1. Introduction

2The counting systems and systems of numerals found in Mande languages are rather heterogeneous, and some of these systems display unique or, at least, typologically rare features. In this paper, we analyse these systems from a typological viewpoint, taking into account ideas formulated in fundamental works which have treated numerals in the languages of the world from universalist, theoretical and typological points of view (Greenberg 1978; Hurford 1975). We begin our analysis by presenting a conceptual framework and terminology for the study of counting systems. In the main part, we investigate eight numeral systems (those of old Bamana, modern Bamana, Boko, Dzuungoo, Mwan, Dan-Gwɛɛtaa, San-Maka, and Soninke) that represent different types of identifiable systems. Next, the data for these and some other Mande languages will be represented in a table, followed by comments on the most interesting features of individual languages. The paper ends with an analysis of general trends and particularities of numeral systems in the Mande language family.

3The main goal of this paper is to represent the Mande language data in a more or less standard format, so that these data might be easily comparable, and to provide a primary analysis. This paper is therefore data-oriented rather than theory-oriented.

4The data represented in this paper are partly drawn from the field materials of the authors (especially for Bamana, Dan-Gwɛɛtaa, Mwan, San-Maka); otherwise, the data are provided with references for their sources.

1.1. Restricted systems and body part counting.

5Typologically, all numeral systems in the languages of the world are either restricted with no arithmetic base or non-restricted.

7We assume that the emergence of numeral systems which have an arithmetic base is a relatively recent phenomenon. According to the World Atlas of Language Structures (Comrie 2005), there still exist “restricted” systems, which use three or four numerals. Larger numbers here are expressed by combinations of these few elements. Within the framework of such a system, the expression of numbers larger than 20 is extremely cumbersome, and as a result, there are limits of the application of exact counting (traces of words with meanings like “the final number” are attested in many languages whose systems are no longer restricted). Such systems are typical of small closed communities, which, until recently, were unconcerned by commercial relations, had no monetary system and conducted a traditional way of life based on hunting and gathering.

8As a rule, these restricted systems coexist with extended body-part counting. In the course of such counting, parts of the body are named in an established order, and the name of a body part is used as the designation of a certain set (“fist” > five). Primarily, fingers and finger phalanges are used for counting.

10Certain terms for numbers in Mande languages can be traced back to body parts: hand, foot, mouth, head, human being, a fact that can be explained in terms of original body-part counting. At the same time, there are terms for higher numerals which go back to the names of certain sets: string of cowries, basket of cola nuts.

11Systems with an arithmetic base (all modern Mande systems belong to this type) are arranged similarly. The structure of a compound numeral in these systems can be described by Bernard Comrie's formula (Comrie n.d.): for a b-based numeral system, (n x b)+m (where m < b). Arithmetic operations are either implied (with the components combined through parataxis) or marked by auxiliary words (connectors).

1.2. Base of the counting system

12Following (Comrie 1997; Comrie 1999), “by the “base” of a numeral system we mean the value n such that numeral expressions are constructed according to the pattern ... xn + y, i.e. some numeral x multiplied by the base plus some other numeral”.

14The multiplicand of the lower order, i.e., the radix (Greenberg's “fundamental base”), defines the type of the numerical system. Different bases can be combined within a single numerative system, rudiments of older systems being preserved. The following radixes are registered in natural languages (this list is certainly not exhaustive):

Radix		Type of the system
4		quaternary
5		quinary
10		decimal
12a		duodecimal
15		pentadecimal (or quindecimal)
20		vigesimal
60		sexagesimal
80		octodecimal
a. “The question is how could a twelve-digit system ever appear? Examples of this count: the division of the year in 12 months, the count of time (12 hours a day), the division of the hour in 60 minutes, and 3600 seconds, the division of the circle in 360 degrees. This system was accepted in Europe, and then spread all over the world. Its origin is usually associated with Babylon and is determined by observations of the movement of the Moon <...> But how the achievements of the Ancient East could reach the remote mountains in the depths of Africa is unclear <...> Therefore, one should search the origin of this system in the hand count, and not fingers, but 12 phalanges” (Olderogge 1982: 28).

15In most instances, base (multiplicands), as well as augends and minuends, are simple (non-derived and non-compound) forms, although some exceptions to this rule can be found. For instance, in Boko there are augends ‘15’ and ‘25’ which are compound words.

16Dmitry Olderogge singled out several counting systems in Africa which are differentiated by the number of “counting thresholds” (i. e., lower multiplicands and augends) they display:

• A four-stage system: thresholds 5, 10, 15, 20. Numbers from 1 through 5 are simple; numbers from 6 through 9 follow the quinary model; numbers from 11 through 14 are derived from 10, and those from 16 through 19 are based on 15. The subsequent numerals are based on 20.
• A three-stage system: thresholds 5, 10, 20.
• A two-stage system: the thresholds are 10 and 20; no traces of the quinary model.
• Complex vigesimal systems “existing among the communities who used to count cowries”. The counting is decimal or/and vigesimal; however, there are special words for 400, 600, etc. (Olderogge 1984: 3–4).

1.3. Arithmetic in counting systems.

17Bernard Comrie (2005) mentions two operations of arithmetic implicitly used in counting systems: addition and multiplication. Quite a few systems also use subtraction, e.g. Latin duo de viginti ‘eighteen’ (2 from 20).‘’

18Division in the numeral systems of natural languages is extremely rare. So, in (Comrie 1997) only Welsh is mentioned in this relation, where hanner cant ‘50’ is ‘half hundred’. According to Greenberg (1990: 280), if the numeral system of a language applies division, «‘half’ is the “unmarked fraction” and is almost always a simple lexical item, often derived from ‘to split’, or 'to break' or the like» (Greenberg adduces Orya (Indo-Aryan) as an exception, where ‘quarter’ is also used in the formation of compound numerals).

1.4. Classifiers or counting words.

19Some languages use classifiers, more or less grammaticalized elements located between the numeral and the noun designating the object counted, along the lines of loaves in two loaves of bread (rather than two breads — although English loaf is not a true classifier word but a noun, in view of the lack of grammaticalization). The term “classifiers” is usually applied to languages of the Far East and Southeastern Asia and to some Amerindian languages, where the elements in question have reached an advanced stage of grammaticalization. Where the level of grammaticalization is more moderate, the terms “counting words” and “measure words” are usually preferred; this is the case in certain Mande languages. We provide information about such words whenever it is available. It is reasonable to think that they are in fact much more widespread, but often passed over in silence by the authors of grammatical descriptions.

1.5. Specialized counting systems.

20Two (or more) different systems may coexist simultaneously in a language, so that the same number can be coded in more than one way (twenty-four and two dozen). Typically, the subsidiary system is used for counting special objects (eggs, notches, stitches in knitting, tube diameter and the like).

23Less clear is the Bamana word for ‘5 francs’, dɔ́rɔ̌mɛ́ (which is much more frequently used than its synonym dálasi). It is generally believed (cf., for example, (Lagarde 1988)) that dɔ́rɔ̌mɛ́ goes back to ‘dirham’, a currency used in Arabic countries (whose name in turn goes back to the Greek drachma). However, the dirham was always a coin of relatively small denomination; its weight was usually close to 3 g of silver. The Spanish dollar in the 18th and 19th centuries ranged between 24 and 30 g of silver. Therefore the dirham could not have been its monetary equivalent. One plausible account then is that the word dɔ́rɔ̌mɛ́, at some point, lost its connection with the dirham and became a default name for a coin, thus becoming a synonym of dálasi.

24In due course this system was sometimes extended, at least in Guro and Mwan (but not in Bamana or Jula), to other currencies. In these languages, not only CFA francs are counted using the “multiply by five” rule, but also dollars, euros, etc.

25In old Bamana, special terms were used when counting cowrie shells (for more detail see the end of section 2.1).

1.6. Internal and external syntax of numerals.

26By “internal syntax of numerals” we mean the manner in which the elements of compound numerals are combined, and by “external syntax”, we mean the way numerals interact with the nouns they determine.

27Unfortunately, we do not have sufficient data on the external syntax of numerals in every language dealt with in this paper. However, we think that there is still some value in providing information on just a limited sub-sample of languages (especially those displaying some unusual features with respect to the common Mande background).

1.7. Etymologies

28Reconstruction of numeral proto-forms is not among the primary tasks of the present paper. However, some potential etymologies of particular forms will be proposed, whenever they are appropriate and useful for understanding the organization of a numeral system. Quite recently, a reconstruction of Mande numerals, against the background of the other Niger-Congo languages, was the subject of a special study by Konstantin Pozdniakov (2018). It seems appropriate to present here Pozdniakov's Proto-Mande etymologies to serve as a point of reference for our “local” etymologies. In some cases, we use brackets to add our own suggestions to the reconstructions (a dot under a letter indicates that the vowel belongs to the –ATR series).

29‘1’ Two roots are candidates for the Proto-Mande level, *do (*dọ́ ) and *kelen, of these *do seems to be better supported.

30‘2’ *pila or *fila (*pìɗà).

• ‘3’ *sakpa/*sagba/*sawa.
• ‘4’ The Proto-Mande form*náání/ nɑ̃ɑ̃i is preserved in the Proto-Western Mande, while *sḭ̀ḭ̀yá̰ is a Proto-Southeastern Mande innovation.
• ‘5’ *dúuru/*sɔ́ɔ́ru (*ɗọ́ọɗụ́ ).
• ‘6’ = ‘5+1’.
• ‘7’ = ‘5+2’.
• ‘8’ unclear (probably ‘5+3’, although other etymologies may also be possible).
• ‘9’ = ‘5+4’ or ‘10-1’.
• ‘10’ *pu/*fu (*pú ); *tan is a Proto-Western Mande innovation stemming from the Niger-Congo root for ‘5’.
• ‘20’ < ‘human being’.
• ‘100’ *kɛmɛ (may be a recent itinerant word) or ‘20x5’.
• ‘1000’ *wulu, *waga (may be recent itinerant words).

1.8. Scope of the study

31We have left certain topics outside our study: in particular, the range of problems concerning ordinal numerals, distributive numerals, and the adverbial use of numerals. We have also left aside some idiosyncratic features of individual numeral systems which have no bearing on our generalizations.

1.9. Mande family

32For the convenience of readers, we reproduce here the genealogical tree of the Mande family, in a form of a hierarchical list (the inner classification follows (Vydrin 2009), in a simplified way). The language names are supplemented by ISO 639-3 codes (in italics). Names of those languages whose numeral systems are analyzed in detail are given in bold; the names of the other languages represented in section 2 are underlined. If a language is known under different names, the alternative names are given in brackets. For some groups containing numerous closely related varieties with a status intermediate between a language and a dialect (such as the Manding group), the lists of varieties given here may not be exhaustive.

• 1. Western Mande
• 1.1. Manding
• 1.1.1. Bamana (Bambara) bam
• 1.1.2. Jula dyu
• 1.1.3. Maninka man
• 1.1.4. Mandinka mnk
• 1.1.5. Xasonka kao
• 1.2. Mokole
• 1.2.1. Mogofin (Mikhifore) mfg
• 1.2.2. Kakabe kke
• 1.2.3. Koranko knk
• 1.2.4. Lele llc
• 1.3. Vai-Kono
• 1.3.1. Vai vai
• 1.3.2. Kono kno
• 1.4. Jogo-Jeri
• 1.4.1. Jogo
• 1.4.1.1. Jogo-Ton
• 1.4.1.2. Jogo-Ligbi lig
• 1.4.2. Jeri jek
• 1.4.3. Jalkunan (Ble) bxl
• 1.5. Southwestern–Susu
• 1.5.1. Susu–Yalunka
• 1.5.1.1. Susu (Soso) sus
• 1.5.1.2. Yalunka (Jalonke) yal
• 1.5.2. Southwestern Mande
• 1.5.2.1. Kpelle
• 1.5.2.1.1. Guinean Kpelle gkp
• 1.5.2.1.2. Liberian Kpelle xpe
• 1.5.2.1.3. Kono knu
• 1.5.2.2. Looma
• 1.5.2.2.1. Liberian Looma lom
• 1.5.2.2.2. Guinean Looma (Toma) tod
• 1.5.2.3. Bandi bza
• 1.5.2.4. Zialo zil
• 1.5.2.5. Mende men
• 1.5.2.6. Loko lok
• 1.6. Soninke–Bozo
• 1.6.1. Soninke snk
• 1.6.2. Bozo
• 1.6.2.1. Bozo-Tigemaxo boz
• 1.6.2.2. Kelengaxo bzx
• 1.6.2.3. Jenaama bze
• 1.6.2.4. Tiemacewe boo
• 1.7. Bobo
• 1.7.1. Southern Bobo bwq
• 1.7.2. Konabere bbo
• 1.8. Samogo
• 1.8.1. Duun dux
• 1.8.2. Dzuun dnn
• 1.8.3. Kpan
• 1.8.4. Banka bxw
• 1.8.5. Seenku (Sembla) sos
• 1.8.6. Jowulu jow
• 2. South-Eastern Mande
• 2.1. Southern Mande
• 2.1.1. Beng nhb
• 2.1.2. Ngen gnj
• 2.1.3. Gbin
• 2.1.4. Gban (Gagu) ggu
• 2.1.5. Wan wan
• 2.1.6. Guro (Kweni) goa
• 2.1.7. Yaure yre
• 2.1.8. Mwan (Mona) moa
• 2.1.9. Dan (Yacouba, Gio) dnj
• 2.1.9.1. Eastern Dan (Yacouba)
• 2.1.9.2. Western Dan (Yacouba)
• 2.1.9.3. Kla-Dan lda
• 2.1.10. Mano mev
• 2.1.11. Tura (Wen) neb
• 2.1.12. Goo
• 2.2. Eastern Mande
• 2.2.1. San
• 2.2.1.1. Southern San sbd
• 2.2.1.2. San-Matya stj
• 2.2.1.3. San-Maya sym
• 2.2.2. Bisa bib
• 2.2.3. Boko-Busa
• 2.2.3.1. Boko bqc
• 2.2.3.2. Bokobaru bus
• 2.2.3.3. Busa bqp
• 2.2.3.4. Illo-Busa
• 2.2.4. Shanga-Kyenga
• 2.2.4.1. Shanga sho
• 2.2.4.2. Kyenga tye

1.10. Orthographic conventions

33All Mande languages are tonal. In our paper, tonal marking follows the IPA system. This system is illustrated in the Table 1, with the example of the vowel a.

Table 1. Tonal marking

Type of system → Tone ↓	2-level system	3-level system	4-level system	5-level system
extra-high			a̋	a̋
high	á	á	á	á
mid		ā		ā
low	à	à	à	à
extra-low			ȁ	ȁ

34If the tonal notation in a consulted source is different, it is converted to the IPA-based model here. Tonal notation in Bamana follows the rules formulated in (Konta & Vydrin 2014).

2. Sample studies: eight numeral systems representing different models

2.1. The modern and old Bamana systems

36In the old Bamana system, the numerals of the first ten are the same as in the modern system. The same is mainly true for the second ten.

Table 2. Bamana numerals, first and second ten

1	kélen
2	fìla
3	sàba
4	náani
5	dúuru
6	wɔ́ɔrɔ
7	wólonfìla, wólonfùla, wólonwùla
8	ségin, séegin
9	kɔ̀nɔntɔ, kɔ̀nɔntɔn
10	tán
11	tân ní kélen   (10+1)
12	tân ní fìla  (10+2)
etc.

37The divergences begin at the end of the second ten, where subtractive numerals emerge. Table 2 presents various forms in the old and modern systems. In the left column, the numerical values are indicated; for the compound numerals, the syntax of the form is clarified in brackets. For numerical values which can be expressed in the old system in two ways (or more), the alternative forms are separated by a tilde (~).

Table 3. The old and modern Bamana systems

	Old system	Modern systema
19	Mùgan kélen ká jɛ̀ ~ mùgan kélen ɲà (20-1) ~ tan ni kɔ̀nɔntɔn (10+9)	tân ni kɔ̀nɔntɔn (10+9)
20	mùgan	mùgan
21	mùgan ni kélen (20+1)	mùgân ni kélen (20+1)
29	mùgan ni tàn kélen ká jɛ̀ ~ mùgan ní tàn kélen ɲà (20+10-1)	mùgân ni kɔ̀nɔntɔn (20+9)
30	mùgan ni tán (20+10)	bî sàba (10x3)
31	mùgan ni tán ni kélen (20+10+1)	bî sàba ni kélen (10x3+1)
38	dɛ̀bɛ fìla ká jɛ̀ ~ dɛ̀bɛ fìla ɲà (40-2)	bî sàba ni séegin (10x3+8)
40	dɛ̀bɛ	bî náani (10x4)
50	dɛ̀bɛ ni tán (40+10)	bî dúuru (10x5)
60	màninkɛ̀mɛ	bî wɔ́ɔrɔ (10x6)
70	màninkɛ̀mɛ ni tán (60+10) ~ kɛ̀mɛ tán ɲà (80-10)~ kɛ̀mɛ tán ká jɛ̀ (80-10)	bî wólonwùla (10x7)
75	màninkɛ̀mɛ ni tán ni dúuru (60+10+5) ~ kɛ̀mɛ dlá ɲà (80-1/2)	bî wólonwùla ni dúuru (10x7+5)
80	kɛ̀mɛ ~ bámanankɛmɛ	bî séegin (10x8)
90	kɛ̀mɛ ni tán (80+10)	bî kɔ̀nɔntɔn (10x9)
100	kɛ̀mɛ ni mùgan (80+20) ~ sìlamɛyakɛmɛ	kɛ̀mɛ
140	kɛ̀mɛ fìla mùgan ká jɛ̀ (80x2-20)	kɛ̀mɛ ni bî náani (100+10x4)
160	kɛ̀mɛ fìla (80x2)	kɛ̀mɛ ni bî wɔ́ɔrɔ (100+10x6)
200	kɛ̀mɛ fìla ni dɛ̀bɛ (80x2+40)	kɛ̀mɛ fìla (100x2)
600	kɛ̀mɛ wólonwùla ni dɛ̀bɛ (80x7+40) ~ màninkaba	kɛ̀mɛ wɔ́ɔrɔ (100x6)
800	bà	kɛ̀mɛ séegin (100x8)
1000	bà kélen ni kɛ̀mɛ fìla ni dɛ̀bɛ (800+80x2+40)~ sìlamɛyaba kélen	bà
a. The numerals for ‘10’ and ‘20’ in modern Bamana, when used in compound numerals which are not divisible by 10 and 20 respectively, appear with a postposed floating low tone; otherwise, they have no floating tone. This suggests that in this context they behave as nouns and therefore they likely carry the tonal article.

38The old Bamana system was characterized by a combination of the following counting systems:

• quinary (within the first ten): in this, both old an new systems coincide. Quinarity manifests itself in the forms 6 wɔ́ɔrɔ and 7 wóronfìla. The latter form can be easily subdivided into two elements, wóron-fìla, where fìla is evidently identical with fìla ‘two’. The other form, wɔ́ɔrɔ ‘6’, is less transparent. Nonethess it could plausibly be subdivided into two elements, wɔ́ɔ-rɔ. In this case, it is highly probable that the second element comes from the reconstructed Proto-Mande numeral *dɔ́ ‘one’. In turn, the first element may be etymologically identical with the element wóron- attested in the numeral for ‘7’ (i.e., ‘6’ *wóron-dɔ > wɔ́ɔrɔ through elision of the foot-internal consonant, contraction of the two-feet structure into one foot with subsequent lenition *-d- > -r- and regressive vocalic assimilation *-oo- > -ɔɔ-). It would be logical to suppose that wóron could be an archaic term for ‘five’;
• heteroradical vigesimal (for tens). Numerals divisible by 20 (40, 60, 80, 100), contrary to expectations, are not formed through multiplication of the numeral for '20', they are represented instead by simple forms (dɛ̀bɛ '40', kɛ̀mɛ '80') or forms derived in idiosyncratic ways: màninkɛ̀mɛ can be interpreted as 'a Maninka hundred', bámanankɛmɛ is 'a Bamana hundred', and sìlamɛyakɛmɛ is 'a Muslim hundred';
• mixed octogesimal and decimal, for hundreds and thousands (the numerical value bà ‘800’ is equal to 80x10).
• In the old Bamana system, the following arithmetic operations were used:
• addition, marked by the coordinative conjunction ni ‘and/with’ (tân ni kélen ‘11’ = 10+1),
• multiplication, expressed by juxtaposition (kɛ̀mɛ fìla ‘160’ = 80 x 2),
• subtraction (‘19’ = 20 – 1, mùgan kélen ɲà, lit. ‘twenty, make one lacking’, or mùgan kélen ká jɛ̀, lit. ‘twenty, let one be lacking’); the subtractable numerals are kélen 1, fìla 2, tán 10 and mùgan 20 (and also ‘half’ = 5). Subtractive numerals are employed in an imperative construction with the verb ɲà ‘lack; take away’ or a subjunctive construction with the verb jɛ̀ ‘miss, be absent; deprive’ (supposedly, preference for one verb or the other depends on the region);
• elements of division: when forming subtractive numerals, ‘5’ can be designated as a ‘half of ten’: dɛ̀bɛ dlá ɲà ‘35’ (‘40 – a half’, lit. ‘forty, make a half lacking’). The word for ‘half’ is normally tíla, but in this context it appears with a voiced initial consonant (although, rarely, it may also be found with a voiceless consonant).

39The term kɛ̀mɛ appears in the numerals for 60, 80 and 100 which are etymologized as follows: màninkɛ̀mɛ ‘60’ lit. ‘a Maninka hundred’; bámanankɛmɛ ‘80’ lit. ‘a Bamana hundred’ (the simple stem, kɛ̀mɛ, can also be used for 80); sìlamɛyakɛmɛ ‘100’ lit. ‘a Muslim hundred’. 100 can also be designated by means of the “Bamana hundred”: kɛ̀mɛ ni mùgan (80 + 20). When kɛ̀mɛ is used in compound numerals, it designates, by default, ‘80’.

40As for the etymology of the numerals for tens, Maurice Delafosse proposed that mùgan ‘20’ could be traced back to the word mɔ̀gɔ ‘human being’ (for the number of a human's fingers and toes, cf. (Delafosse 1955: 520)), and dɛ̀bɛ ‘40’ to dɛ̀bɛn ‘mat’ (for the number of fingers and toes of a couple lying on a mat, cf. (Delafosse 1929: 273–274; Delafosse 1955: 111)). These etymologies may be correct (especially for mɔ̀gɔ, much less clearly for dɛ̀bɛ), all the more so given that irregular form modifications accompany numeral derivations from nouns in many African languages.

41The stem kɛ̀mɛ ‘100/80’ (and various forms which can be traced back to it) is well represented in Mande languages, and there are good reasons to think that it spread through intra-African trade, and primarily the trade in cola nuts. One hypothesis is that it designated a standard cola package (which could vary regionally, leading to different meanings for the “Bambara hundred” and “Maninka hundred” in the old Bamana counting system). Alternatively, kɛ̀mɛ could originally stand for a string of cowrie shells, the predominant pre-colonial currency. Taking into account the role of Soninke which was the dominant language of Ghana (the very first among the medieval empires of Western Sudan), the language of the jùla traders and the source of numerous loans in the sphere of cultural vocabulary in many languages of the region, it can be supposed that the ultimate origin of this stem could be Soninke, where it appears in the forms kàmé (in Western dialects) and kèmé (in Eastern dialects).

42The etymology of the word bà (‘800’ in the old system, ‘1000’ in the modern system) is not clear. It is noteworthy that in Bamana, there is a special word for counting money, wáa ‘1000’ (while in Guinean Maninka and in interethnic Jula, wáa is the general word for ‘1000’) which goes back to the noun wága ~ wáa ‘basket for cola nuts’ (storing 1000 nuts). This stem also appears in many other Mande languages.

44There are several counting wods in Bamana: cɛ̀ ‘man’, mùso ‘woman’, dén ‘child’, ‘fruit’, kílo ‘kilogram’, ɲɛ́ ‘content’, kìsɛ ‘grain’ (used for counting small objects), kùru ‘ball, lump’ (used for counting round objects), kàla ‘stem’ (used to count long objects). They are not yet fully grammaticalized as classifiers and belong to the part of speech of noun.

45External syntax of Bamana numerals. Numerals follow the nouns they determine, and by default, neither noun, nor numeral has plural marker –u (-w in the current orthography). This marker can nevertheless be added to the numeral (1) or to the noun (2) (seemingly without any distinction) to express definiteness.

(1)

Npògotigi`

ka

kán

kà

dí

jɔ́nnì

mà,

cɛ̀

sàba-w

ni

ɲɔ́gɔn

cɛ́?

girl-art

qual.aff

equal

inf

give

who

adr

man

three-pl

and

recp

between

‘From among the three men, to whom should the girl be given?’

46 

(2)

<…>

ù

màa-w

fìla

bɛ́ɛ

jɛ̀-ra

kà

Lànseni

fò,

à

ye

ù

lámìnɛ.

3pl

human-pl

two

all

gather-pfv.intr

inf

Lanseni

greet

3sg

pfv.tr

3pl

answer

‘... both of them gathered to greet Lanseni, and he responded to their greeting’.

2.2. Boko numeral system

48Boko is one of the rare languages in the Mande family where the quinary system continues beyond the first ten since the numerals for ‘15’ and ‘25’ serve as augends (for 16, 18, 26, 27) and minuends (for 14, 24). Otherwise, vigesimality is predominant in Boko.

49Basic numerals. There are non-compound terms (at least, in synchrony) for the numerals from 1 to 5, and also ‘10’ kwȉ; ‘15’ gɛ̰̏ó; ‘20’ bȁó; ‘200’ ‘ɔ̏áȁ.

50In the forms for 6, 7, 8, there is an element sOO- which can be traced back to sɔ̋ó ‘5’. In 6 sɔ́ólɔ́, the element –lɔ́ derives, most probably, from dó ‘1’. In the numerals for 7 and 8, the final elements are identical with the numerals for 2 and 3. Therefore, the numerals of the first ten retain elements of the ancient quinary counting system. In the form kɛ̃́ókwȉ ‘9’, the element kwȉ ‘10’ is easily detectable. The initial element can be subdivided into -ó- which most likely goes back to dó ‘one’, and kɛ̰́ which can be linked to the verb kɛ̃̋ ‘tear, break’. If this etymology is correct, kɛ̃́ókwȉ is a subtractive numeral (10 – 1), lit. ‘break one (from) ten’.

Table 4. Boko numerals

1	dó
2	plá
3	’ȁȁó̰
4	si̋i̋ɔ̰́
5	sɔ̋ó
6	sɔ́ólɔ́ (5+1)
7	sóplá (5+2)
8	swȁȁṍ (5+3)
9	kɛ̰́ókwȉ (–1 10)
10	kwí
11	kúɛ̏dó (10+1)
12	kúɛ̏plá (10+2)
13	kúɛ̏ ’ȁȁɔ̰́ (10+3)
14	gɛ̰̏ó mɛ̏ŋ̏ dó sȁi̋ (15-1)
15	gɛ̰̏ó ((*foot)x1)
16	gɛ̰̏ó ŋ̋ mɛ̏ŋ̏ dó-ȍ (15+1)
17	gɛ̰̏ó ŋ̋ plá ȍ (15+2)
18	bȁó plá sȁi̋ ((20x1)–2)
19	bȁó dó sȁi̋ ((20x1)–1)
20	bȁó (20x1)
21	bȁó ŋ̋ plá ȍ ((20x1)+1)
23	bȁó ŋ̋ ’ȁȁɔ̰́ ȍ ((20x1)+3)
24	bȁásɔ̋ó dó sȁi̋ ((20x1)+5)–1)
25	bȁásɔ̋ó ((20x1)+5)
26	bȁásɔ̋ó ŋ̋ mɛ̏ŋ̏ dó ȍ (((20x1)+5)+1 thing)
27	bȁásɔ̋ó ŋ̋ plá ȍ (((20x1)+5)+2)
28	bȁá kwȉ plá sȁi̋ (20+10–2)
30	bȁá kwȉ (20+10)
35	sómȁ b̏lá ~ sóȁ b̏lá ~ sɔ̰́ȁ b̏lá (–5 40)
36	bȁá kwȉ ŋ̋ sɔ́ólɔ́ ȍ ((20+10)+(5+1))
37	bȁá kwȉ ŋ̋ sóplá ȍ ((20+10)+(5+2))
38	b̏lá plá sȁi̋ ((20x2) – 2)
39	b̏lá dó sȁi̋ ((20x2) – 1)
40	b̏lá (20x2)
45	b̏lá sɔ̋ó ((20x2)+5)
50	b̏lá kwȉ (40+10)
55	sómȁ bȁ ȁȁɔ̰́ (–5 (20x3))
56	b̏lá kwȉ ŋ̋ sɔ́ólɔ́ ȍ ((40+10)+(5+1))
57	b̏lá kwȉ ŋ̋ sóplá ȍ ((40+10)+(5+2))
58	bȁ ȁȁṍ plá sȁi̋ ((20x3)–2)
60	bȁ ȁȁṍ (20x3)
65	bȁ ȁȁṍ sɔ̋ó ((20x3)+5)
70	bȁ ȁȁṍ kwȉ ((20x3)+10)
80	bȁ si̋i̋ɔ̰́ (20x4)
85	bȁ si̋i̋ɔ̰́ sɔ̋ó ((20x4)+5)
90	bȁ si̋i̋ɔ̰́ kwȉ ((20x4)+10)
100	bȁ sɔ̋ó (20x5)
105	bȁ sɔ̋ó sɔ̋ó ((20x5)+5)
120	bȁ sőóló (20x6)
140	bȁ sóplá (20x7)
160	bȁ swȁȁṍ (20x8)
180	bȁ kɛ́ó̰kwȉ (20x9)
200	‘ɔ̏áȁ dó (200x1)
300	‘ɔ̏áȁ dó ŋ̋ bȁ sɔ̋ó ((200x1) + (20x5))
400	‘ɔ̏áȁ plá (200x2)
500	‘ɔ̏áȁ plá ŋ̋ bȁ sɔ̋ó ((200x1) + (20x5))
600	‘ɔ̏áȁ 'ȁȁṍ (200x3)
700	‘ɔ̏áȁ 'ȁȁṍ ŋ̋ bȁ sɔ̋ó ((200x2) + (20x5))
800	‘ɔ̏áȁ si̋i̋ó (200x4)
900	‘ɔ̏áȁ si̋i̋ó ŋ̋ bȁ sɔ̋ó ((200x4) + (20x5))
501	‘ɔ̏áȁ plá ŋ́ bȁ sɔ̋ó ŋ́ mɛ̏ŋ̏ dó ȍ ((200x2) + (20x5) + 1 thing)
750	‘ɔ̏áȁ 'ȁȁṍ ŋ́ bȁ sóplá kwȉ ȍ ((200x3) + (20x7) + 10)
188	‘ɔ̏áȁ dó kúɛ̏plá sȁi̋ ((200x1) – 12)
1000	‘ɔ̏áȁ sɔ̋ó (200x5)
1400	‘ɔ̏áȁ sóplá (200x7)
2944	‘ɔ̏áȁ gɛ̰̏ó dó sȁi̋ ŋ̋ bȁ sóplá ŋ̋ si̋i̋ɔ̰́ ȍ (200 x (15-1) + (20x7) + 4)
50,000	’ɔ̏áȁ sɔ̋sɔ̋ó lɛ́ɛ̏ b̏lá kwí ((200x5redupl. part) (40+10))

51The word for ‘10’ kwȉ can be (very tentatively) linked to the stem for ‘hand’. In modern Boko, ‘hand’ is ’ɔ́, and a preliminary reconstruction for this root at the Proto-Eastern level is *gɔ̌n (cf. Proto-Southern Mande *kɔ̀ ‘hand’ and forms for ‘10’ in Dan, Goo, Tura).

52There are good reasons to think that the stem gɛ̰̏ó ‘15’ may contain an element originally meaning ‘foot’: although the word for ‘foot/leg’ in Boko is gbá, a tentative pre-reconstruction for Proto-Eastern Mande is *gɔ̰̀ or *gwɛ̰̀ ‘foot’, cf. Henning Schreiber's (2008: 323) Proto-Eastern reconstruction *gwaN ‘foot’, cf. also Southern San gòã (Anonym 2003), Lebir gǎm |gànɩ́| (Vanhoudt 1999). The final element -ó in ‘15’ may go back to dó ‘1’, i.e. the term originally meant ‘(two hands and) one foot’, this model for ‘15’ is not exceptional in African languages (Pozdniakov 2018: 120).

53The basic term of the vigesimal system bȁ- (presumably, from bà ‘rope, string’, i.e. ‘string of cowrie shells’) appears in the form bȁó ‘20’, where the final element -ó may go back to dó ‘one’; in bȁásɔ̋ó ‘25’ (cf. sɔ̋ó ‘5’, while the etymology of the linker -á- is unclear); bȁákwȉ ‘30’ (kwȉ < kwí ‘10’; the same linker -á- as in ‘25’); 40 b̏lá (a contractive form of *bȁ ‘20’ + plá ‘2’), and in other numerals divisible by 20.

54An etymology for ‘ɔ̏áȁ ‘200’ may be suggested if we compare this form with those seen in the closely related languages Bokobaru, wàà, and Busa, wàà. It cannot be confirmed that these forms are related to the forms of the type WAA, WAGA for ‘1000’ which are wide spread in Mande languages (usually with high tones), but it can be accepted as a potential hypothesis.

55For counting from 20 through 2000, the vigesimal system is applied quite consisently: 900 ‘ɔ̏áȁ si̋i̋ɔ̰́ ŋ̋ bȁ sɔ̋ó (200х4 + 20х5).

56Arithmetic operations, linkers and inner morphosyntax of compound numerals in Boko.

57In addition, there are certain limitations on the units which can serve as addends. In particular, in the first, second and third tens, ‘4’ si̋i̋ó does not appear in this function, the relevant meanings being expressed via subtraction; ‘6’ and ‘7’ are not used to form ‘16’ and ‘17’, ‘26’ and ‘27’, the formulae “15+1”, “15+2”, “(20+5)+1”, “(20+5)+2” being preferred. By contrast, these units do appear in the compound numerals above 30.

58There are different means used to express addition in the formation of numerals. Juxtaposition is used for the lower compound numerals ‘6’, ‘7’, ‘8’ (the model “5+X”) and higher ones divisible by 5: ‘25’, ‘45’ (the model “20xN+5); ‘30’, ‘50’, ‘70’, ‘90’ (the model “20xN+10”). Juxtaposition is accompanied by fusion: the stems for the augends ‘5’ (in ‘6’, ‘7’, ‘8’) and ‘10’ (in ‘11’, ‘12’, ‘13’) are modified: ‘5’ sɔ́ó → só- or sw-; ‘10’ kwí → kúɛ̏. In ‘6’, the addend stem is also modified: ‘1’ dó → -lɔ́. For higher numerals, there is no fusion, however, the addend ‘10’ changes its tone: kwí → kwȉ.

59For numerals higher than 15 and not divisible by 5 (and not formed by subtraction, see below), addition is expressed by the comitative frame construction ŋ̋ … ȍ ‘with’. To add 1, one must use the numerative word mɛ̏ŋ̏: bȁ sɔ̋ó ŋ̋ mɛ̏ŋ̏ dó ȍ (20x5+1 unit) ‘101’.

61There are three formal means for producing subtractive numerals.

62The most prevalent is the following: the subtrahend follows the minuend and is followed by the postposition sȁi̋ ‘without’ (the subtrahend dó ‘1’ can be preceded by the numerative word mɛ̏ŋ̏):

(3)	gɛ̰̏ó	(mɛ̏ŋ̏)	dó	sȁi̋
	15	unit	1	without
‘14’
(4)	‘ɔ̏áȁ	dó	kúɛ̏-plá	sȁi̋
	200	1	10-2	without
‘188’ = 200x1 – (10 – 2)

63In order to avoid repetition of the postposition sȁi̋ within one compound numeral (when a subtrahend itself contains a subtrahend), the following formula is used:

(5)	ɔ̏áȁ	dó	gɛ̰̏ó	dó	sȁi̋	mɛ̋	kú-ó
	200	1	15	1	without	foc	here-neg
‘186’ = 200x1 – (15–1)

65The third means of formation is applied to only one subtractive numeral, kɛ̰́ó kwȉ ‘9’, lit. ‘cut-one ten’.

66Multiplication is used for higher numerals. Only two basic numerals, bȁ ‘20’ and ’ɔ̏áȁ ‘200’, can function as multiplicands; the multiplication is expressed by juxtaposition (the multiplier follows the multiplicand), e.g. ‘ɔ̏áȁ ‘ȁȁó̰ (200х3) ‘600’. One complex numeral (itself formed through multiplication), ‘ɔ̏áȁ sɔ̋ó (200x5) ‘1000’, can appear as a multiplicand, in which case its second component undergoes partial reduplication, and the multiplier is preceded by the linker lɛ́ɛ̏ ‘part, group’: ’ɔ̏áȁ sɔ̋sɔ̋ó lɛ́ɛ̏ plá (200x5redupl. part 2) ‘2000’; ’ɔ̏áȁ sɔ̋sɔ̋ó lɛ́ɛ̏ ’ɔ̏áȁ sɔ̋ó (200x5redupl. part 200x5) ‘100,000’.

67In Boko, there are three classifiers (numerative words) which may appear when numerals are used as quantifiers: gbɛ̰́ɔ̰́ŋ́, for counting humans; mɛ̏ŋ̏, for other objects; and gɛ̰̏ŋ̏ ‘time’, for actions. The rules of their use and their degree of obligatoriness are not discussed by Jones (1998: 128–129).

(6)	Má	pɔ̋	bɛ̰̋ɛ̰̋	mɛ̏ŋ̏	si̋i̋ɔ̰́	pì-ɔ́	‘ȅ.
	1sg.pfv	thing	alive	object	four	art-pl	see.pfv
‘I saw these four living things’.

2.3. Numerals in Dzuungoo

68Dzuungoo (Samogo group, Western Mande) is spoken in western Burkina Faso. The numerals of this language are represented in Table 5 according to (Solomiac 2014: 152–158).

69In Dzuungoo, elements of a quinary system are practically absent: the only potential survival is seen in the form ɲɛ̀ɛ̀nún ‘7’, in which the element nún may be etymologically connected to nùn ‘5’. This hypothesis, however, is contradicted by its final position in the form (everywhere else in Dzuungoo, just as in all other Mande languages, the larger addend precedes the smaller one, without any exception).

Table 5. Dzuungoo numerals

1	sōō, sɔ̄n
2	fíí
3	zhììgī
4	nààlén
5	nùn
6	tsùnmɛ̄n
7	ɲɛ̀ɛ̀nún
8	ɲáálòn
9	kyèèrón
10	tsyéù
11	téēŋsɔ̄n (10+1)
12	téēŋfíí (10+2)
13	téēŋzhíígī (10+3)
14	téeŋnáàlén (10+4)
15	téēŋnún (10+5)
16	téēŋtsúnmɛ̄n (10+6)
17	téēŋɲɛ́ɛ̀nún (10+7)
18	téēŋáálòn (10+8)
19	téēŋkyéèrón (10+9)
20	mòò
21	móòsɔ̄n (20+1)
22	móòfíí (20+2)
23	móòzhììgī (20+3)
28	móòŋáálòn (20+8)
29	móòkyèèrón (20+9)
30	mòòtsyéù (20+10)
31	mòòtséù kó (dzín) sóó ((20+10)+1)
32	mòòtséù kó (dzín) fíí ((20+10)+2)
40	dzyèè
50	dzyèètsyéù (40+10)
60	mùéy
70	mùéy kó tséù (60+10)
80	cɛ̀ɛ̀n (as determiner), cɛ̀ɛ̀nsōō (in counting)
90	cɛ̀ɛ̀n kó tsyéù (80+10)
100	càànmóò (80+20)
110	càànmóò kó tsyéù ((80+20)+10)
120	cààndzyéè (80+40)
130	cààndzyéè kó tsyéù ((80+40)+10)
140	cɛ̀ɛ̀nfíímòòyáá ((80x2) – 20)
160	cɛ̀ɛ̀nfíí (80x2)
180	cɛ̀ɛ̀nfíí kó mòò
200	cànfóōdzyéè (80x2 + 40)
400	cɛ̀ɛ̀nún (80x5)
800	gúrúsɔ̄n (800x1)
1000	gúrúsɔ̄n kó cànfóōdzyéè (800x1 + ((80x2) + 40)), or wāākērēn, or gbáásōō
1200	ɲɛ̄yn shē zhììgī (cowrie string 3)

71The numerals of the second ten contain the stem téēŋ-, which might be connected with the numeral 10 tsyéù.

72The counting system from 20 through 80 is based on the heteroradical vigesimal principle, and above 80 it is taken over by an octogesimal model. The element cààn ~ càn, which regularly appears in compound numerals (100 (80+20), 110 (80+20+10), 120 (80+40), 130 (80+40+10), 200 (80х2+40) etc.), is formally distinct from the numeral cɛ̀ɛ̀n ‘80’, but they are, without any doubt, etymologically related. In one numeral, cɛ̀ɛ̀nfíí ‘160’ (80x2), the original form cɛ̀ɛ̀n ‘80’ appears unmodified.

73When counting thousands (and higher), Dzuun people cross over to a decimal system and use only the Jula loan wāākērēn (< wáa kélen), or the Dzuun calque gbáásōō. The expression ɲɛ̄yn shē ‘cowrie string’ is used as a base with the numeral value ‘400’ only in the compound numeral ‘1200’.

74Arithmetic operations, linkers and inner morphosyntax of compound numerals in Dzuungoo

75Addition. There are three formal models of addition:

76a) through fusion of the summands, with one or both components undergoing modification in the great majority of cases. This model is applied to the numerals:

• from 11 through 19, which are accompanied by both segmental and suprasegmental modification of the stem ‘10’ (tsyéù → téēŋ-) and by tonal rising on the first syllable of the component for the units (zhììgī → zhíígī ‘3’, kyèèrón → kyéèrón ‘5’, etc.). The numeral ‘11’ is an exception, because the name of the unit is modified segmentally, and not tonally (sōō → sɔ̄n). Solomiac's hypothesis is that the final element –ŋ of the form téēŋ- may go back to the connector *kó whose high tone may also have been retained as a floating high tone at the end of téēŋ-;
• from 21 through 29, where the smaller addend is not modified, while the name of the larger summand (the augend) is tonally modified (mòò → móò- ‘20’). Solomiac attributes this to the omission of the connector *kó;
• ‘100’ (80+20) and ‘120’ (80+40), where the first summand is segmentally modified (cɛ̀ɛ̀n → cààn- ‘80’), and the second one raises its initial tone. In the numeral cànfóōdzyéè ‘200’ (80x2 + 40), the first addend, although complex, is segmentally modified (cɛ̀ɛ̀nfíí ‘160’ → cànfóō-), while the second one maintains its form.

77b) Through simple juxtaposition of the summands (without any formal modification of the components), for the numerals ‘30’, ‘50’.

78c) With the connector kó ‘and’ between the addends, the addends are not modified in form. This model is used for linking tens and units in numbers higher than 30 (31 = (20+10)+1, 32 = (20+10)+2, etc.; 110 = (80+20)+10, 130 = (80+40)+10), but also for ‘70’ (60+10) and ‘90’ (80+10) where both addends are simple. When the second (smaller) addend is a unit, it can be preceded by the word dzín meaning ‘unit’ (original meaning: ‘child, fruit, grain’). Addition with kó is also used for higher numerals, e.g.:

(7)	cànfóōdzyéè	kó	cɛ̀ɛ̀nfíímòòyáá	kó		sóó
	80x2+40	and	80x2-20	and		1
‘341’ (200+141)
(8)	cɛ̀ɛ̀nnún	kó	cànfóōdzyéè	kó	cɛ̀ɛ̀nfíímòòyáá		kó	sóó
	80x5	and	80x2+40	and	80x2-20		and	1
‘741’ (400+200+141)

79Subtraction in compound numerals is limited in its use. In fact, the only attested case is represented by the numeral for 140 derived from 160 through subtraction (by means of the marker yáá ‘without’): cɛ̀ɛ̀n-fíí-mòò-yáá 140 (80х2-20).

81The study of higher numerals in Dzuungoo is complicated by the fact that practically all speakers are bilingual in Jula, and most often they count in Dzuungoo up to 29 and then switch into Jula (Solomiac 2014: 152).

82Besides the traditional compound numeral gúrúsɔ̄n kó cànfóōdzyéè (800+80x2+40) following the octogesimal model, 1000 can be designated in the modern language by the Jula loan wāākērēn or by the better-integrated form gbáá (sōō) ‘(one) thousand’. Thus, elements of a decimal model have begun to penetrate the numeral system of Dzuungoo.

83External syntax of numerals

84Numerals follow the nouns they determine. In this construction, as a rule, the noun cannot carry any other determiner, and the numeral has none either. More rarely, the numeral can have a plural marker or a definite article, both of which express the same meaning of definiteness.

2.4. Numerals in Mwan

85Mwan (South Mande < Southeastern Mande) is spoken by some 17,000 people in the central part of Côte d’Ivoire. The data were collected by Elena Perekhvalskaya during her fieldwork. See also (Perekhvalskaya 2017: 726–727).

Table 6. Mwan numerals

1	dō
2	plɛ̄
3	yāgā
4	yīzīɛ̄
5	sóó
6	slɔ́ádō (5+1)
7	slɔ́áplɛ̄ (5+2)
8	slɔ́áā (5+3)
9	slɔ́áyīzīɛ̄ (5+4)
10	vū
11	vū tā do (10+1)
12	vū tā plɛ̄ (10+2)
13	vū tā yāgā (10+3)
20	mīā dō (20x1)
22	mīā dō ɓɛ̄ plɛ̄ (20x1 + 2)
30	mīā dō ɓɛ̄ vū (20x1 + 10)
35	mīā dō ɓɛ̄ vū tā sóó (20x1 + 10 + 5)
40	mīā plɛ̄ (20x2)
50	mīā plɛ̄ ɓɛ̄ vū (20x2 + 10)
57	mīā plɛ̄ ɓɛ̄ vū tā slɔ́áplɛ̄ (20x2 + 10 + (5+2))
60	mīā yāgā (20x3)
70	mīā yāgā ɓɛ̄ vū (20x3 + 10)
79	mīā yāgā ɓɛ̄ vū tā slɔ́áyīziɛ̄ (20x3 + 10 + 9 (= 5+4))
100	kɛ̀mɛ̀ dō (100x1)
120	kɛ̀mɛ̀ dō mīā dō (100 + 20)
140	kɛ̀mɛ̀ dō ké mīā plɛ̄ (100x1 + 20x2)
200	kɛ̀mɛ̀ plɛ̄ (100x2)
248	kɛ̀mɛ̀ plɛ̄ mīā plɛ̄ ɓɛ̄ slɔ́ā (100х2 + 20х2+8 (= 5+3))
338	kɛ̀mɛ̀ yāgā ké mīā dō ɓɛ̄ vū tā slɔ́á (100x3 + 20x1 + 10 + 8 (= 5+3))
565	kɛ̀mɛ̀ sóó mīā yāgā ɓɛ̄ sóó (100х5 + 20х3+5)
787	kɛ̀mɛ̀ slɔ́áplɛ̄ mīā yīzīɛ̄ ɓɛ̄ slɔ́áplɛ̄ (100х7 + 20х4+7 (= 5+2))
1000	wáá dō (1000x1)
2400	wáá plɛ̄ kɛ̄mɛ̄ yīzīɛ̄

86The numeral system of Mwan combines vigesimal (in the first hundred) and decimal (above 100) models. The numeral bases are mīā ‘20’, kɛ̀mɛ̀ ‘100’ and wáá ‘1000’. Obviously, the decimal model should be regarded as predominant in Mwan, while the elements of the vigesimal model may have been borrowed from the neighbouring Guro language (cf. Table 10). It should be noted that the Mwan and Guro communities are connected by intense commercial and marital ties; Guro speakers are much more numerous, and many Mwan speak some Guro. However, since the forms of the numerals for 20 in Mwan (mīā) and Guro (yọ̄) are not related, we can speak of borrowing of the model, but not of the forms (structural borrowing, in Haspelmath's (2009) sense).

87The other augends are sóó ‘5’ and vū ‘10’, both of Proto-Mande origin. We can speculate that the Mwan word for 20, mīā, might be etymologically related to the word mɛ̄ɛ̄ ‘human’. Kɛ̀mɛ̀ and wáá have most probably been borrowed from Manding.

88Addition. Several models of addition are used:

• for the numerals ‘6–9’, the model is slɔ́-á-N. The element slɔ́- certainly stands for ‘5’ and provides evidence for the reconstruction of a word-internal consonant in this numeral, *sóló (lost in the independent numeral sóó ‘5’). The second element, -á- goes back, most probably, to the superessive postposition tā. The third element is the second addend, which may be fused (slɔ́ā ‘8’) or not (slɔ́ádō ‘6’, slɔ́áplɛ̄ ‘7’, slɔ́áyīzīɛ̄ ‘9’);
• for numerals above 10, the connector for the names of tens and units is a full-fledged superessive postposition tā;
• for those numerals from 30 onwards divisible by 10 but not divisible by 20, the model is mīā N ɓɛ̄ vū (i.e., 20 N ɓɛ̄ 10, or “20 N + 10”). The connector ɓɛ̄ goes back to the noun ɓɛ̄ ‘grain’, this word is also used in singulative meaning (‘single object out of a mass or pair’: ylɛ̀ ɓɛ̄ dō ‘one eye’, zlɔ̀ ɓɛ̄ ‘a bee’);
• for numerals above 100, ké (a comitative preposition) serves to connect names of hundreds and names of tens.

89Multiplication is expressed through simple juxtaposition of the multiplicand and the multiplier. The multiplicands in Mwan are mīā ‘20’, kɛ̀mɛ̀ ‘100’, wáá ‘1000’.

90No subtraction is used in the formation of compound numerals.

91Money counting has the same features as in numerous other languages of West Africa: when combined with the noun gɔ̀lí ‘money; 5 franc coin’ (or even in its absence, when money is meant), a numeral designates a quantity multiplied by five: gɔ̀lí kɛ̀mɛ̀ dō ‘500 francs’ (lit. ‘100 5-franc coins’); wáá yīzīɛ̄ ‘20,000’ (lit. ‘4 thousands’).

92External syntax. Numerals follow the determined noun which appears without the plural marker (mū): mɛ̄ɛ̄ plɛ̄ ‘two people’.

2.5. Dan-Gwɛɛtaa numerals

93Dan-Gwɛɛtaa is one of the numerous varieties of Dan (Southern Mande) spoken in the west of Côte d’Ivoire. The Dan-Gwɛɛtaa data were collected by Valentin Vydrin during his fieldwork; also see (Vydrin 2017).

Table 7. Dan-Gwɛɛtaa numerals

1	dō
2	plɛ̀ ~ pȅȅdʌ̄
3	yȁȁgā
4	yȉȉsīɤ̄
5	sɔ̋ɔ̋ɗű
6	sɒ̋ɒ̏dō (5+1)
7	sɒ̋ɒ̏plɛ̀ (5+2)
8	sa̋ȁgā (5+3)
9	sűȅsīɤ̄ (5+4)
10	kɒ̏ɒ̏ŋ̏ dō (10x1)
11	kɒ̏ɒ̏ŋ̏ dō ɤ̄ gā dō (10x1 + 1)
12	kɒ̏ɒ̏ŋ̏ dō ɤ̄ gā plɛ̀ (10x1 + 2)
13	kɒ̏ɒ̏ŋ̏ dō ɤ̄ gā yȁȁgā (10x1 + 3)
20	kɒ̏ɒ̏ŋ̏ plɛ̀ (10x2)
21	kɒ̏ɒ̏ŋ̏ plɛ̀ ɤ̄ gā dō (10x2 + 1)
22	kɒ̏ɒ̏ŋ̏ plɛ̀ ɤ̄ gā plɛ̀ ~ pȅȅdʌ̄ (10x2 +2)
30	kɒ̏ɒ̏ŋ̏ yȁȁgā (10x3)
40	kɒ̏ɒ̏ŋ̏ yȉȉsīɤ̄ (10x4)
90	kɒ̏ɒ̏ŋ̏ sűȅsīɤ̄ (10x9)
100	kʌ̄ŋ̄ dō (100x1)
102	kʌ̄ŋ̄ dō wāȁ plɛ̏ (100x1 and 2)
111	kʌ̄ŋ̄ dō wāȁ kɒ̏ɒ̏ŋ̏ dō wāȁ dō (100x1 and 10x1 and 1) ~ kʌ̄ŋ̄ dō ɤ̄ kɔ̏ dō wāȁ dō (100x1 its hand x1 and 1)
200	kʌ̄ŋ̄ plɛ̀ (100x2)
1000	gblɯ̋ dō (1000x1)
2000	gblɯ̋ plɛ̀ (1000x2)
2222	gblɯ̋ plɛ̀ wāȁ kʌ̄ŋ̄ plɛ̀ ɤ̄ kɔ̏ plɛ̀ wāȁ ɤ̄ gā plɛ̀ (1000x2 and 100x2 its hand x2 and its bone x2)
10,000	gblɯ̋ kɒ̏ɒ̏ŋ̏ dō (1000x10x1)
100,000	gblɯ̋ kʌ̄ŋ̄ dō (1000x100x1)
1,000,000	gblɯ̋ ɓɛ̃̄ɛ̃̏ dō

95Addition manifests itself through several models:

• for the numerals 6 through 9, the model is reconstructible as *5 tȁ X, where X = 1 through 4 (see above);
• for numerals above 10 and indivisible by 10, the model is: “Nx10 ɤ̄ gā N”; ɤ̄ gā means literally ‘its bone’ (ɤ̄ is a singular reflexive pronoun, while gā is a polysemous noun whose basic meaning is ‘bone’ or ‘grain’, but also ‘unit’, cf. (Erman 2005));
• for numerals above 100, the previous model (construction with the singular reflexive pronoun) competes with the regular coordinative construction expressed by the coordinative personal pronoun wāȁ (on coordinative pronouns in Eastern Dan, see (Vydrin 2010)); for more detail see (Vydrin 2017: 487–488). In the construction with the reflexive pronoun, the word for ‘10’ is kɔ̏ ‘hand / pair of hands’, rather than kɒ̏ɒ̏ŋ̏ (the latter is certainly derived from kɔ̏, although it is unclear what element it has fused with).

96Multiplication is expressed through simple juxtaposition of the multiplicand and the multiplier. The multiplicands in Dan-Gwɛɛtaa are kɒ̏ɒ̏ŋ̏ ‘10’, kʌ̄ŋ̄ ‘100’, gblɯ̋ ‘1000’, gblɯ̋ ɓɛ̰̄ɛ̰̏ ‘1,000,000’.

97Subtraction and division are absent from the Dan-Gwɛɛtaa numeral system.

98External syntax. The basic syntax of the numerative construction is the same as elsewhere in the Mande family: the numeral follows the noun, and, most often, the plural marker ɗṵ̏ is absent: sààgā sɔ̋ɔ̋ɗű ‘5 arrows’. However, this marker can appear after the noun or after the numeral, and in both cases it expresses restrictive meaning: ɓɛ̰̄ sɔ̋ɔ̋ɗű-ɗṵ̋ = ɓɛ̰̄-ɗṵ̏ sɔ̋ɔ̋ɗű ‘exactly 5 persons’.

99In complex numerals, the count noun is repeated after each component (9a), with the exception of those cases where the reflexive pronoun is used as connector (9b)

(9a)

tɔ̏

kʌ̄ŋ̄

dō

wāȁ

tɔ̏

kɒ̏ɒ̏ŋ̏

dō

wāȁ

tɔ̏

dō

chicken

100

1

they.and

chicken

10

1

they.and

chicken

1

or

(9b)

tɔ̏

kʌ̄ŋ̄

dō

ɤ̄

kɔ̏

dō

ɤ̄

gā

dō

chicken

100

1

refl.sg

hand

one

refl.sg

bone

one

‘111 chickens’

100There are several counting words in Dan-Gwɛɛtaa which diverge in their degrees of grammaticalization (all can also be used as regular nouns). The most advanced in this direction (although still not quite obligatory in their use) are:

• gā ‘single object out of a mass/group’ (original meaning: ‘bone, grain, drop’). This word has further grammaticalized as a connector in compound numerals, see above;
• ɓɛ̄ ‘oblong object’ (original meaning: ‘fruit’);
• ɗʌ̰́ ‘living being’ (original meaning: ‘child’).

101Even less grammaticalized in the function of classifiers or measure words are the following nouns: kpɤ̄ ‘ball’ (for spherical objects), gèè ‘(dead) body’ (for carcasses of animals), bɤ̋ɤ̋ ‘bunch, bundle’ (for carcasses of animals), ya̰̋ ‘eye’ (for the content of receptacles).

2.6. San-Maka numerals

102San-Maka (Eastern Mande < Southeastern Mande) is spoken by some 230,000 people in the province of Nayala in Burkina Faso (Berthelette 2001). The data were collected by Elena Perekhvalskaya during her fieldwork; additional data are drawn from (Paré 1999). The tilde ~ separates off phonetic variants.

Table 8. San-Maka numerals

1	gōōŋ̀
2	páā
3	sɔ̄ɔ̄
4	síí
5	sórō
6	sɔ̄rɔ̄
7	sɔ̄bāā
8	kíwísí
9	mānāgɔ̀rɔ̄
10	fù
11	fù lù gōōŋ̀ (10+1)
12	fù lù páā (10+2)
13	fù lù sɔ̄ɔ̄ (10+3)
20	fòó
21	fòó ŋ́ gōōŋ̀ (20+1)
25	fòó ŋ́ sórō (20+5)
30	fù-sɔ̄ɔ̄ ~ pù-sɔ̄ɔ̄ (10x3)
33	fù-sɔ̄ɔ̄ ká sɔ̄ɔ̄ (10x3+3)
35	fù-sɔ̄ɔ̄ ká sórō (10x3 + 5)
37	fù-sɔ̄ɔ̄ ká sɔ̄bāā (10x3 + 7 (5+2))
40	fù-síí ~ pù-síí (10x4)
50	fù-sórō ~ pò-sórō (10x5)
57	fù-sórō ká sɔ̄bāā (10x5 + 7 (5+2))
60	fù-sɔ̄rɔ̄ ~ pɔ̀ sɔ̄ɔ̄ (10x6)
70	fù-bɔ̄-mā-lɛ́ (-10)
80	pāābīɛ̄ ~ bīɛ̄
90	pāābīɛ̄ ká fù (80+10)
100	pāābīɛ̄ ká fòó, bīɛ̄ fòó (80+20)
150	pāābīɛ̄ ká fù-bɔ̄-mā-lɛ́, bīɛ̄ fù-bɔ̄-mā-lɛ́ (80+70 (-10))
151	pāābīɛ̄ ká fù-bɔ̄-mā-lɛ́, bīɛ̄ fù-bɔ̄-mā-lɛ́ ká gōōŋ̀ (80+70 (-10)+1)
160	bīɛ̄-páā (80x2)
800	dúú gōōŋ̀ (800x1)
64,000	dúú-pāābīɛ̄ (800x80)
576,000	dúú-bīɛ̄ mānāgɔ̀rɔ̄ (800x720 (80x9))
639,279	dúú-bīɛ̄-mānāgɔ̀rɔ̄ ká fùbɔ̄mālɛ́ ká mānāgɔ̀rɔ̄ (800x799 {720 (80x9) + 79 [(80-10) + 9]}).
1,000,000	mílíɲɔ́ɔ́nɛ́ góōŋ̀
1,000,000,000	mílíáárè góōŋ̀

104The form for ‘8’, kíwísí (in other San varieties cígísí, kígísí, tīgīsī (Berthelette 2001: 62)) is opaque; it is certainly neither ‘5+3’, nor ‘10-2’. The only plausible interpretation along these lines would be ‘4x2’ (cf. síí ‘4’), otherwise it should be regarded as a non-derived word of unknown origin.

105The numeral for ‘9’, mānāgɔ̀rɔ̄, is most probably ‘10 – 1’ in origin, in which case the element -gɔ̀rɔ̄ can be interpreted as an archaic form of ‘1’, and the initial element mānā- is opaque (at least at the current stage of our knowledge of San-Maka).

107The peculiarity of the San-Maka system is a complete lack of 100 and 1000 count bases.

108In addition, the following strategies are used.

109a) in the formation of the numerals 6 and 7, we can postulate addition without a connector (a fact obscured by fusion);

110b) with the linker lù, in the formation of the numerals 11-19: fù lù sɔ̄ɔ̄ ‘13’ (10+3);

111c) with the linker ŋ́ (a possessive marker), for the numerals 21-29: fòó ŋ́ sɔ̄ɔ̄ ‘23’ (20+3); fòó ŋ́ kíwísí ‘28’ (20+8);

112d) with the linker ká (a comitative marker) for the addition:

• of units to tens in numerals bigger than 30: fù-sɔ̄ɔ̄ ká sórō ‘35’ (10x3 + 5); fù-sórō ká sɔ̄bāā ‘57’ (10x5 + 7 (5+2));
• of the tens to 80: pāā-bīɛ̄ ká fù-bɔ̄-mā-lɛ́ ‘150’ (80+[80-10]).

113multiplication is expressed by juxtaposition. The multiplicands are 10, 80 and 800. The multipliers with 10 are limited to 3 through 6. With 80, multipliers 2-9 are used. With 800, the multiplier can be any number from 1 to 799: dúú fù-síí ‘32000’ (800x40); dúú-bīɛ̄-mānāgɔ̀rɔ̄ ká fùbɔ̄mālɛ́ ká mānāgɔ̀rɔ̄ ‘639200’ (800x799 {720 (80x9) + 79 [(80-10) + 9]}).

114Subtraction in San-Maka is used in the formation of the numeral ‘70’, fù-bɔ̄-mā-lɛ́ (lit. “ten take.away-me-on”) , i.e. ‘minus ten’: in this construction the numeral pāā bīɛ̄ ‘80’ is not named, but implied.

115Counting words. In San-Maka there is one counting word, māŋ́ ‘thing’, for all types of counted objects. It is used with every numeral except for gōōŋ̀ ‘one’. Moïse Paré considers this classifier to be optional (Paré 1999: 81), but in Elena Perekhvalskaya's field data it appears consistently, except in the context of nouns denoting people: nɛ̄lɔ́ páá ‘two daughters’, but cōŋ̄ màŋ́ páá ‘two houses’. The counting word also appear when numerals are used in predicative function:

(10)	Māā	cōŋ̄	nə̄	màŋ́	sɔ̄ɔ̄.
	1sg-poss	house	ipf	thing	three
‘I have three houses (my houses are three)’.

116External syntax. Numerals follow the determined noun which appears without the plural marker ŋ̄/ -nə́, see examples above.

2.7. Soninke numerals

117Soninke is a Western Mande language spoken by more than 2 millions people in Mali, Senegal and Mauritania, but also in the Gambia and Burkina Faso. Soninke was the dominant language of the first great empire of Western Sudan, Ghana (or Wagadu), and exerted a tangible influence on the languages of the region. Soninke differs considerably from other Mande languages in many respects, this divergence is particularly visible in terms of the morphology and syntax of the numeral system. Our analysis of Soninke numerals is based primarily on (Creissels & Urmanchieva 2017: 261–262), but also on (Diagana 1995: 150–159) and (Diagana 1994: 249–274). When the sources provide different forms (which may be due to dialectal divergence), we give the forms from (Creissels & Urmanchieva 2017).

Table 9. Soninke numerals

1	báané
2	hílló
3	sìkkó
4	nàxàtó
5	kàràgó
6	tùnmú
7	ɲérú
8	ségú
9	kàbú
10	tánmú
11	tánmú dò báané
20	tánpíllé
30	tánjíkké
40	tánnáxáté
50	tánkárágé
60	tándúmé
70	tánɲéré
80	táncégé
90	tánkábé
100	kàmé
200	kàmó hìllì
1000	wújìné
6000	wújùnú tùnmì
1,000,000	mílíyò

119Arithmetic operations in the Soninke numeral system are trivial. Addition is used to combine numerals of different orders. It is always marked by the standard coordinative/ comitative preposition/conjunction `dó. Multiplication is applied to the bases tánmú/tán- ‘10’, kàmé/kàmó ‘100’, wújìné/wújùnú ‘1000’ and mílíyò ‘1,000,000’; the forms tán-, kàmó, wújùnú appear in compounds formed through multiplication (kàmó and wújùnú are, in fact, regular plural forms of kàmé and wújùné). In compound numerals with the multiplicand tán ‘10’, most multipliers appear in modified forms: they have high tone (whatever the tone of the original form may be), and the initial consonant undergoes alternation (in Soninke there is a regular alternation of initial consonants after the nasal element: N + s → nc, N + h → np, etc. However, consonant alternations in the names of tens are sometimes in disagreement with the general rule.). Ex. (Creissels & Urmanchieva 2017: 262):

(11)	wújùnú	tùnmì	dó	kàmó	hìllì	dó	tán-jíkké	dò	tùnmú
	1000	6	and	100.pl	2	and	10-3	and	6
‘6236’

• báané ‘1’ behaves as an adjective: it follows the determined noun which appears in its non-autonomous form, and the numeral assumes HL tone: yíttè ‘tree’ + báané ‘1’ → yíttí báanè ‘one tree’;
• numerals from 2 through 10 follow the determined noun, they assume a low tone, and their final vowel is replaced by –i. The determined noun appears in its plural form: yàxàrê ‘woman’ (sg.), yàxàrû ‘women’ (pl.), ɲérú ‘7’ → yàxàrú ɲèrì ‘seven women’;
• numerals from 20 and above, very atypically for Mande languages, precede the determined noun which appears in its singular form and assumes a tonal contour typical of the head noun in a genitival construction (LH): yúgúxásè ‘elderly man’ (sg.), yúgúxásò ‘elderly men’ (pl.), tánpíllé ‘20’ → tánpíllé yùgùxàsé ‘twenty elderly men’. One could interpret such constructions as genitive, where the numeral appears as a dependent noun; in this case, the literal translation would be something like “a score of elderly men”. However, this interpretation is contradicted by the syntax of compound numerals which combines the first and the second strategies above: the part for the tens (hundreds, thousands...) precedes the noun (which appears in plural form, if not followed by ‘1’), and the part for the units follows it, while the preposition/conjunction `dó ‘and’ is placed between the higher numeral and the count noun.

(12)	wújúnè	dó	kàmó	sègì	dó	tóngú	ɲèrì
	1000	and	100.pl	8	and	hoe.pl	7
‘1807 hoes’ (Diagana 1995: 157)

3. A general survey of numeral systems in Mande family

121Table 10 displays the main characteristics of the numeral systems of 30 Mande languages (including the 8 languages analyzed above whose names are in bold) representing all 11 groups of the family. The total number of languages in the Mande family is between 60 and 70 (the definitive figure depends on decisions on the status of closely related varieties), so, in our sample, about half of the Mande languages are represented. There are good reasons to think that the languages remaining outside our sample might not display characteristics radically different from those included in our survey; in other words, we consider our survey to be quite representative.

122Table 10 is followed by concise comments on any interesting points of the numeral systems of those languages not analyzed above.

General comments on Table 10.

123The column “Numerals ‘5+N’” is relevant only for languages with elements of quinarity (even if marginal and/or synchronically obscure). In this column, numerals formed through addition with the augend ‘5’ are presented.

126In the column “Connectors”, connective elements are provided, with their literal meanings or etymologies and comments on their spheres of use. By default, the connectors are used in addition constructions; in the exceptional case of a multiplicative connector, this fact is specially mentioned.

Table 10. Summary table

Language	Simple numerals (<10)	Numerals ‘5+N’	Multiplicands	Augends	Subtrahends & division	Connectors
Bamana (old system)	1 kélen 2 fìla 3 sàba 4 náani 5 dúuru 8 séegin 9 kɔ̀nɔntɔn	6 wɔ́ɔrɔ 7 wóronwìla	80 kɛ̀mɛ Num 800 bà Nom	5 wÓ- 10 tán Num 20 mùgan Num 40 dɛ̀bɛ Num 60 màninkɛ̀mɛ Num 80 kɛ̀mɛ Num 100 sìlamɛyakɛmɛ	Sbtr.: 1, 2, 10, 20 Division: 5 dlá (5 = 10/2)	ni ‘and’
Bamana (modern system)	1 kélen 2 fìla 3 sàba 4 náani 5 dúuru 8 séegin 9 kɔ̀nɔntɔn	6 wɔ́ɔrɔ 7 wóronwìla	10 bî Irr 100 kɛ̀mɛ Num 1000 bà Nom 1000 wáa Nom (for money) 1000000 míliyɔn Nom	5 wÓ- 10 tán Num 20 mùgan Num 100 kɛ̀mɛ Num	—	ni ‘and’
Mandinka (Creissels & Sambou 2013)	1 kílîŋ 2 fùlá 3 sàbá 4 náaní 5 lúulu 8 sáyí 9 kònóntó	6 wóoró 7 wórówùlá	10 tâŋ Num 100 kèmé Num ~ Nom 1000 wúlî ~ wílî Nom 1000000 mílíyóŋ Nom	5 wó- 10 tâŋ Num 20 mùwáŋ Num 100 kèmé Num ~ Nom	—	níŋ ‘and’
Kakabe (Vydrina 2017)	1 kélen` 2 fìla 3 sàba 4 náani 5 lɔ́ɔlu 8 ságin 9 kɔ̀nɔntɔ`	6 wɔ́ɔrɔ 7 wɔ́rɔwila	10 bíi` Irr 100 kɛ̀mɛ Num 1000 wáa Nom 1000000 míliyɔn Nom	5 wɔ́- 10 tán` Num 20 mùgan Num 100 kɛ̀mɛ Num	—	nín ‘and’
Vai (Welmers 1976)	1 lɔ̀ndɔ́ 2 fè'á 3 sàkpá 4 náánì 5 sóó'ù	6 sôŋ lɔ̀ndɔ́ 7 sôŋ fè'á 8 sôŋ sàkpá 9 sôŋ náánì	20 mɔ̀ (bándé) (‘entire person’) Num 100 hɔ́ndɔ̂ Nom 1000 táúsùŋ Nom?	5 sôŋ Num 10 tâŋ Num 20 mɔ̀ (bándé) (‘entire person’) Num	—	ø (<20), àkɔ́ ‘after it’ (> 20)
Jogo-Ton (Sapozhnikova 2017)	1 dīɛ̄ 2 fálá 3 sègbā 4 nāānì 5 sōolò	6 māarɔ́ 7 màalā 8 màsēgbá 9 mànānī	20 kēlē Irr 100 kēmē Num 1000 wúlú Nom	5 mà- 10 tán, táá Num 20 kēlēmò Num 100 kēmē Num	— Division: 50 kēmē tárá (50 = 100/2)	tɔ́ ‘and’
Jogo-Ligbi (Persson & Persson 1976; Persson & Persson 2007)	1 díɛ́n 2 fàlà 3 sigbá 4 náani 5 sóoló	6 mɔ̀ɔdó, 7 màúlà, 8 màsègbá, 9 màdáané	20 cɛ̌lè 100 cɛ̀mé 1000 wúlú	5 mà- ~ mɔ̀ɔ- 10 táàn 15 tígán 20 kèlɛ̀mɔ́ Num 30 túdó ~ tírró 100 cɛ̀mé	— Division: 50 cìmé tàdá, kyémé tàrá (50 = 100/2)	tɔ́ ‘and’
Jalkunan (Heath 2017)	1 dúlì 2 flā 3 sīgbō 4 nāānī 5 sóóló	6 mī-īlō 7 mà-álā 8 mà-sīgbō 9 má-nānì	20 jáān Irr 1000 wàʔà Nom, búúlì Nom	5 mi- ~ ma- 10 táá/tàà Num 20 jálámà Num	—	rá (11-19) ŕ (N20+10) tú (N10 + units, >20)
Susu (Shluinsky 2017; Touré 1994)	1 kérén 2 fìrín 3 sàxán 4 náání 5 súúlí	6 sénní (?) 7 sólófèrè 8 sólómásàxán 9 sólómánáaní	10 tòngó Irr 100 kɛ̀mɛ́ Num 1000 wúlù Nom	5 sóló- 10 fùú Num 20 mɔ̀xɔ́ɲɛ̀ń Num 100 kɛ̀mɛ́ Num	—	nùń ‘and’
Kpelle (Konoshenko 2017)	1 tááŋ́, tɔ̀nɔ ~ tànɔ́ŋ́ 2 hwèèlɛ̌ ~ hwèèlɛ́ 3 hààbǎ ~ hààbá 4 nááŋ́ 5 lɔ́ɔ́lí	6 mɛ̀ídà 7 mɛ̀ìhwéélɛ̀ 8 mɛ̀ìháábà 9 mɛ̀ìnááŋ́	10 pòǔ Num 100 ŋwúŋ̀ Nom 1000 wáá Nom	5 mɛ̀í Num 10 pòǔ Num	—	kɔ́ú ‘bone, grain, unit’ for Nx10+N, Nx100+Nx10 ø for Nx1000 + Nx100
Looma (Mishchenko 2017; Mishchenko p.c)	1 gílà(g) 2 félé(g) 3 sáwà(g) 4 náánì(g) 5 dɔ́ɔ́lù(g)	(6 dòzìtà) 7 dɔ́fèlà 8 dɔ́sáwà	10 pù(g) Irr 100 ṵ́(g) Nom 1000 wà(g) Nom 1000000 wùlù(g) Nom	5 dO- 10 pùgɔ̀ Num	9 tàwù(g) (*‘1-10’)	mázú ‘on, upon’ for Nx10+N, ø for all upper orders
Loko (Vydrin & Morozova 2017)	1 ngèrá(ŋ), ngìlá(ŋ) 2 fèlè(ŋ), fèè(ŋ) 3 cáwá(ŋ), sáwá(ŋ) 4 nááí(ŋ), náí(ŋ) 5 ndɔ́(ŋ), ndɔ́ú(ŋ)	6 ngɔ́hita(ŋ) 7 ngɔ̀fɛ̀là(ŋ) 8 ngɔ̀sákpá(ŋ)	10 pubu Irr 100 kɛmɛ Nom 1000 wul Nom	5 ngɔ- 10 pùù(ŋ) Num	9 ká-rábúú(ŋ), raabu(ŋ) (*‘1-10’)	mah̃u ‘on, above’
Bozo-Tigemaxo (Blecke 1996; Blecke p.c.)	1 sana; kɔ ~ kuɔ̃ ~ kuɔ in compounds 2 fende ~ pende 3 sio ~ siyo 4 nara 5 kɔlɔ 6 tuumi 7 jeeni 8 seki 9 kiɛwi ~ kiawi	—	10 tã (20=10x2, 30=10x3) Num 100 kɛmɛ Num 1000 gulu Nom	10 tã Num 40 lɛwɛ Num 60 mɛnikɛmɛ 80 kiɛnima ~ bambarakiɛrima 100 silambekiɛrima	70 = 80-10 kiɛnima tã kẽ	yee ‘and’
Soninke (Creissels & Urmanchieva 2017)	1 báané 2 hílló 3 sìkkó 4 nàxàtó 5 kàràgó 6 tùnmú 7 ɲérú 8 ségú 9 kàbú	—	10 tán- Irr 100 kàmé Num 1000 wújìné Num	10 tánmú Num 100 kàmé Num 1000 wújìné Num	? 10 kàbú (*‘1-10’)	`dó ‘and’
Dzuungoo (Solomiac 2014)	1 sōō, sɔ̄n 2 fíí 3 zhììgī 4 nààlén 5 nùn 6 tsùnmɛ̄n 7 ɲɛ̀ɛ̀nún 8 ɲáálòn 9 kyèèrón	—	80 cɛ̀ɛ̀n Nom	10 tsyéù/téēŋ Num 20 mòò/móò Num 40 dzyèè Num 60 mùéy Num 80 cààn Num 800 gúrú-sɔ̄n Num	140 = 80x2-20 cɛ̀ɛ̀nf́íímòòyáá	ø (11-30, 50, 100, 120) kó ‘and’ (>30)
Seenku (Prost 1971)	1 sòèn 2 fi̋ 3 sùɛ̏ 4 nȁà 5 nò 6 tsȉìn 7 nyèèn 8 kȁà 9 kȍomèn	—	20 te̋gé Irr 200 bi̋ Irr (2000 dan)	10 tő Num 20 fűe̋ Num	—	nè ‘with’ (20xN + 10), wɛ́ɛ̏ (special postposition for numerals, 10xN+unit)
Jowulu (Carlson 1993; Prost 1958)	1 tḛ̏ḛ̏nȉ 2 fúúlì 3 byàì (Prost dyue) 4 píírèì 5 tá̰á̰ 10 byḭ̏	6 tá̰á̰mànì (Prost tonte)	20 yàà Irr 1000 wáá Nom	5 tá̰á̰ (5+1=6) 10 bȕ Num 20 kȍnȉ Num	7 jɔ̰̀ɔ̰̀-pɔ́nȉ (-3) 8 fúl-pɔ́nȉ (-2) 9 tḛ̏ḛ̏-pɔ́nȉ (-1)	nȉ ‘plus’ (11-19) fá ‘add’ (20xN + N)
Bobo (le Bris & Prost 1981)	1 tèlè, tàlá 2 pə̄lā 3 sàà 4 nàà 5 kòò	6 kònālā, 7 kòpə̄rā, 8 kòrōsòòn, 9 kòrōnòòn	20 kìē Irr 100 zɔ̀, zɔ̀ō Irr 1000 sānā	5 kò- ~ kòrō- 10 fūn Num 20 kiòrō Num 100 zɔ̀lō Num 1000 sànōn	—	ø for 11-19, 20xN + 10, 100xN + 10 nē nɔ̄mā ‘with child’ for the others
Beng (Paperno & Maloletnyaya 2017)	1 dō 2 plāŋ̄ 3 ŋā̰ŋ̄ 4 síéŋ́ 5 sɔ́ŋ́	6 sɔ́dɔ̄ 7 sɔ́plā 8 sɔ́wà 9 sīsí	10 bū Num 100 làà Nom 1000 kɛ́ŋ́ Nom	5 sɔ́- 10 bū Num	—	àsḭ́ŋ́ for units nà̰ ... lō ‘and’ for upper orders
Gban (Fedotov 2017)	1 dó 2 fɛ̰̋ḭ̋ 3 yȉȁ 4 zȉe̋ 5 sűű	6 sɛ̋dò, 7 sɛ̋fɛ̰̋ḭ̋, 8 sɛ̋ȁȁ, 9 ti̋zȉe̋	10 vȕ Num 100 lű Nom 1000 wlű Nom 10,000 kpë Nom 1,000,000 mlḭ̀ɔ̰́ Nom	5 sɛ̋- 10 vȕ Num	—	ɛ̏kpȁȁ ‘it is on it’ or wli̋ ‘bone, fruit’ for units ø for upper orders
Guro (Kuznetsova 2008; Kuznetsova & Kuznetsova 2017)	1 dụ̄ 2 fíé 3 yāá, yāáká 4 zḭ̀ḛ́ ~ zḭ̀yḛ́, zḭ̀ɛ̰̀kɛ̰́ 5 sólú	6 sụ̄ẹ̄dụ̄ ~ sụ̄ẹ̄lụ̄ 7 sụ̄làyíé 8 sụ̄lāá 9 sụ̄làzḭ̀ḛ́	20 yọ̄ Nom 200 wúlù Nom 2000 bàà Nom	5 sụ̄- 10 vū Num	—	lâ ~ tâ ‘on, above’, for 6-9, 11-19 wọ́lẹ́ ‘bone’ for 20-29 é tâ lē ‘on its...’ after bàà or wúlù
Yaure (Kushnir 2017; Kushnir p.c.)	1 tụ̏ 2 fííli̋ ~ fi̋li̋ 3 yààgá 4 sḭ̀ḭ̀jḛ̀ 5 sóólú	6 šẹ́ẹ́dụ̏ 7 šálàvi̋li̋ 8 šálàà 9 šálàsḭ̀ḛ̀	10 fȕ Num 100 yȁȁ Nom 1000 kpȉ Nom	5 šá- 10 fȕ Num (20 yọ̏ Nom)	—	là ‘on’ for 7-9 (also 6?) tá ‘on’ for 11-19 wẹ́lẹ́ ‘bone’ for 21-99 ø for ‘100xN + 10xN’ and for ‘1000xN + 100xN’ pá multiplier for 30-90
Mwan (Perekhvalskaya 2017)	1 dō 2 plɛ̄ 3 yāgā 4 yīzīɛ̄ 5 sóó	6 slúádō 7 slúáplɛ̄ 8 slɔ́ā 9 slúáyīziɛ̄	20 mīā Nom 100 kɛ̀mɛ̀ Nom 1000 wáá Nom	5 slú- 10 vū Num	—	-á- for 5-9 tā ‘on’ for 11-19 ɓɛ̄ ‘grain’ for ‘20N+10’, ‘20N+N’ ké ‘with’ for ‘100xN + 10xN’ ø for higher orders
Dan-Gwɛɛtaa (Vydrin 2017)	1 dō 2 plɛ̀ ~ pȅȅdʌ̄ 3 yȁȁgā 4 yȉȉsīɤ̄ 5 sɔ̋ɔ̋ɗű	6 sɒ̋ɒ̏dō 7 sɒ̋ɒ̏plɛ̀ 8 sa̋ȁgā 9 sűȅsīɤ̄	10 kɒ̏ɒ̏ŋ̏ Nom/ kɔ̏ 100 kʌ̄ŋ̄ Nom 1000 gblɯ̋ Nom 1,000,000 gblɯ̋ ɓɛ̰̄ɛ̰̏ Nom	5 sO-	—	*tȁ ‘on’ for 6-9 ɤ̄ gā ‘its grain/unit’ for 11-99 wāȁ ‘and’ >100
Kla-Dan (Makeeva 2017)	1 dò 2 plɛ̏ ~ pȅȅdʌ̀ 3 yȁȁgà 4 yȉȉsìè 5 sɔ̋ɔ̋lű	6 sűa̋àdò 7 sűa̋ȁplɛ̏ 8 sűa̋ȁ 9 sűa̋ȁyȉȉsìè	10 kȕȁŋ̏/kɔ̏ŋ̏ Nom 100 kʌ̀ŋ̋ Nom 1000 wa̋a̋ Nom	5 sűa̋-	—	-ȁ- < *tȁ ‘on’ for 6-9 (à) gà ‘(its)grain/unit’ > 10
Mano (Khachaturyan 2017)	1 dōó 2 pèèlɛ̄ 3 yààkā 4 yììsɛ̄ 5 sɔ́ɔ́lī	6 sáládō 7 sálápèèlɛ̄ 8 sálákā, sáláyììkā 9 sɛ́lɛ́sɛ̀, sɛ́lɛ̄ìsɛ̀	10 vṵ̀ Nom 100 ŋwū Nom 1000 wáá Nom	5 sá-	—	-lá- for 6-9 wɛ́lɛ́ ‘bone, grain’ for ‘10xN + 1xN’ là ‘on’ for ‘100xN + 10xN’, ‘1000xN + 100xN’
Tura (Idiatov 2005; Idiatov & Aplonova 2017)	1 dó 2 pȉȉɗɛ́ 3 yȁká 4 yȉsɛ́ 5 sʋ̋ɗʋ̋, sőőɗű	6 sa̋ȁdó, 7 sa̋ȁpȉȉɗɛ̋, 8 sa̋ȁká, 9 sɔ̋ìsɛ́	10 kúȁ Irr (< 100), kɔ̏ɔ̏ (> 100) 100 kȁ̰ḭ́ Nom 1000 wa̋a̋ Nom	5 sa̋ȁ-/sɔ̋- 10 bùù Num	—	nḭ̀ wɛ̋ɛ̋ ‘and grain’ for ‘10xN + 1xN’ nḭ̀ for upper orders
Goo (Aplonova & Vydrin 2017)	1 dọ́ / dọ́ọ́ 2 pẹ̏ẹ̏lʌ́ / pẹ̏ẹ̏lʌ̏ 3 yȁȁká / yȁȁkȁ 4 yȉȉsʌ́ / yȉȉsʌ̏ 5 sőőlú / sőőlű	6 sa̋ládọ́ / sa̋ládọ̏ 7 sa̋lápẹ̏ẹ̏lʌ́ / sa̋lápẹ̏ẹ̏lʌ̏ 8 sa̋la̋ȁká / sa̋la̋ȁkȁ 9 sẹ̋ẹ̏sᴧ́	10 gọ́ Irr 10 kɔ̀ɔ̀ Nom (in compounds ‘100N+10N’) 100 kɛ̏ȉŋ́ Irr 1000 gbi̋li̋ Nom	5 sA̋- (10 gọ̋ọ̏ Nom) 100 kɛ̏ȉŋ́lȉ Num	—	-lȁ- for 6-9 wɛ̋ɛ̋ ‘grain’ for ‘10xN+1xN’ wáȁ pɔ̰̀ ~ wàȁ pɔ̰́ɔ̰́ ‘and thing’ for ‘100xN+1xN’
Wan (Nikitina 2017; Nikitina ms.)	1 dō 2 pīlɔ̄ŋ 3 ã̊ 4 si̊i̊yá 5 sɔ̊ɔ̊lů 6 wã́ã́ŋ 7 síāã̊ 8 séŋyã̄	9 sɔ̄lásīyá	(3 ã̊) (4 si̊i̊yá ) (5 sɔ̊ɔ̊lů) 20 mlīŋ Nom 100 cɛ̀mɛ̰̀, kɛ̀mɛ̰̀ Nom 1000 wågå	(4 si̊i̊yá) (5 sɔ̊ɔ̊lů) 10 sɔ́ŋyɔ̊lů Num	—	tā ‘on’ for 11-19 ɔ̄ŋ ‘each other’ for ‘20xN+10’
Boko (Jones 1998)	1 dó 2 plá 3 ’ȁȁó̰ 4 si̋i̋ɔ̰̋ 5 sɔ̋ó	6 sɔ́òlɔ́ 7 sóòplá 8 sɔ́ȁàȍ̰	20 bȁ Nom 200 ɔ̏áȁ Nom	10 kwȉ/kúɛ̏- 15 gɛ̰̏ó 25 bȁásɔ̋ó	1, 2, 5; up to 25	ø for ‘6-8’, ‘20xN+5’, ‘20xN+10’ ŋ̏ … ȍ ‘with’ for >15 (not divisible by 5)
San-Maka (Paré 1999)	1 gōōŋ̀ 2 páā 3 sɔ̄ɔ̄ 4 síí 5 sórō 8 kíwísí 9 mānāgɔ̀rɔ̄	6 sɔ̄rɔ̄ 7 sɔ̄bāā	10 fù Num 80 pāābīɛ̄, bīɛ̄ Num 800 dúú Nom	5 sɔ̄- 10 fù Num 20 fòó Num 80 pāābīɛ̄ Num	70 (80-10) fù-bɔ̄-mā-lɛ́	ø for ‘6, 7’ lù for ‘11-19’ ŋ́ for ‘21-29’ ká for ‘10xN+1xN’ (>30), ‘80xN + 10xN’

127Comments on individual languages (not analyzed specially in section 2).

3.1. Mandinka

128Unlike in other Manding varieties, in compound numerals from 30 to 90 divisible by 10, it is the regular form tâŋ that is used, rather than *bî.

129In a numerative construction (noun + numeral), a numeral can have the plural marker –lu, always preceded by the definite article –o, to express definiteness. The same meaning can be expressed by the article –o without the plural marker (Creissels & Sambou 2013: 221–22).

3.2. Vai

130The numeral system, as represented in (Welmers 1976: 67-73), is very transparent. The forms for ‘6–9’ are made by simple juxtaposition of numerals (although ‘5’ appears in an abridged form, sôŋ instead of sóó'ù). ‘20’ mɔ̀ bándé (lit. ‘entire person’) is a combination of two separate words, rather than one compound word, and in numerals divisible by 20, the multiplier is inserted between mɔ̀ and bándé, e.g.: ‘40’ mɔ̌ fè'á bándé, ‘80’ mɔ̌ náánì bándé.

131Beginning from 100, the system is decimal; the numerals for 100 and 1000 are English loans.

3.3. Jogo-Ton

132The autonomous form for ‘20’, kēlēmò, could be etymologised as a compound whose initial element, kēlē, might go back to the widespread Western Mande form *kélén ‘1’ (while in modern Jogo ‘1’ is dīɛ̄), and mò is a shortened form of the word mɔ̀gɔ̀ ~ mògò ‘human being’. I.e., kēlēmò could originally mean ‘one human being’, which would be very much in line with the general tendency in the Mande family, and the multiplicand form kēlē could be regarded as a result of subsequent truncation. There are however some weak points in this etymology: taking into account the regular syntax of numerals, one would expect to find the elements in the inverse order (along the lines of *mò-kēlē).

133As in Old Bamana, Jogo uses division: ‘50’ is designated as kēmē tárá, lit. ‘half (tárá) of a hundred’.

3.4. Jogo-Ligbi

134The numeral tígán ‘15’ probably contains an archaic stem *kan or *gan ‘foot, leg’ (cf. 2.2 for a discussion of the etymology of ‘15’ in Boko). However, in modern Ligbi the word for ‘foot, leg’ is kpɔ́, and the same form is attested in all the other languages of the group (Jogo, Jeri, Jalkuna). Therefore, if tígán is a relatively recent Jogo-Ligbi innovation, the element ‑gán is difficult to interpret as ‘foot’. According to (Persson & Persson 2007), there is a special word for ‘30’ (túdó or tírró), which may be an element of a heterogeneous quindecimal system (30 = 15x2).

135What was said about the etymology of ‘20’ in Jogo-Ton can also be applied to the Jogo-Ligbi forms kèlɛ̀mɔ́ / cɛ̌lè.

3.5. Jalkunan

136The tones borne by numerals undergo significant modifications depending on the context. Space constraints prevent us from presenting these modifications here; for the details see (Heath 2017: 90–95).

137The forms for ‘20’, jálámà / jáāⁿ, may go back to a proto-form similar to those in Jogo and Ligbi (*kélé-mɔ̀), although the sonorization of the initial consonant (*k- > *c- > j-) is yet to be explained. The irregular form for ‘20’ jén appearing in the numeral ‘30’, jén-tàà, may display the original vowel of the form ‘20’.

138One of the terms for ‘1000’, wàʔà, seems to be a Manding borrowing. The other, búúlì, also means ‘finger’ (Heath 2017: 94), and may be a remnant of the extended bodypart count.

139The connectors rá and ŕ are most probably allomorphs of one and the same morpheme (Heath, p.c.), which may be identical to the article-like noun suffix in Jalkunan which has the same forms.

3.6. Susu

140The form tòngó used as the base for the names of tens is homonymous with the verb tòngó ‘take’ (Touré 1994: 166), which can presumably be explained through particularities of Susu finger counting.

141The form for ‘20’, mɔ̀xɔ́ɲɛ̀ń, looks very much like the word for ‘human being’ in Manding and other languages of the Great Manding macro-group (Mokole, Vai-Kono, Jogo-Jeri); in Susu the corresponding word may be etymologically related too, but it has different vocalization: mìxí ‘person’. A borrowing from Manding or Mokole cannot be excluded.

3.7. Kpelle (Guinean variety)

142ŋwúŋ̀ ‘100’ probably goes back to ŋwùŋ̌ ‘head’, although the tonal difference between these forms is yet to be explained.

143The use of the connector kɔ́ú is not limited to units. For higher orders, addition is expressed by juxtaposition; optionally, the counted noun can be repeated before each order. When counting years, the postposition púlû ‘after, behind’ can be optionally inserted between the thousands and the hundreds (Konoshenko, p.c.):

(13)	gɔ̀láŋ	ŋwáá	tɔ́nɔ́	!púlû	ŋùŋ́	mɛ̀ìnááŋ́	kɔ́ú	pòú	mɛ̀ìnááŋ́	kɔ́ú	lɔ́ɔ́lí
	year	1000	1	after	100	9	grain	10	9	grain	5
‘the year 1995’

3.8. Looma (Guinean variety)

144In the form dòzìtà ‘6’, dò- goes back to dɔ́ɔ́lù(g) ‘5’, while the element -zìtà may contain the archaic stem for ‘1’, -ta, cf. tá ‘certain, some’ (cf. also Kpelle tááŋ́ ‘one’). The element -zì- may be etymologically related to the inessive postposition that appears in different Looma dialects in the forms sù/zù and sì/zì.

145Just as in Guinean Kpelle, ṵ́(g) ‘hundred’ is similar to ṵ̀(g) ‘head’; the Kpelle and the Looma forms are undoubtedly identical etymologically (Looma tones are regularly the inverse of the tones found in Kpelle). In all likelihood, this word for ‘100’ can be explained as reflecting an extended body-part system (unfortunately, we have no information on the ways in which body parts are used for counting among different Mande peoples).

146In Guinean Looma, both stems for ‘1000’ found widely across the Western Mande languages, wà(g) and wùlù(g), do appear; however, the latter has been reinterpreted as ‘1,000,000’.

147There is one (at least etymologically) subtractive numeral: 9 tàwù(g), where -wù(g) < pù(g) ‘10’, and the initial element tà- may go back to the word tá ‘certain, some’ (cf. Kpelle tááŋ́ ‘one’). If this etymology is correct, this form violates Greenberg's (1990: 278) 10th generalization: “Subtraction is never expressed by the mere sequence of the subtrahend and the minuend”.

3.9. Loko

148In the form ngɔ́hita(ŋ) ‘5’, the element –hita(ŋ) can be interpreted in the same way as –zìtà in Looma (Loko h : Looma z is a regular correspondence), while ngɔ́- is an archaic stem for ‘5’ used in compounds.

149There are special forms for two numerals used in counting: kárábúú(ŋ) ‘9’ and kɛ́pù(ŋ), kápùù ‘10’. The etymology of the element ká-/kɛ́- is not clear.

150The reduplicated form pubu that appears in compound numerals divisible by ten (pubu fee ‘20’, pubu naai ‘40’, etc.) is certainly the plural form of pùu(ŋ) ‘10’; in Loko reduplication is a regular means of plural formation.

151There is one numeral which is subtractive by origin, 9 rábúú(ŋ). It seems to be composed according to the same model as tàwù(g) ‘9’ in Looma.

3.10. Bozo-Tegemaxo

152Oddly, ‘20’ is tã fende ‘10x2’ and ‘30’ is tã sio ‘10x3’, while the other numbers divisible by 20 are designated by special words, and ‘50’ = ‘40+10’ (lɛwɛ ye tã), ‘90’ = ‘80+10’ (kiɛnima ye tã). This fragment of the counting system, most probably, is borrowed from Old Bamana (or, alternatively, one could speculate that ‘40’ was borrowed by Old Bamana from Bozo). There is only one subtractive numeral, 70 = 80 – 10 kiɛnima tã kẽ lit. “80 ten lack”.

3.11. Seenku

153The word for ‘20’ used as an augend, fűe̋, could be a loan from the Eastern Mande languages, cf. San-Maka fòó. The origin of the multiplicand for ‘20’, te̋gé, is not quite clear; an ancient borrowing from Senufo or Teen (Gur family) cannot be excluded (cf. footnote 10). ‘200’ is expressed by a compound te̋gé tő ‘20x10’, and the base bi̋ ‘200’ is used only in multiplication, beginning from 400 (bi̋ fi̋ ‘200x2). dan ‘2000’ looks like a numeral base; however, it is not used as such, because for higher numeral Seenku speakers usually switch into Jula.

154The numerals follow the determined nouns, which appear in their plural form (with the exception of those preceded by the numeral ‘1’). In compound numerals, the determined noun is repeated before the component belonging to each order.

3.12. Jowulu

155Quinary counting is represented only by the form tá̰á̰mànì ‘6’ where the initial element is certainly tá̰á̰ ‘5’, but the final element is unclear. In (Prost 1958) we find the form tonte ‘6’ whose final element seems to be more transparent and can be traced back to the numeral ‘one’, tḛ̏ḛ̏nȉ (Carlson 1993), têna (Prost 1958).

156One would expect a numeral base ‘200’, but in (Carlson 1993) numerals between 100 and 1000 are missing. The form yàà ‘20’ used in compounds (‘40’, ‘60’, ‘100’, etc.) means ‘rope’ (evidently it goes back to the name for a bunch of cowrie shells). The subtractive numerals are jɔ̰̀ɔ̰̀-pɔ́nȉ ‘7’, fúl-pɔ́nȉ ‘8’ and tḛ̏ḛ̏-pɔ́nȉ ‘9’, see discussion of these forms in (Pozdniakov 2018: 224).

3.13. Bobo

158The numeral ‘1’ is tèlè when counting, and tàlá when it determines a noun.

3.14. Beng

159In the numerals 6 through 9, compounding is accompanied by a loss of nasalization, cf.

• ‘2’ plāŋ̄ vs. ‘7’ sɔ́plā,
• ‘3’ ŋā̰ŋ̄ vs. ‘8’ sɔ́wā,
• ‘4’ síéŋ́ vs. ‘9’ sīsí,25
• and also ‘5’ sɔ́ŋ́ vs. ‘6’ sɔ́dō.

160However, an alternative interpretation is also possible: in most other South Mande languages, numerals 2 through 5 have no nasal elements. It is probable that the Beng numerals for 6 through 9 have retained the original forms of the root, while numerals 2 through 5 have undergone nasalization.

161Besides sɔ́wà (which is used as a simple form), there is an alternative form for ‘8’, kēŋēsíéŋ́, which appears in compound numerals (e.g. bū kēŋēsíéŋ́ ‘80’, kēŋēsíéŋ́-kēŋēsíéŋ́ ‘eight by eight, in eights’, a distributive meaning). Evidently, the final element of this form, síéŋ́, stems from síéŋ́ ‘4’, and kēŋēsíéŋ́ might be a multiplicative form, 4x2. However, the etymology of the initial element, kēŋē-, remains obscure.

162The form èbū ‘10’ appears in counting, otherwise, the shorter form bū is used.

163Numerals for the second and the third tens display an optional tendency towards fusion: būwlāŋ̄ ~ būplāŋ ‘20’, bṵ̄ā̰ŋ̄ ~ būŋā̰ŋ ‘30’.

164An alternative form for ‘1000’ is attested, là bū (100x10); it is used in the “forest dialect” of Beng.

3.15. Gban

165The numerals 6 through 8 have an initial element sɛ̋- which probably results from a merger of the numeral ‘5’ sűű with an undetermined locative postposition. The irregular consonant t- in the numeral ti̋zȉe̋ ‘9’ can probably be explained through regressive consonantal dissimilation (accompanied by a regressive assimilation of the vowel): *sɛ̋-zȉe̋ > ti̋zȉe̋.

166In complex numerals ending in ‘10’, the numeral ‘10’ appears in the “heavy” form vȕcɛ̰̀, e.g.: lű dò vȕcɛ̰̀ ‘110’. The meaning and etymology of the element cɛ̰̀ is unclear (it is perhaps a multiplier ‘1’).

167According to Maxim Fedotov (2017: 917), the word for ‘100’ may go back to lű ‘mouth’; if so, this may be another manifestation of extended bodypart counting.

3.16. Guro

168The numerals 3 and 4 have “heavy” forms, yāáká ~ yāāká and zḭ̀ɛ̰̀kɛ̰́, used in isolation or in the final position of compound numerals, which seem to express an emotional evaluation of the quantity: “as many as 3 (4)” (Kuznetsova 2008: 413). Comparison with other Southern and Eastern Mande languages suggests that the “heavy” form for ‘3’, yāáká, may be the original one (in which case yāá should be regarded as a shortened form), but for zḭ̀ɛ̰̀kɛ̰́ ‘4’ this is less probable (cf. however the Eastern Mande forms from the Boko-Busa cluster: Boko si̋i̋ɔ̰́, Bokobaru síígɔ̃).

169Wúlù ‘200’ seems to be a Manding loan. The semantic shift for wúlù (1000 > 200) may be explained either in terms of the discrepancy between the Jula decimal model and the Guro vigesimal model (i.e., its status as a numeral base has been maintained, but with a change to its numerical value), or through a confusion between money count (where the basic unit is 5 Fr) and object count; cf. (Zepp 1983: 149–150) on similar cases of confusion between 200 and 1000 in different languages of Côte d’Ivoire. Bàà ‘2000’ is homonymous with the word for a basket made of palm leaves (supposedly used in earlier times for carrying cola nuts). Yɔ̄ ‘20’ might go back to yɔ̄ ‘oil palm’, although it is not quite clear which components of the oil palm were counted.

170Noun precedes numeral. In compound numerals, the noun can be repeated before each order (in which case the connective element é tâ lē can be omitted), although it is rarely repeated before the number for the lowest order (units).

171The plural marker lṵ̂ is most often absent from the numerative, however, it can be present following either the noun or the numeral. It marks focus.

3.17. Yaure

172This language is the closest relative of Guro, so the forms and meanings of the numerals up to 20 are very similar to those found in Guro. The divergence concerns the numeral system after 20: it is consistently vigesimal in Guro and consequently decimal in Yaure. The word for ‘1000’ kpȉ (in other sources, kpȉȉ) is of unknown origin (cf. however kpë ‘10,000’ in Gban which could be cognate with kpȉ ).

173Yaure seems to be the only Mande language to make use of a connector (pá, most probably from the verb pá ‘take’) marking multiplication for tens: fȕ pá yààgá ‘30’, fȕ pá šálàsḭ̀ḛ̀ ‘90’. In the higher orders, multiplication is expressed by simple juxtaposition: yȁȁ fííli̋ ‘200’, kpȉ fííli̋ ‘2000’.

174The plural marker lṵ̀ can appear after the noun or after the numeral; it may also be absent.

3.18. Kla-Dan

175The Kla-Dan system is very close to that of Dan-Gwɛɛtaa, there is only one tangible divergence which concerns the word for 1000, wa̋a̋, borrowed in Kla-Dan from Manding. At the same time, the word gblɯ̋ (whose cognate yielded the term for ‘1000’ in Dan-Gwɛɛtaa) exists in Kla-Dan with the meaning of ‘temporary bag made of a palm leaf’.

3.19. Mano

176In the numerals ‘6-8’, the a-vocalization of initial syllables indicates the presence of a postposition (most probably, là ‘on’) as a connector between two simple numeral stems. In ‘9’, the change *a > ɛ is due to regressive assimilation.

3.20. Tura

178There are three forms for ‘10’: bùù is used as an augend in the numerals from 10 through 19; kúȁ is a multiplicand ‘10’ in the numerals from 20 through 90; kɔ̀ɔ̀ is a multiplicand ‘10’ in compound numerals above 100. Both kúȁ and kɔ̀ɔ̀ go back, undoubtedly, to the noun kɔ̀ɔ̀ ‘hand, pair of hands’. The compound numeral for ‘100’ is kȁ̰ḭ́ dḭ̏nḭ̏ instead of the expected kà̰ḭ́ dó. Dḭ̏nḭ̏ is, to use Idiatov's (2005: 54) term, a “specified noun” meaning generally ‘the other (of two)’.

3.21. Goo

179Numerals 1 to 8 have different forms when used in the attributive construction (in Table 10 they are given before the slash), and in counting (after the slash). ‘10’ also has two forms: gọ̋ọ̏dọ́ in the attributive construction, gọ̋ọ̏dọ̏ in counting (this fact can be regarded as an argument for an interpretation of these forms as monolexemic in synchrony, rather than as a combination of two lexemes ‘10’ and ‘1’).

181One would expect the word for ‘1000’ gbi̋li̋ to stem from the noun for ‘cola basket’ (as in Dan-Gweetaa), however the name for ‘cola basket’ in Goo has a different form, gblì ~ gbìì (although its lexical tone is yet to be verified). One cannot exclude the possibility that gbi̋li̋ ‘1000’ in Goo has been borrowed from Dan.

182In compound numerals, the counted noun is repeated before each order, except for the units, where the numerative word wɛ̋ɛ̋ ‘bone, grain’ appears instead:

(14)	kwẹ̀ẹ̀	gbi̋li̋	dọ́	kwẹ̀ẹ̀	kɛ̏ȉŋ́	sẹ̋ẹ̏sᴧ́	kwẹ̀ẹ̀	gọ̋	sẹ̋ẹ̏sᴧ́	wɛ̋ɛ̋	sa̋ládọ́
	year	1000	1	year	100	9	year	10	9	bone	6
‘the year 1996’

3.22. Wan

184The pair numerals (6, 8, 10) seem to be archaic plurals (or duals) formed through suffixation or infixation of a nasal syllabic (probably accompanied by an undetermined vowel; in any case, this derivation is accompanied by irregular modifications of the stems):

• ‘3’ ã̊ — ‘6’ wã́ã́ŋ,
• ‘4’ si̊i̊yá — ‘8’ séŋyã̄,
• ‘5’ sɔ̊ɔ̊lů — ‘10’ sɔ́ŋyɔ̊lů.

185Infixation is an extremely untypical procedure in Mande, although transfixation as a marker of superintensive form of adjectives is attested in Dan-Gwɛɛtaa (Vydrine 2007:84ff). If this infixation were present in only one pair of numerals (especially in the pair ‘3 — 6’ where infixation is not evident, because of the monosyllabic character of the numeral ‘3’), it would be appropriate to discard it as a fortuitous resemblance; however, its appearance in three pairs of numerals makes the assumption of a fortuitous resemblance rather improbable.

186Finally, síāã̊ ‘7’ looks like a combination of ‘4’, si̊i̊yá, and ‘3’, ã̊. The model “7 = 4+3” is not only exceptional in Mande, but is also rare in Niger-Congo as a whole (Pozdniakov 2018: 282).

187To sum up, the counting system of Wan for the second five looks (etymologically) as follows:

• 6=3x2; 7=4+3; 8=4x2; 9=5+4; 10=5x2.

188The even numerals are formed through multiplication by two, and the odd numerals through addition of the largest possible items. Although very unusual for the Mande family, this system is not exceptional among the languages of the world. In fact, it is close to Type II of Seidenberg's “Neo-2” system (Hurford 1987: 250–251; Hanke 2010: 71–72), with the sole difference that the even numerals are derived by multiplication by two (expressed though infixation), rather than reduplication.

189The term for ‘20’, mlīŋ, can be compared with mlītī ‘people’ ((Nikitina ms.); cf. mī ‘human being’); as in many other Mande languages, it may stem from the noun for ‘human being’.

4. Major characteristic features of Mande numeral systems

4.1. Classification of numerical systems into types

190If we proceed from the principle formulated in 1.3 (”the multiplicand of the lower order, i.e., the radix, defines the type of the numerical system”), the following numerical systems exist in the Mande family:

• decimal (Modern Bamana, Mandinka, Kakabe, Susu, Kpelle, Soninke, Beng, Gban, Yaure, Dan-Gwɛɛtaa, Kla-Dan, Mano, Tura, Goo),
• pure vigesimal (Seenku, Jalkunan, Jowulu, Guro, Boko),
• mixed decimal and vigesimal (Vai, Jogo, Ligbi, Bozo-Tigemaxo, Bobo, Mwan, Wan),
• octogesimal (Old Bamana, Dzuungoo),
• mixed decimal and octogesimal (San-Maka).

191By “mixed decimal and vigesimal”, we mean systems where “the numbers up to 99 are expressed vigesimally, but the system then shifts to being decimal for the expression of the hundreds, so that one ends up with expressions of the type x100 + y20 + z” (Comrie 2005). In the mixed decimal and octogesimal system of San-Maka, counting is decimal up to 79, and then switches to octogesimal (x80 + y10 + z).

192The Bozo-Tigemaxo system is peculiar: it is decimal up to 39 and from 100 on, but it is “heteroradical vigesimal” in the intermediate span (x + 10 + y, where x is divisible by 20). The same “heteroradical vigesimal” component is also present in both octogesimal systems (Old Bamana, Dzuungoo).

193Vigesimal models (pure or mixed with decimal ones) do not predominate in Mande, but they are well represented; they are characteristic of more than a third of the languages in our sample. It is interesting that vigesimal models are dispersed over diverse genetic groups and are not concentrated in one or two closely related ones. At the same time, nearly all Mande languages where vigesimal systems are attested (with the exception of Vai) are in close contact with Kru, Gur and/or Senufo languages (or such contact can be supposed at an earlier stage). This serves a crucial point in favor of the argument that the vigesimal model in Mande is contact-induced.

195In two languages, Jogo-Ligbi and Boko, elements of a pentadecimal system (built on base 15) have been identified: the word for 15 is not compound (at least in synchrony), it appears as an augend for the numerals from 16 through 19, and in Boko it also acts as a diminuend for 14. Pentadecimal systems are very rare in the languages of the world.

4.2. Elements of the quinary model and other models within the first ten

196No Mande numerical system is quinary in the strictest sense, given that a numeral for ‘5’ does not serve as a multiplicand in any language. However, it is very frequent as an augend, and there are reasons to think that this situation can be reconstructed at the Proto-Mande level. Quinary models never extended beyond the first ten, and in many languages the quinarity eventually faded away altogether: original components have often fused and lost their distinctiveness, and some numerals from 6 through 9 have lost their quinary nature. Remnants of this model are found in 25 out of 30 systems of our sample, and are missing only in Dzuungoo, Seenku, Bozo-Tegemaxo and Soninke (Wan, with its idiosyncratic model for the numerals of the second five, stands apart).

197Exceptional in Mande are the remnants of models with multiplicand 4: the form for ‘8’ used in compound numerals in Beng, kēŋēsíéŋ́, presumably ‘4x2’ by origin, see 3.14; ‘8’ in San-Maka, kíwísí, probably also ‘4x2’ by origin, see 2.6; the Wan form for ‘8’, and also for ‘6’ and ‘10’, see 3.22. However, these remnants of “Neo-2” structures are not so rare in Niger-Congo; in particular, this model (mainly for 4x2 = 8) is attested in some Gur languages (Natioro, Wara, Viemo, some Grusi languages), in numerous Adamawa languages, in Ijo, in the Ka-Togo and Na-Togo groups of the Kwa family, etc. (Pozdniakov 2018: 282–288), in Ekoi (Bantoid) (Hanke 2010: 71).

198Beng and Wan are Southern Mande languages which stand apart from the rest of the group; according to glottochronological analysis, their separation from the Proto-Southern Mande languages took place some 23-25 centuries ago, and the separation of San from its closest relative Bisa lies within the same temporal range (Vydrin 2009). The presence of non-quinary models in these languages might be interpreted as evidence for the lack of stability of count systems, even within the first ten, at that temporal distance.

4.3. Different forms for bases and corresponding numerals

199A phenomenon that often disguises the presence of quinarity (both for speakers and for researchers) is a mismatch between the forms for the numeral ‘5’ and the element referring to 5 in the numerals 6 through 9. In many languages (Vai, Susu, Looma, Bobo, Beng, Gban, Guro, Yaure, Mwan, Dan-Gwɛɛtaa, Kla-Dan, Mano, Tura, Goo, Wan, Boko, San-Maka) such formal differences result from assimilative processes accompanying the fusion of the elements and do not represent a great problem. There are however cases where we are undoubtedly dealing with different stems, see Table 11.

Table 11. Different stems for ‘5’ as a numeral and as an augend

	Bamana	Mandinka	Kakabe	Jogo-Ton	Jogo-Ligbi	Jalkunan	Loko	Kpelle
5	dúuru	lúulu	lɔ́ɔlu	sōolò	sóoló	sóóló	ndɔ́(ŋ)	lɔ́ɔ́lí
5 in compounds	wóron-, wɔ́ɔ-	wóro-, wóo-	wɔ́rɔ-, wɔ́ɔ-	māā-, màà-	mɔ̀ɔ-, màú-, mà-	ma-, mi-	ngɔ-	mɛ̀í-, mɛ̀ì-

201The stem MA- likely goes back to the superessive postposition broadly represented in the languages of the Mande family and reconstructible at the Proto-Mande level as *ɓã̀; the use of superessive adpositions as linkers in compound numerals is a well-attested phenomenon in Mande and elsewhere. If so, the original form of the numerals for 6-9 in Jogo, Ligbi, Jalkunan, Kpelle could be built on the model of “5 MA N”, with the subsequent truncation of the initial element.

Table 12. Variability in stems formed from the same roots

Language	10		20		80		100
	sole or augend	multipli­cand	sole or augend	multipli­cand	sole or augend	multipli­cand	sole or augend	multipli­cand
Jogo			kēlē	kēlēmò
Ligbi			kèlɛ̀mɔ́	cɛ̌lè
Jalkunan			jálámà	jáān, jén
Looma	pù(g)	pùgɔ̀
Loko	pubu	pùù(ŋ)
Bozo-Tigemaxo							silambe­kiɛrima	kɛmɛ
Soninke	tánmú	tán
Dzuungoo					cààn	cɛ̀ɛ̀n
Bobo			kìē	kiòrō			zɔ̀, zɔ̀ɔ̄	zɔ̀lō
Goo	gọ̋ọ́	gọ́ < 100

Table 13. Different stems in decimal and vigesimal systems

Language	10		20
	sole or augend	multiplicand	sole or augend	multiplicand
Modern Bamana	tán	bî
Kakabe	tán	bíi`
Susu	fùú	tòngó
Jowulu	byḭ̏	bȕ	kȍnȉ	yàà
Seenku			fűe̋	te̋gé
Tura	bùù	kúȁ
Goo	gọ̋ọ́	kɔ̀ɔ̀ (>100)

203With the exception of the Samogho languages (Jowulu and Seenku), for which our knowledge is very limited, the use of different roots for the same numerical values can be explained more or less satisfactorily. In Manding and Kakabe, BI-forms for ‘10’ most probably go back to the Proto-Mande root *pú, while tán may be a Western Mande innovation going back to the Proto-Niger-Congo stem for ‘5’ (Pozdniakov 2018: 224–226). In Susu, fùú is certainly an ancient form, and tòngó is a new one, probably resulting from the grammaticalization of the verb tòngó ‘take’. In Tura, bùù also represents the Proto-Mande root *pú, and kúȁ is an innovation of the Proto-Dan-Tura level going back to the word for ‘hand, arm; pair of hands’ (in Tura, kɔ̀ɔ̀); recall that in Dan, reflexes of *pú have been replaced by the form derived from “hand” (in Dan-Gwɛɛtaa, kɒ̏ɒ̏ŋ̏ or kɔ̏) in all contexts. In Goo, the “hand” form has replaced the original stem for ‘10’ in multiplicand function, but this form is used only in compound numerals above 100. In lower numerals another stem has emerged, which probably has the etymology ‘(small package of) cola nuts’.

4.4. Status of numerical bases

204Higher numerical bases tend to be more similar to nouns, and smaller bases behave more as true numerals (in the sense formulated in the general comments preceding Table 10). The only exception is the Goo language (Aplonova & Vydrin 2017: 461) where gọ̋ọ́ ‘10’ (used in the first hundred) and kɔ̀ɔ̀ ‘10’ (used in the second hundred and higher) are nouns (15, 16), while kḭ́ŋ́lḭ̏ ‘100’ is a numeral (16).

(15)	kwi̋í	gọ̋ọ́	dọ́	wɛ̋ɛ̋	pẹ̀ẹ̀lʌ́
	basket	10	1	bone	2
‘12 baskets’ (10x1 + 2)
(16)	kḭ́ŋ́lḭ̏	kɔ̀ɔ̀	dọ́
	100	10	1
‘110’ (100 + 10x1)

205This unusual situation in Goo might be explained by the fact that both words for ‘10’ have etymologies more or less transparent for the speakers of this language (‘cola/package of cola’ and ‘pair of hands’), cf. 3.21, and kḭ́ŋ́lḭ̏ ‘100’ is most probably a Manding loan, and is etymologically opaque for speakers.

206Another peculiar case is represented by Boko where the numeral for ‘15’ probably contains the stem for ‘1’, cf. 2.2, and can thus be regarded as the lowest base displaying nominal characteristics.

• 10 — Dan-Gwɛɛtaa, Kla-Dan, Mano
• 20 — Guro, Yaure, Mwan, Wan, Boko
• 80 — Dzuungoo
• 100 — Vai, Kpelle, Looma, Loko, Beng, Gban, Tura
• 800 — Old Bamana, San-Maka
• 1000 — Modern Bamana, Kakabe, Jogo, Jalkunan, Susu, Bozo-Tigemaxo, Jowulu
• > 1000 — Soninke

208The situation in Mandinka is ambiguous (kème ‘100’ can behave as a numeral or as a noun). In Soninke, all bases, including wújìné ‘1000’, behave as numerals.

209A correlation emerges: languages of Sahelian communities with historical engagement in long-distance trade and state-building tend to transform counting bases into numerals to a higher degree. In Soninke, the language of the very first great empire of the Western Sudan, even ‘1000’ is a numeral. In Manding and Bozo (but also in other Sahelian languages), ‘100’ is a number, and only ‘1000’ is a noun. In languages spoken in the forest zone by communities which, until the most recent times, abstained from large-scale commerce, the “nominal threshold” is either ‘100’ (Southwestern and some Southern Mande languages) or ‘20’ (most of Southern Mande languages, Boko), or even ‘10’ (Dan, Mano). This correlation is highly natural: the more a language community is exposed to commerce, the more prevalent higher numeral bases are, because their frequent use facilitates their transformation into “true numerals”.

4.5. Subtraction and division

210Subtraction has been attested in nine languages, that is, in almost one third of our sample. However, in only two systems (Old Bamana, see 2.1, and Boko, see 2.2) is this operation productive. As mentioned in 2.2, some violations of the Greenbergian generalizations concerning subtraction are attested in the Boko system.

211In Looma and Loko, subtraction is used only for (etymologically opaque) ‘9 = 10-1’, and in Jowulu it is used only for ‘7’ (= 10-3), ‘8’ (=10-2), ‘9’ (= 10-1). In Bozo and San-Maka the subtractive numeral is ‘70’ (= 80-10), and in Dzuungoo it is ‘140’ (= 80x2 – 20); these cases can be understood in the context of the octogesimal systems, where the fundamental base is 80. In Dzuungoo, most probably, a subtractive numeral ‘70 = 80-10’ also existed at an earlier stage, but it has been ousted by a non-subtractive form.

212Although subtraction is not particularly rare in Mande counting systems, the internal syntax of subtractive numerals is idiosyncratic for every language, and as such is hardly reconstructible at the proto-level.

213Division is found in three numeral systems: Old Bamana, Jogo-Ton, Jogo-Ligbi. All these cases confirm Greenberg's (1990: 280) observations that the denominator of the fraction is always 2 (or a power of 2), and also, that division is usually found in vigesimal systems.

214The coexistence of systems with different bases and of various arithmetic operations in one language makes possible different designations for one and the same number. So, in the archaic Bamana system, 70 can be expressed as màninkɛ̀mɛ ní tán (60+10), as kɛ̀mɛ tán ɲà (80–10) or as kɛ̀mɛ tán ká jɛ̀ (80–10). In Dzuungoo, 1000 can be designated by the compound numeral gúrú-sɔ̄n kó cànfóōdzyéè (800x1 + ((80x2) + 40)) or by the Jula loanword wāākērēn.

4.6. Inner syntax of complex numerals. Connectors

215The order of components in complex numerals formed by means of multiplication and addition is the same in all Mande languages: Base x a + b, where a is a multiplier, and b is the addend.

216Multiplication is expressed paratactically, with the solitary exception of the Yaure language where we find a multiplication connector pá (”10 pá N” for the numerals from 30 to 90 divisible by 10).

217As for addition, it can be expressed by various means in Mande languages. Four models predominate statistically:

• simple juxtaposition, without any morphological marking:33 Vai, Looma,34 Kpelle, Dzuungoo, Bobo, Gban, Yaure, Mwan, Boko. This is a rather infrequent model for addition, and it is never the only model used for addition in a language. Such a marginal position of additive parataxis can be explained by its predominance in multiplication;
• a coordinative-comitative construction:35 Modern and Old Bamana, Mandinka, Kakabe, Jogo, Ligbi, Jalkunan, Bozo-Tigemaxo, Soninke, Dzuungoo, Seenku, Jowulu, Beng, Mwan, Dan-Gwɛɛtaa, Tura, San-Maka, Boko. This model predominates in the languages of the savanna zone, while it is rarer and in fact rather marginal in the forest zone (its expansion in Tura might be explained by Manding influence);
• superessive postpositions or extended expressions with such postpositions (in some of the languages, the postposition is fused): Jalkunan (?), Looma, Loko, Gban, Guro, Yaure, Mwan, Dan-Gwɛɛtaa, Kla-Dan, Mano, Goo, Wan. The use of superessive adpositions as connectors for numerals (also attested in other languages of the world, although not very frequently) is very well represented in Southern Mande languages and in Southwestern Mande, but practically absent in the rest of the family;
• a noun ‘bone, unit’ (which is also used as a numerative), sometimes ‘3sg/refl + grain’ (‘its grain’): Kpelle, Gban, Guro, Yaure, Mwan, Dan-Gwɛɛtaa, Kla-Dan, Mano, Goo, Tura. A connector kai going back to the noun for ‘bone, grain’ appears in Liberian dialects of Looma (Vydrin 1987: 46). In most languages, this connector appears only before the terms for units, but in Kpelle it precedes both units and tens, and in Mwan the connector ɓɛ̄ (presumably from ɓɛ̄ ‘grain’) is preceded by mīā ‘20’, while vū ‘10’ is linked to the units by another connector, tā (a superessive postposition). As we have demonstrated, the connector ‘bone, grain, unit’ is very well attested in the Southern Mande group, and its existence in Kpelle and Liberian Looma may result from Southern Mande influence.

218There are also some models attested only in single languages:

• the nominal expression ‘its child’ (Bobo);
• the expression ‘behind it’ (Vai);
• the reciprocal pronoun (Wan);
• an article-like suffix (Jalkunan);
• the verb ‘add’ (Jowulu);
• the possessive marker (San-Maka; however, possessive markers in Mande languages often go back to locative postpositions).

219Some connectors are of unknown origin: àsḭ́ŋ́ in Beng; wɛ́ɛ̏ in Seenku, a specialized postposition used only in compound numerals; lù in San-Maka.

220In certain languages, only one additive model is used in all cases. These are: Old and Modern Bamana, Mandinka, Kakabe, Jogo, Ligbi, Susu, Bozo-Tigemaxo, Soninke. In this group we find languages of large communities which have been involved in long-range trade for many centuries. In languages of the forest zone, we find multiple models for addition. Presumably, the presence of such “less standardized” systems correlates with a lower level of involvement in commercial relations and, therefore, less frequent use of higher numerals in the daily life of the communities.

4.7. External morphosyntax of numerals

4.7.1. Word order.

221The basic word order in the numerative construction in Mande is “Noun + Numeral”. A notable exception from this rule is represented by Soninke where numerals from 20 precede the noun, and in compound numerals, where elements for higher orders (from 20 on) precede the noun, while the elements for smaller numbers follow it (see 2.7).

222At the same time, certain elements resembling the Soninke model can also be found in other languages of the family. In particular, in compound numerals in Guinean Kpelle (Maria Konoshenko, p.c.), the noun can be optionally repeated between the orders of thousands and hundreds (17), although, unlike in Soninke, the noun obligatorily appears to the left of the entire compound numeral.

(17)	pɛ́lɛ́	wáá	hwéélɛ̀	(pɛ́lɛ́)	ŋùŋ	lɔ́ɔ́lí
	house	1000	2	house	100	5
‘2500 houses’

223Repetition of the noun before each order of numerals is also the norm in Dan-Gwɛɛtaa (see 2.5), Goo, Guro and Seenku.

4.7.2. Plural markers.

224By default, in Mande numerative constructions, neither the noun, nor the numeral carries a plural marker. There are two exceptions in this respect: Soninke, where nouns appear in their plural forms before the numerals from 2 through 9, and Seenku, where nouns are always pluralized before numerals higher than 1.

225At the same time, in many languages the plural marker can be added to either the noun or the numeral, and in both cases the semantics of definiteness is expressed. This phenomenon is attested in Bamana (2.1), Mandinka (3.1) and Dzuungoo (2.3), and its existence cannot be excluded for some other languages (for which the necessary evidence may be missing at this time). In Dan-Gwɛɛtaa, a plural marker can be added to any element of the numerative construction, but the semantics of this is different from Bamana or Dzuungoo; it expresses restrictive meaning (”exactly the indicated number of items”). In Guro, a plural marker in a numerative construction expresses focus.

5. On the historical evolution of Mande counting systems

226One can make some suggestions concerning the sequence of emergence of different counting bases in Mande languages (which can probably be extended to other African languages as well).

• The quinary system is based on counting on one's fingers: it is the most archaic model, which emerged before commerce appeared.
• There are also traces of non-quinary models for the numbers of the first ten (in particular, a probable instance of te “Neo-2” strategy in Beng, San-Maka, and especially in Wan). These sporadic non-quinary elements seem to be quite archaic.
• The presumable etymological links of the numerals for ‘100’ with ‘mouth’ in Gban and with ‘head’ in Mano, Kpelle and Looma are indicative of the existence, at an earlier stage, of extended body-tallying count systems similar to those described for Haruai by Comrie (1999).
• The decimal model is an extension of the quinary system. In languages of large communities with a long history of state-building and involvement in long-distance trade, quinary systems tend to become blurred or disappear altogether.
• The pentadecimal and vigesimal models can be considered as further extensions of “finger count systems”; more precisely, they can be defined as “finger and toe count systems”. In our sample, they are represented in 12 languages (out of 30). In many Mande languages, elements of vigesimal systems are rudimentary: there is a special word for ‘20’ (often going back to a lexeme for ‘human being’), while other numerals divisible by 20 belong to the decimal system (40 = 4x10, 60 = 6x10, etc.). Those languages which have more substantial (”extended”) vigesimal systems may follow two different models: numerals divisible by 20 are formed through multiplication of the numeral ‘20’ (40 = 2x20, 60 = 3x20, etc., as in Guro, Mwan, Bobo, etc., a “true vigesimal system”) or they are represented by different stems (as in the old Bamana system, Dzuungoo, Bozo, a “heteroradical vigesimal system”). The languages with extended vigesimal systems belong to different Mande groups, a fact which makes problematic the reconstruction of such a model for Proto-Mande (and even for the proto-languages of the lower-level groups within Mande). Instead, the geographical distribution of these languages makes plausible the hypothesis that the vigesimal model was borrowed from Senufo and/or Gur languages (where this model is much more widespread); in South Mande languages, the vigesimal model can also be explained as the result of Kru influence (on the spread of vigesimal systems in Kru, see (Zepp 1983)).
  The elements of pentadecimal systems in Jogo-Ligbi and Boko, isolated within the Mande family, may also result from borrowing from Gur languages or Senufo (cf. (Carlson 1994) on the presence of the pentadecimal model in some Senufo languages).
• The octogesimal models (old Bamana, Dzuungoo, San-Maka) stem from the commercial relations of the pre-colonial period: they reflect a count based on bunches of cowrie shells (the predominant pre-colonial currency) or packages of cola nuts (the main item of intra-African trade).
• The consistent decimal systems are the most innovative: they continue to expand and in the process are ousting more archaic systems, cf. the replacement of the old Bamana system by the new one.

227The dynamics of the evolution of numerical systems is also reflected in the use of connectors. In languages spoken by communities involved in large-scale trade for many centuries, it is typical to have one universal connector (normally a coordinative/comitative conjunction/preposition), while languages spoken in the forest zone, not used until very recently for large-scale trade, tend to employ various connectors for different orders.

228