Non-exhaustive connectives

1.1 Overview

Coordinating connectives have been widely studied in a typological perspective, as witnessed by the large amount of literature available (see Haspelmath 2004, 2007; Longacre 1985; Mauri 2008a; Mithun 1988; Payne 1985; Stassen 2001 among others). Yet, their classification has been mainly organized on the basis of the semantic relation they encode, typically leading to identify conjunctive, disjunctive, adversative and in some cases conclusive connectives, roughly corresponding to ‘and’, ‘or’, ‘but’, and ‘so’, respectively. The cross-linguistic variation that can be observed, once the analysis becomes more fine-grained, has led scholars to distinguish between different subtypes within each type, with disjunctive connectives being classified as ‘interrogative’ or ‘simple’ (cf. Haspelmath 2007; Mauri 2008b) or adversative connectives being classified as ‘counterexpectative’, ‘corrective’ and ‘oppositive’ (cf. Malchukov 2004). However, little if no attention has been devoted to those coordinating connectives whose semantic core is not a specific semantic relation, but rather a property having to do with the set they create, which can be a closed or an open set. Despite their shy presence in the literature, such connectives do exist and constitute a challenge for the existing accounts of coordinating connectives.

The aim of this paper is to provide the first systematic description and typological study on non-exhaustive connectives, that is, coordinating connectives encoding the openness of the set they create. Based on a 215-language sample, we will i) provide a semantic and structural definition of non-exhaustive connectives, ii) we will describe their cross-linguistic distribution and variation, and iii) we will describe the attested diachronic patterns leading to their development.

We will start by discussing the notion of non-exhaustivity in purely semantic terms (Section 1.2), briefly exploring the linguistic strategies that can be employed to express this function. Section 2 will be devoted to the methodological issues raised by the definition and search for non-exhaustive connectives across languages. Special attention will be paid to the problems that the analysis of an underdescribed phenomenon such as the one at issue may raise, discussing in detail the language sampling technique adopted and the criteria we followed to manage borderline cases, namely cases in which the grammar was not clear or consistent in the description of the phenomenon.

Once the theoretical and methodological premises are set, we will provide a synchronic picture of the actual distribution of non-exhaustive connectives in our sample (Section 3). Based on the analysis of the 35 languages for which at least one non-exhaustive connective has been found, it will be argued that non-exhaustive connectives show quite homogenous distributional properties and frequently consist of multifunctional morphemes. Such multifunctionality is explored in its diachronic implications in Section 4, where the attested sources are exemplified and discussed. Some of the diachronic patterns identified appear to be frequent, while others are attested in a few languages, but they all point to a unitary consistent picture, that will be the object of Section 5. Based on the synchronic and diachronic evidence of this study, we will argue that list constructions play a crucial role in the development of non-exhaustive connectives. In particular, we will argue that they arise in lists which first express non-exhaustivity as a construction, and only later the markers are reanalyzed as non-exhaustive connectives themselves. Furthermore, the attested sources reveal the close connection between non-exhaustivity on the one hand, and epistemic uncertainty and exemplification on the other hand, which address respectively the motivations and the tools underlying the expression of non-exhaustivity in discourse. Some conclusory remarks are sketched in Section 6.

1.2 The notion of non-exhaustivity and its linguistic expression

A definition of non-exhaustivity is difficult to find in the literature, where this notion has been mainly referred to as opposed to exhaustivity, especially within formal approaches to focus particles and negative polarity items (cf. Chierchia 2004, 2006; Chierchia et al. 2009; Giannakidou 2016; Lin and Giannakidou 2015). In the literature on focus markers we find a definition of exhaustive identification which provides us with an operational starting point: Kiss (2010, on exhaustive focus operators in Hungarian) defines exhaustive identification as operating “on the set of contextually determined elements for which the predicate of the sentence can potentially hold”, exhaustively identifying “the proper subset for which the predicate actually holds, excluding the complementary subset” (Kiss 2010: 68).

Non-exhaustivity is explicitly examined by Giannakidou (2016), who analyzes it in terms of referential vagueness, whereby a given referential expression is indeterminate as far as the value of a given indefinite element is concerned. Moreover, she argues that non-exhaustivity is closely linked to the speaker’s epistemic stance, either because of indifference or because of ignorance. A more comprehensive discussion is provided by Xiang (2016) on non-exhaustive answers, who distinguishes three types of non-exhaustivity: i) mention-some, where only one of the possible answers for a given question is provided; ii) partial readings, namely answers to questions including for example; iii) choice readings, that is, answers where the speaker is free to choose one referent out of many.

In the light of the existing literature, we define non-exhaustivity as a referential property of a plural set (cf. also Mauri et al. 2019a). The set comprises contextually determined elements for which a specific predication holds, and non-exhaustivity operates on the set by opening it to referentially vague additions. If we call ‘Set’ the plural set and X₁ X_{2 …} X_n its members, we can say that a non-exhaustive linguistic expression makes reference to a set S comprising X_n items and at least one further implicit and referentially vague additional item X_n+i (with i being a natural number ≥1):

As argued in detail by Mauri and Sansò (2018, 2019, while non-exhaustivity in itself is only concerned with the quantification of the set denotation, the interpretation of non-exhaustivity in discourse requires a further step, namely accessing context to identify the relevant property shared by the set members, making it possible to identify specific values for X_n+i. In other words, the notion of non-exhaustivity as such only implies the existence of additional set members, thus concerning their amount, rather than their identification. However, the communication and interpretation of non-exhaustivity do also require the possibility to identify the additional set members, which crucially does not equate to the necessity to do so. A non-exhaustive construction can indeed be processed if the hearer identifies the property or frame that allows to discriminate between possible and impossible set members, but the exact identity of the set members may remain non-specific. Mauri and Sansò (2018) discuss this process in terms of ad hoc categorization, since the abstraction of the contextually relevant property leads to the construction of a contextually determined category.^[1]

The identification of X_n+i requires what Barotto and Mauri (2018) describe as a process of indexical saturation, whereby what has to be saturated is precisely the value of the shared property P, not the reference of each set member. Let us consider example (2):^[2]

According to Barotto and Mauri (2018), once the hearer has to react to a sentence like (2), she is aware that X_n+i has to be assigned a specific value, her problem is how. The solution comes from context and from the frame it activates, which may lead to identify the specific property or properties that allow X_n+i to be part of the set: if the frame is ‘we are preparing an artichoke quiche’, then the underlying property will be ‘ingredient necessary for an artichoke quiche’, and eggs would be a good candidate for X_n+i. Non-exhaustivity thus implies the existence of some exceeding quantity, which can only be interpreted by accessing context, abstracting the relevant property from the explicit set members and identifying the additional one(s) through analogical reasoning.

Non-exhaustivity is involved in a number of different linguistic phenomena, either as an entailment or as the core meaning being coded. Non-exhaustivity is indeed entailed by the semantics of several constructions directly encoding other meanings, such as exemplifying markers (e.g. for example, such as, including, cf. Lo Baido 2018) and additive focus-sensitive particles (e.g. also, too, cf. De Cesare 2010; König 1991; Ricca 2017). Although these constructions do not explicitly encode the existence of other elements beyond those mentioned, their core meaning entails non-exhaustivity. This becomes evident if we remove the exemplifying marker in (3b) and the additive focus-sensitive particle in (4b), and if we explicitly deny the reference to Set(X_n+i), leading to contradictory statements (3c, 4c):

The entailment of for example and also is that the elements mentioned (e-mail spam in (3) and catering services for special functions in (4)) are not the only ones for which what is predicated holds true, that is, in both cases the existence of X_n+i is entailed.^[3] In the case of exemplifying constructions, non-exhaustivity is necessary in order to conceive exemplification in itself, because examples, by their very nature, must be representative of a larger set of similar items.

In addition, languages show a wide range of strategies that directly encode non-exhaustivity, that is, they have it as their core meaning. The strategy that has received more attention in the literature is general extenders (see Overstreet 1999), that is, words or small expressions (e.g. and such) that occur at the end of a list to indicate the existence of X_n+i, as in (5):

Mauri and Sansò (2018) provide a broad typological account of other less common strategies that are used in the languages of the world to encode non-exhaustivity: i) heterogenous plurals, such as similative plurals (Daniel and Moravcsik 2013) and associative plurals (Corbett 2000: 101; Moravcsik 2003), ii) a specific type of reduplication called ‘echo-reduplication’ (see also Barotto and Mattiola 2018), and iii) so-called non-exhaustive connectives. Heterogeneous plurals (cf. Mauri and Sansò 2018: 19, 2019) are special types of plural that refer to heterogeneous sets of referents. Let us consider (6).

In Cavineña, the plural marker -kwana does not make reference to many karetu ‘carts’, but to a wider set of elements X_n+i that are not identical but only similar to the one mentioned (‘cart and everything’). In our view, as will become clear in Section 1.3, general extenders and similative plurals are only distinguished based on syntactic and distributional features, having to do with the ability of the marker to take scope over a single noun (similative plurals) or a whole list (general extenders).

Echo-reduplication is rather different, in that it involves the “reduplication of a word, with replacement of the onset or, sometimes, vocalism or internal material in one copy” (Inkelas 2014: 170; Stolz 2008). Let us consider an example from Kannada, where this construction involves the reduplication of an element through the replacement of the first consonant and vowel of the noun with the sequence gi- or gi:-.

Similarly to heterogeneous plurals, echo-reduplication has scope over a single item and makes reference to a wider set of elements that are semantically related to the one mentioned (cf. Inkelas 2014: 171). What is different is the phono-morphological mechanism at work.

In addition to the linguistic constructions described above, there is (at least) one more attested strategy to explicitly encode non-exhaustivity, the so-called non-exhaustive connectives, to which the rest of this paper is dedicated. Whereas reduplication and similative plurals have been described in the typological literature, albeit not extensively as far as non-exhaustivity is concerned, non-exhaustive connectives are characterized by a descriptive and theoretical gap, which makes it especially interesting and urgent to provide a comprehensive cross-linguistic study.

4.1 Heterogeneous plural > nEx connective

We observe a recurrent multifunctionality pattern involving plural markers and nEx connectives, with clear evidence supporting a diachronic change from the former to the latter. The plural markers involved are mainly ‘heterogeneous plurals’, which Mauri and Sansò (2019 and 2021) define as plural markers referring to heterogeneous sets of items, i.e. items that are similar or associated to the one(s) explicitly mentioned (contrary to additive plurals, commonly referring to homogeneous sets). This label encompasses both associative plurals (i.e. expressions meaning ‘X_[person] and associates’) and similative plurals (i.e. expressions meaning ‘X and things like that’, see Daniel and Moravcsik 2013). We can observe this multifunctionality pattern in example (19) from Burmese (Sino-Tibetan), where the marker toú is used with non-human referents to signal similative plural (cf. 19a) and with human referents to signal associative plural (cf. 19b),^[12] but it is also attested as nEx connective to link elements in open-ended lists (cf. 19c).

In two cases, namely in Jamsay Dogon and Mundari, we can identify a plural source for the connective, but we do not have clear evidence supporting the heterogeneous nature of the set it creates. In the case of Jamsay Dogon, we were indeed not able to find examples of bé as heterogeneous plural, although we know that in other Dogon languages (e.g. Tommo So) additive plural and heterogeneous plural are encoded by means of the same element (cf. mbe, diachronically related to bé). In the case of Mundari, we observe that -ko is attested as third person plural clitic pronoun (‘they’), as additive plural primarily used with animate nouns (e.g. hoṛo ‘man’ hoṛo-ko ‘the men, people’, lit. ‘man-they’, see Cook 1965: 120), and as nEx connective primarily used with inanimate entities. Nouns denoting inanimate entities indeed usually do not take the plural suffix, unless they are used in enumerations and -ko is “suffixed to each of the objects in the whole series enumerated” (Cook 1965: 120). Mauri and Sansò (2021) argue that third person plural pronouns are a frequent diachronic source for additive and associative plurals. Therefore, it is likely that the pronoun is the original source out of which the plural function first and the nEx connective later on developed.

Table 3 illustrates the languages for which the pattern at issue is observed, indicating the connectives and the diachronic source identified. As the column ‘Synchronic layering’ shows, in 12 out of 13 connectives the source and the target function co-exist in synchrony (see Hopper 1991), and only for Alabama we were not able to find explicit information on synchronic layering.

The diachronic pattern leading from (heterogeneous) plurals to nEx connectives is the most frequent in our sample. The high frequency of this pattern can (at least partially) explain why nEx connectives are often restricted to noun phrases (cf. Section 3) and are rarely used to link verbal phrases and clauses.

As noted in Section 2 for Kanuri, the development of nEx connectives out of heterogeneous plurals likely requires an intermediate step, in which speakers create lists of juxtaposed plurals. This intermediate step can be observed for instance in Tommo So (Dogon, McPherson 2013), where we find a similative plural^[13] marker mbe (cf. 20a), which can be repeated after each list item in enumerations (20b), favoring its reanalysis as nEx connective (see Section 5).

While McPherson (2013) does not acknowledge the status of mbe as nEx connective and only provides examples like (20b), showing its repeated use within lists, the discussion by Corbett (2000, citing Plungian 1995) makes the functional extension more evident, with the use of the general extender ‘and similar things’ in the translation of (21):

Interestingly, in Jamsay Dogon (Dogon), we can find a morpheme bé which originated from a plural marker and is also used to create “open-ended lists”, according to Heath (2008: 272):

Therefore, it appears that in some Dogon languages there is a (similative) plural marker bé/=mbe that tends to occur in non-exhaustive lists, which at some point has been reinterpreted as a nEx connective.

A similar phenomenon is also attested in four Sino-Tibetan languages (Burmese, Haka Chin, Mizo, and Tshangla), where we find nEx connectives deriving from related similative (and in some cases also associative) plural markers. The most interesting case is that of Haka Chin. Peterson and VanBik (2004) note the existence of two post-nominal elements -teè and -poòl that seem to function “as marking collectives” (Peterson and VanBik 2004: 351) and as “representative conjunction” markers (i.e. nEx connectives), even though their exact function is hard to identify due to their low frequency in corpus data. Haka Chin also has a clear plural marker -hnaà, which is not used as a nEx connective. Looking at the examples provided by Peterson (2003) and Peterson and VanBik (2004), we believe that only -teè is a nEx connective originated from a similative plural:

Example (23) is not easy to interpret. According to Peterson and VanBik, all the listed expressions (vok-teè-pool, vok-teè-hnaa, etc.) can be translated as ‘pigs and such’. Interestingly, the only element that all these expressions have in common is the marker -teè. If we take into account also evidence collected from the other Sino-Tibetan languages (Burmese, Mizo, and Tshangla, see Table 3), we are inclined to think that the ‘and such’ meaning of the sentence is provided mainly by -teè, which behaves here like a similative plural, while the other two markers (-pool and -hnaa) behave like additive or collective markers. The marker -teè is instead clearly employed as a nEx connective in (24).

The change from heterogeneous plurals to nEx connectives is not particularly surprising. As noted in Section 1.3, both strategies encode non-exhaustivity and can be used to make reference to sets of similar elements. In this diachronic path, what allows for the development of nEx connectives is a scope reanalysis, whereby a marker having scope only over one item is reanalyzed as having scope over a list. Interestingly, additive plural markers are attested as sources only in two languages (Jamsay Dogon and Mundari), but in both cases there are elements suggesting that the markers at issue may function as both additive and heterogeneous plurals (see discussion above). This suggests that it is non-exhaustivity itself, more than plurality, that triggers the change.

4.2 Irrealis (interrogative and dubitative marker) > nEx connective

The second most frequent multifunctionality pattern is that in which nEx connectives derive from irrealis markers, in particular interrogative markers and dubitative markers (e.g. ‘maybe’, ‘perhaps’). With the term ‘irrealis marker’ we refer to any linguistic element that marks a given state of affairs as belonging “to the realm of the imagined or hypothetical”, thus constituting “a potential or possible event but […] not as an observable fact of reality” (Elliot 2000: 67, cf. also Mauri and Sansò 2012).

This pattern is attested in 8 languages, and we have enough evidence to treat it as diachronically oriented. Table 4 illustrates the languages for which the pattern at issue is observed, showing the connectives, the diachronic source identified and the presence of synchronic layering.

Let us consider the case of Ambonese Malay and Papuan Malay, where the interrogative marker ka has developed into a nEx connective, as shown in (25a–b) and (26a–b):

The pattern above is noteworthy because the interrogative marker ka develops into a disjunctive marker in several languages of the Vehicular Malay sub-family (Austronesian), such as Baba Malay (Lee 2014; Lim 1988) and North Moluccan Malay (Hayami-Allen 2001).^[14] However, its use as a dedicated nEx connective is apparently attested only in Ambonese Malay and Papuan Malay. Interestingly, along with ka, some of these languages have another disjunctive marker ato/atau. According to Lee (2014: 311), ka is derived from Min Nan Chinese (Sino-Tibetan), while ato/atau is derived from Malay (Austronesian). Looking at the examples provided by descriptive grammars, it seems that different languages have assigned to these markers different types of disjunctive functions. For instance, in Baba Malay (see Lee 2014: 312), they are used interchangeably or even together to mark disjunction. On the other hand, in Papuan Malay they are clearly different, with ato being used as a marker of exhaustive disjunction (cf. 27) and ka being used as a marker of non-exhaustive disjunction (cf. 26b, see Kluge 2017: 542–543). In Ambonese Malay, ato/atau is not attested.

Beyond interrogative markers, also dubitative markers can develop a non-exhaustive connective function. In Martuthunira (Pama-Nyungan), the conditional marker wii ‘if, maybe’, expressing epistemic doubt (cf. 28a), can also be used as a nEx connective, employed by speakers to indicate “a progressive widening of the set of objects out of which something may be chosen” (Dench 1995: 181) as shown in (28b).

The same pattern is attested in Murriny Patha (Southern Daly), where dubitative elements are described as following each item within a list encoding non-exhaustive disjunction (cf. Walsh 1976: 245–246):

At first sight, the patterns attested in Martuthunira and Murriny Patha may resemble the one described in Section 2, where we argued that Dyirbal yamba should not be considered a nEx connective, on the basis of the defining parameters we employed. Nevertheless, the same parameters suggest that kamayya in Murriny Patha and wii in Martuthunira are instead cases of nEx connectives. Contrary to Dyirbal, Murriny Patha indeed has a fully-fledged system of connectives, including the conjunctive connective yi ‘and’ and, more importantly, an exhaustive disjunctive connective a ‘or’. In Martuthunira too, the picture is relatively clear, because the status of wii as a (disjunctive) connective is not controversial as that of Dyirbal yamba, as witnessed by the fact that in some occurrences it is even glossed as ‘or’ (cf. Dench 1995: 79).

While all the cases discussed so far are instances of multifunctionality, thus showing synchronic layering, there are also elements that have fully developed into nEx connectives and have completely lost their original function. This is the case of Japanese ya, which is one of the most grammaticalized connectives in our sample. In Old and Early Middle Japanese, ya was indeed used in yes/no questions to seek the hearer’s confirmation and in rhetorical questions (Frellesvig 2010: 253), for instance:

Through Late Middle Japanese, it was then reinterpreted as a marker of uncertainty ‘I wonder’, frequently used in combination with modal forms (Frellesvig 2010: 359, e.g. ya in combination with modal forms of the existential verb ar- > yaraũ ‘is it, I wonder’ > nEx connective yara), although by the end of the period, ya had developed into a nEx connective. In Contemporary Japanese, the function of ya is to indicate non-exhaustive lists of items (cf. Chino 2001: 41).

Another nEx connective that does not show synchronic layering is toka in Japanese. Toka should be analyzed as the combination of the comitative marker to and the interrogative and disjunctive marker ka (see Ohori 2004: 50). While there is no question that the meanings of these components have an important role in determining the overall meaning of toka, its main function in Contemporary Japanese is to encode non-exhaustivity.

4.3 Additive marker (>) nEx connective

Another pattern attested in our sample (albeit with a lower frequency compared to the previous two) concerns elements that can be used both as additive markers (‘also/too’, ‘even’) and nEx connectives. Four unrelated languages, namely Matsés, Ingush, Sedang and Southern Aymara, show this type of multifunctionality.

In Matsés (Pano-Tacanan) chedo functions as an additive marker (‘too/also’) “indicating that the noun phrase is adding to a listing commenced in a preceding sentence” (Fleck 2003: 806), as shown in (31).

The additive marker chedo ‘too’ is also attested together with the emphatic suffix -bi, leading to the scalar meaning ‘even’ (cf. 32a) and, in some cases, the sequence chedo-bi may also function as a connective to perform what seems to be an emphatic conjunction (‘both A and B’ cf. 32b, see Haspelmath 2004: 10).

In addition, chedo is also employed as a non-exhaustivity marker, that is, to explicitly indicate that the marked element(s) is part of a larger category (cf. Fleck 2003: 714).^[15] In particular, chedo can be used as a nEx connective following each element of the list (cf. 33a) and as a general extender following the last element of the list (cf. 33b).

A similar multifunctionality is also attested in Ingush (Nakh-Daghestanian), where the clitic =’a can be used as coordinative conjunction to mark “incomplete, distributive, or exemplifying listing” (Nichols 2011: 525) as shown in (34a), but it can also function as an additive marker, having the meaning ‘also’, ‘too’, ‘even’ and occurring “in this emphatic or focal sense as a singleton” (Nichols 2011: 527), as shown in (34b).

According to Nichols (2011: 525), =’a is also employed for emphatic coordination (‘both…and’). However, there are no clear examples of this use.

Finally, a similar situation is observed in Sedang (Austroasiatic), where the marker hiã appears to have both the meaning of ‘also’^[16] and the meaning of ‘etcetera’ (cf. Smith 1975: 130). The gloss ‘etc.’ is probably an attempt by Smith to convey the non-exhaustivity meaning, because hiã is indeed employed as a nEx connective following each element of the list, rather than as a general extender:

Finally, the same multifunctionality pattern is attested also in Southern Aymara, where the nominal suffix -sa can be used as an additive (emphatic) marker (e.g. ut-sa ‘even (your) house’, see Coler 2014: 550), but also as a means to achieve non-exhaustive listing of items (cf. Ilsa-s(a) Justina-s(a) ‘Elsa and Justina (among others)’, Coler 2014: 679).

As already noted, in the three cases discussed above we do not have explicit information about the diachronic development of the nEx connectives. Nevertheless, across languages we find evidence of additive markers developing into coordinators (cf. Mithun 1988: 341–342). For instance, in Vlax Romani (Indo-European,), the additive focus-sensitive marker vi can be used as a coordinator “to mention another entity of concern to the actual statement” (Wagner 2012: 410), and is employed within open lists in discourse, as in (36).

However, in the case of Vlax Romani, the available data do not allow us to decide whether we are faced with a discourse pattern or with a nEx connective at full right. For this reason, contrary to the cases discussed before, it was not considered in our analysis.

4.4 Similative marker > exemplification > nEx connective

A further diachronic source attested in our sample are similative markers, that is, linguistic elements that encode an approximate similarity between two items (cf. Haspelmath and Buchholz 1998; Treis and Vanhove 2017). This pattern is attested in two languages: come in Italian and khale in Kombai. In both cases, we can still observe synchronic multifunctionality.

According to de Vries (1993: 49–50), khale is a noun meaning ‘kind, sort’. When it is used in combination with another noun, it may mean ‘like, as, resembling’ thus behaving like a similative marker, as shown in (37).

Nevertheless, khale has also developed “a conjunction-like grammatical meaning” (de Vries 1993: 50), to encode an inclusive non-exhaustive disjunction:

In Italian, we observe a parallel pattern, with the similative marker come ‘like’ being used in non-exhaustive lists to link the list items (cf. Goria and Mauri 2018):

Contrary to the diachronic paths discussed in the previous sections, the relationship between similative markers and non-exhaustivity may seem less straightforward. We have indeed to identify an intermediate step, necessary to explain this development, which in our view coincides with the function of exemplification. Similative markers are indeed often employed to signal a relation of similarity between a hypernym and one or more hyponyms, which are taken as exemplars (Barotto and Mauri 2018: 117). In turn, as noted in Section 1.2, exemplifying markers entail the notion of non-exhaustivity, since the item marked as example has to be conceived as representative of a larger set.

Let us consider the case of Italian in detail, for which we have corpus evidence from spoken language. In (40), the similative marker come indicates that ‘cats’ is an example of the larger set ‘animals’:

The usage of come to express exemplification is also found in lists, as in (41), where the speaker repeats come before each list member in order to stress the relationship with the hypernym, thus preparing the ground for its reanalysis as nEx connective:

Starting from list constructions as the ones in (41a), we can assume two intermediate steps: i) first, the speaker omits the coordinating connective (as in (41b)), then ii) she omits the hypernym and, consequently, the first come (i.e. the one linking the hypernym to the list members, as in 39). The resulting construction is a list of two examples linked by come. This process can be schematized as follows:

All these stages are attested synchronically in spoken (and some of them also in written) Italian. Such corpus-based evidence is of great help in explaining the development of come in Italian and provides a key to interpret the data attested for khale in Kombai, for which less information is available.

4.5 ‘Say’ > exemplification > nEx connective

Exemplification is likely to play a role also in the case of Usan (Nuclear Trans New Guinea), where the verb qâmb may be used both with its lexical meaning ‘to say’ and as a nEx connective. As already observed for many of the cases described until here, as a nEx connective qâmb follows each member of the open list. Let us consider three examples:

In (43a) qâmb is used with its lexical meaning ‘to say’, while in (43b) it is used to introduce indirect speech, showing the development of “conjunctive-like” functions. Interestingly, in (43c), qâmb is used to indicate the existence of further similar items beyond ‘yam’ and ‘mugam’. Reesink (1987: 84) notes that this usage of qâmb “resembles the use of English ‘say’ when one wants to give an example”.

Instances of ‘say’ verbs employed to mark exemplification are indeed attested across languages, albeit with different degrees of grammaticalization. For instance, in Japanese, the expression to itta ‘said, such as’ (a combination of quotative marker to and the past tense of the verb iu ‘to say’, cf 44a) has grammaticalized into an exemplifying marker that can be used to link a hypernym to its exemplifying hyponym, as in (44b):

In Italian, diciamo ‘we say’ and per dire ‘(in order) to say’ are frequently used to mark exemplification (Lo Baido 2018). For instance,

Interestingly, while in the cases just mentioned, ‘say’ verbs develop into exemplifying markers (that is, in exemplifying contexts, they can be paraphrased as ‘for example’), in Usan qâmb developed into a connective linking examples, that is, a nEx connective (Reesink 1987: 84).

4.6 Heterogeneity marker > nEx connective

NEx connectives may also develop from linguistic elements that explicitly encode a variety of elements (e.g. ‘various’). This pattern is attested in Hokkaido Ainu (Ainu). In Hokkaido Ainu, the noun usa ‘various’ is commonly used as a prefix before other nouns meaning ‘various X’.

According to Refsing (1986: 164), when usa is used after two or more nouns in a row, it indicates “an open-ended string of elements”:

Although we were not able to find further evidence of this pattern in other descriptive grammars of Ainu languages, the usage of usa ‘various’ as a nEx connective is indeed attested in the Glossed Audio Corpus of Ainu Folklore (Nakagawa et al. 2016), an annotated corpus of Ainu folktales with translations into Japanese and English, recorded between 1977 and 1983.^[20] Let us consider the occurrences in (48) and (49):

Two aspects are particularly interesting about this occurrence. First, in the English translation, the meaning ‘various’ is completely absent, even though usa is repeated three times. Even more importantly, in the Japanese translation, the list of animals (which is marked by usa in Ainu) is marked by the nEx connective dano in Japanese. Let us consider another instance from the corpus:

Again, as in (49), in the English translation there is no evidence of the meaning ‘various’ and in the Japanese translation, the list is marked by the nEx connective ya. Thus, evidence from the Glossed Audio Corpus of Ainu Folklore confirms that when usa is used to join two or more elements, it loses its original meaning ‘various’ and functions as a nEx connective, indicating the existence of further similar elements. The absence of non-exhaustivity in the English translation is probably due to the attempt to provide a faithful, almost literal translation, which led the translator to choose among the connectives available in English (none of which conveys non-exhaustivity).

The relationship between nEx connectives (and non-exhaustivity in general) and heterogeneity is noteworthy. First of all, as argued for heterogeneous plurals, nEx connectives are used to designate inherently heterogeneous sets, where set members are not identical but are grouped together on the basis of some kind of similarity (often contextually dependent, see Mauri 2017 for the usage of nEx connectives to make reference to ad hoc categories). Secondly, since words like ‘various’ encode not only heterogeneity but also numerosity (that is, their meaning can be paraphrased as ‘many different X’, cf. Garassino 2009), their repetition inside lists seem to suggest that the set encompasses more members than those explicitly stated. Barotto (2019) indeed suggests that the function of ‘various’ in the context of listing is employed to further highlight the non-exhaustive nature of the encoded set. Interestingly, the mechanism she observes is the same for nEx connectives, although performed by lexical means. In this regard, the fact that, over time and in specific contexts, words conveying heterogeneity can grammaticalize into nEx connectives confirms Barotto’s account of their role as markers of non-exhaustivity.

4.7 Preference > exemplification > nEx connective

Recently, Italian developed a nEx connective, which is still restricted to the colloquial variety, namely piuttosto che (cf. Brucale 2012, Mauri and Giacalone Ramat 2015 for a detailed diachronic analysis). Originally this connective had a preferential meaning ‘rather than’ but is nowadays attested both with its source value and as a marker of non-exhaustivity in lists. It can only be used when the speaker aims to name some potential exemplars of an open set, as in (50), in order to construct (or imply) a higher-level category (‘old nostalgic things’ in 50). As a nEx connective, piuttosto che cannot occur in alternative questions aimed at a choice, which by definition imply an exhaustive list of alternatives.

How does a preferential marker, which is typically used to oppose two alternatives (in a closed set) happen to encode non-exhaustivity? As argued by Mauri and Giacalone Ramat (2015), the change took place in free-choice contexts where the two alternatives refer to non-specific, replaceable items, as the ones exemplified in (51) and (52), by two famous Italian authors of the 19th century:

In these cases, by virtue of the free-choice context and their non-specific reference, the two alternatives are interchangeable: the preference of a over b is equivalent to preference of b over a: ‘one property or the other’, ‘one or another’. The free-choice context further implicitly conveys that they are randomly picked exemplars within a larger set. The assumption of a larger set is made explicit in (52), where the two alternatives are preceded by the overt indication ‘con mezzi diversi’ ‘through different means’, among which the author picks ‘one or another’. Preference between mutually replaceable items in a free-choice context is thus interpreted as exemplification, and the preferential connective is reanalyzed as a connective linking examples of a larger set, i.e. a nEx connective.

4.8 Other cases

This section is devoted to cases that are not supported by clear diachronic evidence, either concerning the source or concerning the path, despite the presence of potential hints. In two languages, non-exhaustive connectives appear to originate from a verbal suffix encoding aspectual categories, namely Andoque and Japanese, and in both cases the diachronic path that led to a nEx connective is not straightforward to understand.

In Andoque, Landaburu (1979: 153) notes that the progressive suffix -añe/-eñe (exemplified in (53a)) can be attached to nouns to encode “les termes successifs d’une énumération”, as in (53b):^[21]

As can be observed in (53), the progressive suffix -añe/-eñe can only be used together with the nominalization marker -ʌ. Interestingly the enumerative function of -añe/-eñe can be performed only with nominal bases and cannot be used to enumerate verbs. Furthermore, the same structure is attested to denote free-choice reference, that is, to designate a random element within the base’s denotation (hí-ʌ ‘thing’, hí -añe -ʌ ‘anything, whatever’, cf. Landaburu 1979: 154).

We can put forward two hypotheses. In the first one, we can think of an intermediate step between the progressive and the enumerative function, in which the element acquires an additive meaning ‘again’ (i.e. progressive > ‘again’ > additive > nEx connective), thus resembling the path described in Section 4.3. The second hypothesis sees the free-choice meaning as the intermediate step, which comes to be reanalyzed as an exemplification marker and ultimately as nEx connective stage (i.e. progressive > free-choice > exemplification > nEx connective), through a process that is reminiscent of the one described in Section 4.8 for Italian piuttosto che. Yet, despite many pages devoted to the description of this marker, Landaburu (1979) does not provide enough diachronic evidence to make one of these hypotheses something more than speculation.

The case of Japanese is different, because here the source meaning is not progressive, but rather perfective aspect. The converb -tari indeed originates from a perfect auxiliary that was used to denote the completion of an event in Old Japanese (cf. Frellesvig 2010; Narrog 2012; Shinzato 2005). In Contemporary Japanese, -tari has lost its aspectual perfective meaning, developing into what Narrog (2012: 145–148) calls a ‘subordinating mood’. He notes that tari “is not specifically an irrealis subordinate mood”, but it leads nonetheless “to a lower factuality of the event portrayed, since through -tari the event becomes marked as unspecific” (Narrog 2012: 147). This lower factuality is probably what enables the modern functions of tari to emerge, namely non-exhaustive enumeration and hedging. Once again, this may be reminiscent of the pattern described in Section 4.2, where we discussed how irrealis markers, that is, markers encoding a low factuality of the item in their scope, come to be reanalyzed as nEx connectives.

It is noteworthy that tari is the only nEx connective restricted to verbal phrases and clauses, as noted in Section 3, and this is clearly a direct consequence of its being a verbal suffix. Nevertheless, it should be noted that this is not a general rule: as noted above, -añe in Andoque originated from a verbal suffix as well, but, as a connective, it can be used to link noun phrases.

Moving away from verbal suffixes, an isolated case is the one attested in Mpade, where the consequential marker kanía (‘therefore’, 54a) can be employed to mark non-exhaustive lists (Allison 2012: 529, 54b). It is likely that some intermediate discourse function of progression or sequentiality has allowed the successive development of kanía into a nEx connective, through reference to a general concept of ‘there is more to come’, which may in turn have been reinterpreted as a marker of non-exhaustivity. Unfortunately, the grammar does not provide enough diachronic data to confirm this.

The last case that is worth mentioning is Mandarin Chinese, where a is described as a nEx used to indicate open sets (cf. Zhang 2008: 137):

Beyond marking open-ended lists, a is also frequently used as an interjection or as a final-sentence particle conveying pragmatic meanings (see Po-Ching and Rimmington 2004). This may point to a diachronic path in which the interjection at some point starts to be systematically employed as a hesitation marker in contexts where the speaker has processing difficulties or is uncertain of what she is saying. The speaker’s uncertainty marker may then have been reanalyzed in lists as a nEx connective.