Verbal numeral classifiers in languages of Eastern Eurasia: A typological survey

Abstract

This article presents a typological survey of verbal numeral classifiers (VNCs) in languages of Eastern Eurasia. As classifiers of occurrence (e.g., to V once, to V twice), VNCs are prevalent in languages of East Asia and Southeast Asia, yet the phenomenon has been largely overlooked in typological studies that have overwhelmingly focused on the adnominal use of numeral classifiers. Analyzing a sample of 543 linguistic varieties, the study categorizes the language families of Eastern Eurasia into three groups based on their use of VNCs: 1) VNCs consistently present in all languages (e.g., Tai-Kadai); 2) VNCs present in some languages (e.g., Sino-Tibetan); and 3) VNCs consistently absent (e.g., Yukaghir). Additionally, the meanings of VNCs are shown to follow a non-random distribution by centering on certain semantic fields, such as quantifying hits with ‘verbs of violence’. The study aims to offer the first step toward an areal typological study of VNCs, highlighting the importance of including this phenomenon in future descriptive work.

Full text

Research Article
Sami Honkasalo* and Chingduang Yurayong
Verbal numeral classifiers in languages of
Eastern Eurasia: A typological survey
https://doi.org/10.1515/opli-2024-0021
received July 19, 2024; accepted July 22, 2024
Abstract: This article presents a typological survey of verbal numeral classifiers (VNCs) in languages of Eastern
Eurasia. As classifiers of occurrence (e.g., to V once, to V twice), VNCs are prevalent in languages of East Asia
and Southeast Asia, yet the phenomenon has been largely overlooked in typological studies that have over-
whelmingly focused on the adnominal use of numeral classifiers. Analyzing a sample of 543 linguistic varieties,
the study categorizes the language families of Eastern Eurasia into three groups based on their use of VNCs:
1) VNCs consistently present in all languages (e.g., Tai-Kadai); 2) VNCs present in some languages (e.g., Sino-
Tibetan); and 3) VNCs consistently absent (e.g., Yukaghir). Additionally, the meanings of VNCs are shown to
follow a non-random distribution by centering on certain semantic fields, such as quantifying hits with ‘verbs
of violence’. The study aims to offer the first step toward an areal typological study of VNCs, highlighting the
importance of including this phenomenon in future descriptive work.
Keywords: verbal numeral classifiers, classifiers, event quantification, Eastern Eurasia, areal typology
1 Introduction
The present study offers an areal typological survey of verbal numeral classifiers (VNCs), namely classifiers of
occurrence, in languages of Eastern Eurasia. While nominal numeral classifiers (NNCs) have been subject to
extensive previous research, VNCs that frequently occur in East and South-East Asian languages have not been
studied to the same extent from a comparative-typological viewpoint. Previous studies addressing the phe-
nomenon have been mostly limited to the systems present in the most dominant individual languages of the
region, Mandarin Chinese in particular. It is no exaggeration to say that rarely has such a regionally ubiquitous
phenomenon been neglected in typological research literature. To address this gap, the present study offers the
first typological study of VNCs in Eastern Eurasia, utilizing a representative sample of surveyed languages
from all 23 major language families and language isolates (i.e., one-language families) of the region discussed
further in Section 1.3.
The study has the following three primary goals: First, based on the surveyed languages, we formulate a
more precise and typologically oriented definition for VNCs and differentiate the phenomenon from related
linguistic concepts, since the term has been used inconsistently in various differing and ill-defined ways in
earlier research, including its application to morphosyntactic phenomena that should not be considered
prototypical numeral quantification.
Second, the study examines the distribution of VNCs among the 23 surveyed language families of Eastern
Eurasia and demonstrates how languages and language families of the region can be grouped into three main

* Corresponding author: Sami Honkasalo, Department of Languages, Faculty of Arts, University of Helsinki, Helsinki, Finland; Research
Institute for Languages and Cultures of Asia, Mahidol University, Nakhon Pathom, Thailand, e-mail: samihonkasalo@gmail.com
Chingduang Yurayong: Department of Languages, Faculty of Arts, University of Helsinki, Helsinki, Finland; Research Institute for
Languages and Cultures of Asia, Mahidol University, Nakhon Pathom, Thailand
Open Linguistics 2024; 10: 20240021
Open Access. © 2024 the author(s), published by De Gruyter. This work is licensed under the Creative Commons Attribution 4.0
International License.

categories based on whether VNCs can be identified. The distribution of prototypical VNC systems centers
around Southeast and East Asia. In language families, such as Tai-Kadai and Hmong-Mien, VNCs are ubiquitous
and show great semantic breadth. In contrast, many of the surveyed language families contain languages, such
as Mangghuer (Mongolic) that have simple systems of verbal classifiers that in several cases result from
language contact while many other languages in such families lack verbal classifiers. Finally, when moving
outside the core distribution area of verbal classifiers, languages and language families, such as Uralic
Yukaghir, with no systems of VNCs become dominant.
As the third primary goal, the study investigates the semantics of VNCs in the surveyed languages. It
demonstrates that the meanings of VNCs in the surveyed languages across language families manifest non-
random distribution by clustering around specific semantic core fields. For instance, in addition to a generic
VNC present in every language with a grammaticalized system of such classifiers, languages with full VNC
systems often include one or several classifiers to quantify and classify hits with ‘verbs of violence’.
1.1 Terminology
Three terminological points need clarification. To begin with, ‘Eastern Eurasia’refers to the Eastern reaches of
the Eurasian continent that comprises East Asia, Central Asia, and Mainland Southeast Asia, regions where
language contact has historically been frequent.
1
In particular, Mainland Southeast Asia and East Asia form the
geographic core area for the distribution of both NNC and VNC systems. While the main island groups next to
the Eurasian mainland and in historical contact with it, such as the Japanese archipelago, Taiwan, and Hainan,
are included, the Eastern maritime regions, such as Maritime Southeast Asia and the Pacific islands, remain
outside the scope of investigation. Individual languages from the Indian subcontinent are also included and all
language families indigenous to India are surveyed, but the Indian subcontinent remains outside the focus of
the present study. In the territory corresponding to Russia, we include languages and language families spoken
to the east of the Ural Mountains, a traditional geographic boundary between Europe and Asia. While this
artificial boundary does not necessarily correspond to any linguistic reality, it is nevertheless necessary for
delimiting the scope of the study. Also, we exclude languages of Europe and Southwest Asia. It should be noted
that the selection aims not to deny the possible existence of VNCs outside the investigated region.
2
For a
universal typology, further scholarship is consequently needed to confirm the global distribution of VNC
systems in languages of the world.
Second, the present study addresses a phenomenon for which we coin the term ‘exact and adjustable
event quantification’, since VNCs constitute one possible device for this purpose. Exact and adjustable quanti-
fication is a subtype of quantification that fulfills two conditions, namely 1) the attribution of an exact
numerical value, and 2) the possibility to adjust such value. Languages of the world possess a wide range
of means for event quantification, but not all of these attribute an exact and adjustable numerical value to the
event, such as the English adverb sometimes in She sometimes goes out for a walk in the evening. Such non-
exact and non-adjustable devices of event quantification remain outside the scope of the study. They have
been widely discussed from the viewpoint of aspect, pluractionality, reduplication, and adverbs in previous
research. On the other hand, VNCs constitute a grammatical device for exact and adjustable event quantifica-
tion, since an exact numerical value is attributable to event occurrence and the value can also be adjusted, as

1The study is limited to Eastern Eurasia only. This is not intended as a statement concerning the global distribution of verbal
numeral classifiers that may exist also in some languages of the Americas, for instance. Rather, the geographic boundaries of the
study resulting in an areal convenience sample are due to its pilot nature: We wished to focus on East and Southeast Asia that form
the distributional core for verbal numeral classifiers as identified in earlier research and to further investigate the spread of the
phenomenon in the adjacent areas.
2As shown by typological studies of classifiers, such as Gil (2013) and Her and Li (2023), among the regions adjacent to Eastern
Eurasia, nominal numeral classifiers can be identified in languages of Maritime Southeast Asia, and the Pacific, although more
sparsely than in their core distributional center investigated in the present study. As discussed in Section 2.2, verbal numeral
classifiers may exist in languages with nominal numeral classifiers and not vice versa. Consequently, a possibility for discovering
verbal numeral classifiers in languages of the regions excluded in this study cannot be ruled out.
2Sami Honkasalo and Chingduang Yurayong

with the Mandarin Chinese classifier cì ‘time’in qù sāncì去三次(go three CLF.time) ‘to go three times’where
the numeral sān‘three’can be productively replaced by sì 四‘four’, for instance.
In addition to VNCs, other devices for exact and adjustable event quantification exist in languages of
Eastern Eurasia. These devices, found in Eastern Eurasian languages that lack VNCs, are not covered in the
present study but are briefly introduced below (refer to Yurayong et al. 2024, Section 4, for more details). The
main types include counted nouns, iterative numerals, and verbalized numerals. Counted nouns refer to
nouns that can be counted directly as a device of exact and adjustable quantification, such as English ‘time’
and its Written Tibetan counterpart thengs ‘time’(1).
(1) Counted nouns: Written Tibetan (Sino-Tibetan)
nga rgyagar=la thengs gsum ‘gro myong yod.
1SG India=DAT time three go EXP.PRF AUX.EGO
‘I have been to India three times.’(constructed)
In turn, iterative numerals are a denumeral category (Fradin 2015) used for indicating exact and adjus-
table quantification. For instance, Alytor derives iterative numerals by means of the suffix-saŋattaching to a
numeral, as in (2).
(2) Iterative numerals: Alytor (Chukotko-Kamchatkan)
ŋəra-saŋjat-ə-tkənedel
j
a-k.
three-ITER come-E-IMPF:2SG.Sweek-LOC
‘You come three times a week.’(Nagayama 2003, 81)
Finally, as a rare strategy in languages of Eastern Eurasia, verbalized numerals are attested in the
Formosan Austronesian languages that attach verbalizing prefixes to numerals for deriving verbs that indicate
frequency. For instance, Thao uses verbalizing prefixes, such as mu- (3).
(3) Verbalized numerals: Thao (Austronesian)
Mu-turu-z iza yaku mu-tusi
VBLZ-three-SUFF already 1SG.NOM VBLZ-there
Lipún m-riqaz nak a azazak.
Japan AF-see 1SG.POSS LNK child
‘I have been to Japan three times to visit my daughter.’(Blust 2003, 1021–2)
The phenomenon termed ‘verbal numeral classifiers’in the present study is known by several names in
previous research. To begin with, many specialists of Sino-Tibetan and Southeast Asian languages have used
the term ‘verbal classifier’to refer to classifiers of occurrence (e.g., Aikhenvald 2000, 9, Paris 2013, Zhang
2017a), illustrated by Thai ครั้ง kʰráŋ‘time’in expressions such as สองครั้ง sɔ
ːŋkʰráŋ‘twice, two times’and สาม
ครั้ง săːmkʰráŋ‘three times’, respectively. In addition, the research literature contains particularly frequent
mentions of ‘verbal action classifiers’(e.g., Bai 2019, 295–6, Chirkova 2009, 35, Daudey 2014, 146, LaPolla and
Huang 2003, 68, Tunzhi 2019, 238–9, Zhang 2013, Zheng 2016, 116). Other similar terms include ‘activity classi-
fiers’(Ahrens and Huang 2016, 174–5, Massupong 1982, 115) and ‘measures for verbs of action’(Chao 1968, 627).
The term ‘verbal action classifiers’is commonly used by Chinese scholars or other scholars operating
primarily in the linguistic context of China. Together with ‘activity classifier’,itreflects a Chinese language
term for the phenomenon. The Chinese research tradition frequently uses the term dòngliàngcí 动量词
translatable as ‘action classifier’or ‘verbal classifier’. It contrasts with wùliàngcí 物量词‘object classifier’
used for NNCs. The term Dòngliàngcí 动量词can be found in most Chinese publications dealing with VNCs,
Verbal numeral classifiers in languages of Eastern Eurasia 3

such as in multiple volumes of the influential series of descriptive grammars, Zhōngguó xīnfāxiàn yǔyán
yánjiūcóngshū中国新发现语言研究丛书 Newly Found Minority Languages in China Series.
Less frequently, the term ‘(verbal) event classifier’is also attested (Kobuchi-Philip 2021). Finally, high-
lighting a functional perspective, the term ‘adverbial classifiers’appears infrequently in previous studies
(Thurgood et al. 2014, Donlay 2015, 235).
Many of the used terms present certain challenges. For instance, while the term ‘verbal action classifier’
avoids the polysemic confusion of the term ‘verbal classifier’, not all instances of use are limited to action verbs
only. Also, as will be discussed in the following section, the term ‘verbal classifier’is polysemous and already
used differently in the established research tradition of native American languages. Consequently, based on
the established term ‘numeral classifier’(e.g., Aikhenvald 2000), we adopt the term ‘verbal numeral classifier’
to specify numeral classifiers that occur in the verbal context.
3
Similarly, we use the term ‘nominal numeral
classifier’in the nominal contexts. The terminological choice also reflects our theoretical interpretation that
numeral classifiers branch into two main subcategories: nominal and verbal. While ‘verbal numeral classi-
fiers’as a term has not been in frequent use, it is not an innovation coined here, but can also be found in
previous research, such as Bisang (2018, 276) and Zhang (2017a).
1.2 Previous studies of VNCs in languages of Eastern Eurasia
VNCs are often omitted in descriptive studies. For instance, Iwasaki and Ingkaphirom’s (2005) representative
grammar of Thai lacks any mentions of VNCs, even though Thai possesses a rich system of VNCs. Even when
the phenomenon is explicitly identified, it is often given superficial treatment with few if any examples of use.
Also, verbal classifiers may be present in the examples of descriptive grammars without receiving any
dedicated discussion, which results in documentation without description. Consequently, the present study
identifies VNCs not only from direct mentions but also indirectly, despite the risk of misinterpretation that is
mitigated by the specialist survey.
Verbal classifier languages have been subject to highly uneven attention regarding the phenomenon. Of
the surveyed languages, the most substantial research concerning VNCs has focused on Mandarin Chinese.
Previous studies, such as that of Chao (1968, 627–30), identify tens of different VNCs in the language. Other
representative studies of the phenomenon include the pioneering work of Bhaskararao and Joshi (1985) on
Newar (Sino-Tibetan, Newaric) and Gerner (2009) on Kam (Tai-Kadai, Kam-Sui). Among recent studies, Li (2020,
166–232, 369–81) not only offers a particularly detailed study concerning the NNCs of Zauzou (Sino-Tibetan,
Loloish), but also discusses the VNC system at great length, showing the way and providing a standard for
future grammar writing.
The uneven and often insufficient treatment of VNCs has direct relevance to the limitations of this study.
To enhance our understanding of this frequent yet paradoxically neglected grammatical phenomenon in
languages of Eastern Eurasia and beyond, more fieldwork-based in-depth research from lesser-researched
languages is needed. The authors hope that the present study with an areal typological approach lays the
groundwork for such research.

3An anonymous reviewer proposed the term ‘event quantification classifier’as an alternative. The term is informative as a label
describing the function of the classifiers. Following a similar functional approach in terminology, nominal numeral classifiers could
be termed ‘entity classifiers’to maintain consistency in the principles of nomenclature. In this study, we nevertheless prefer the
terms ‘verbal’and ‘nominal numeral classifiers’that focus on syntactic relations. Also, the choice of the term ‘verbal numeral
classifiers’highlights the classifier type’s close relationship with ‘nominal numeral classifiers’, the classifier type better known from
earlier research.
4Sami Honkasalo and Chingduang Yurayong

1.3 Methodology and structure of the article
4
The study is based on an analysis of 543 linguistic varieties
5
from Eastern Eurasia. It aims to include languages
from all major language families of the region and from diverse branches of the families. The following 23
language families are included in the study: Ainuic, Amuric, Austroasiatic, Austronesian, Burushaski,
Chukotko-Kamchatkan, Dravidian, Great Andamanese, Hmong-Mien, Indo-European (Indic and Tocharian
branches), Japonic, Koreanic, Kusunda, Mongolic, Nihali, Ongan, Sino-Tibetan, Tai-Kadai, Tungusic, Turkic,
Uralic, Yeniseian, and Yukaghir. In addition, available data from the regional creole languages are also taken
into account.
Rather than following random sampling suited for formulating statistical generalizations, the sampling of
languages included herein is pragmatic based on the availability of descriptive data. Several closely related
languages in genealogically relatively compact language families were investigated since distinct patterns of
language contact may have yielded different results in those languages regarding the presence of VNCs. A case
in point is the Mongolic language family where Khalkha lacks a grammaticalized classifier system, yet such a
system with both NNCs and VNCs exists in Mangghuer due to Chinese influence. Finally, reflecting Dahl’s (2001,
1456) insight that areal typology manifests both a synchronic and a diachronic side, the survey also includes
historical languages of Eastern Eurasia in addition to contemporary languages of the region, since they
provide valuable information concerning the diachronic development of VNC systems and explaining the
diffusion of the phenomenon.
6
Methodologically, the data collection and analysis rely on three approaches. First, as the cornerstone of
this study, we surveyed existing descriptive grammars and text collections from among the regional languages.
Since VNCs are often neglected to varying degrees in grammar writing, this step involved developing hypoth-
eses and dealing with uncertainties regarding the analysis of several phenomena.
7
As Section 3 will show, due
to the fragmentary and limited nature of the existing data, some of such uncertainties remain to be answered

4In the discussion that follows, verbal numeral classifiers are named descriptively with the format CLF.X where X briefly char-
acterizes the nature of the event, as in CLF.time (the generic undifferentiated verbal numeral classifier) and CLF.step (steps taken
with a movement verb as the predicate). When more semantic detail is necessary, further distinctions are included with the colon
CLF.X:Y, as in CLF.hit:knife (stabs with a knife with a verb of violence as the predicate), as in Stau æ-ndəɲcʰə (one- CLF.hit:knife hit) ‘to
stab once’. Due to the nature of the phenomenon and semantic domains associated with it, some examples unavoidably include
instances of aggressive and violent behavior. Finally, the paper contains examples from three kinds of sources identifiable as
follows: Examples from previous published research are attributed to their sources while examples received via the survey or an
interviewed native speaker of a language are attributed to the contributor, unless they have been published previously in another
context. Examples with the label ‘constructed’originate from the authors of the study. To harmonize data from various sources
with different research traditions, we have implemented slight adjustments to the glossing conventions followed by each con-
tributor and the surveyed original sources.
5The term ‘linguistic varieties’is used in lieu of ‘language’here, since some of the surveyed varieties can be argued to be dialects of
a language. While the issue is universal in linguistics, in the Eastern Eurasian context, the Greater Himalayan region contains many
linguistic cases where drawing the line between a language and a dialect is often challenging. Two main approaches to distinguish
languages from dialects have been attempted to draw a line between languages and dialects, namely mutual intelligibility and
‘counting sameness’when measuring different domains of the linguistic system (Wichmann 2019). As part of the latter, Wichmann
uses a computational method based on the Normalized Levenshtein Distance, reaching the conclusion that the language-dialect
dichotomy has an observable basis. Some scholars nevertheless advocate the interpretation of languages and dialects as socially
constructed entities without clear boundaries; refer to Weber and Horner (2012, 27–38) and Romaine (2000, 10) for the critique of
the notion of countable distinct languages.
6When offering a visual geographic overview of the language sample and the presence of verbal numeral classifiers from the
viewpoint of the present time, only contemporary linguistic varieties are included in Figures 1 and 2. In turn, a brief diachronic
overview is offered in Section 5.
7The topic of verbal numeral classifiers is often not discussed directly in descriptive studies. Even when they are claimed to exist,
grammars frequently omit examples of practical use that could serve for identifying the morphosyntactic behavior of the proposed
verbal numeral classifiers further. This has required reading the descriptive grammars interpretively as source data and analyzing
whether certain descriptions and examples are interpretable as instances of verbal numeral classifiers. Therefore, the classifica-
tions and analyses of individual language varieties in this study should not be taken universally as claims by the original authors,
but in many instances as our interpretations.
Verbal numeral classifiers in languages of Eastern Eurasia 5

in future work. Following, we conducted a VNC questionnaire distributed among linguists researching lan-
guages of Eastern Eurasia. The brief questionnaire serves the function of supplementing the data obtained
from written research outputs. It thus served as a supplementary tool for clearing points of doubt, not as a
primary research method. Finally, as an additional step to disambiguate the collected data, we informally
interviewed several native speakers of regional languages who shared their native intuition regarding VNCs
and event quantification in their respective native languages. The remaining challenges are further discussed
in Section 3.
The results from the three steps above were collected into an event quantification marker database (see
the supplementary file). In the database, each language variety was assigned a language family and geogra-
phical coordinates. As the first step, we identified whether the language possessed verbal and nominal
classifiers. When the answer to both was positive, to the extent that this was possible, we investigated the
semantic domains where the VNCs occur. For languages with no VNCs, we aimed to determine the strategy
followed for exact and adjustable quantification. Finally, the completed database was used for drawing
typological generalizations.
The present study follows the following structure: In Section 2, we will offer a definition of VNCs and
equally illustrate several morphosyntactic phenomena that should be excluded from the definition to keep it
meaningful. Following, Section 3 introduces the language sample serving as the major data source and
provides an overview of both nominal and verbal classification in the surveyed languages. Analyzing the
data, Section 4 shows that typologically, the semantics of VNCs in the surveyed languages follow a non-random
distribution by centering around certain core semantic domains discussed in detail. Section 5 briefly delves
into the diachronic aspects of VNCs in languages of Eastern Eurasia and discusses the role of language contact
in their diffusion. We propose that both NNCs and VNCs originate from the Tai-Kadai languages and have
spread widely via language contact. Finally, Section 6 with a conclusion summarizes the central findings of the
study and discusses its limitations.
2Defining VNCs
This section contrasts VNCs with their nominal counterparts and thus starts with a brief discussion on
grammaticalized nominal classification devices in Section 2.1. Then, in Section 2.2 it presents a detailed
definition of VNCs since the term has been used polysemously in earlier studies. It is argued that not all
instances that have been described as verbal or event classifiers in earlier studies should be included inside
the scope of the definition to keep it typologically meaningful.
In brief, VNCs are classifiers of occurrence. In other words, they indicate the exact and adjustable
frequency of an event expressed by the predicate verb, e.g., once, twice, thrice, as -li ‘time’in (4) from
Geshiza. As discussed in detail in Section 4, in addition to a generic VNC ‘time’as illustrated in the example,
languages with VNCs may make additional different degrees of semantic distinctions regarding the occurrence
of an event, e.g., ‘steps’or ‘hits with the fist’.
(4) Geshiza (Sino-Tibetan)
sopʰoæ-li dæ-ɕʰoŋzda.
Suopo one-CLF.time PFV-go.PST.1 EXP
‘I have been to Suopo once.’(Honkasalo 2019, 330)
2.1 Grammaticalized nominal classification devices
Nominal classification and NNCs have received extensive typological interest. Research on VNCs builds on this
research tradition. Consequently, before proceeding further and offering a more detailed definition of VNCs, it
6Sami Honkasalo and Chingduang Yurayong

is necessary to contrast them with NNCs, a prominent noun categorization device. Furthermore, in what
follows, we will briefly investigate other common related noun categorization devices.
8
Some grammatical devices for categorizing nouns and nominals exist in almost all languages (Aikhenvald
2006, 463). Gender systems in languages, such as Arabic, German, and Spanish, categorize nominal referents
into grammatical genders based on various parameters, including animacy and sex. Such systems are char-
acterized by grammatical agreement, exemplified by Spanish where both the indefinite article un/una and the
adjective caro/cara ‘expensive’agree in gender with the masculine head noun apartamento ‘apartment’(5a)
and the feminine head noun casa ‘house’(5b).
(5a) Spanish (Indo-European)
un apartamento car-o
ART.INDF.Mapartment.M.SG expensive-M.SG
‘an expensive apartment’(constructed)
(5b) una casa car-a
ART.INDF.Fhouse.F.SG expensive-F.SG
‘an expensive house’(constructed)
Many linguists see gender as a subtype of noun classes with few available categories. Swahili provides an
illustrative example of an extensive noun class system and the resulting agreement. The noun ki-tabu ‘book’
belongs to the seventh noun class along with many other tools. Both the adjectival modifier and the predicate
verb must agree with the noun by means of the agreement prefixki-, as in (6a). In contrast, m-toto ‘child’is a
noun of the first class for people, and the agreement surfaces as m- for the adjectival modifier and as a- for the
subject marker in the verb, illustrated in (6b).
(6a) Swahili (Niger–Congo)
Ki-tabu ki-zuri ki-li-kuwa meza-ni.
7-book 7-good 7-PST-EXV table-LOC
‘The good book was on the table.’(constructed)
(6b) M-toto m-zuri a-li-kuwa nyumba-ni.
1-child 1-good 1-PST-EXV home-LOC
‘The good child was at home.’(constructed)
Finally, NNCs are a noun categorization device that co-occurs with accompanying numerals and/or quan-
tifiers and classifies nominal referents into different semantic categories. The classificatory criteria may be
based on animacy, shape, and other inherent properties (Aikhenvald 2006, 466). NNCs form a system at a mid-
way point between lexical (e.g., English two head of cattle) and grammatical (e.g., gender and noun classes)
means of nominal classification (Grinevald 1999, 2000). The two main categories of NNCs are sortal and
mensural classifiers, although the latter is sometimes treated separately and the criteria distinguishing the
two have been debated. Sortal numeral classifiers categorize nouns in terms of their inherent properties, such
as animacy and shape (Aikhenvald 2000, 115). For instance, in Xong, the classifier ngonl marks the referent,
namely mioul ‘fish’, as an animate entity (7), unlike the classifier hnanf, for instance, which is used for
inanimate referents in the semantic field of clothing (Sposato 2015, 258–9).

8The treatment here is not intended as exhaustive, but merely to give sufficient background information placing numeral
classifiers into a larger context and to illustrate how verbal numeral classifiers differ from nominal classification devices. In
addition, noun classifiers, classifiers in possessive constructions (relational, possessed, and possessor classifiers), locative classi-
fiers, and deictic classifiers exist in languages of the world (Aikhenvald 2000, 2006, 2019).
Verbal numeral classifiers in languages of Eastern Eurasia 7

(7) Xong (Hmong-Mien)
Wel naond aod-ngonl mioul neind liox guaot!
1SG ASSOC one-CLF.animate fish DEM.PROX big pass
‘This fish of mine is really big!’(Sposato 2015, 229)
Mensural numeral classifiers, also known as measure words, individuate entities in terms of quantity
(Lyons 1977, 464). In contrast to sortal classifiers that categorize nouns based on their inherent properties, they
categorize based on temporal states. In other words, they “contain information about how the referent is
measured”(Aikhenvald 2006, 466). In (8), the Xong mensural numeral classifier zhet ‘CLF.bowl’indicates that
the cooked rice that the speaker ate was in a bowl and the speaker ate two bowls of such rice.
(8) Xong (Hmong-Mien)
Tat-hnef wel nonx oub-zhet hlit.
this-CLF.day 1SG eat two-CLF.bowl cooked.rice
‘I ate two bowls of rice today.’(Sposato 2015, 235)
Since the present study focuses on VNCs, rather than NNCs, we adopt the following broad criterion for the
presence of NNCs: A language is defined as having grammaticalized NNCs if it has at least one sortal numeral
classifier and at least some nouns in the language are typically counted with the help of classifiers. The size of
the classifier inventory does not matter, and classifiers need not be mandatory, but a prototypical classifier
inventory contains several classifiers so that it can serve the function of classifying nouns. When counting,
NNC insertion may be compulsory (e.g., Mandarin Chinese) or optional (e.g., Khmer), especially under lan-
guage-specific conditions, such as in the context of large numbers. Systems that deviate from the prototype
above and yet show similarity with it are defined as limited classifier systems. For instance, a language might
be in the process of classifier genesis, as Kurtöp (Hyslop 2017, 148–9). Alternatively, it might possess counters
for the subset of specific nouns, such as humans, which serve the function of noun categorization, as in Ainu
(Tamura 2000, 255–9). Systems with mensural quantifiers for entities of low countability, such as English
bottles and ounces in three bottles of Coke and five ounces of gold dust, respectively, occur in most or all
languages (Gil 2013) and are excluded here. Like many other linguistic phenomena, however, NNCs form a
continuum from prototypical to limited or emerging systems, which occasionally makes it difficult to establish
whether a language is a ‘classifier language’or not in borderline cases.
NNCs show strikingly uneven geographic distribution with a core area in East and Southeast Asia
extending both eastwards and westwards (Gil 2013). At the time of writing, the research database of Her
and Li (2023) qualifies as the most extensive cross-linguistic source of information concerning NNCs, and the
authors identify over 700 classifier languages from among languages of the world, also confirming the earlier
findings of strikingly uneven areal distribution. Hölzl and Cathcart (2019) show how classifiers become less
frequent when moving from East to West in Eurasia. In all, as confirmed by the present study, the distribution
of NNCs greatly overlaps with that of VNCs. The two are nevertheless not identical, since only a subset of
languages with NNCs also possess VNCs. Consequently, verbal numeral classifier languages form a smaller
subset of all classifier languages.
2.2 Defining VNCs
The notion of verbal classifiers has been used in various ways in different linguistic approaches. For instance,
in the research tradition of indigenous languages of the Americas and Papua, verbal classifiers refer to “mor-
phemes which occur on the verb and characterize a core argument in terms of its shape, form, consistency, and
other semantic properties”(Aikhenvald and Dixon 2011, 157). To illustrate, Imonda, a Papuan language, possesses
a wide range of verbal classifiers that attach prefixes to the verb. In (8), the classifier u- indicates that the referent
tõbtõ ‘fish’belongs to the group of small animals, such as frogs and fish (Seiler 1985, 121). Since such devices do
not mark event frequency, they are not discussed in the present study. Rather, they constitute yet another form
8Sami Honkasalo and Chingduang Yurayong

of noun categorization device with agreement (Aikhenvald 2000, 151). Against this backdrop, we propose a
terminological distinction, namely ‘verbal classifiers’to be used in the meaning illustrated above and ‘verbal
numeral classifiers’for grammaticalized classifier devices for counting and classifying events.
(9) Imonda (Border/Upper Tami)
tõbtõ ka-m u-ai-h-u.
fish 1-GL CLF.small.animal-give-REC-IMP
‘Give me the fish!’(Seiler 1985, 120)
To avoid terminological confusion, it is necessary to define VNCs in a fashion that succinctly, yet accu-
rately characterizes their behavior from a cross-linguistic viewpoint in Eastern Eurasia. Below, we offer three
criteria for defining VNCs. The first criterion defines the classificatory device as dependent on a grammati-
calized NNC system. Following, criteria two and three address the syntax and semantics of the classifiers,
respectively. In addition to showing what VNCs are, the criteria also illustrate what they are not. Multiple
linguistic phenomena have either been included as instances of VNCs or could potentially be conceptually
conflated with it. The end of the section summarizes the discussed criteria.
2.2.1 Criterion 1: Dependence on nominal numeral classification
VNCs constitute a grammaticalized classifier system and thus differ from lexical quantification devices, such as
counted nouns (e.g., English ‘time’in She saw the movie three times). To qualify as a VNC, a morpheme needs to
fulfill the basic properties of classifiers in a language. The definition should not be taken as a requirement for
fully identical morphosyntactic behavior, an impossibility due to among other things the different syntactic
distribution of NNCs and VNCs. Rather, the emphasis lies on similarities between the two. In other words, both
VNCs and NNCs are subclasses of the word class ‘classifiers’.
The criteria of classifierhood must be established on a language-specific basis and as an example, com-
monly include (at least moderately) free co-occurrence with a numeral. To illustrate further, Indonesian has
NNCs, such as ekor used for animals. The numeral satu ‘one’appears in a prefixed form se- attached to a
classifier, including both the sortal and mensural subtypes. Consequently, tendangan ‘kick’fails to qualify as a
VNC in the language, since it cannot host the prefixal form of the numeral and should thus be considered a
counted noun (10), unlike the genuine VNC kali ‘time’that behaves like other classifiers in the language by
hosting the numeral prefix, as in se-kali ‘one time, once’.
(10) Indonesian (Austronesian)
Atlet memberikan satu/*se- tendangan kepada lawan-nya.
athlete give one kick to opponent-3SG
‘The athlete gave a kick to his/her opponent.’(Syafruddin)
The classifier system of a language may have NNCs and lack VNCs, but no opposite systems where VNCs
exist while NNCs are lacking were identified in this study (see also Vittrant and Tang 2021). To illustrate,
English has no grammaticalized system of nominal numeral classification, and ‘time’does not qualify for a
VNC. Rather, in terms of its word class properties, it behaves like other nouns of the language and thus
constitutes a lexical means for achieving the same function, namely to indicate the frequency of an event.
At the same time, Japanese in example (11) contains a VNC -do ‘time’, since the language possesses a system of
NNCs and in terms of its word class, -do shares the basic properties and behavior of classifiers in the language,
rather than that of prototypical nouns.
Verbal numeral classifiers in languages of Eastern Eurasia 9

(11) Standard Japanese (Japonic)
彼女は東京を三度訪れた ことが ある。
kanojo=wa tokyo=wo san-do otozure-ta koto=ga aru.
she=TOP Tokyo=ACC three-CLF.time visit-PST NMLZ=NOM EXV
‘She has visited Tokyo three times.’(constructed)
Importantly, the requirement above signifies that similar to prototypical NNCs, VNCs must be compatible
with a wide numeral range, not only with the number one, for instance. The choice of the term ‘wide’is
intentional: A wide numeral range does not necessarily mean the same as a limitless numeral range, since
among others, pragmatic and cultural reasons may limit the numeral range a certain VNC typically co-occurs
with. Stau possesses both NNCs and VNCs as closed word classes, -ɳɖʐə ‘time’illustrating the latter (12). In
addition, many verbs may undergo semelfactive-like derivation in which the number prefixæ- ‘one’attaches
to the unconjugated verb stem followed by the light verb v-ræ (13). Unlike in the case of VNCs with productively
adjustable numeral value, however, such semelfactives that closely resemble classifiers cannot replace the
numeral prefixæ- by any other number. Consequently, ɣnə-‘two’in (14) results in an ungrammatical for-
mulation, while in (12), the numeral prefix can be freely adjusted, as in ɣnə-ɳɖʐə ‘two times’. Ergo, classifier-
like semelfactive formulations of the type (13), sensu stricto, fall outside the scope of the prototypical verbal
numeral classification.
(12) Stau (Sino-Tibetan)
ŋa kapəla gɛ-ʁjɛdə-ŋoŋ.stɛwu æ-ɳɖʐə ɮaŋ.
1Sforehead ADJZ-good PFV-COP.1 Stau one-CLF.time come.PST.1
‘I was lucky. I came (i.e., I was able to come back) to Daofu once.’(personal fieldwork)
(13) tʰe=qʰeŋevdʑə=gi æ-skʰəri də-re.
DEM=LOC 1SG.GEN friend=DAT SEM-shout IMP-LV.2SG
‘Call to my friend over there!’(personal fieldwork)
(14) *tʰe=qʰeŋevdʑə=gi ɣnə-skʰəri də-re.
DEM=LOC 1SG.GEN friend=DAT two-shout IMP-LV.2SG
Intended meaning: ‘Call twice to my friend over there!’(rejected)
Paris (2013, 270) divides VNCs into ‘weak’and ‘strong’. Cases restricted to co-occurrence with the number
‘one’belong to ‘weak classifiers’while ‘strong classifiers’co-occur with the full range of numbers. From this,
we can see that Paris’s‘weak’classifiers correspond to the semelfactive quasi-classifiers of this article. They do
not quantify the times of occurrence with the possibility of adjustability in the number value and are thus
excluded from the present study, an issue which will be discussed in more detail in criterion 3. Classifier-like
semelfactives, however, exist in several languages of Eastern Eurasia and would thus form a fruitful ground
for a separate typology study. Example (15) from Japanese illustrates what Paris’s categorization labels ‘weak
verbal classifiers’and what the present study calls semelfactive quasi-classifiers. The word koe 声literally
means ‘voice’and it is used idiomatically in the expression hito-koe wo kakeru ‘to call (someone)’. The numeral
is fixed and cannot be replaced. Interestingly, the number ‘one’occurring in such expressions may play the
role of marking infiniteness, rather than counting.
(15) koe 声‘voice’
hito-koe 一声‘one call’
*futa-koe 二声intended: ‘two calls’
*hyaku-koe 百声intended: ‘a hundred calls’
Finally, the similarity between NNCs and VNCs implies that like the former, the latter are generally
compatible with interrogation by an interrogative pro-form that often carries the basic meaning ‘how
10 Sami Honkasalo and Chingduang Yurayong

much, how many’. The two typically share at least one interrogative pro-form, as in Thai: กี่ kìː‘how many’,กี่
คน kìːk
h
on ‘how many people’vs. กี่ครั้ง kìːkʰráŋ‘how many times’. In other words, interrogatability con-
stitutes a shared feature between NNCs and VNCs, illustrated further in Table 1. It follows that incompatibility
with an interrogative generally indicates that a form is not a VNC. This is connected to VNCs’semantic
functions of quantifying events discussed in criterion 3. However, since the alternative devices for VNCs,
such as counted nouns and iterative numerals, may also be compatible with interrogation, co-occurrence
interrogation does not necessarily confirm the status of a morpheme as a VNC.
To conclude the first criterion and to inspect the similarity and overlap in VNCs and NNCs, three distribu-
tional possibilities exist for a numeral classifier regarding its co-occurrence (or lack thereof) in nominal and
verbal contexts. A numeral classifier may be dedicated for either nominal or verbal use only or it may be
compatible with both. Matthews and Leung (2004) illustrate how Cantonese Chinese exhibits all these three
patterns. For instance, seng
1
声‘voice’is used exclusively for vocalizations, such as calls and instances of cough.
On the other hand, faai
3
塊‘slice’shows exclusive nominal use only for counting slices of objects, such as a cake.
Finally, numeral classifiers, such as caan
1
餐‘meal’show compatibility with both nominal and verbal use.
2.2.2 Criterion 2: Syntactic link to the verb phrase
NNCs belong syntactically to noun phrases (NPs) where they may either modify a head noun or substitute for
it, depending on the language in question. In (17) from Thai, the classifier เลมlêm used for books modifies the
head noun หนังสือ nǎŋsɯ
ː‘book’together with which they form a classifier phrase (CLFP), a subunit of NP.
(16) Central Thai (Tai-Kadai)
[[หนังสือ สาม เลม]
CLFP
นั้น]
NP
หนัก มาก
[[nǎŋsɯ
ːsǎːm lêm]
CLFP
nán]
NP
nàk mâːk.
book three CLF.book DEM.DST be.heavy very
‘Those three books are very heavy.’(constructed)
In contrast, VNCs manifest a syntactic connection with the predicate verb. They combine with a numeral
to form a verb CLFP that serves the function of a complement to the verb. Regarding their syntactic function,
VNCs resemble adverbs at least in some of the investigated languages, which is the reason for the term
‘adverbial classifiers’occasionally met in the research literature. In Stau, the generic VNC -ɳɖʐə ‘time’that
attaches to its numeral host ɣsu ‘three’directly precedes the verb in (17a). The adverb vɛrvi ‘slowly’in (17b)
shows identical syntactic placement by directly preceding the verb.
Table 1: Compatibility of both NNCs and VNCs with interrogative pro-forms
Language Interrogative Nominal example Verbal example
Japanese nan(i) nan-biki nan-po
‘what’‘how many small animals’‘how many steps’
Korean myŏtmyŏt-myŏng myŏt-bŏn
‘how many’‘how many people’‘how many times’
Mandarin Chinese jǐjǐběnjǐquán
‘how many’‘how many volumes of books’‘how many hits with a fist’
Geshiza xazi xazi-qʰa xazi-rəu
‘how many’‘how many stick-like objects’‘how many hits with an axe’
Thai kìːkìːkʰan kìːkâːw
‘how many’‘how many cars’‘how many steps’
Vietnamese bao nhiêu bao nhiêu người bao nhiêu lần
‘how many’‘how many people’‘how many times’
Malay berapa berapa orang berapa kali
‘how many’‘
how many people’‘how many times’
Verbal numeral classifiers in languages of Eastern Eurasia 11

(17a) Stau (Sino-Tibetan)
vdzi ɣsu-ʁeɣsu-ɳɖʐə nə-ɕə-sə.
person three-CLF.person three-CLF.time PFV-go.3-IFR
‘Three people went three times.’(Gates 2021, 225, transcription harmonized)
(17b) vɛrvi tə-ɕən.
slowly IMP-go.2
‘Go slowly!’(Gates 2021, 124)
It is important to distinguish VNCs from NNCs occurring in or as an object argument to a verb. As Gerner
(2014, 288–9) demonstrates, Nuosu allows some ‘container nouns’, such as ‘mouth’, to be used as classifiers.
Such classifiers, however, modify a noun that may be present or omitted, as in (18). While frequent omissions
of the possible head noun complicate the quick identification of such cases, they must nevertheless be
excluded from VNCs, since they do not modify the verb, rather than a head noun that may be present or
omitted.
(18) Nuosu (Sino-Tibetan)
tsʰɨ³³ (dza³³) tsʰɨ²¹ ba³³ɬɔ²¹ dzɯ³³.
3SG food one CLF.mouth eat
‘He ate two mouthfuls of food.’(Gerner 2014, 289)
Somewhat distinctly from the above, NNCs indicating frequency should not be conflated with VNCs that
indicate occurrence. In Mandarin Chinese, the classifier cì 次‘time’functions as the generic VNC for iterations
(19a). The same classifier, however, is polyfunctional by also occurring in NPs, as in (19b) where in jǐcì huìtán
几次会谈‘some/several talks’cì indicates that the head noun ‘talk’took place several times. In sum, the
requirement for a syntactic link with the verb phrase rules out (19b) as an instance of VNC, yet (19a) qualifies.
As stated earlier, this reflects how a single classifier may function both in the nominal and verbal domains.
(19a) Mandarin Chinese (Sino-Tibetan)
我去过日本两次。
wǒqù guò Rìběnliǎng cì.
1SG go EXP Japan two CLF.time
‘I have been to Japan twice.’(constructed)
(19b) 他们 进行了几次会谈。
tā-men jìnxíng le jǐcì huìtán.
3-PL hold ASP some CLF.time talk
‘They held several talks.’(Li and Thompson 1981, 110)
Also, the study omits what Gerner (2014,289–91) calls ‘verbal auto-classifiers’,namely‘verbs which serve as their
own phase and event counter’.CoinedbyMatisoff(1973, 89) for Lahu (Sino-Tibetan, Lolo-Burmese), auto-classifiers in
the nominal context refer to scenarios where some nouns function as their own classifiers, such as yɛtê yɛ‘one
house’in Lahu. The term ‘repeaters’is also used synonymously. In the verbal context, the term has been used for
expressions, such as Mandarin Chinese kàn (yí) kàn 看(一)看(look (one) look) ‘to have, take a look’.
9
To distinguish
the nominal and verbal domains, Liu (2018) labels the former ‘auto nominal classifiers’and the latter ‘auto verbal
classifiers’with the ‘reduplicants coding lexical or grammatical aspects’.Inlanguages,suchasMandarinChinese,
auto-verbal classifiers do not quantify events and show compatibility only with the numeral yī一‘one’,whichmakes
them functionally semelfactives (see criterion 1) and further highlights their difference from auto-nominal classifiers.

9It has been widely discussed whether the presence of the numeral yī一‘one’included in parenthesis changes the meaning of the
expression. However, we will not explore this issue further, as it is tangential to the main arguments of the present study.
12 Sami Honkasalo and Chingduang Yurayong

In some languages, auto-verbal classifiers may possibly be fully productive in the sense of being compatible with a
broad base of verbs and serve the function of adjustable event quantification. Adjustability in event quantification,
namely whether the accompanying numeral value can be modified or not, is typically not discussed in descriptions
of VNCs. This issue thus requires more investigation and would benefit from a dedicated study.
2.2.3 Criterion 3: Semantic function of quantifying events
VNCs quantify events rather than referents, the domain of NNCs. Consequently, due to their distinct function,
even when they show similarity with NNCs, they cannot be considered a nominal categorization device.
Duratives, namely expressions indicating temporal lengths of time units, such as ‘two days’,‘three weeks’,
and ‘five years’remain outside the scope of the definition, since they do not indicate frequency. In many of the
surveyed languages, duratives are either temporal nouns or temporal adverbs. In others, they appear as ‘self-
referential’classifiers, for instance in Geshiza, Mandarin Chinese, and Thai (Table 2).
Self-referential classifiers count themselves (Konnerth 2014, 142), rather than quantifying events. Also
called ‘independent classifiers’by Haas (1942, 204–5) in the context of Thai, they are close to mensural nominal
classifiers, since rather than serving the role of noun categorization device, they measure the quantity of time
units, albeit self-referentially (see Honkasalo 2019, 328). The term ‘time classifier’is also occasionally used for
the classifier type (Zhang 2014, 901). In Karbi, self-referential classifiers, such as arnì ‘day’and jó ‘night’, appear
(20). Similarly, วัน wan ‘day’in Thai counts itself, as in สองวัน sɔ
ːŋwan (two CLF.day) ‘two days’. Repeated
formulations, such as *วันสองวัน *wan sɔ
ːŋwan (day two CLF.day), with the intended meaning ‘two days’are
not used.
(20) Karbi (Sino-Tibetan)
aphi atum ejo arnisi dokokta
a-phì a-tūme-jó arnì-sī
10
dō-kòk=tā
POSS-grandmother POSS-PL one-CLF.night CLF.day-one stay-back=ADD
“ehem chedamnang erit chedamnang […]”
e-hēm chV-dām-nāng e-rīt chV-dām-nāng
1PL.INCL-house RR-go-HORT 1PL.INCL-field RR-go-HORT
‘The grandmother and grandfather stayed just one night and one day, and then (the old man) said:
let’s go home, let’s go to our property […]’(Konnerth 2014, 143)
In addition to the semantic function of quantifying events with an exact and adjustable numerical value,
VNCs also manifest a delimiting function by setting a boundary for an event (Matthews and Leung 2004, 453).
To illustrate, in (21a), the bare verb haang
4
行‘to walk’describes the event as an unbounded event with no VNC
present. In contrast, the introduction of the VNC bou
6
步‘CLF.step’in (21b) delimits the event and converts the
Table 2: Examples of self-referential classifiers in Geshiza, Mandarin Chinese, and Thai
Type Geshiza Mandarin Chinese Thai
day -sni tiān wan
night -rja wǎnkʰɯːn
week n/a n/a ʔaːtʰít, sàpdaː
month -slən/a dɯən
year -ko nián piː

10 Konnerth (2014, 143) points out that the enumeration is irregular and surfaces instead of the expected *e-nì.
Verbal numeral classifiers in languages of Eastern Eurasia 13

act of walking into a bounded event by establishing a boundary to its occurrence. As the authors further argue,
this also corresponds to transforming the event from an atelic into a telic one.
(21a) Cantonese (Sino-Tibetan)
行
haang
4
walk
‘to walk’(Matthews and Leung 2004, 453, Chinese character added)
(21b) 行兩步
haang
4
loeng
5
bou
6
walk two CLF.step
‘to walk two steps’(Cantonese; Matthews and Leung 2004, 453, Chinese characters added)
2.2.4 Summary of the criteria for defining VNCs
In conclusion, the present study defines prototypical VNCs as follows: VNCs are a grammatical classificatory
device for quantifying events in some languages that also have NNCs. VNCs manifest a strong syntactic link
with the predicate verb, rather than with an NP that is the domain of NNCs. VNCs show productive compat-
ibility with wide numeral ranges, rather than co-occurring with fixed numbers only. Via negativa, the fol-
lowing are excluded from the definition of prototypical VNCs: 1) formulations occurring with fixed numbers
only, typically ‘one’; 2) NNCs occurring as the object argument of a verb regardless whether a head noun is
present or not; 3) verbal ‘auto-classifiers’where a verb is repeated to serve as its own counter; and 4) duratives
appearing typically as self-referential classifiers to indicate temporal lengths without quantifying events.
3 Language sample of the present study
This section offers an overview of the distribution of numeral classifier systems in the surveyed 23 language
families. The 11 major phylogenetic units of Eurasian languages in the present study include Austroasiatic,
Austronesian, Dravidian, Hmong-Mien, Indo-European (mostly Indo-Iranian), Mongolic, Sino-Tibetan (also known
alternatively as Trans-Himalayan, Tibeto-Burman), Tai-Kadai, Tungusic, Turkic, and Uralic. Additionally, in the
figures that follow, we include under an umbrella unit ‘Paleo-Asiatic’the following 12 smaller genealogical units:
Ainuic,Amuric,Burushaski,Chukotko-Kamchatkan,GreatAndamanese,Japonic,Ongan,Koreanic,Kusunda,
Nihali, Yeniseic, and Yukaghir, and also take the major creoles and pidgins of the region into account. In total,
the sample contains 543 distinct contemporary linguistic varieties.
The analysis of the languages and the language families reveals that language families of Eastern Eurasia
form three core groups regarding the presence and absence of VNCs. First, Hmong-Mien, Japonic, Koreanic,
and Tai-Kadai are the only language families where VNCs are consistently present among all member lan-
guages of the respective families. Also, only languages of these language families, together with some branches
of Sino-Tibetan (e.g., Qiangic), possess fully fledged VNC systems. In contrast, in the remaining language
families, grammaticalized VNCs are generally rudimentary if they are present.
Second, in Austroasiatic, Austronesian, Indo-European, Mongolic, Sino-Tibetan, and Turkic, verbal classi-
fiers are present only in some, but not in all languages of the respective language families. Finally, in Ainuic,
Amuric, Burushaski, Chukotko-Kamchatkan, Dravidian, Great Andamanese, Kusunda, Nihali, Ongan, Tungusic,
Uralic, Yeniseian, and Yukaghir, no systems of VNCs were identified.
Summarizing the findings, Table 3 provides an overview of the surveyed languages in terms of the
presence or absence of VNCs.
Following, Figure 1 illustrates the geographic distribution of the surveyed languages. In turn, Figure 2
illustrates the distribution of classifier systems in Eastern Eurasia. Red dots on the map indicate the surveyed
14 Sami Honkasalo and Chingduang Yurayong

languages with both NNCs and VNCs present; blue dots mark languages with only NNCs being identified; and
black dots code the languages with no classifiers of either kind.
Since there is also a significant internal variation in each phylogenetic unit as well as ambiguity observed
in the published grammatical descriptions, we subsequently discuss three worthwhile observations from
the data.
The first issue concerns the nature of data sources in historical languages. For instance, it is impossible to
ascertain whether the absence of VNCs in the Shang Chinese oracle bone inscriptions in general results from a
lacuna in the surviving inscriptions or from a genuine lack of this grammatical feature. At the same time, while
the surviving materials from Old Korean are insufficient for determining how the language counted verbal
events, the lack of NNCs helps us deduce that VNCs must also have been absent.
Second, drawing the line between devices of exact and adjustable quantification, particularly in the case
of counted nouns and VNCs, is often challenging. While not unique in this respect, Modern Khmer presents
such a problem for analysis, since dɑːŋ‘time’is analyzable either as a VNC or as a counted noun, namely a
noun that is counted directly by means of a numeral. Old Khmer with the oldest historical documents available
shows incipient classifier genesis, but the language likely had no classifier system in its earlier stages that have
not been recorded. Also, it is historically clear that Khmer has used the device of counted nouns for exact and
adjustable event quantification. Against this backdrop and due to the scarcity of usable morphosyntactic clues,
Table 3: Languages of Eastern Eurasia and VNCs
Typ. group Phylogenetic units Remarks on distribution
Group 1 Hmong-Mien, Japonic, Koreanic, Tai-Kadai VNCs consistently present across the
units
Group 2 Austroasiatic, Austronesian, Creoles and pidgins, Indo-European, Mongolic,
Sino-Tibetan, Turkic
VNCs present in some languages of
the units
Group 3 Ainuic, Amuric, Burushaski, Chukotko-Kamchatkan, Dravidian, Great
Andamanese, Kusunda, Nihali, Ongan, Tungusic, Uralic, Yeniseian, Yukaghir
VNCs consistently absent across the
units
Figure 1: Geographical locations of the surveyed languages (map base © ArcGIS).
Verbal numeral classifiers in languages of Eastern Eurasia 15

it remains challenging to determine whether dɑːŋhas been reanalyzed as a VNC or whether it remains as a
counted noun. In all, dɑːŋeither continues as a counted noun or has become reanalyzed as a classifier
sometime at the Middle Khmer stage or after when the classifier system emerged in the language. The issue
requires further investigation.
As the mirror image of the above, some earlier research has used the label ‘verbal numeral classifier’to
devices of exact and adjustable quantification that the criteria established in the present study analyze differ-
ently. In Jingpaw for instance, làŋis often described as a VNC, but the morpheme is likely a counted noun
forming right-headed noun-noun compounds typical to the language; it also shows different word order (NUM-
làŋ) when compared with NNCs (CLF-NUM; Kurabe, personal communication, July 25, 2022). Consequently,
following the criteria for establishing VNCs adopted in this article (Section 2.2.1), a non-VNC interpretation is
preferred.
The third area of potential confusion is what counts as a classifier system in contrast to sporadic counting
devices which can be alternatively analyzed as counted nouns. For example, China Coast Pidgin shows limited
use of a general classifier piece(e) that originates from Chinese influence (Ansaldo et al. 2010; Shi 1991, 20). This
limited use of a counter is not considered a fully-fledged classifier system here. Likewise, in Sri Lankan
Portuguese Creole, Smith (1977, 148–9, 2013) shows how pesaam ‘person’occasionally occurs as a classifier
for humans. This limited use of a counter is not interpreted as a fully-fledged classifier system in the pre-
sent work.
We propose a two-part universal based on the respective sizes of the VNC and NNC inventories. First,
complex NNC systems correlate with the presence of verbal numeral systems. The most complex VNC systems
of Tai-Kadai, Hmong-Mien, and some branches of Sino-Tibetan, such as Qiangic and Sinitic, invariably occur in
languages with complex NNCs. Conversely, very simple NNC systems correlate with the lack of VNCs. Second,
in a language with both subtypes of classifiers present, the NNC subsystem typically shows richer formal
complexity than its verbal counterpart. While the universal is not absolute, it is nevertheless a very strong
tendency: The only identified counterexamples originate from Sinitic contact varieties that have greatly
simplified their systems of NNCs, such as Kazakhstani Gansu Dungan (Honkasalo 2024), but at the same
time, retain systems of VNCs.
Figure 2: Core distribution of VNCs (map base © ArcGIS).
16 Sami Honkasalo and Chingduang Yurayong

4 Typological trends in lexical meanings of VNCs
The present study shows that rather than following random distribution, lexical meanings of VNCs in lan-
guages of Eastern Eurasia are restricted to particular semantic fields. This parallels many other grammatical
phenomena. For instance, lexicalizations of negative senses center around a limited number of semantic
domains (Veselinova 2013).
The semantic types discussed below are established with the help of two criteria. Primarily, we have
formulated hypernymic categories that encompass a wide range of subtype-level manifestations of VNCs
(apart from the category of ‘generic’) identified in the source materials. For instance, classifiers of hits with
first, kicks with feet, and bites of mouth, etc. have been grouped under the semantic category ‘instrumental A:
body parts as classifiers with “verbs of violence”’. Secondarily, we have attempted to create categories for
widely attested semantic types of VNCs present in the source materials.
Table 4 summarizes the findings. The ten language families with identified VNCs can be grouped further
on the elaborateness of the respective systems. The widest semantic range of VNCs is found in Tai-Kadai,
Hmong-Mien and Sino-Tibetan, particularly in the Qiangic and Sinitic branches. In contrast, the Japonic and
Koreanic language families show less elaborate systems of VNCs. The VNC systems of Indo-European, Mon-
golic, and Turkic are of very simple nature, highlighting their origin due to language contact. Also, the systems
in Austroasiatic and Austronesian include only an identified generic classifier and, in a few cases, additionally
an identified locomotive classifier, typically for steps. Due to the nature of existing grammatical descriptions,
many instances in these two language families remain opaque regarding whether a lexical item is best
analyzed as a classifier or a counted noun. Consequently, Austroasiatic and Austronesian are clearly not
core languages for VNCs, and further research may result in narrowing down the range of language with
VNCs. These observations are also directly linked with the diachrony of VNCs, as will be discussed in Section 5.
We propose a hierarchy between the semantic fields for predicting the occurrence of a classifier type:
onomatopoeic >instrumental >locomotive >generic. While this hierarchy should not be taken as an absolute
universal, it illustrates the general tendencies in languages with verbal classifiers. If a language possesses the
onomatopoeic classifier type, all other types are likely to be equally present (e.g., Thai). In turn, if onomato-
poeic classifiers are lacking, but the instrumental type is present, a language likely also possesses locomotive
and generic verbal nominal classifiers (e.g., Geshiza). Furthermore, if a language lacks both onomatopoeic and
instrumental classifiers, yet possesses the locomotive type, it will additionally have one or several generic
classifiers (e.g., Dungan).
4.1 Generic VNCs and other general classifiers
All languages with grammaticalized verbal numeral classification include a classifier with a broad functional
range for quantifying times of occurrence. Such a classifier is termed ‘generic verbal numeral classifier’in the
Table 4: Semantic ranges of the VNCs in the analyzed language families
Language family Generic Locomotive Instrumental B Instrumental A Onomatopoeic
Tai-Kadai ✓✓ ✓ ✓ ✓
Hmong-Mien ✓✓ ✓ ✓ ✗
Sino-Tibetan ✓✓ ✓ ✓ ✗
Japonic ✓✓ ✓ ✗ ✗
Koreanic ✓✓ ✓ ✗ ✗
Austroasiatic ✓✓ ✗ ✗ ✗
Austronesian ✓✓ ✗ ✗ ✗
Indo-European ✓✓ ✗ ✗ ✗
Mongolic ✓✓ ✗ ✗ ✗
Turkic ✓✓ ✗ ✗ ✗
Verbal numeral classifiers in languages of Eastern Eurasia 17

present work, and Table 5 lists some such classifiers in the surveyed languages. Moreover, in some languages,
such as Jruq/Loven (Austroasiatic, Bahnaric) ruoh ‘time’, the generic VNC is the only identified VNC that exists
in addition to more numerous NNCs. Due to its generic nature, the general VNC may replace other more
specific classifiers in some languages, as in Bhaktapur Newar (Joshi 1984, 188).
The following examples illustrate generic VNCs in use. To begin with, Gan Chinese uses tshi
213
次as a
generic VNC (22).
(22) Gan Chinese (Sino-Tibetan)
我去过三次。
ŋo
34
tɕhiɛ
44
-kuo san
34
tshi
213
.
1SG go-EXP three CLF.time
‘I’ve been there three times.’(Li 2018, 141)
In a similar fashion, Zoulei, the VNC vie
31
counts the frequency of events (23). Stieng and Korean use wət
(24) and -bŏn(25), respectively.
(23) Zoulei (Tai-Kadai)
vlei
31
na
13
i
13
lɔ
55
vəɯ
13
ʁəɯ
31
əɯ
55
vie
31
.
year DEM.PROX LSG return go home one CLF.time
‘This year, I went home once.’(Li et al. 2014, 129)
(24) Stieng (Austroasiatic)
kənɔŋdi-cənam ɓaːnbɘːh ti psaːr-snual ɓaːn di-wət.
in one-year TAM:get come to market-Suol TAM:get one-CLF.time
‘I could come to Snuol market once a year.’(Bon 2014, 128)
(25) Standard Korean (Koreanic)
리더는하루에백번싸운다
lidŏ-nŭn haru-e paek bŏnssau-nda.
leader-TOP day-DAT hundred CLF.time fight-PRS.DECL
‘The leaders fight hundred times a day’(title of a book by Jo U-Sŏng 2019)
Even though the generic VNCs are often used for counting trips and spatial movement, they are not
dedicated trip classifiers (Section 4.4), but typically co-occur with a wide set of predicates. In Jiaomuzu Situ
Gyalrong, the generic classifier cʰais attested as a counter of various kinds of events, such as marrying (26a)
and killing (26b) in addition to trips (26c), highlighting its generic nature.
Table 5: Examples of generic VNCs in the surveyed languages
•Austroasiatic: Aslian: Semelai tən; Bahnaric: Jruq/Loven ruoh, Stieng wət; Katuic: Kuy ɗɐːŋ,Pacoh kəntɨʔ; Khasi-Palaungic: Mok pɔk;
Khmuic: Mlabri thɯɯ; Pearic: Chong dɔːŋ; Vietic: Vietnamese lần
•Austronesian: Malay kali, Urak Lawoi’kali
•Japonic: Japanese: Old Japanese 多妣 -taNpyi, Standard Japanese 回-kai; Ryukuan: Yuwan Amami 回-kəi, Irabu 回-kwai,
Okinoerabu 回-(k)koi
•Koreanic: Standard Korean: 번-bŏn
•Sino-Tibetan: Gyalrongic: Geshiza -li, Stau -ɳɖʐə,Khroskyabs -sti, Jiaomuzu Situ Gyalrong cʰa; Karenic: Sgaw Karen ɓlɔ1; Lolo-
Burmese: Bisu tsaŋ
55
,Cosao thɔ
55
,Enu/Ximoluo phv
33
,Khatso thɤ55, Lalo phuq, Lisu xua
35
,tʂo
33
,tʂɒ
33
(dialectal variation), Yao’an
Lolo bol; Newaric: Bhaktapur Newar kAlA/kA, Pahari Newar -pu; Qiangic: Yongning Na ʂɯ33, Pumi/Prinmi xɜjF; Sinitic: Gan
Chinese 次tshi213, Mandarin Chinese 次cì, Shaowu Min Chinese 回fei22, Southern Min Chinese 擺pai3, Xiang Chinese 回,次
fei13, tshɪ45; unclear affiliation: Tujia qie2
•Tai-Kadai: Kam-Sui: Kam tau⁵³; Kra: Lakhi ɕɔN
22
,A’ou/Zoulei vie
31
, Tai: Lao ເທື
່ອtʰʉa, Thai ครั้ง kʰráŋ,Neo-Ahom pak,Aiton thaa1,
Lungchow baːt4, Nung bạt
18 Sami Honkasalo and Chingduang Yurayong

(26a) Jiaomuzu Situ Gyalrong (Sino-Tibetan)
stonmoŋkəɲes cʰana-ˈa-va-w pəʒərnə-ŋanəŋka-jn.
wedding two CLF.time PRF-NEV-do-3SG again PRF-divorce-3PL
‘She married twice and divorced again.’(Prins 2016, 272)
(26b) kə-cʰa=tʃetəwaʔm na-ˈa-sat-w.
one-CLF.time=LOC bear PRF-NEV-kill-3SG
‘He killed a bear once.’(Prins 2016, 271)
(26c) wuɟo-ɲo lhase kə-cʰaji-ˈa-tʃʰi-jn.
3-PL Lhasa one-CLF.time PRF-NEV-go
1
-3PL
‘They went to Lhasa once.’(Prins 2016, 271)
A question arises whether a language may possess two or several generic VNCs. In languages such as
Japanese, two classifiers show particularly wide usability. The classifiers -kai and -do show frequent inter-
changeability, yet a closer inspection reveals that in contrast to -kai, which is compatible with any numeral, the
functional range of -do is mostly limited to small numerals (Kobuchi-Philip 2021, 182, see also Tao 2013
concerning the distribution of the two). Preferring the more generic over the more specific and restricted,
-kai thus constitutes the generic VNC in Japanese. Similarly, while both ครั้ง kʰráŋand หน hǒn(originally
meaning ‘path, way’) can count a wide variety of events in Thai, the former far exceeds the latter in frequency
and is consequently considered more basic. In sum, while it is theoretically possible that a language possesses
more than one generic VNC, in practice, differences in functional range and frequency can be generally used to
argue for the primacy of one form over the others. Due to the limited source materials, however, in some
instances, it has not been possible to determine whether one of the several generic VNCs is more basic than the
other(s). Detailed future studies of individual languages will cast more light on the issue and help to solve such
uncertainties.
In some languages with NNCs, it is possible to specify the referential value of a noun with the choice of a
classifier. For instance, Zhang (2013, 400) illustrates this from Ersu where the word ndzɿ‘buckwheat’is
compatible with a wide array of classifiers, each of which results in foregrounding some distinct semantic
aspects of the word, although broadly from the same semantic field. When used with the classifier -pu ‘plants’,
ndzɿtə-pu refers to ‘a buckwheat plant’. Co-occurring with the classifier -pa ‘roundish objects’, however, ndzɿ
ta-pa means ‘a buckwheat seed’. Other examples by Zhang are offered in (27).
(27) no classifier: ndzɿ‘buckwheat’
-pu ‘plants’ndzɿtə-pu ‘a buckwheat plant’
-pa ‘roundish objects’ndzɿtɑ-pa ‘a buckwheat seed’
-tɕʰo‘bundles’ndzɿtə-tɕʰo‘a bundle of (harvested) buckwheat’
-tɕʰo‘pieces of land’ndzɿtə-tʂʰo‘piece of land where buckwheat grows’
A parallel phenomenon is identifiable in some languages with VNCs. In other words, the choice of
a classifier with a single verb may specify the semantics of the verb. To illustrate, in Mandarin Chinese,
the verb dǎ打‘to hit’has a very general meaning, to the extent that it has grammaticalized into a
light verb, as in dǎdiànhuà 打电话‘to call’, literally ‘to hit a phone’. The verb shows compatibility with a
multitude of VNCs, each of which foregrounds a certain semantic value from the wide semantic field of the
verb. With the generic VNC cì 次‘time’,asindǎyí cì 打一次‘to hit once’the meaning remains general. With the
classifier quán 拳‘hits with fist’,dǎyì quán 打一拳acquires the meaning ‘to hit with a fist once’. Additional
examples are listed in (28).
Verbal numeral classifiers in languages of Eastern Eurasia 19

(28) no classifier: dǎ打‘to hit’
cì 次‘generic’dǎyí cì 打一次‘to hit once’(general)
quán 拳‘hits with the fist’dǎyì quán 打一拳‘to hit with a fist once’
qiāng 枪‘shots with a gun’dǎyì qiāng 枪打一枪‘to shoot once’
bāzhang 巴掌‘slaps with the palm’dǎyì bāzhang 打一巴掌‘to give a slap with the palm’
Finally, in the non-generic VNCs discussed below, a selectional relation holds between the predicate verb
and its accompanying classifier. In other words, the use of many non-generic VNCs is controlled by the
predicate verb, similarly to the selectional relation between an NNC and the head noun. To illustrate, in
Geshiza, the classifier -ba(r) ‘CLF.step’requires a predicate that is a verb of movement, such as v-tɕə ‘to take
steps’. In contrast, the classifier -bi ‘CLF.hit:stick’only accepts the light verb v-ra ‘to hit’as the predicate.
4.2 Instrumental classifiers type A: body parts as classifiers with ‘verbs of violence’
In many languages of East and Southeast Asia, VNCs exist for counting hits, either with body parts
used instrumentally or with an instrument proper. In the first type, a classifier has been formed for
hits with various body parts predicated by verbs of violence. This semantic subtype of VNCs can be identified
in Tai-Kadai, Hmong-Mien, and Sino-Tibetan. Languages show major variation regarding the subdivisions
in this category. To illustrate, Pahari Newar (Sino-Tibetan, Newaric) applies the versatile classifier -thu for
violence perpetrated with body parts, such as blows, kicks, and scratches (Shrestha 2010, 218). In contrast,
Geshiza with a wide spectrum of subdivisions for instrumental VNCs possesses the dedicated classifiers -rgu
‘hits fists’,-rtsʰokicks with the feet, -ltə‘hits with the head’, and -wʈʂæ‘clawings with claws’(Honkasalo
2019, 329).
To illustrate the use of the first subtype of instrumental classifiers, Wadu Pumi counts strikes with the fist
involving vertical movement with the VNC -bi and slaps with -lɑ
(Daudey 2014, 146–7) (29).
(29) Wadu Pumi (Sino-Tibetan)
tsɑ
w=sətsɑwʑìŋwèŋkʰì=bù,
pound=CONTR.TOP pound can CUST.EXCL time=TOP
sòŋ-bí=nòŋ=bù sòŋ-lɑ
tɕəqʰùmədzə
three-CLF.pummel=COORD=TOP three-CLF.slap do need GNOMIC
ʻ(…) You can pound, but you will have to thump three times and slap three times (…)ʼ(Daudey 2014, 147)
Similarly, Khroskyabs counts hits with the palm with the classifier, as in təlpʰæʁ(30).
(30) Khroskyabs (Sino-Tibetan)
næ-ɕôŋ=ɕəvæ gəvə
=təcə
mɲədə
-təlpʰæʁo-tsʰóŋ.
PST-go.II.1SG=CONJ wife=DEF like.this one-CLF.hit:palm PST-hit.II.1SG
‘When I left, I hit my wife with the palm like this.’(Lai)
Finally, Khatso uses the classifier tɕo
24
to mark the frequency of kicks with the feet (Donlay 2015, 237) (31).
(31) Khatso (Sino-Tibetan)
i
33
ŋa
33
tɛi
31
tɕo
24
tɕhɛ
55
wa
323
.
3SG 1SG one CLF.kick Kick PFV
‘He kicked me once.’(Donlay 2015, 237)
20 Sami Honkasalo and Chingduang Yurayong

4.3 Instrumental classifiers type B: tools and objects with ‘verbs of violence’
In the second instrumental type, the VNC counts hits by tools and objects. This group resembles group A
closely, but the hits occur by means of an instrument proper, not a body part used as an instrument. Examples
of common semantic domains include hits or cuts with a knife, strokes with a rod, and gunshots. Stabbing or
cutting with a knife in particular constitutes a common type of VNC contained in this semantic category. In
Zhaozhuang Bai, the classifier ta
35
counts the number of cuts with a knife (32).
(32) Zhaozhuang Bai (Sino-Tibetan)
sɛ
44
sa
55
ta
35
cut three CLF.cut:knife
‘to cut three times with a knife’(Zhao 2012,75–88)
In turn, Stau uses the classifier -ndəfor counting hits with a knife (33). Importantly, while the instrumental
classifiers often derive from instrument nouns cross-linguistically, the classifier bears no direct connection
with the noun pərzi ‘knife’and remains synchronically opaque.
(33) Stau (Sino-Tibetan)
pərzi a-qʰa=kʰææ-ndənə-ɲcʰə-sə.
knife one-CLF.stick=INSTR one-CLF.hit:stab PFV-LV.3-IFR
‘S/he hit it with a knife once.’(personal fieldwork)
Also, gunshots constitute a cross-linguistically common semantic field for the instrumental VNCs. In Xong,
the classifier put is used for counting the number of gunshots (34). Similarly, Mandarin Chinese uses the
classifier qiāng 槍for counting gunshots (35).
(34) Xong (Hmong-Mien)
Beul baond aod-ngonl nbat-doub dox bub-put, beul
3 shoot one-CLF:animate pig-earth that three-CLF:gun 3
deit jix daos.
still NEG
1
die
‘He shot the wild pig three times (with a gun), but it still didn’t die.’(Sposato 2015, 278)
(35) Mandarin Chinese (Sino-Tibetan)
大林對著小鳥打了四槍。
Dàlí duìzhe xiǎoniǎodǎle sì qiāng.
Dàlín to small bird hit ASP four CLF.gunshot
‘Dàlín shot the small bird four times.’(Zhang 2017b, 629)
Since firearms are a relatively new technology, these classifiers, at least with their current meanings, lack
great antiquity. Also, while firearms have largely replaced the bow and arrow as the primary trajectory
weapon, we may speculate that historically, classifiers for counting the arrows shot may also have existed,
although this remains speculative and not confirmed by any data at the moment.
4.4 Locomotive type
The locomotive type of VNCs measures the frequency of events that involve movement. By far, the most
frequently attested classifier measures the number of steps taken when walking, as in (36) with ʔjaːm
5
‘CLF.step’from Bouyei. In many languages, it constitutes the only instance of a locomotive VNC. In addition,
Verbal numeral classifiers in languages of Eastern Eurasia 21

the frequency of jumps, runs, rounds around an object, trips, and round trips, among others are measured
with VNCs of this semantic category.
(36) Bouyei (Tai-Kadai)
ku
1
naːi
5
tɯ
2
ðaːi
4
ʔjaːm
5
lo
1
tu
3
piaːi
3
mi
2
ʔdai
4
leu
4
.
1SG tired really CLF.step one all walk NEG can already
‘I am so tired that I cannot move one more step.’(Burusphat et al. 2001, 559)
When counting trips, languages may make a distinction between simple trips with no regard to return and
return trips where the start and end points are identical. Offering an illustration, simple one-way trips are
counted with the classifier thɔŋ
5
in Jianghua Mien (37a). In contrast, the language applies a distinct classifier
dzuŋ
6
for return trips (37b).
(37a) Jianghua Mien (Hmong-Mien)
miŋ
2
ji
2
thɔŋ
5
ta
2
ji
2
thɔŋ
5
go one CLF.one.way.trip come one CLF.one.way.trip
‘to go and to come once’(Zheng 2011, 162)
(37b) ming
2
i
1
dzuŋ
6
mai
3
lɔ
4
pwə
8
go two CLF:round.trip NEG find meet
‘to go and come back twice without finding (the person one was looking for)’(Zheng 2011, 162)
4.5 Onomatopoeic type
Onomatopoeic VNCs imitate the sound that results from the action or event of a verb. They exist at least in
some languages of the Tai-Kadai family, but among all semantic types of VNCs, they have received by far least
attention. In Thai, Noss (1964, 108–9) identifies this group of morphemes as classifiers, and they are also
mentioned in other previous research (e.g., Lu 2012, 229–30) for Thai. To illustrate, sounds of laughter in
Thai can be counted with the onomatopoeic VNC ฮาก hâːk(38a) while instances of flatulence receive a
dedicated onomatopoeic classifier ปาด pâːt(38b).
(38a) Standard Thai (Tai-Kadai)
หัวเราะ สอง ฮาก
hǔarɔsɔ
ːŋhâːk
laugh two CLF.ONOM:laughter
‘to laugh twice’(Lu 2012, 229–30)
(38b) ตด สอง ปาด
tòt sɔ
ːŋpâːt
fart two CLF.ONOM:fart
‘to fart twice’(constructed)
Some other Thai onomatopoeic words commonly observed in the adjustable event quantification struc-
ture are, for instance, เฮือก hɯ
ak for breathing and attempting to accomplish something, ฟอด fɔ
ːtfor pecking
someone on the cheek, and กรึ๊บ krɯ
pfor sipping a drink.
22 Sami Honkasalo and Chingduang Yurayong

4.6 Others
Like in the case of NNCs (Aikhenvald 2000, 273), languages may have specific VNCs attested less frequently in
other languages. Such classifiers are placed in the category of ‘other’in the present work. However, applying
the criteria described at the beginning of this section may warrant establishing new semantic categories as
more data becomes available in the future.
To illustrate, Rabha possesses a dedicated classifier sákho- for the frequency of bearing offspring in
humans and batri- for animals (39).
(39) Rabha (Sino-Tibetan)
batri-sa sábra pra-a.
CLF.bear.offspring.animal-one child give.birth-PRS
‘to bear offspring (animal) once’(Joseph 2007, 462)
Vocalizations, such as calls, shouts, and other productions of noise, frequently appear subject to counting,
which may even justify treating them as a semantic subtype of VNCs when more data become available.
Vocalizations are illustrated in Khatso (40), Kam (41), and Namuzi (42).
(40) Khatso (Sino-Tibetan)
za
31
ni
24
za
31
jo
35
ŋ

31
tɕi
55
v

24
wa
323
.
girl CLF two CLF.shout call PFV
‘(He) shouted twice (at) the girl.’(Donlay 2015, 237)
(41) Kam (Tai-Kadai)
mau
33
pan
55
sam
35
so
33
.
3SG call three CLF.voice
‘He called three times. (lit. He called three voices).’(Gerner 2009, 729)
(42) Namuzi (Sino-Tibetan)
Âmî-m û sô-qo luó-ndzhuò, sô-kì ló-gô dzhuò.
ʔa˥mi˥-mu˥so˥-qo˧lu
o˧˥-ndʐu
o˥˧ so˥-ki˥˧ lᵘo˧˥ go˥dʐu
o˦˨.
mother-SG three-CLF.voice up-call three-CLF.time up-turn hither
‘Mother has called [him] three times, [so he] three times turned back.’(Pavlík 2017, 125)
5 Brief diachronic remarks
It remains outside the scope of this pilot study to delve deeply into the history of VNCs. The following illustrates
the authors’general interpretation of the collected data.
The question regarding the origin of VNCs connects closely to that of NNCs since the former are dependent
on the latter, as explained earlier in this article. In addition, based on the proposed NNC primacy, we argue
that NNCs predate VNCs that emerge from the former through analogy. In other words, an existing classifi-
catory device is extended from the nominal into the verbal domain. A hypothesis of polygenesis would
postulate that verbal and numeral classifiers emerged in languages of Eastern Eurasia independently from
each other. However, as illustrated in Section 4, the great degree of cross-linguistic similarity in the semantics
of VNCs hints that such systems result from language contact and have spread through diffusion. Conse-
quently, a hypothesis of monogenesis cannot be ruled out, and it has been offered in earlier research, e.g.,
Her and Li (2023) and Janhunen (2000).
Classifiers are likely a later innovation in the Sino-Tibetan language family. Shang Chinese, the Sino-
Tibetan language with earliest written records as the oracle bone inscriptions, did not have compulsory sortal
Verbal numeral classifiers in languages of Eastern Eurasia 23

numeral classifiers while some ‘proto-classifiers’, such as 人‘person’, and some mensural classifiers (Wang
1994,77–8, 171–2), including 玉‘a pair of jade’,朋‘a pair of shell’, and 丙‘pair of horses’(see also Takashima
1984-1985, 231–3) are identifiable in the language. Consequently, two possible sources emerge for classifier
genesis: Hmong-Mien and Tai-Kadai. Classifiers are universal in languages of both language families the
languages of which contain both VNCs and NNCs. Classifiers are postulated for Proto-Tai-Kadai (DeLancey
1986). Moreover, in his pioneering comparative work of classifiers, Jones (1970, 11) proposes that Tai languages
are the possible original source of the areal phenomenon. Complex classifier systems, however, are also
universally attested in Hmong-Mien languages, and the language family thus cannot be ruled out as a possible
source of classifier systems without further evidence.
In both language families, VNCs show a wide range of semantic subtypes, but only Tai-Kadai contains
onomatopoeic classifiers that were not identified in the analyzed Hmong-Mien languages. The wider semantic
range may indicate Tai-Kadai primacy in classifier genesis. It may also, however, result from limited descrip-
tions, since onomatopoeic phenomena that require a high level of mastery in a language are often neglected to
various extents in descriptive grammar. The issue of classifier origins becomes further entangled with that of
the origin of complex number systems, since classifier systems are dependent on developed numeral systems.
Hmong-Mien numerals are borrowed from Sino-Tibetan with the native system comprising only ‘two’,‘three’,
and possibly ‘one’(Ratliff2010, 214). While it is possible that borrowing resulted in the replacement of earlier
native numerals, Proto-Hmong-Mien may also have been a language with a simple number system (Ratliff
2010, 214). In Tai-Kadai languages, Chinese numerals have been borrowed in the Kam-Tai branch whereas Hlai
and Kra maintain the (more) original numerals. It thus seems that proto-Tai-Kadai possessed a more evolved
number system fertile for the development of a classifier system.
While VNCs likely originate from Tai-Kadai, Chinese has played a major role in spreading VNCs in
languages in Eastern Eurasia in later times. Starting with the Japonic languages, the native VNC 多妣
taNpyi ‘time’has fallen into disuse save in some fixed expressions, such as hito-tabi ‘once’. All productive
VNCs originate as Chinese loans, such as Standard Japanese 回-kai ‘time’,度-do ‘time’, and 歩-ho ‘step’,
corresponding to 回huí, 度dù, and 步bù in Modern Mandarin, respectively. Likewise, the Vietnamese generic
lần‘times’traces its origins to the Chinese lún 轮‘round, turn’(Alves 2007, 223) and the Korean generic 번-bŏn
‘time’originates from the Chinese fān番‘turn, time’.
Along the northern and north-western contact zone of Sinitic and Altaic-type languages, particularly the
ones spoken in the Qinghai and Gansu areas, some non-Sinitic languages have borrowed their VNCs from
Chinese (further discussion on the contact aspects of classifiers in Yurayong et al. 2024). Even Sinitic varieties
with generally simple classifier systems, for instance, Wutun (Sino-Tibetan, divergent) with tang ‘time’cf.
Mandarin Chinese tàng 趟‘CLF.trips’, have retained VNCs, as discussed earlier in this study. As for the Altaic-
type languages, classifier systems are generally not common, but the counted noun ‘time(s)’could have existed
among the earliest counting devices. While the function of the limited classifiers across Altaic-type languages is
primarily dedicated to counting entities or general verbal events ‘time(s)’, classifiers for counting more
versatile types of verbal events are also observed in a limited fashion, particularly in those languages spoken
in the Gansu-Qinghai areas. Salar with limited NNCs may have taken steps toward the emergence of VNCs
(43a–43b) and Mangghuer has undoubtedly done so (44). Notice that while Salar uses native Turkic lexical
items, the Mang