidence that oppositive contrast is fuzzy. In languages with more than one relevant marker or construction, fuzziness can also be demonstrated within one language. This is illustrated here with Popti’, a language with several different phenomena sensitive to oppositive contrast.

Popti’ has basic VSO-order (Craig 1977: 8) with preverbal negation. A sentence-final first person clitic =an, glossed fin.1, indicates that some subject, object or possessor in the sentence is first person. This marker also occurs at the end of the topic intonation unit (Aissen 1992: 61). Put differently, in some uses =an is a kind of selective cumulating with, and restricted to, first person. However, preverbal position of non-predicative phrases can also be triggered by the oppositive contrast marker wal (Craig 1977: 37), but this construction differs from topicalization in that pronouns can occur preverbally, as in (29), which is not allowed in topicalization (Craig 1977: 56).

Wal can be preceded by yaj/yaja’ ‘but’, but even bare yaj(a’) can attract initial NPs. In addition, the particle xin ‘then’ may follow an =an-clitic on an initial phrase (Day 1973: 97). While all these different phenomena are associated with initial non-predicative phrases, their distribution does not match exactly and cannot be explained by one single pre-established category such as a Top(ic)P(hrase)), testifying to the fuzziness of the oppositive contrast domain.

It may be assumed that initial oppositive phrases are the most important phenomenon type characterizing oppositive contrast cross-linguistically. Non-predicative initial phrases occur in an overwhelming majority of all VSO/VOS-languages considered in Section 4.2, and there is reason to believe that non-predicative initial phrases also occur in many other languages. The phenomenon just happens to be less salient in subject-initial languages where contrasted subjects are initial anyway. Oppositive contrast markers and selectives are more restricted in use: oppositive contrast markers have been identified in more than a seventh of the languages and selectives have so far been identified only in less than a tenth of the languages of the Bible corpus. Note that there are only very few exceptions to (27) “Oppositive contrast markers are immediately adjacent to oppositive phrases in prototypical oppositive contrast contexts”. This means that oppositive contrast marking will be induced by syntax to a very large extent. If many languages have some sort of oppositive contrast markers, as shown in Section 4.3, this is because various kinds of markers can assume this function if they happen to be associated with contrast-sentence-initial oppositive phrases. From this follows the assumption that many markers sensitive to oppositive contrast will not be dedicated to oppositive contrast in all of their functions, such as the particle Popti’ xin ‘then’ in (29).

A further illustrative example is the emphatic second position clitic že in Slavic languages which differs in its propensity to be associated with oppositive contrast from language to language. It is not formally ideal for signaling oppositive phrases, because it usually occurs after the first word. It is most strongly associated with oppositive contrast in Church Slavonic, where it is a calque from Koine Greek de(2), and in those Russian translations that are most strongly influenced by Church Slavonic. It is also common in Ukrainian (often reduced to ž, see example (14)). However, its borrowed version in Erzya Mordvin žo is rather strongly associated with oppositive contrast (high loading in PCA-analysis). Not incidentally, Erzya Mordvin žo is noun- and postposition-phrase-final (l’ija-tn’e-n’ marto žo [other-pl.def-gen with but.opp] ‘but with the others …’ 42008010), except if a relative clause follows. Put differently, the syntactic position of Erzya Mordvin žo affords oppositive contrast more easily than Slavic že.

We can conclude that although oppositive contrast can be conceived of as a semantic domain, it is syntactically co-conditioned. Oppositive contrast provides evidence that functional domains are not pre-established extralinguistic semantic domains, but units reflecting non-arbitrary relationships between meaning and form.

This raises the question as to how non-arbitrary relationships between meaning and form should be explained, a question that can only be touched upon here. The most often adduced candidates are iconicity (Haiman 1985) and frequency, but these are not the only options. Gibson’s (1979) notion of affordance (see also Du Bois 2014: 367) is a further candidate. The notion of affordance implies that properties of objects can be immediately meaningful to observers without recourse to previous experience (such as a stone directly “affords” graspability and throwability). For contrast constructions this means that an emphatic particle following a non-predicative initial phrase in a contrast sentence directly affords oppositive contrast, which can explain why oppositive contrast markers are often emergent (weakly conventionalized).

5.2 The consequences of fuzziness of the oppositive contrast domain

The results of this study show that the oppositive contrast domain is fuzzy with scalar contexts with unexpected oppositive phrases such as (30) reflecting the most prototypical examples. Contrast is a rhetorical discourse relation, which is why it is important to consider real corpus examples rather than isolated constructed examples such as (1). Not incidentally, all examples on the oppositive pole of the oppositive–counterexpectative dimension in Section 4.1 are from direct speech. The major discourse function of the most prototypical oppositive contrast examples in the NT is explanation where some sort of wise speaker corrects what is rhetorically framed as an addressee’s misconception. New common ground is intersubjectively established stepwise. Example (30) is part of Jesus’ explanation to Simon the Pharisee why a woman, who was known to be a sinner, acted in an impeccable way.

Fuzziness of the domain is mainly due to degree of emphasis. Strength of emphasis presupposes degree of intensity on a scale. The most prototypical strong oppositive contrast contexts have multiple scales, such as the three scales in (30), shown in Table 8.

To summarize, gradualness of oppositive contrast between contrasted phrases is mainly rhetorical in character and highly context-dependent (cannot easily be addressed in isolated examples without context). It is strongly connected to degree of emphasis of oppositive phrases (see also Section 2.2).

5.3 Limitations and outlook

This study has many limitations. The functions that could be established on the oppositive–counterexpectative dimension (Table 5) are dependent on what kind of examples happen to occur in the NT. There are only few examples with strong oppositive contrast, such as (30), and they all happen to have a woman as the surprising oppositive referent (interestingly, oppositive contrast is also a relevant domain for incipient feminine gender markers, see Wälchli 2019: 65). Further research is needed to explore more details of the oppositive–counterexpectative distinction. However, if there are strong limitations on contexts available in the NT, this data source is more useful than reference grammars and dictionaries alone where examples are usually few and usually presented without discourse context.

This study focuses on a restricted region of the similarity space of contrast (for a larger picture, see Malchukov 2004). Concessive and restrictive contexts and contexts intermediate between contrast and coordination were not considered.

Most importantly, there is also a need for many language-specific studies that investigate oppositive contrast in original texts, taking into account that a large amount of language-internal variability may be expected. Very few markers sensitive to oppositive contrast have so far been subject to more detailed study, mainly from Indo-European languages, for instance Polish a (Andrason 2020) and Latin autem (Kroon 2019, ch. 10).