Singular-plural verb stem alternation: uncovering global and local drivers of typological variation

1.1 History of study

Singular-plural alternation was probably first mentioned by Boas (1911a: 381) in his Tsimshian sketch under the term “irregular plural”, but little attention was paid to the phenomenon until the 1980s. As noted by Booker (1982), “even the most comprehensive typology of North American Indian languages (Sherzer 1976) fails to mention number suppletion”. It is known by many terms, including “singular versus plural roots” and “number-paired roots” (Kinkade 1981), “number suppletion” (Booker 1982),^[3] “singular-plural root suppletion” (Veselinova 2003), “verbal number pairs” (Veselinova 2006: Chapter 7), and is sometimes reported under the general phenomenon of “verbal number” (Corbett 2000; Frajzyngier 1985; François 2019; Mattiola 2019, 2020; Veselinova 2003, 2013).^[4]

Several cross-linguistic works on verbal number have appeared since the early 1980s. Booker (1982), Krasnoukhova (2022), and Mithun (2022) focus on North or South American languages exclusively, while Durie (1986) and Veselinova (2013) are global surveys of the phenomenon. All studies including North American languages emphasize its overwhelming presence there, and Mithun (2022) demonstrates how it is an areal pattern across families in Northern California.^[5] Some works also cover more qualitative and/or theoretical topics (Corbett 2000: 243–246; Durie 1986; Frajzyngier 1985; François 2019; Mithun 1988, 2022; Veselinova 2006).

Sg-Pl alternation usually targets both transitive and intransitive verbs (e.g. Mithun 1988, 2022), and less frequently intransitive verbs only (Booker 1982: 20; Durie 1986: 356, 366). The absolutive pattern of the phenomenon has been emphasized since the earliest works: it encodes the number of intransitive subjects and that of transitive objects (Durie 1986: 357; Frajzyngier 1985: 98; Mithun 1988: 214).^[6] Durie further observes that this pattern occurs irrespective of the dominant case marking or agreement pattern of the language, and points to a semantic motivation, namely the higher relevance of the “affected argument” (see also Mithun 1988: 214).

The system typically encodes a bipartite opposition: singular versus non-singular or individual versus collective. Booker (1982: 24) and Durie (1986: 356) both note that sometimes languages additionally encode a tripartite opposition, such as Karok (isolate; United States). Veselinova (2006: 151) reports one language (Maricopa, Cochimi-Yuman; United States) with only tripartite oppositions. The number of verb pairs in a given language varies, from a few to a few dozen (Durie 1986: 356; Krasnoukhova 2022: 642; Veselinova 2006: 153).

In terms of verb semantics, most authors emphasize the salience of motion (spontaneous motion like ‘go’, and caused motion like ‘take’),^[7] posture (both spontaneous and caused, like ‘lie’ and ‘lay’ respectively) and other frequently-reported semantics which do not fall into a broader category, like ‘die’ and ‘kill’ (e.g. Mithun 1988: 232), and sometimes also ‘cry’ (e.g. Kinkade 1981: 262) and ‘sleep’ (e.g. Veselinova 2013). Some languages display such an alternation in some adjective roots, often ‘small’ and ‘big’ (Booker 1982; Durie 1986: 361; François 2019: 349, 351; Krasnoukhova 2022: 651; Mithun 1988: 218; Veselinova 2006: 154).

1.2 Different approaches

While there has been agreement on some aspects of Sg-Pl alternation – its prevalence in North America, absolutive patterning, the greater frequency of intransitive verbs, and the semantic skew of alternating verbs – there have also been considerable differences in how the phenomenon has been approached. The very definition of Sg-Pl alternation has varied substantially across researchers.

All authors consider verb pairs which display unexpected formal differences (often referred to as “proper suppletion”), and most also include verb pairs where the alternation is encoded by a non productive (plural) morpheme, irregular ablaut or reduplication included. These non-suppletive pairs are irregular in that their plural morphology is restricted to a closed set of verbs, much like the suppletive pairs. Veselinova (2006: 158–169) considers the difference between alternating verbs which form exceptions to otherwise regular agreement paradigms and verbs which are not part of a paradigm, concluding that these are two ends of a continuum (p. 173). This distinction between paradigmatic (“suppleting”) and non-paradigmatic (“non-suppleting”) alternation is not made by all authors.

Perhaps the most significant discrepancy is the semantics of “verbal number” (see footnote 4). While some authors distinguish between participant and event number (Corbett 2000; Durie 1986; Krasnoukhova 2022; Mithun 2022), others consider them jointly (Frajzyngier 1985: 98; Mattiola 2019, 2020; Mithun 1988; Veselinova 2006, 2013). Only a few individual language studies like Heath (1980) or François (2019) clearly set apart the two semantic domains. In others it is more complex, as in Canela-Krahô (Nuclear-Macro-Je; Brazil), where Miranda (2014: 134) separates the two, although he acknowledges that plurality of arguments is also considered by the speakers as indicating a plurality of events – e.g. killing several people can be considered as a single killing event or as several distinct killing events (Maxwell Miranda, pers. comm. December 2022, in the same spirit as Frajzyngier 1985: 98). In still other studies, sources disagree on the semantic interpretation of the verb pairs within a single language, as is the case for Everett (2006: 348–355) and Storto (2014: 409–410) on Karitiâna (Tupian; Brazil). This is one reason why Corbett (2000: 248–249) and Krasnoukhova (2022: 641–642) classify the phenomenon into three subtypes: event plurality only, argument plurality only, and a last mixed type. Interestingly, Müller (2014) and Krasnoukhova (2022: 643) note that “proper suppletion” necessarily expresses the number of participants, sometimes along with that of events.

Another theoretical disagreement concerns the syntactic or lexico-semantic nature of the phenomenon. Several studies, starting with Boas (1911b: 381), insist that the phenomenon is distinct from agreement (i.e. echoing the nominal number marked on NPs), and is rather associated with a lexical choice based on the overall semantics of the clause. The three main reasons for an analysis in terms of semantic selection (rather than syntactic redundancy) are:

1. In many cases, no (explicit) NPs are present to agree with the verbal number (e.g. Frajzyngier 1985: 96–97; Mithun 1988: 212);

2. Syntactic incongruence occurs in several languages (e.g. I 1sg-walk: dual with him vs. one of us 1non.sg-walk: sg in Navajo (Athabaskan; United States) as in Durie 1986: 358; see also Mithun 1988: 215);

3. Sg-Pl alternation is part of a larger phenomenon targeting not just participant number but also specific participant shapes (Mithun 1988: 214, 2022).

Additionally, François (2019: 357) describes how Sg-Pl alternation in Hiw [hiww1237] (Austronesian; Vanuatu) follows semantic rules that are specific to verbal number (sg-pl), and only partially coincide with nominal number (sg-non.sg). Note however that Veselinova (2013) shows that analyses in terms of lexical selection or agreement might both be present within a given family, e.g. in Athabaskan, where the syntactic incongruence described in Navajo (see (2) above) is impossible in Slavey.

For any typological survey, the consistent coding of variables across potentially very different languages is fundamental, and it requires strict and clear definitions. We now describe how we divide the conceptual space, i.e. the “comparative concepts” (Haspelmath 2010) we use in our survey.

4.1 Distribution

Sg-Pl stem alternation is fairly frequent, occurring in a third of our sample (109/325 languages). Its distribution across macroareas is highly uneven, with North America greatly overrepresented (Figure 2). The raw counts and frequencies of Sg-Pl alternation are given in Table 6, while Table 7 presents a stochastic subsampling method, described below.

Figure 2:

Sg-Pl verb stem alternation.

As discussed, our sample is heavily skewed toward the Americas, which happen to occupy opposite ends of the distribution. We have tested and corrected for any resulting biases in two independent ways: first, by rebalancing the raw frequencies so that macroareas are weighted evenly (presented in the last row of Table 6), as described in this paragraph, and second, by performing random subsampling (Table 7), as described further below. Rebalancing the raw frequencies in Table 6 by macroarea (i.e. weighting each macroarea by 1/6) gives only a slightly lower global average of 31 %. Rebalancing without the outlier of North America gives a still lower average of 27 %. If Central America (defined as north of South America and south of the Chihuahuan Desert) is separated from North America, 60 % (52/86) of the remaining languages of North America have at least one verb pair, while the languages of Central America display a much more globally typical 23 % (7/30). In other words, Sg-Pl alternation is more than twice as frequent in North America than in the rest of the world, a matter we will return to in the discussion.

A different balancing method (Table 7) was implemented, following concerns from a reviewer about remaining macroareal biases. We generated 250 independent, random subsamples of 15 unrelated languages per macroarea (90 languages in total), guaranteeing that there was no phylogenetic or geographic bias in a given sample. From these samples, we calculated the mean and standard deviation for the frequency of the presence of Sg-Pl alternation in each macroarea. As seen by comparing Tables 6 and 7, the resulting means are nearly identical to the overall frequencies, though standard deviations vary by macroarea. The global standard deviation is however notably low. This procedure confirms that the American skew of our sample is not biasing macroareal differences. These frequencies are similar to the figures in Veselinova (2006: 156), except that we find a much greater presence of Sg-Pl alternation in Papunesia.

As will be detailed in Section 5.1, the consistency of Sg-Pl alternation within a family is somewhat mixed. In some families, all or almost all languages display the phenomenon (e.g. the Athabaskan-Tlingit-Eyak, Uto-Aztecan, and Otomanguean families), in others only half of the languages display it (e.g. the Pano-Tacanan family), in still others it is even less (e.g. the Algic, Salishan, and Tupian families).

Within a language, typically only a handful of verbs alternate. Figures 3 and 4 group languages according to the number of verb pairs that alternate by suppletion and by regular pattern, respectively. Recall that we have excluded derived verb pairs.

Figure 3:

Root suppletion counts by language (N = 103).

Figure 4:

Regular pattern counts by language (N = 28).

Root suppletion is the most common way of forming Sg-Pl verbal pairs (present in 104/109 languages in our sample). The mode number of suppletive pairs is 6 and the mean is 9, with half the languages having 7 or fewer suppletive pairs. The mean is substantially higher than the mode because of a long tail distribution, with a handful of North American (and one Papunesian) languages having very large numbers of verb pairs: at the extreme tail, 34 in Hopi (Uto-Aztecan), 46 in Comanche (Uto-Aztecan), and 61 in Western Keres (Keresan). Sg-Pl alternation by a regular pattern has a similar distribution, although this strategy is much less common overall (32/109 languages).^[16] The mode number is 3, and the mean is 18, with half the languages having 9 or fewer verbs following a regular pattern. The mean is much higher than the mode once again because of a small number of outliers, in particular Creek (Muskogean) and Laal (isolate) with over one hundred pairs.

As for alternation in verbal morphology, of the 33 languages in our sample which display it, six lack Sg-Pl alternation in main verbs, a pattern we expected to be much rarer. Four of these languages are in or near the Andes (Huallaga Huánuco Quechua, North Junín Quechua, Urarina, and Wichí Lhamtés Nocten), possibly indicating a regional historical process. The other two are far apart (Kusunda in Eurasia and Northwest Maidu in North America).

Our sample confirms the preponderance of Sg-Pl stem alternation in North America. However, it is still common in the rest of the world, occurring in about one in four languages in our sample outside of North America. As expected for a typological feature which is definitionally lexically limited, it typically occurs in a handful of verbs in a language, and significantly, suppletion is much more common than limited regular patterns.

4.2 Syntactic profile

Previous researchers (e.g. Booker 1982: 20) have noted that Sg-Pl alternation targets either only intransitive verbs or both intransitive and transitive ones. We confirm this tendency, with 33/109 languages having only intransitive alternating pairs, and 70/109 having both intransitive and transitive pairs. However, five languages in our sample show alternation only for transitives.^[17] Of these, only the extinct and little-documented language Esselen (isolate; United States) is from North America. The remaining languages are in Papua New Guinea (Daga, Nimboran, and Yau-Nungon) and the Andaman Islands (Akabea), regions where Sg-Pl alternation has been scarcely investigated. The cross-linguistic tendency for intransitive verbs may not hold for the whole of Papunesia, where the languages in our sample never display intransitive-only alternation, but either transitive-only or both.

We confirm the overwhelming tendency for Sg-Pl alternation to be triggered by S and P roles (Corbett 2000: 253; Frajzyngier 1985: 98; Mithun 1988: 214). In 62 out of 75 languages with transitive alternating pairs, only the P role is targeted. However, in five languages, the A role is targeted: Laal (isolate; Africa), Murriny Patha (Southern Daly; Australia), Akabea (Great Andamanese; India), Seri (isolate; United States), Malinaltepec Me’phaa (Otomanguean, Mexico); and in six languages, some transitive pairs target the A role, and others the P role: Ket (Yeniseian; Russia), Coahuilteco (isolate; United States), Hopi (Uto-Aztecan; United States), Imonda (Border; Papua New Guinea), Jaminjung-Ngaliwurru (Mirndi; Australia), Wari’ (Chapacuran; Brazil).^[18] No language in our sample displays verb pairs which could alternatively express A or P plurality.^[19]

Verbal morphology follows a similar absolutive pattern. The associated motion suffix pair -m(m)ed (sg) versus -op(po) (pl) ‘go.to.do’ in Tohono O’odham may co-alternate with a lexical verb pair (here, me’a ∼ kokthai ‘kill’), yielding the four possible combinations illustrated in (6). The alternating verb pair targets the number of P, and the alternating suffix pair seems to target A. However, the suffix actually targets the S role of an associated intransitive Motion Event, the expression of which is grammaticalized.

4.3 Number distinctions

Bipartite number distinctions are overwhelmingly more present than tripartite ones, with 97/109 languages showing bipartite only distinctions, 14 showing both bi- and tripartite distinctions, and two showing tripartite distinctions only. Among the bipartite number distinctions, the semantics is overwhelmingly Singular-Plural (see Table 8).^[20]

The Singular-Plural number distinction refers to a strict singular versus non-singular opposition, Individual-Collective opposes an individual acting alone versus a group acting collectively, and Paucal-Many a small versus large number. Tripartite distinctions generally add a dual rather than a paucal as an intermediate step between singular and plural (see Appendix A for the detailed distribution of tripartite patterns).

Individual-Collective number systems being less frequent in the literature, we illustrate with two examples from a single paragraph in a Nuu-chah-nulth (Wakashan; Canada) text. In (7), the geese fly up from a beach as a group and the collective form for ‘fly’ huʔak is thus used. In (8), the geese have been blinded by fires and no longer fly collectively, and the individual form mataa is used despite a plural subject. That is, collectivity, rather than plurality, determines the choice of verb form.

The Nuu-chah-nulth verbs mataa and huʔak are thus basic-level terms, unlike the English verbs fly and flock, which are in a hypernym/hyponym relationship. While fly is agnostic about the individuation or collectivity of its (possibly plural) subject, flock necessitates a collective subject.

Such a determination is dependent on access to large corpora, which are not available for many languages. Even when Sg-Pl alternation is discussed in a grammar, thorough semantic discussion is often lacking. Following Corbett (2000: 250) and the observations of Boas (1911a: 381) about the importance of collectivity in many American languages, the individual-collective opposition could be more frequent than our results suggest.

The presence of multiple number distinctions within a single language is rare (18/109), and most typically involves the addition of a tripartite distinction to an otherwise bipartite system (14/18 mixed systems, presented in more detail in Appendix A). Only a few languages cumulate different bipartite systems without any tripartite ones, like Yuracaré (isolate; Bolivia). Van Gijn (2006: 191–192) shows how the verb pairs for ‘go’, ‘arrive’, and ‘go across’ strictly follow a Singular-Plural semantic distribution, while other verb pairs like ‘fall’ follow a Paucal-Many distribution, and are grammatical with low numerals.

4.4 Lexical semantics

The lexical semantics of verb pairs follows previous observations. We report here only on the figures for suppletion for the 103 languages where we have lists of suppletive pairs, since looking at root suppletion only or at all stem alternations makes little difference to the overall picture.^[22] We distributed verb pairs into five semantic categories: Motion, Posture, Location, Frequent Verbs, and All Others, as illustrated in Table 9. Motion, which includes both spontaneous and caused motion verbs (like ‘go’ and ‘take’), is the most predominant semantics, both in terms of number of languages (86/103) and total number of suppletive pairs (379/928).

The Posture category, which includes both spontaneous posture (like ‘lie’) and caused posture verbs (like ‘lay’), is the second most predominant semantics, again both in terms of number of languages (67/103) and tokens (190/928). Note that verbs like ‘put X (lying)’ or ‘lay X’ express both caused motion and caused posture, and were sometimes not easy to categorize. We followed the cues given in the translation in the sources, unless the verb form indicated the opposite analysis.

The Location category includes location verbs (like ‘be there’ or ‘exist’) and caused location verbs (like ‘hold’ or ‘grab’). Again, there is an overlap with posture verbs which might be used in basic location and existential constructions. Posture to location/existence being a known grammaticalization path (e.g. Newman 2002), we favored Posture over Location if posture was listed in the translation. The Location category occurs in 35 languages and 50 tokens, and stands thus far behind Motion and Posture. It is notable that only six languages^[23] have neither Motion, Posture nor Location.

The Frequent Verbs category includes a heterogeneous group of frequently attested verbs already noticed in the literature. They have highly specific semantics – ‘die’, ‘kill’, ‘sleep’, ‘cry’, and ‘eat’ – and their frequencies are given in Table 10. The first two (‘die’ and ‘kill’) are almost as frequent as the entire category of Location, and each appear in 30 languages.

The All Others category includes all verbs which did not fit into one of the four previous categories and contains a variety of semantics. We looked through the semantics of individual verbs and created five post hoc subcategories that could accomodate approximately half of the tokens: Destruction, Property, Transfer, Emission, and Perception verbs (Table 11).

Destruction verbs (e.g., ‘hit’, ‘pierce’, ‘break’) are the most frequent, with 42 tokens across 25 languages. Note that ‘kill’ and ‘die’, which are grouped with Frequent Verbs, could be included in this category.^[24] Property verbs (e.g. ‘be big’, ‘be dirty’) are also common, with 39 tokens across 15 languages. Another 17 languages could be added to this count, if suppleting adjectives are considered in addition to verbs. The Property category has been noted by Booker (1982), although she only comments on ‘big’ and ‘small’. Transfer verbs (e.g. ‘give’, ‘get’) have only 19 tokens across 15 languages, but several Motion verbs like ‘throw’ or ‘bring’ could be included in this category.

Verbs of Emission (e.g. ‘talk’, ‘speak’) and Perception (e.g. ‘see’, ‘hear’) respectively amount to 23 tokens across 15 languages and eight tokens across five languages. Although basic and common verbs in most languages, they hardly supplete for number when compared with verbs from other categories, even Destruction and Property. This discrepancy points to at least a partially semantic-based, rather than a purely frequency-based, explanation for verbs targeted by Sg-Pl alternation, to which we return in the discussion. We were unable to define further categories with more than two lexical items and occurring in more than a handful of languages.

5.1 Inheritance

The extreme geographic skew toward North America raises the prospect that this feature may be especially present in certain language families. The language families in North America could have inherited this feature, either from the known proto-languages, or perhaps from a theoretical common ancestor of North American languages that lies beyond the limit of the comparative method.^[25]

At the continent-wide level, the inheritance model is a poor explanation of the difference between North America on the one hand and Central and South America on the other. If Sg-Pl alternation is a long-term stable feature reflecting deep shared ancestry, then the sharp typological shift that occurs south of the Chihuahuan Desert is a problem. If North American families inherited this feature from the distant past, then that process should affect languages further south as well.

The inheritance model also cannot explain the distribution within known families in North America. As noted in Section 4.1, Sg-Pl alternation is often unevenly distributed within a family. Though this feature has not been a focus of reconstruction efforts, we will briefly review two detailed studies on Uto-Aztecan (Haugen and Everdell 2015) and Salishan (Kinkade 1981), and a less detailed study on Wakashan (Inman 2021). These, as well as the results of our survey, suggest that most alternating verbs have developed after these languages differentiated.

The Uto-Aztecan family provides the strongest potential example for widespread inheritance of Sg-Pl alternation. Of the eight Uto-Aztecan languages in our sample, six have Sg-Pl alternation, including all languages from the Northern branch (Cahuilla, Hopi, Comanche, and Northern Paiute), and two from the Southern branch, members of the Cahitan (Yaqui) and Piman (Tohono O’odham) subbranches. Only the Aztecan languages in our sample lack the phenomenon (Pipil and Tetelcingo Nahuatl). A look in grammars of Southern, non-Aztecan languages outside of our sample confirms the presence of Sg-Pl alternation in Southeastern Tepehua [sout2976] (Tepehuan subbranch; Willett 1991: 44, 125, 171), Huarijio [huar1255] (Tarahumara-Guarijo; Félix Armendáriz 2005: 116), and Huichol [huic1243] (Corachol; Bierge 2017: 130).

Despite this abundance of Sg-Pl alternation within the family, Haugen and Everdell (2015) reconstruct only two verb pairs in the proto-language: ‘die’ and ‘kill’. While the phenomenon was thus present to at least some degree in Proto-Uto-Aztecan, it has greatly developed in many languages of the Northern branch (Comanche, Hopi, and Northern Paiute all have at least 18 suppleting verb pairs) as well as in (Southern but geographically northern) Yaqui (20 verb pairs), while it remained small or dwindled out elsewhere. Interestingly, the branch where Sg-Pl alternation is not attested, Aztecan, is entirely spoken within Central America.

The distribution of the phenomenon in Salishan is very similar. Kinkade (1981) documents between six and 23 verbal pairs in seven Salishan languages, all in the Interior or Tsamosan branches. He did not find many examples among Central Salish languages or in Bella Coola, a distribution also reflected in our sample. Kinkade finds only a few cognates among the Interior languages, and in only a few cases was he able to trace the alternation to the phonological reduction of an attested regular plural process. He concludes that this phenomenon, despite its abundance in some languages, is relatively recent within the family, and proposes that it was borrowed, possibly from Athabaskan.

Though less robustly documented, the same pattern of recent spread seems to hold true for Wakashan. Inman (2021) reports on plural processes in Nuu-chah-nulth verbs and nouns and briefly compares them to other Wakashan languages. Nuu-chah-nulth has four suppletive pairs,^[26] and Kwak’wala is reported to have two. Alternating pairs are (so far) not documented for the rest of the family. Similar to the other families discussed above, there are no cognate suppletive roots and the feature seems to be a more recent development, rather than inherited from Proto-Wakashan.

Finally, as alluded to in Section 4.1, four out of the five Otomanguean languages in our sample have Sg-Pl alternation, which represents a disproportionate number of the Central American languages with the phenomenon. However, only the semantics ‘sit’ is present across three of them (Nellis and Nellis 1983: 322; Suárez 1983: 161; Veerman-Leichsenring 1991: 119), and there is no obvious cognacy among the forms (Eric W. Campbell, pers. comm., March 2023).

Sg-Pl alternation can undoubtedly be inherited. However, even in cases in which it has persisted in a family through time, the individual verb pairs are not necessarily preserved, and cognates within families are very rare.

5.2 Areal spread

Inheritance being insufficient to describe the distribution of Sg-Pl alternation, we now turn to borrowing and linguistic areas as an explanation, as proposed by Kinkade (1981) for Salishan and by Mithun (2022) for several unrelated Californian languages.

The mechanism proposed by Mithun (2022) is that second language speakers, accustomed to Sg-Pl alternation in their L1, overuse hypernyms in their L2 to signify plurality, which over time leads to the spread of Sg-Pl alternation within a given linguistic area. The same process, of course, happens in the other direction, with L1 speakers (of a language without Sg-Pl alternation) learning to speak an L2 (with Sg-Pl alternation), and bringing this feature home to their L1. Importantly, despite this multilingual mechanism, the data reveal no form borrowing across languages in any of the areas investigated below.

We will examine this process of areal spread by looking at the specific semantics of suppletive verb pairs in five different regions: the well-established linguistic areas of the Pacific Northwest, California, and the Pueblo area, all of which are in North America, as well as the dense clusters of Sg-Pl alternation revealed in the Guaporé-Mamoré region and Eastern Papua. Recall that we only collected underived verb pairs, which might diminish the chance of having similar semantics across languages. One can easily imagine that a language derives an additional suppletive pair with regular valency changing markers to copy alternation occurring in two underived suppletive pairs in the neighboring language.

We are also only presenting suppletive alternations within each area. As we have noted, suppletion is the most typical form of alternation worldwide, and while this restriction makes our data easier to interpret, it runs the risk of missing contact events that occur between suppletive pairs and regular patterns.^[27] Comparing suppletive pairs will however capture any Sg-Pl alternation that spreads according to an areal mechanism of hypernym generalization, as outlined in Mithun (2022).

5.2.1 Pacific Northwest

The Pacific Northwest has long been recognized as a linguistic area (e.g. Boas 1898, 1911c; Jacobs 1937; Thompson and Kinkade 1990), but its boundaries are not well-defined. We will consider the set of languages provided in Beck (2000). Like much of North America, these languages frequently have root suppletion. Interestingly, the semantics of alternating verbs is frequently shared between neighboring languages, but not necessarily across the entire region. Figure 5 is a map of all the Pacific Northwest languages in our sample. Table 12 presents the suppleting verbal pairs. Semantics that are shared between neighbors or near-neighbors, regardless of their family, are bolded.

Figure 5:

Languages in the Pacific Northwest. Those without Sg-Pl alternation are: 6 Bella Coola (Salishan), 7 Lillooet (Salishan), 10 Halkomelem (Salishan), 11 Kutenai (isolate), 13 Northwest Sahaptin (Sahaptian), 15 Central Kalapuya (Kalapuyan), 17 Molale (isolate), 19 Siuslaw (isolate).

Table 12:

Suppletive alternations in the Pacific Northwest.

#	Language	Family	Suppletive verb pairs
1	Eyak	Na-Dene	Motion: go, run Posture: sit, stand, lie (prone), extend
2	Tlingit	Na-Dene	(Caused) Motion: walk, run, fly Posture: be sitting, sit down, stand, lie Frequent: cry
3	Nisga’a	Tsimshianic	(Caused) Motion: go/walk, go out, go up a slope, leave, be/come back, run, flee, ride, put away, take, throw in a fire Posture: sit, stand, lie, hang Caused Location: hold/carry Frequent: die, kill, cry, eat Other: grow, be small
4	Southern Haida	Haida	Motion: go, fly Posture: sit Frequent: kill
5	Southern-Coastal Tsimshian	Tsimshianic	(Caused) Motion: go, go out, leave, run, start off, escape, drift away, put away, put down, take in (Caused) Posture: sit, stand, lie, place upright Frequent: die/kill, eat Other: be dirty, be easy, be ripe (of a rain cloud), be sour, be strong, tear out and turn over, grow, name, wake up, wipe
8	Kwak’wala	Wakashan	Posture: sit, stand
9	Nuu-chah-nulth	Wakashan	Motion: run, fly Frequent: cry, sleep
12	Upper Chehalis	Salishan	(Caused) Motion: run, walk, fall, drop, throw Posture: sit, stand, lie/put down Caused Posture: grab Frequent: kill, cry
14	Clatsop-Shoalwater Chinook	Chinookan	Posture: stand Location: be (there) Frequent: die, kill, cry
16	Alsea-Yaquina	isolate	Motion: dance, enter, enter water, jump Posture: sit, stand Frequent: die, cry, sleep (verb 1), sleep (verb 2) Other: agree
18	Hanis	Coosan	Motion: fly/jump, travel Posture: sit, lie Frequent: die, kill, sleep Other: do, speak
20	Takelma	isolate	Posture: sit, stand
21	Klamath-Modoc	Isolate	(Caused) Motion: run, throw Posture: sit/live/stay, stand, lie down, float Frequent: die, kill Other: give, hit/shoot
22	Karok	Isolate	(Caused) Motion: swim, fly, climb, throw Posture: sit
23	Shasta	Shastan	Motion: run (verb 1), run (verb 2), walk, swim, jump, fall, go out to camp, move suddenly, move along Posture: sit, stand, lean Caused Location: hold Frequent: sleep Other: look, bite, hunt/dig/mine

The semantics of most verbs pairs shared between neighbors are globally typical of Sg-Pl alternation (‘go’, ‘run’, ‘sit’, ‘stand’, etc.); however, some are regionally specific, such as ‘jump’ in Alsea-Yaquina and Hanis, and ‘throw’ in Klamath-Modoc and Karok. Some sub-regions share more verb semantics than others: The alternating pairs in the northern end of the linguistic area (roughly Eyak to Southern-Coastal Tsimshian note the relative phylogenetic homogeneity) repeat quite frequently language-to-language, and pairs are shared to a slightly lesser degree in the south (roughly Clatsop to Shasta, note the relative phylogenetic diversity). In the central region of the Pacific Northwest, traditionally thought of as the core of the linguistic area (the Salishan, Wakashan, and Chimakuan families, or roughly Bella Coola to Upper Chehalis, see Beck 2000), there are more languages lacking Sg-Pl alternation altogether.

In addition to these similarities between neighboring languages, there is also significant diversity in the region. Despite belonging to the same family, Nisga’a and Southern-Coastal Tsimshian have a large number of non-shared suppletive stems. In the southern sub-region, there are a large number of idiosyncratic suppletive alternations. Overall, the picture suggests both areal spread and independent innovation.

5.2.2 California

Much like the Pacific Northwest, the exact boundaries of the California linguistic area are somewhat fuzzy, and many sub-areas have been proposed (Northern California, Northwest California, Clear Lake, the South Coast Range, and Southern California-Western Arizona, see Hill 2019). Since our sample is relatively sparse, we will report on Sg-Pl alternation among all languages spoken in what is today California, plus Northern Paiute (due to its proximity to Washo). Note that Karok and Shasta are also present in the Pacific Northwest grouping given above. Our data are presented in Figure 6 and Table 13.

Figure 6:

Languages in California. Those without Sg-Pl alternation are: 6 Achumawi (Palaihnihan), 7 Northern Yukian (Yuki-Wappo), 8 Wintu (Wintuan), 9 Northwest Maidu (Maiduan), 11 Wiyot (Algic), 12 Lake Miwok (Miwok-Costanoan), 16 Southern Sierra Miwok (Miwok-Costanoan), 19 Salinan (isolate), 20 Mutsun (Miwok-Costanoan), 22 Northern Yokuts (Yokutsan), 23 Ineseño (Chumashan). Note: Northwest Maidu has however three bound (verbal) stems (wO-, bO-, and he-) which regularly alternate by vowel lengthening and which we considered in feature SgPl.19 in the supplementary materials (see the Data availability statement) on Sg-Pl alternating morphology. Their semantics is unclear.

Table 13:

Suppletive alternations in California.

#	Language	Family	Suppletive verb pairs
1	Karok	Isolate	(Caused) Motion: swim, fly, climb, throw Posture: sit
2	Shasta	Shastan	Motion: swim, run (verb 1), run (verb 2), walk, jump, fall, go out to camp, move suddenly, move along Posture: sit, stand, lean Caused Location: hold Frequent: sleep Other: look, bite, hunt/dig/mine
3	Yurok	Algic	Motion: arrive Posture: sit, stand (inanimate)
4	Hupa-Chilula	Na-Dene	(Caused) Motion: run, move, move something somewhere, carry Posture: stand, lie down, lie motionless, have a position Location: stay/live Frequent: kill
5	Chimariko	isolate	(Caused) Motion: run, carry (a firm object), carry (a load), carry (a long object), carry (a small object) Posture: sit
10	Yana	isolate	Motion: go (M), go (F), go/move, run Posture: sit, sit/stay/live Other: laugh, awaken, be strong, be long
13	Southeastern Pomo	Pomoan	(Caused) Motion: run/fly/flow, walk/go, swim, roll/fall, float, crawl, come to rest on water, come to rest (long object), plop down, place X on ground, place X on surface Posture: sit (non-long object), sit down, stand (long object), lie (long object)
14	Kashaya	Pomoan	(Caused) Motion: go, crawl/swim, carry (any-shaped object), carry/hold (long object), carry/hold (non-long object), implant (long object), put (long object), put (non-long object) Posture: sit (on ground), sit (on platform), stand (on feet), stand (as tree), lie, hang Frequent: kill Other: shoot, give
15	Wappo	Yuki-Wappo	(Caused) Motion: go away, come out, run, walk, carry Posture: sit, stand, lie Frequent: die, kill, sleep Other: sing
17	Washo	isolate	Motion: run, fly, swim, fall off, fall over Posture: sit, stand, lie
18	Northern Paiute	Uto-Aztecan	(Caused) Motion: go, come, enter, return, travel, fly, put Posture: sit, stand, lie Frequent: die, kill Other: talk, give, make a hole in, open
21	Esselen	isolate	Caused Motion: bring (cause to be carried)
24	Cahuilla	Uto-Aztecan	(Caused) Motion: fall, put, take Posture: lying flat Location: exist, stay Frequent: die Other: shoot, light a fire

As in the Pacific Northwest, there are regions in California where languages share verbal semantics, although these are often the typical semantics of the phenomenon as a whole. However, there are also more idiosyncratic verbs (e.g. ‘carry’ in Hupa-Chilula and Chimariko, as well as in Kashaya and Wappo, and ‘crawl’ in Southeastern Pomo and Kashaya). Once again the picture is mixed, revealing both cross-linguistically shared semantics and considerable variation.

5.2.3 Pueblo

The Pueblo linguistic area belongs to a well-established cultural area of the historic pueblos of the American Southwest (Eggans 1979). It consists of Zuni (isolate), Hopi (Uto-Aztecan), Navajo (Athabaskan-Eyak-Tlingit), and the Kiowa-Tanoan and Keresan families (Campbell 1997: 339), although it should be noted the Navajos arrived in the area only just before the Spaniards (Eggans 1979). From the perspective of Sg-Pl alternation, these languages do not look markedly different from neighboring languages. However, we focus here on only the Pueblo languages, since they are known to have been in intensive contact. These languages also have some of the largest inventories of verb pairs.^[28] Because of the tight-knit nature of this linguistic area, we consider every language to be near-neighbors (Figure 7 and Table 14).

Figure 7:

Languages in the Pueblo linguistic area.

Table 14:

Suppletive alternations in the Pueblo linguistic area.

#	Language	Family	Suppletive verb pairs
1	Hopi	Uto-Aztecan	(Caused) Motion: arrive, enter, come out, fall/drop, run, climb up, climb down, dance, wander, escape, throw, put, put into a rigid/enclosed container, put on top, carry (animate), carry (inanimate), bring back (animate), fetch, get extracted Posture:: sit, stand Frequent: die, kill, cry, sleep, eat Other: receive, give, break (linear segments), break/crack, dress, talk about, harvest/pick from tree, laugh
2	Arizona Tewa	Kiowa-Tanoan	(Caused) Motion: fall, throw away Posture: sit, lie, stand up Other: get, hunt/dig/mine
3	Navajo	Na-Dene	(Caused) Motion: go, run, throw, lead Posture: sit, lie down Other: die, kill
4	Zuni	isolate	(Caused) Motion: run away, take, put against, put down Posture: sit, stand, lie (verb 1), lie (verb 2), hang (Caused) Location: be among, be inside, hold Frequent: kill, sleep Other: give
5	Western Keres	Keresan	(Caused) Motion: go, arrive, enter, leave, walk, fall in, fall off, climb, pass by, step down, fly, rise, take, throw, throw away Posture: lie down, lean, stand up, be standing, float Location: dwell Frequent: die, cry, sleep, spend the night, eat Other: say, speak up, hear, listen, look for, buy, lose, find, be closed (eyes), finish, have it closed, have luck, know, know how, tend, nurse, release, wait for, wake up, win, grow, grow old, be lost, be enclosed, be stuck, be big, be fat, be hungry, be strong, be a fast runner, be talkative, be pregnant, be born, be first, drink
6	Taos Northern Tiwa	Kiowa-Tanoan	Unknown (mentioned in grammar but no list given)

The languages with relatively small inventories (Arizona Tewa and Navajo) share all or almost all of their verb pairs with their neighbors, including both more common semantics like ‘sit’ and ‘stand’ and relatively less common semantics like ‘throw’. Languages with very large inventories (Hopi and Western Keres) share some semantics with their neighbors but have a much larger inventory of idiosyncratic pairs.

5.2.4 The Guaporé-Mamoré region

The cluster of languages around the Brazilian state of Rondônia is a part of the proposed Guaporé-Mamoré linguistic area (Crevels and Voort 2008), and stands out in our sample as a hotspot of Sg-Pl alternation in South America. This is one of the clearest clusters of the phenomenon among a phylogenetically diverse group outside of North America (Figure 8 and Table 15).

Figure 8:

Languages in the Guaporé-Mamoré region. Those without Sg-Pl alternation are: 2 Apurinã (Arawakan), 3 Ese Ejja (Pano-Tacanan), 4 Cavineña (Pano-Tacanan), 5 Cayubaba (isolate), 6 Movima (isolate), 7 Trinitario-Javeriano-Loretano (Arawakan), 10 Kanoê (isolate), 11 Djeoromitxí (Nuclear-Macro-Je), 13 Pirahã (isolate), 18 Lomeriano-Ignaciano Chiquitano (Chiquitano), 19 Lakondê (Nambiquaran), 20 Mundurukú (Tupian), 23 Parecís (Arawakan).

Table 15:

Suppletive alternations in the Guaporé-Mamoré region.

#	Language	Family	Suppletive verb pairs
1	Madi	Arawan	(Caused) Motion: fall, put in water, put inside, put on ground/surface, take out, throw Posture: sit/stay, stand, lie on ground, lie on surface, lie on water, hang, tie onto a hook (Caused) Location: be inside, hold in hand Frequent: kill Other: roast/fry, pierce, be big
8	Yuracaré	isolate	(Caused) Motion: go, go in, fall, take, throw Posture: lie (down) Frequent: die
9	Itonama	isolate	Motion: fall
12	Wari’	Chapacuran	(Caused) Motion: go, arrive, fall, fly, run, take, throw Posture: stand, lie down Location: be at/live Frequent: die, cry Other: be big, be stuck
14	Karo	Tupian	Motion: run, come back Frequent: cry, sleep
15	Gavião Do Jirapaná	Tupian	Motion: go Frequent: die, kill Other: beat, knock down, make
16	Kwaza	isolate	(Caused) Motion: move, cross Posture: sit, lie down Other: end, sting on head
17	Mekens	Tupian	Motion: go, go/come, walk Posture: sit, stand, lie
21	Rikbaktsa	Nuclear-Macro-Je	Motion: run Frequent: kill
22	Irántxe-Münkü	isolate	Location: live

Though the inventory of verb pairs is much smaller in this region than it is in the Pueblo linguistic area, a very similar pattern emerges: Small inventories tend to be fully or nearly-fully shared with neighboring languages (e.g. Rikbaktsa, Mekens, Itonama), while larger inventories have more idiosyncratic semantics (e.g. Wari’, Kwaza).

5.2.5 Coastal Eastern Papua

Because of the skew of our sample, our data in the Americas are much more densely sampled than in the other four macroareas. However, Papunesia is a small macroarea with high linguistic diversity and thus provides us a relatively dense sample of the island of Papua and nearby New Britain. It is also the macroarea with the second most frequent occurrence of Sg-Pl alternation. We can thus report on the phenomenon in Eastern Papua, where the same pattern of shared semantics and internal diversification recurs (Figure 9 and Table 16).

Figure 9:

Languages in Eastern Papua and New Britain. Those without Sg-Pl alternation are: 2 Sentani (Sentanic), 5 Bukiyip (Nuclear Torricelli), 6 Taiap (isolate), 7 Duna (isolate), 8 Alamblak (Sepik), 9 Yimas (Lower Sepik-Ramu), 11 Menya (Angan).

Table 16:

Suppletive alternations in Eastern Papua and New Britain.

#	Language	Family	Suppletive verb pairs
1	Nimboran	Nimboranic	(Caused) Motion: climb, put, take, throw
3	Imonda	Border	(Caused) Motion: follow, put, put up, put into fire, stick into (Caused) Posture: sit, sit down, lie, stand, hang up (Caused) Location: hold, hold/possess Frequent: sleep, eat Other: get, give, cut, hear, see, fill in, make netbag, dislike
4	Bauni	Sko	Frequent: die Other: get
10	Yau-Nungon	Nuclear Trans New Guinea	Caused Motion: follow, take/pick up, take away Other: beat
12	Sulka	isolate	(Caused) Motion: arrive, put (away) Other: get, cook, dig, scrape
13	Daga	Dagan	Caused Motion: throw, put down Frequent: kill/hit, Other: get
14	Grass Koiari	Koiaran	Caused Motion: put Posture: sit Other: get, become/be, defecate, urinate
15	Tabo	isolate	(Caused) Motion: appear, ascend, descend, jump, put, take Posture: sit Location: stay Frequent: kill/strike Other: grow, scream, small
16	Marind	Anim	(Caused) Motion: be in movement, come, go/leave, climb, run, run away, put (horizontal item), put (vertical item), put inside, throw Posture: be sitting, sit down, be standing, stand up, be lying, lie down Caused Location: catch, grasp (horizontal item), grasp (vertical item) Frequent: die, cry, eat Other: hit, shoot, call (someone’s name), become, father (a child), give birth to, plant coconut

These languages mostly share the semantics of ‘put’, and the less common semantics of ‘get’ (the only other similar verbs are ‘get’ in Arizona Tewa and ‘receive’ in Hopi, both in the Pueblo region, see Section 5.2.3). The only two occurrences of suppletion for the word ‘follow’ in our sample occur in this region, in Yau-Nungon and Imonda, and though these two languages are relatively far apart, there may be intervening speech communities not in our sample which would reveal a chain of languages with alternating verbs for ‘follow’. Also, unlike other regions, the above languages lack common spontaneous motion verbs such as ‘go’ and ‘come’ and have very few posture verbs, except for Marind and Imonda. Recall also that transitive-only languages have almost only been observed in this region, which also has no intransitive-only languages (see Section 4.2). A more targeted study in Papunesia would likely be fruitful in uncovering more details about contact-induced Sg-Pl alternation.^[29]

There may be other clusters of languages displaying Sg-Pl alternation in regions where our sampling is too sparse to make strong observations. Possible locations include Northwestern Australia, the East African Rift Valley, and the Caucasus.

Regardless of the region, suppletive alternations overwhelmingly fall into the semantic domains listed in Section 4.4. The areal picture that emerges from the individual regions is a mixture of two trends: shared verbal semantics, likely through direct contact, and significant internal variation, especially once the inventory of alternations becomes large (more than eight or so verb pairs). The first trend indicates direct calquing of the semantic organization of the lexicon, as noted by Mithun (2022) for California. This frequently happens at the sub-areal level, with close neighbors sharing semantics which often do not extend to the area as a whole. The second trend steps in once the feature is sufficiently present in a language: Sg-Pl alternation can then diffuse across the lexicon without the need for semantic calquing. We finally want to reiterate the absence of form borrowing across any of the areas investigated.

6.1 Syntax and semantics together

The strong global preference for Sg-Pl alternation to follow an S/P pattern, and to target the semantics of Motion, Posture, and Location, together suggest a cognitive or a functional motivation at least partially driving the phenomenon. Durie (1986: 357) and Mithun (1988: 214) propose that the S/P pattern arises from the semantics of the targeted entity, namely the “affected argument.” However, many verbs with highly affected patients such as ‘hit’, ‘kick’, and ‘slap’ are infrequent in our sample. The presence of a highly affected argument is thus insufficient to explain the semantic profile of alternating verb pairs. We instead believe that a significant motivation is the visual configuration of central entities in Motion Events.^[30]

Motion, Posture, and Location events are all Motion Events in the sense of Talmy (1972), in that they all involve a Figure being located or moving with respect to a Ground. The Figure is thus very central, and has an asymmetric relationship with the Ground: the Figure is typically more movable and smaller than the Ground, of greater concern, and less immediately perceivable but more salient once perceived (Talmy 2000).

By definition, Motion Events are spatially oriented and thus highly visual. If the Figure contains more than one entity, their configuration (grouped, dispersed, etc.) is consequently more salient than that of the arguments of basic perception and emission events. The significance of configuration in Motion Events is so great that Motion lexemes frequently target specific configurations even in languages with no Sg-Pl alternation. For instance, the English word scatter specifies that the moving P is both plural and has a resulting dispersed spatial configuration, which distinguishes it from drop, which lacks both numerical and configurational specification of its arguments. (Note that drop and scatter are not an example of a Sg-Pl verb pair, as only scatter specifies number. Further, transitive scatter must be intentional, while drop may be unintentional.) Number, as a necessary component of configuration, is thus also more likely to cause lexical differentiation among Motion Events. What sets apart Sg-Pl alternation is that the lexemes are pairs contrasting number only.

In transitive Motion Events, it is typically the P argument (and not the A) which is the Figure, and according to which the alternation is triggered. The A argument is however sometimes the Figure as in ‘A crosses a bridge’ and ‘A lives at X’. Interestingly, transitive ‘cross’ and ‘live at’ are two of these A-triggered verb pairs, along with ‘chase’, where A is also moving (following the Figure).^[31]

Spatial configuration may also be relevant for three of the “Frequent Verbs”. Both ‘die’ and ‘kill’ typically involve a significant postural change in the entity passing away. (This comes of course in addition to the cognitive salience of the number of entities dying.) The verb ‘sleep’ is typically associated with the lying posture (CLICS lists colexifications of ‘lie down’ or ‘lie (rest)’ with ‘sleep’ in 225 languages; Rzymski et al. 2020). By contrast, other semantic classes like cognitive and mental states (e.g. ‘think’, ‘dream’) and basic perception or sensation (e.g. ‘see’, ‘hear’, ‘feel’) lack salient spatial configuration, which may explain their infrequency in our list of verb pairs.

The spatial configuration of objects is probably not the only motivation. The cultural significance of certain activities may be important, as already noted by Mithun (1988: 232). This is true for both ‘sleep’ and ‘eat’, which can have culturally distinct meanings for an individual as opposed to a collective (e.g. a family or village). Other somewhat-common categories like Destruction and Property verbs do not seem to be straightforwardly related to either culture or configuration.

Note that Mithun (2022: 375) considers not only number, but also animacy, shape, and/or consistency among the features relevant to the verb semantics in the Californian area. Our data suggest that animacy, shape and consistency of the entity are not particularly relevant in the verb semantics globally, since they were only present in Yana, Chimariko and Kashaya (North America), Marind (Papunesia), and Kaingang (South America).^[32] We therefore believe that these three parameters are areally relevant lexico-semantic classifications in the Californian languages, and these semantics thus are sometimes also present among Sg-Pl verb pairs. In other words, we think these are two separate typological phenomena, independently present in California. Our sample is however not as dense as hers in the region, and our methodology excludes derived and compositional verb pairs. Thus, where Mithun (2022: 387–8) counts several verbs (still pointing to some compositionality), we count only one alternating root.

Finally, we cannot rule out the possibility that our definition, which specifically targets argument rather than event number, affected our semantic results. The most relevant number for events like ‘hit’ and ‘beat’ may be the repetition of the event, while for Motion Events, it is the argument number. The semantics of lexical alternation strictly according to event number remains to be systematically surveyed.

6.2 Explaining the macroareal distribution

The link between the local influence driving the adoption of particular alternating pairs (Section 5.2) and the macroareal level (Section 4.1), at which North America stands out as a whole, remains somewhat unclear. Previous researchers have proposed various mechanisms for the overwhelming presence of Sg-Pl alternation in North America, but we find all of these proposals problematic.

Mithun (1988: 228) suggests that this feature is linked to number marking in North America, which is typically verbal rather than nominal, and expresses distribution rather than simple plurality. Under certain circumstances morphological distributivity can shift to plurality when it becomes lexically incorporated into the verb, a process she documents for some Iroquoian languages. From our survey, this pathway appears to be fairly atypical. As explained in Section 4.1, suppletion is far more frequent than lexically limited regular patterns: 75/109 languages have only suppletive pairs, only 5/109 languages have only regular patterns, and regular patterns overall are less common than suppletion. If (lexicalized) distributive morphology generates lexical alternations, it appears to be a relatively uncommon process. However, it remains possible that a category of distributive number may contribute to a cognitive organization of number as a verbal (rather than a nominal) property, which then would contribute to the formation of Sg-Pl alternation.

This is related to another proposal, namely that the absence of nominal number marking promotes number marking (including suppletion) on the verb. One suchproposal is given by Frajzyngier (1985: 99–100) (criticized in Durie 1986: 367), who suggests an inverse correlation between productive nominal number marking and the marking of a plural P role on the verb. To our knowledge, no hypotheses of this sort have been tested on typological data at scale. To do so, “lack of nominal number” would need to be narrowly defined: for example, is obligatory number marking for humans and optional number marking for animals and inanimates considered to be a lack of nominal number or not? For Frajzyngier, the association is with productive nominal plurals, and not the presence of any plural marking at all. Our survey did not keep track of the presence of distributive number nor of obligatory nominal number marking. We are inclined to agree with Durie (1986) and consider these phenomena to be of different natures, but this remains to be systematically investigated.

Another explanation that is sometimes proposed for various typological phenomena is the mode of subsistence (agriculture vs. hunter-gatherers) and the size of the speaker population. Given the presence of Sg-Pl alternation in North America and the historic dominant subsistence mode on that continent (which in many places depended on following a “yearly round” of seasonally-available resources, rather than settled agricultural plots, see Smith 2006), it is worth addressing briefly. We are fairly confident that subsistence mode and population size do not play a strong role in driving this distribution. As noted, the Pueblo linguistic area includes languages with some of the largest inventories of suppletive verbs in our sample. This is a region which has had a robust agricultural system and settled cities since at least 500 BC, with the beginnings of this system stretching back at least a millennium before (Woodbury and Zubrow 1979), and which may have been the origin of the domesticated turkey (Munro 2006). The cluster of languages in Central Mexico, which includes many long-agricultural Otomanguean groups (Campbell 2017), is another piece of counterevidence. If the Caucasus proves to hold as much Sg-Pl alternation as our limited sample in Eurasia suggests, this would be another region of long-settled agriculturalists. In much of the world and especially in our sample, a sedentary and agriculture-centered lifestyle is relatively new. The greater presence of Sg-Pl alternation among historical hunter-gatherer groups is most probably a reflection of both our sample (which excludes LOL languages, see Dahl 2015) and the global distribution of linguistic diversity (i.e. denser among recent hunter-gatherer populations).

Another possibility for the strong tendency for this feature’s presence in North America is some as-yet unrecognized cultural feature which promotes the development of these verbal pairs. While this is possible, we are at a loss to hypothesize what this cultural trait might be, nor how it might spread across a continent independently of the language contact necessary for forming a linguistic area. Notably, we were unable to find any trace of Sg-Pl alternation in records of Plains Indian Sign Language (Clark 1885; Mallery 1880/1978), a trade language over a large part of interior North America (but see Samarin 1987 for questions about its robustness at the time of European contact). Incredible as it seems across a continent as large and geographically varied as North America, the best explanation we have for the presence of Sg-Pl alternation in this macroarea is a long-standing series of contact events neighbor-to-neighbor and linguistic area-to-linguistic area. Sg-Pl alternation would thus have spread through the continent one neighbor and one area at a time, perhaps along with the spread of language families carrying this feature (such as Uto-Aztecan), which is particularly prone to being borrowed thanks to its functional properties.