SETSWANA

NOUN CLASSES

Conceptual Categories Marked by Grammar?

Kari-Anne Selvik

Cand. philol. thesis

Department of Linguistics
University of Oslo
1996
 SETSWANA
NOUN CLASSES

Conceptual Categories Marked by Grammar?

Kari-Anne Selvik

Cand. philol. thesis

Department of Linguistics
University of Oslo
1996
Acknowledgments

First and foremost I would like to thank Professor Rolf Theil Endresen, my tutor,
for all the help and constant support that he has lent me in the course of writing
this thesis.
I would also like to thank Professor S.A. Swanepoel who guided me
through a period of Setswana studies at Potchefstroom University, and later
helped me check a large data sample. The whole Swanepoel family should be
thanked for their kind hospitability during my and my family's two stays in
South Africa.
Dr. Joseph Tsonope has provided me with invaluable help in the process of
arranging my research stay in Botswana. He has also showed his kind interest in
my project.
During my research stay in Botswana, many people contributed to my
project. The headmasters of Gaborone Secondary School and Maru A Pula kindly
allowed me to use students from their schools in a psycholinguistic test. In
particular, I would like to thank the two Gaborone Secondary School teachers Mr.
N. Mompei and Mrs. B. Sibiya for providing practical help with organising the
test. I would also like to thank the students and the other test subjects for
willingly lending their time to participate in the test.

Several others have contributed in different ways to my project. I would like to

thank the following:
- Elie Dib Wardini who has introduced me to the Filemaker database system.
- Senior Lecturer Hanne Gram Simonsen who helped and supported me in the
course of constructing the psycholinguistic test, while my tutor was abroad.
- Simon Chilembo who was of invaluable help in the process of constructing
nonsense nouns for the test.
- Senior Lecturer Grete Usterud Fenstad who has conducted a significance test of
my test results.
- Andrew Baarson who has provided corrections in my use of the English
language.
-Finally, I would like to thank the Department of Linguistics for a supp ortive
environment.

The Norwegian Research Council has supported my project financially, and the
authorities in Botswana provided me with a research perll'!it.
Table of contents

Chapter 1. Introduction ................................................................................................ 1

Chapter 2. Previous research.......................................................................................5
Chapter 3. Theoretical approach ............................................................................... 13
Chapter 4. Analysis .................................................................................. .................... 23
4.1. Introductory remarks ............................................................................... 23
4.1.1. Selection of data .......................................................... ................ 24
4.2. Class 3 .......................................................................................................... 27
4.3. Class 5 ..........................................................................................................42
4.4. Class 6 ............................................................... ...........................................59
4.5. Class 7 .......................................................................................................... 73
4.6. Concluding remarks ........................ .... .................................... .................85
Chapter 5. Psycholinguistic testing of the hypotheses .........................................89
5.1 . Introductory remarks ...............................................................................89
5.2. The pilot test ............ .............................. .................................................... 92
5.3. Construction of the final tests ................................................................ 95
5.4. Data collection ...........................................................................................98
5.5. Presentation of tests and results ............................................................99
5.5.1. Test 1 .......................................................................... ................. 100
5.5.2. Test 5 ........................ ............... ........... ......................................... 108
5.5.3. Test 7 ........................................................................................... 112
5.5.4. Test 8 ........................................................................................... 116
5.6. Discussion .................................................................................................120
Chapter 6. Conclusion .............................................................................................. 127
REFERENCES .......................................................................................... ........... ........129
APPENDICES .............................................................................................................. 133
Appendix I. Noun lists ................................................................................. 133
Appendix II. Tests, part 1 and part 2 ........................................................... 157
Appendix III. Statistical Model and Method ............................................ 179
Appendix IV. Description and results of the tests 2- 4 and 6 ............... 181
Chapter 1. Introduction

The aim of this thesis is to investigate the semantics of Setswana noun classes. Is
it possible at all to arrive at semantic analyses of these classes?
If it is:
- what will the semantic «content» of the classes be, and
- what «kind of» semantic characterisation will account for this zn the most
plausible way?

Setswana noun classes

Setswana is one among several Bantu languages in Southern Africa. It is spoken
by about 4 million speakers in Botswana (where it is the national language) and
South Africa (where it is one among several official languages).
A characteristic trait of all the Bantu languages is their highly elaborate
noun class systems.
Dixon (1982:160) has suggested that a noun class can be characterised as
follows:
We can say that the category of noun classes is (1) a grouping of all the nouns
of a language into a smallish number of classes, (2) so that there is some overt
indication of the class of a noun within any sentence in which it occurs, (3) and
this indication is not entirely within the noun-word.

In Setswana all nouns are members of a noun class. The classes can be established
on the basis of the concordia! system in syntactic structures. We find concordia!
elements in verbs, pronouns, adjectives, demonstratives, numerals etc. In
addition, most noun classes have a particular class prefix attached to the nouns
themselves. In many Bantu languages it is not uncommon that the noun prefix
and the concordia! elements suggest two different noun classes. In Setswana,
however, this phenomenon is close to non-existent.

Mo-simane yo mo-leele o - rekisa se-lepe se se-nnye sa Tebogo.

cl.l-boy cl.l eLl-tall cl.l sell cl.7-axe cl.7 c1.7-little cl.7 Tebogo
(The tall boy sells Tebogo' s little axe.)

Ba-simane ba ba-leele ba - rekisa di-lepe tse di-nnye tsa Tebogo.

cl.2-boys cl.2 cl.2-tall cl.2 sell cl.8-axes cl.8 cl.8-small cl.8 Tebogo
(The tall boys sell Tebogo's small axes.)

- 1-
The table below shows the noun prefixes belonging to the 16 noun classes in
Setswana. 1 The prefixes in the same row in the table refer to singular and plural
forms of the same noun stems (thus, for example: monna (man, class 1) - banna
(men , class 2), selepe (axe, class 7) - dilepe (axes, class 8), etc.) The prefixes in the
column 'other classes' refer to the «infinitive class» (15) and three locative classes
(16-18).
I have adopted the numerical labelling that was established by Meinhof
(1899), and is now used as a standard for all Bantu languages. Thus, the table
contains the standard presentation of these noun classes.

singular classes plural classes other classes

cl. number prefix cl. number prefix cL number prefix
1 mo- 2 ba-
1a -- 2a bo-
3 mo- 4 me-
5 le- 6 rna-
7 se- 8 di-
9 N- 10 diN-
11 lo- 10 diN-
14 bo- 6 rn a-
15 go-
16 fa-
17 go-
18 mo -
Table 1. Noun class prefixes

Whether or not, or to what degree, the Bantu noun classes reflect semantic
categories, is a long debated subject. Some have more or less rejected the idea that
these classes are semantically based. Others have suggested that the system was
semantically transparent at an earlier stage in history, and that the present
system, although it may show a few traces of semantics, largely has become a
formal system. Contini-Morava (1994:6) puts forward the following reasons to be
sceptical towards such a position: Firstly, no modern language with such

1 1a and 2a should not be counted as separate classes, as their concordia! elements are the same as

those of class 1 and 2. Note also that class 6 and class 10 are mentioned twice in the table . They
should, of course, not be counted twice.

-2-
elaborate noun class systems as we find in Bantu, is attested with completely
transparent semantics. Hence, it seems implausible to attribute a property to an
ancestral language that has not been attested in any language that we have direct
knowledge about. Secondly, how is it possible for the Bantu noun class systems to
persist essentially intact, at least formally, for so long (some 3000 years or more),
even after they apparently have become «messed up systems» full of anomalies?
In other words: why hasn't the system broken down, and led to a reduction of
classes, leaving the speakers with a couple of transparent classes like animate -
non-animate for instance?
Like Contini-Morava (1994), my own hypothesis has been that the noun
classes in modern Bantu languages to a large extent are semantically organised. Of
course this is not a new or original standpoint: Quite a number of bantuists and
linguists have struggled to disentangle the apparently confusing semantics of
various classes in different Bantu languages.

Outline of the thesis

This thesis has been structured as follows:

Chapter 2
presents brief accounts of previous and contemporary research on the
semantics of Bantu noun classes.
Chapter 3
introduces the theoretical framework that I have used as a starting point for
my own analyses.
Chapter 4
provides suggestions for semantic analyses of four selected noun classes. These
suggestions are compared to the contributions of other bantuists and linguists.
Chapter 5
describes and discusses a psycholinguistic experiment and its results.

At last, some final conclusions are drawn in chapter 6.

Throughout the text I have used double quotes (" ") for quotations only. Single
quotes (' ') have been used for terms (theoretical terms as well as terms for
semantic groups established in the thesis), whereas double brackets (« ») have
been used in all other cases.

-3-
-4-
 Chapter 2. Previous research

A number of bantuists and linguists have struggled to disentangle the seemingly

confusing semantics of the noun classes in different Bantu languages.
Some have written extensive grammars, where a description of the noun
classes is only a minor part of the work. Others have explicitly aimed at an
analysis of the semantics of the noun classes as such.
It would be impossible within the limits of this thesis to provide an
extensive overview of all the work that has been done on Bantu noun class
semantics.
I have chosen to relate my own work to a small selection of authors, who
are either proponents of certain viewpoints typical for their time, or who have
provided the field with novel works.

The nineteenth century bantuists seem to have taken it for granted that the noun
classes were based on semantics.
Torrend (1891) was one of the first who wrote an extensive comparative
grammar of southern African languages, including a presentation of the noun
classes of 36 Bantu languages.2 He explains his view on the noun classes in this
way:
In the Bantu languages we find no genders based on sex, but instead other
genders or classes of substantives, based principally, as I hope will appear in
this chapter, on the degree of unity and consistency of those things of which
they are the names, as determined by their natural position and shape, their
proper motions, effects, relative strength, etc.
(Torrend 1891:63)

I interpret him as referring to «things as they appear in the real world».3 It seems
that he is expecting the noun classes to reflect certain «natural groups of things in
the world». But that does not (perhaps surprisingly) lead him to present a pure
«taxonomy of ' kinds' », in the usual sense. On the contrary, he obviously feels free
to connect different «groups of things» based on different types of similarities (cf.
chapter 2). For example, when he sketches the semantics of class 3, the way of

2 Bleek (1862) was even earlier, and he is more widely known, but I have chosen to refer to Torrend
in my thesis, as his semantic descriptions are more exhaustive.
3 As mentioned in the introduction I reserve double quotes (" ") for quotations and single quotes (' ')

for terms. Double brackets(« »)are used elsewhere.

-5 -
describing different «sub-groups», all belonging to the class, indicates that he in
fact implicitly employs metaphorical meaning extensions:4

[... ] the substantives which belong to this class are principally: -

1. The names of such complete trees and plants as stand up without support [...]
2. The names of such tools or artificial objects as remind one of the form of a
tree by having branches or bushy parts[ ... ]
3. The human and animal body [... ] as also such of its parts as branch off in
some manner, growing out into accessory parts, or move up and down[ ... ] The
same may be said of the similar parts of trees .
(Torrend 1891:79)
Meinhof (1906) who has been considered the «father of Bantuistics», treats the
semantics of the noun classes in much the same way as Torrend does.
Meinhof does not explicitly clarify his conception of the noun class
semantics, but he organises his treatment of each class according to claims about
which semantic groups that the class exhibits. He uses data both from the
reconstructed Urbantu and from several contemporary Bantu languages.
Like Torrend, he makes use of different types of meaning relations,
assuming, it seems, that different «groups of things» within the same class are
semantically connected. From the way he describes the class meanings it is clear
that he believes that each class originally had one single meaning, that only later
has extended into several different related meanings. For example, he claims that
class 6 is "urspriinglich dualisch" (originally dualic), and that "[a]us dem
dualischen Sinn hat sich ein kollektiver entwickelt", and "[a]us der kollektiven
Bedeutung ist die abstrakte abzuleiten" (Meinhof 1906:8-9).5 Sometimes it seems
that he is lead into deep waters, because he feels obliged to connect even the less
easily connectable groups of nouns. For example, he claims that class 5 nouns are
the singular counterparts of the "dualisch" (dualic) class 6 (1906:8). Confronted
with the fact that most (if not all) Bantu languages include 'fruits' in class 5, he
suggests the following connection between the «dual» and the fruits:

Friichte zerfallen entweder von selbst in zwei Teile oder werden vom
Menschen, da er zwei Hi:inde hat, regelmaBig in zwei Teile zerlegt.
(Meinhof 1906:10)6

4 I have not included the exhaustive list of his "class 3 sub-groups", as this is only meant as an
illustration of the type of groups that he proposes.
5 In English translation: "from the dualic sense a collective has developed", and "from the
collective meaning the abstract is derived".
6 In English translation: "Fruits either divide into two parts by themselves, or are regularly
divided into two parts by humans, because tltey ltave two ltands."

-6-
As the field of linguistics developed, and structuralist approaches became the
«leading trend» (in the beginning of this century), it must have become
increasingly difficult to operate with the somewhat «loose» type of semantic
explanations that many of the «old bantuists» had used. If one could not establish
one semantic feature (or bundle of features) that represented the necessary and
sufficient criteria shared by all members in a class (as in a «classical» or
Aristotelian category), one would have to conclude that the class was not
semantically based.
For some decades the whole issue of the semantics of the Bantu noun
classes was questioned among many linguists. It was argued that the noun classes
represented purely formal systems. An often quoted proponent of such a view is
Richardson (1967). He rejected not only semantic descriptions of contemporary
Bantu languages, but even the idea that the classes could have a semantic origin:
He concludes a discussion with the following statement:

[... ] it is impossible to prove conclusively by any reputable methodology that

nominal classification in Proto-Bantu was indeed widely based on conceptual
implication [... ]
(Richardson 1967:378)

However, it seems that such an absolute rejection of semantic content in the

noun classes has been difficult to live with for the many authors of Bantu
grammars and textbooks.
Cole's (1955) standard grammar of Setswana is not an exception:

Although most of the noun classes are miscellaneous in content, certain trends
of significance are recognizable. Thus most personal nouns are included in
Class 1 [1/ 2], most names of animals in class 5 [9 / 19], and most abstract nouns in
class 7 [14/ 6], though such nouns are also found in other classes. Sometimes the
same stem may be used in different classes, resulting in a change or
modification of its significance [... ].
(Cole 1955:69)

The typical approach, it seems, has been pure listing of noun groups described as
semantically similar. Cole's «semantic description» of the content of class 3/4 and
5 I 6 represents an example:

Nouns of this class are miscellaneous and impersonal. They include names of
parts of the body, tools and instruments, animals, a few special terms applied
to domestic animals, natural phenomena, and most names of trees and plants.
(Cole 1955:75, about class 3 / 4)

- 7-
 This miscellaneous class includes, among others, names of parts of the body,
animals, plants, natural phenomena, some collective nouns, and a number of
personal nouns.
(Cole 1955:79, about class 516)

Denny and Creider (1986) introduce a new and radically different approach to the
noun class semantics. They criticise the typical conclusion of the time, that:

[... ] while all or most members of certain natural classes of entity may be
placed in the same class (familiar examples are the placement of trees in
class 3 I 4 and fruits in class 5 I 6, the classes themselves have no overall
intrinsic semantic content (with the frequent exception of class 112 for
humans).
(Denny & Creider 1986:217)
(My italics)

They present a study of Proto-Bantu (PB) noun forms, and conclude from it that:

[ ... ] PB noun prefixes realized a semantic system where each prefix was
associated with a particular characteristic meaning. [... ] [W]e will be
claiming that the bulk of the noun prefixes were associated with
configurational or shape meanings.
(Denny & Creider 1986:217)
(My italics)

They use as «direct evidence» an examination of Proto Bantu vocabulary, and as

«indirect evidence» a discussion of noun classifier sets found in other languages
throughout the world.
The figure below reproduces Denny and Creider's (1986:219) presentation of
Proto-Bantu noun class semantics for count nouns7:

count

kind

•nima~ct
~
~ded ~
~
unit collection
~
human animal
718

~ 10 J/2 9/tO
unit collection unit collection non-extended extended
5,14 6 3 4 9 11

7 They present the semantics of mass nouns in a separate diagram.

- 8-
Batibo (1987) represents the rather commonly held view that the contemporary
Bantu noun classes are «distorted» versions of an earlier salient system. He uses
Proto-Bantu material, and aims at showing that the ancient Bantu had developed
a noun class system where each class exhibited only one «class meaning».
He suggests the following «class meanings» for the classes 1/2 - 7/8 (Batibo
1987:18):

cl. 1/ 2: "human"
cl.3/4: "planted"
cl. 5 : "itematized"
cl. 6 : "amassed"
cl. 7/8 : "marked"

In his conclusion Batibo (1987:21) writes that his:

[... ] study has shown that by u sing the typically Bantu noun stems which are
assumed to represent the earlier Bantu nouns, it is possible to arrive at common
semantic features for each cla ss or grammatical gender. [... ] It has been
demonstrated that the noun class systems in the contemporary languages have
lost the original regularity due to both formal and semantic shifts.
(My italics)

Spitulnik (1987:7) rejects the position that the " 'random', 'non-systematic' state of
present day Bantu languages is [... ] a semantically white-washed relic of a more
homogeneous proto-system [... ]".
She also questions the «tradition» of accounting for the semantic
organisation of the noun class systems by "giving an inventory of the typical
referents of the morphologically distinct classes" (Spitulnik 1987:8). (Cf. the «old
bantuists»)
Spitulnik (1987:8-11) gives Denny and Creider credit for their "thought-
provoking study", and characterises their approach using a «feature analysis» as
more promising than the «natural kind approach», "since at a higher level of
abstraction (describing features of referents rather than types of referents) a wider
range of semantic relations can be covered". While also admitting that some of
their conclusions are supported by her own work on ChiBemba, sh e criticises
their work because their analysis is only based upon "questionable proto-forms

- 9-
with questionable proto-class assignments"8, completely leaving out
contemporary Bantu languages, and she questions "the advantage of reducing the
noun class system to sets of binary oppositions".
In her own contribution, the terms 'notional value' and 'notional core' are
central:
We can characterize Bantu noun classes as having central notio11al values, i.e.
notional values which quite pervasively characterize the denotata of a class.
However, for most classes there are a number of notional values, no one of
which appears to dominate over the others.[ ... ] Since these central notional
values are often semantically linked to one another, together they are like a
' notional core' of a form class. We sh ould note however, that th is type of
'notional core', unlike those of Indo-European gender systems:
i. usually consists of a number of notional values, not just one, and
ii. characterizes a great m ajority (if not all) of the denotata of the class in
question.
Together, (i) and (ii) yield a picture of Bantu n oun class systems as intricately
structured by notional values, w ith a minimum of randomness to noun class
membership.
(Spitulnik 1987:25)

As an example, the figure below reproduces Spitulnik's representation of the

'notional core' of ChiBemba class 3/4 with all the 'notional values' and the
associations between these values:

'extending over time' 'in ground'

(Spitulnik's figure 9, 1987:61)

Contini-Morava (1994:6) also suggests that the problems of characterising the

semantics of the Bantu noun classes "lie not with the languages, but with the
assumptions about the nature of linguistic categorisation that are brought to bear

8 She goes into details about why she claims that the proto-form s and the proto class assignments
are doubtful. I will not include those details here.

- 10-
on this question". She rejects the idea that one has to "abandon the search for
semantic coherence and settle for a heterogeneous list" when no "set of common
properties shared by all nouns in a given class" can be found.
Contini-Morava points out that within the framework of Cognitive
Grammar (d. Lakoff 1987; Langacker 1987, 1990; Rudzka-Ostyn 1988):

(it) has been argued that membership in a given linguistic category (for
example, a noun class) may be based on multiple criteria, including 'family
resemblances', metaphor, and metonymy, and that linguistic categories may
exhibit an internal structure in which some members of the category are more
central, or prototypical, and others are more peripheral.
(Contini Morava 1994:6)

Based on a noun database of contemporary Swahili, and with Cognitive

Grammar as a starting point, she presents semantic networks ("schematic
representations of semantic structure") for some of the Swahili noun classes.
The following example is a graphical representation of the semantic
network for Swahili class 7:

utilitarian objects small enough to hold in hand

l ··········
•mall entdit in gene" I ................................

I I I . . . . . . . . . . . . . . . . . . . ..
I
small
I
small immature pieces/ small concrete
artifacts animals object
things.,,
,,..
~
assoc. with
,,. ,,.•' l
:
'••,,,.,, verbs (product,
,,.•'' i '••,,, implement, etc.)
Il l II I
···'•', ~
~
...,..,
''•,,,

p;~t of
• • 0 '•

shortened pointed

/)"8' ;~:::</ •ub•t•~: . . . . . . . . . . . '

---
people with
···· ail~ents
associated
physical with body
defect/lack parts
similarity I
manner

(Contini M orava's class 7 figure, 1994:13)

-11-
 The top-to-bottom organization of the diagram moves from the more general
to the more specific, but the diagram is not intended to be a 'taxonomy' in the
technical sense[ ... ]. I have borrowed the conventions used by Langacker (1988)
for the representation of a linguistic category. Langacker defines two basic
types of semantic relationships among elements in a category: (a) relations of
'schematicity' in which one element is an 'elaboration' or 'instantiation' of
another, more abstract element (represented by solid lines in the diagram); (b)
relations of 'extension', in which some feature specifications are suspended or
modified, while other features are retained (represented by dotted lines in
the diagram).
(Contini Morava 1994:10-11)

The main difference between Spitulnik's and Contini-Morava's approaches is

that Contini-Morava introduces a notion of schematicity (the idea that some
elements in a category may be more generalised or abstract than others) and the
possibility that some category member(s) may be more central than others.9

This has been a brief presentation of some approaches to the semantics of Bantu
nominal classes.
It is interesting to note that the earliest and the most recent approaches
referred to here have something in common: the implicit or explicit use of
different principles of semantic associations that bind together various groups of
nouns within each single noun class. That is a viewpoint that will be carried on
in the analyses in this thesis. There is, however, a big difference between the early
and the recent approaches: Whereas «the old bantuists» seemed to believe that
their semantic descriptions pointed out groups of objects or phenomena in the
«real world», linguists like Spitulnik (1987) and Contini-Morava (1994) make
clear that they are not describing groups of «real-world objects or phenomena»,
but rather the semantic organisation of human concepts.
I will base my own analyses on the assumption that noun class semantics is
not about grouping and labelling «real-world objects or phenomena». Rather, I
believe that it reflects parts of human conceptualisation.
In chapter 3 I will present briefly parts of the theoretical framework that is
known as Cognitive Grammar. In chapter 4 I will make use of some of the
«theoretical tools » provided by such a framework, when I suggest semantic
analyses of Setswana noun classes.
However, together with my analyses, I will refer to and discuss ideas and
insights of all the aforementioned bantuists and linguists.

9 See chapter 2 for a further explication of 'schematicity' and 'centrality'.

-12-
Chapter 3. Theoretical approach

Introduction
It is a startling fact that such a semantically intriguing area as the Bantu nominal
class semantics is something to which relatively little attention has been paid;
much less than one could expect considering the large number of Bantu
languages spoken by millions and millions of speakers, and the nominal class
system exhibiting such a prominent part of these languages.
The reason can be ascribed, I believe, to the commonly held view (in
linguistics) that categories are to be defined in terms of a set of necessary and
sufficient features shared by all category members, and that these features
correspond to measurable and objective aspects of external reality. I will refer to
this view of categories as the 'classical approach'.
With such an approach as a starting point (explicitly or implicitly) it is
more or less self-evident that one has to give up a semantic description of the
noun class categories: What could possibly be the «objectively» common features
of a morula tree, a pestle, a dance, a leg and a pot used for cooking medicine?
These are all examples of (translated) Setswana class 3 nouns. Failing to find the
common features that would be the necessary and sufficient defining criteria for
class membership, would leave us with the conclusion that these class 3 nouns
are not members of the category 'class 3' by virtue of any semantic principle.
The 'classical approach' to categories includes an implicit assumption that
the categories we are dealing with have an existence «out there in the real world»
independent of the minds and conceptual apparatus of the human beings that
perceive them.
Taylor (1989) labels the 'classical approach' to categories the «criterial-
feature approach», and suggests that within such a framework

[...] features in a sense pre-exist the entities they define: categories result from
the combination of features selected from a pre-existing universal inventory,
and things are assigned membership in a category by a process of 'checking off'
their feature specifications.
(Taylor 1989:524-5)

Seeing that the 'classical approach' to categories inevitably leads to the conclusion
that most Bantu nominal classes, because of their apparently rather
heterogeneous nature, can not be described in term of semantics, we are faced by
two options: Either we have to settle for the «arbitrariness conclusion», which is
clearly unsatisfactory considering the rather commonly expressed intuition

-13-
(among grammarians and language users alike) that there is «something
semantic» about the noun classes. Or we have to take another look at categories
and how they may be structured.
In this chapter I will present the starting point from which my own
suggestion for a semantic treatment of the Setswana noun classes departs.
The noun classes are examples of linguistic catego ries. Choosing a
conception of 'category' in general is essential for any treatment of a
phenomenon like nominal classes.
In doing so, I will inevitably touch upon some very central philosophical
and linguistic debates concerning categorisation in general. I will, of course, not
try to embrace neither the width nor the depth of any of these debates. That
would reach far beyond the scope of this thesis, which aims at throwing some
new light upon the semantics of the Bantu noun classes. Rather, I have restricted
this presentation to covering briefly only those fundamental assumptions
concerning categories, and those «theoretical tools» that have enabled me to
present a suggestion for a semantic analysis of the Setswana noun classes. The
choice of approach has been governed by an evaluation of earlier attempts to
«solve» the problem of nominal class semantics in Bantu languages, and the
dissatisfaction with the «arbitrariness conclusion».

«Family resemblance»
A relatively recent alternative to the 'classical approach' to categories, is an
approach based on networks of related senses (d. Lakoff 1987, among others).
Although I have labelled this alternative «relatively recent», it is clear that it is
built on many important insights stemming from linguists, psychologists,
anthropologists and philosophers, not all of whom are «recent».
That some categories may be structured as "complicated networks of
similarities" rather than in terms of a set of common features, was suggested by
the philosopher Wittgenstein as early as 1953. His conception of such categories is
commonly known and referred to as the 'family resemblance' approach to
categories.
His famous and much-quoted passage about the category 'games', illustrates
in an illuminating manner what he means by a "network of similarities":

Consider for example the proceedings that we call "games". I mean board-
games, card-games, ball-games, Olympic games, and so on. What is common to
them all? - Don't say: "There must be something common , or they would not be
called 'games' " - but look and see whether there is anything common to all. -
For if you look at them you will not see something th at is common to all, but

- 14 -
 similarities, relationships, and a whole series of them at that. To repeat:
don't think, but look! - Look for example at board-games, with their
multifarious relationships. Now pass to card-games; here you find many
correspondences with the first group, but many common features drop out, and
others appear. When we pass next to ball-games, much that is common is
retained, but much is lost. - Are they all 'amusing'? Compare chess with
noughts and crosses. Or is there always winning and loosing, or competition
between players? Think of patience. In ball games there is w inning and
loosing; but when a child throws his ball at the wall and catches it again,
this feature has disappeared. Look at the parts played by skill and luck; and
at the difference between skill in chess and skill in tennis. Think now of games
like ring-a-ring-a-roses; here is the element of amusement, but how many
other characteristic features have disappeared! And we can go through the
many, many other groups of games in the same way; can see how similarities
crop up and disappear.
And the result of this examination is: we see a complicated n etwork of
s imilarities overlapping and criss-crossing; sometimes overall similarities,
sometimes similarities of detail.
I can think of no better expression to characterize these similarities than
"family resemblances"; for the various resemblances between members of a
family: build, features, colour of eyes, gait, temperament, etc. etc. overlap and
criss-cross in the same way.- And I shall say: 'games' form a family [... ]
(Wittgenstein 1958:31-2, first edition 1953)

In other words, Wittgenstein suggests that we can not find a uniform set of
features that constitute the meaning of a linguistic expression. What we can find
is a network of similarities between different uses of an expression. 10

Prototypes
While Wittgenstein (1958, first edition 1953) pointed out the possibility that a
category might be internally structured as a network of similarities, others have
focused on the possibility that that within such networks some member(s) might
be more central than others. The most central member can be called a prototype.

10 I have noted that Wierzbicka (1990:356-8) asserts that "the time has come to re-examine his

[Wittgen stein's] doctrine of 'family resemblances' which have acquired the status of
unchallengeable dogma in much of the current literature on meaning", and that she claims that
it is possible to define the concept of ' game' in terms of essential components that all 'games'
have in common. I will , however, neither go into details about the nature of the definition
that she proposes, nor take a stand on her critique, as my intension here only is to draw
attention to a central precursor of the conception of 'semantic networks'. And I still think that
Wittgenstein's example represents an elucidation of a phenomenon that exists in languages,
irrespective of the actual example: 'games'.

-15-
 According to Tsohatzidis (1990:1) a prototype approach to linguistic categories
involves two central claims:
1) "[F]or the most part, linguistic categorization exploits principles that are not
specific to language but characterize most, if not all, processes of cognition."
2) "[O]ne of the basic principles according to which cognitive and (under their
influence) linguistic categories are organized is the prototype principle - that is to
say, a principle whereby elements are assigned to a category not because they
exemplify properties that are absolutely required of each one of its members, but
because they exhibit to a greater or lesser extent (or are simply expected to exhibit
to a greater or lesser extent) certain types of similarity with a particular category
member that has been (naturally or culturally) established as the best example (or
prototype) of its kind."
The psychologist Rosch (1973), because of her extensive experimental
research, has become known as the initiator of this theoretical tradition. The
tradition is, however, not unambiguous. According to Taylor (1989:59) " [t]here are
two ways in which to understand the term 'prototype'. We can apply the term to
the central member, or perhaps to the cluster of central members, of a category.
Thus, one could refer to a particular artefact as the prototype of CUP.
Alternatively, the prototype can be understood as a schematic representation of
the conceptual core of a category. On this approach, we would say, not that a
particular entity is the prototype, but that it instantiates the prototype."11
It is, however, not automatically clear exactly what kind of status a
«schematic prototype» has: for example in Langacker's glossary (1987:492) he
defines a prototype as "[t]hat unit in a schematic network which is naturally most
salient, most often thought of, most likely to be chosen as representative of the
category." But then he adds: "In a generalized sense, the term is also adopted for
the standard in a categorizing relationship based on extension rather than
schematicity." (Cf. the section below for a clarification of terms.) As far as I can
understand, there is a conflict between these two definitions: The second
definition will always be compatible with the first one, but the opposite doesn' t
need to be true. If one postulates a «psychologically salient» prototype in a
category, it will of course by definition be possible to claim that this prototype

11 In a footnote Taylor (1989:59-60) calls attention to Rosch' later apparent rejection of both these

alternatives, as she according to him "content[s] herself with a characterization of

prototypicality merely in terms of its experimental effects (membership judgements, reaction
times, etc.)".

- 16-
represents the standard to which some other member in the category is
compared. However, in a large network consisting of several comparisons, one
will easily find, within the more peripheral parts of the network, standards to
which something else is compared, but which exhibit no saliency.
I will touch upon this issue later in this theses, but I will not make it a
major question. That would require a study of another kind, and would reach far
beyond the scope of this thesis, which aims at suggesting semantic structures for
selected Setswana noun classes. While the question of prototypicality clearly
touches my treatment of the classes, it is not the main point of concern.

Schematic networks
In his Foundations of Cognitive Grammar (1987 and 1991) Langacker aims at an
alternative theoretical approach to language in general, built on the general
assumption that linguistic organisation can be characterised in terms of cognitive
processing. He admits that cognitive grammar is "not a finished or formalized
theory" and is "more realistically viewed as an evolving conceptual framework".
(Langacker 1987:2) Nevertheless, he has certainly, I would argue, provided the
field with some «theoretical tools» that, among other things, have proved useful
in trying to come closer to an understanding of the fascinating, but apparently
chaotic field of Bantu nominal class semantics.
As will become clear in chapter 4, the semantic studies of Bantu noun
classes may benefit from Langacker' s conception of schematic networks.
A schema for Langacker (1987:37) is:

an abstract characterization that is fully compatible with all the members of

the category it defines [... ]; it is an integrated structure tha t embodies the
commonality of its members, which are conceptions of greater specificity and
detail that elaborate the schema in contrasting ways.

A schematic network, according to him (Langacker 1987:380) includes:

a finite set of nodes, which can be linguistic structures of any sort. Every node
is linked to at least one other through a categorizing relationship, of which
there are several types: full schematicity (at varying elaborative distances);
partial schematicity (with varying degrees of [standard / targe t]
discrepancy); or a bidirectional relationship decomposable into component
relations of the aforementioned kinds.

In the following I will refer to the two relation types 'full schematicity' and
'partial schematicity' as instantiation relationships and extension relationships
respectively.

-17-
Extension relationships, according to Langacker {1991:548), consist of a
"categorizing relationship involving some conflict in specification between the
standard and the target". (A standard is here understood as a basis for comparison
or categorisation, while a target is the unit that is being evaluated relative to the
standard.)
As a notational convention, Langacker uses broken-line arrows to
represent relationships of extension, and solid arrows to represent instantiation
relationships:

TREES ································HI••· LONG BODY PARTS

In the example above, the direction of the «extension arrow» shows that 'trees'
are represented as the standard to which 'long body parts' is compared.12 The
relationship is clearly not one of instantiation ('full schematicity'), as 'long body
parts' can not be considered a subcase of 'trees'. Neither can 'trees' be considered a
subcase of 'long body parts'. Rather, there are obvious conflicts in the
specifications of the two concepts. Nevertheless, they can be compared, and it is
not difficult to perceive similarities between them. One such similarity is that the
notion 'long' can be extracted from both concepts.

An instantiation for Langacker (1987:489) is an elaboration of a more abstract

schema "consistent with its specifications, but [... ] more fully and precisely
specified". In other words it represents a «subcase of a schema». Thus, in the
example above the shared quality of being 'long' can be represented as another,
and more generalised schema ('long'), and 'trees' and 'long body parts' can be
represented as instantiations of that more abstract schema:

LONG

TREES .................................1111· LONG BODY PARTS

figure 1

12 Since the aim here only is to clarify the reading of the networks that I will propose later, I will

not discu ss the reasons for choice of direction of the arrow in this particular example.

-18-
We can of course elaborate the two instantiations of 'long things' further as well.
Then we have a graphical representation of a tiny part of what I will later suggest
is the schematic network of Setswana class 3:13

LONG

TREES ................................. 111••· LONG BODY PARTS

PINE OAK BIRCH THUMB LEG BACK

figure 2

It is of great importance to note that even if such a representation looks

equivalent to a taxonomic hierarchy with superordinate and subordinate nodes,
all the nodes are to be understood as conceptual schemas: Even the more specific
'oak' is to be understood as a generalised conception, and not as one or all
concrete oak(s) in the real world. The «hierarchy» is only a graphical method of
conveying degrees of specificity and generality.
In sum: an abstract schema reflects a commonality between two (or several)
more elaborated concepts. This commonality is extracted by the language user on
the basis of a comparison between a 'standard' (or prototype in the most
generalised sense) and a 'target'. When the comparison results in a judgement of
similarity (of a certain type and degree), a more generalised (more abstract)
schema can be posited. In other words: Langacker (1987:372) claims that three
cognitive entities figure in a similarity judgement: the prototype, the entity that is
compared to the prototype, and finally: a representation of their perceived
similarity (referred to as SCHEMA in the figure below):

13 I have left out the extension relationships between the most specific concepts in this example, as

that is not needed for this general presentation of the model.

-19-
 SCHEMA

PROTOTYPE .................................lilt•· EXTENSION

figure 3

Thus, we see that Langacker's model provides a synthesis that treats the prototype
model and the model based on schematicity "as special cases of a unified
phenomenon and relates them in integral fashion to the network conception of
complex categories" (Langacker 1987:371).
For convenience, the convention is that instantiation relationships are
represented vertically, and extension relationships horizontally. That eases the
interpretation, especially in very complex networks of relations.

Metaphor, metonymy and synecdoche

According to Taylor (1989:122) metaphor and metonymy are "two of the most
important processes whereby different meanings get associated". The association
processes referred to are equivalent to what Langacker (1987) refers to as relations
of extension. So in a schematic network, we need the notions of metaphor and
metonymy to explicate in more detail the different extension relationships.
While originally being a term from traditional rhetoric (referring to figural
speech like «she has a heart of stone»), metaphor has become an object of
research for «cognitive linguists» (see for example Lakoff and Johnson 1980).
After outlining and criticizing the tradition of capturing the essence of metaphor
by a notion of «violation of selection restriction», Taylor (1989:132-3) writes:

[... ] the cognitive paradigm sees metaphor as a means whereby ever more
abstract and intangible areas of experience can be conceptualized in terms of
the familiar and concrete. Metaphor is thus motivated by a search for
understanding. It is characterized, not by a violation of selection restrictions,
but by the conceptualization of one cognitive domain in terms of components
more usually associated with another cognitive domain. (My italics.)

Metonymy has in traditional rhetoric been defined as "a figure of speech

whereby the name of one entity el is used to refer to another entity e2 which is
contiguous to el" (Taylor 1989:122). When somebody says «Peter really likes milk;
every morning he drinks two glasses», we easily understand that he is not
literally drinking the glasses, but their content.

-20-
 Taylor (1989:122-30) argues that "the essence of metonymy resides in the
possibility of establishing connections between entities which co-occur within a
given conceptual structure" (my italics), and that metonymy's place in language is
of much greater importance than the definition above indicates. He claims that
"metonymy turns out to be one of the most fundamental processes of meaning
extension".
The particular case where reference to the whole is made by reference to a
salient part (like referring to a person by the term «curlyhead») is also sometimes
subsumed into the category 'metonymy'. In my analyses I shall refer to this, not as
metonymy, but as synecdoche.

Polysemy
Polysemy traditionally has been regarded as the phenomenon that one
word/lexical item has a range of different, but related meanings.
More recent research, however, indicates that grammatical categories also
may exhibit such clusters of related meanings (cf. for example Lakoff 1987 and
Croft 1990).
The 'schematic networks' that are proposed in my semantic analyses of the
Setswana noun classes, can be seen as graphical representations of polysemou s
categories. Thus, the networks should be read as 'polysemy charts'. In this case the
elements that exhibit a polysemous structure, are the noun class prefixes (or
rather, the whole noun class concordia! system).

In the introduction to this chapter I gave a few examples of Setswana class 3

nouns that immediately might seem difficult to connect to the same category
(class 3) on a semantic basis: morula tree (morula), pestle (motsh e), dance
(mmino), leg (moletse), and pot used for cooking medicine (morurelo). In chapter
4 we will see that it is possible to point out motivations for the connection of all
these nouns (or rather: the concepts that these nouns refer to) within a large
'network of similarities'. To do so, we need the notions of schematicity, prototype,
instantiation and extension relationships (including metaphorical and
metonymical extensions) that has been briefly presented in this chapter.

-21-
-22-
Chapter 4. Analysis
4.1. Introductory remarks
In this chapter I will present my suggestions for semantic analyses of the
Setswana noun classes 3, 5, 6, and 7. The point of departure has been an
assumption that adopting Cognitive Grammar's conception of categorisation and
semantic networks (as briefly accounted for in the preceding chapter) might be
useful for the understanding of Bantu nominal class semantics.
For each class I have suggested a network of semantic connections. I first
present the complete network as a graphical representation, and then go into
more detail about each group of concepts and the different semantic relations that
exist between them.
In the graphical representations 'schemas' are represented as square boxes.
These are connected by broken-line arrow, representing extension relationships,
and solid-line arrows, representing instantiation relationships. It will be clear
from the text, that the extension arrows don't necessarily imply the same kind of
extension relationships. The text will clarify in each case the nature of the
semantic extension in question (whether it is based on metaphor or metonymy,
what the basis for the metaphor/metonymy is, etc.).
The most abstract (or generalised) schemas in each class are the topmost
ones in the figures. They represent generalisations over many other schemas, and
are not themselves represented as instantiations of any other schema. I have
called them 'class schemas', because they can be used as general characterisations
of the classes' semantics. I have also used the term 'local schema' in a few cases.
The local schemas also represent generalisations over two (or more) other
schemas, but they are not important to the overall semantic characterisation of
the classes.
For each class I have suggested one schema as a class prototype. In doing so,
it is evident that I have chosen a schematic interpretation of the notion
'prototype' (cf. chapter 3). I have not, however, equated 'prototype' with a
'standard for comparison' . If standard and prototype are interpreted as the same
phenomenon, a semantic network will exhibit multiple prototypes. In one
respect that might be justified; for in most 'similarity judgements' one entity is
the standard to which another entity is compared. Thus, the standard might often
represent the most typical (prototypical) instantiation of the abstract schema
emerging from the comparison. However, with such a perspective we will miss a
conspicuous aspect of the semantic patterns represented here as «class networks»,

- 23-
namely the tendency that one schema in the network seems to be the main point
of departure for the other extensions. Accordingly, I have posited one prototype
schema for each class (indicated in the figures by boxes with thick lines).
In some cases there are additional independent reasons for suggesting that
a schema has the character of a prototype. Such reasons could be: productivity,
saliency, relative size (number of instantiations), and so forth. Whenever such
factors are relevant, they are pointed out in the text. That such factors can be
found at all, suggests that my main interpretation of 'class prototype' (that the
prototypes mainly can be established on the basis of central position within the
schematic network) perhaps is unnecessarily weak.
Within the proposed schematic networks one could easily «break up» the
least generalised schemas and then suggest additional generalisations
(represented as new schemas) over smaller groups of concepts. My network
representations are of course not meant to be the full and only representations. I
have focused only on those generalisations that seem to offer very general
semantic characterisations of the classes.
In the text, the notions 'group of concepts', 'group of notions', and 'group of
nouns' are mingled together. They all refer to the same: groups of concepts as
mental represen tations .

In the coming sections, I begin the presentation of each noun class by referring to
the semantic characterisation given in the standard grammar of Setswana. Then I
briefly present (chronologically) some extracts from the works of those bantuists
and linguists whose works I have referred to in chapter 2, before I explicate my
own suggestions for semantic analyses. After the treatment of each class I discuss
my own contribution in relation to the contributions provided by the other
authors referred to. A more general conclusion is drawn in the last section of the
chapter.

4.1.1. Selection of data

The analyses are built on a study of a random sample of nouns from the
Setswana - English - Setswana dictionary (Matumo (comp.) 1993). In order to
work with multiple categorisations of each noun (in the process of searching for
larger «class patterns»), I established a data base that enabled me to sort
information tagged onto each noun in many different ways.
I started by entering each lOth. noun of all the singular classes plus the non
count class 6 into the data base, that is: class 1 (and la), 3, 5, 6, 7, 9, 11, 14. After the

-24-
 first tentative attempts to arrive at relatively consistent analyses, I realised,
however, that the project would benefit from limiting the number of classes
under consideration, while enlarging the data material for the selected classes. I
chose to work with the classes 3, 5, (non count) 6, and 7. The data base entries for
these classes were doubled; the data material now consists of a random sample of
each 5th. of these nouns in the dictionary, altogether 620 nouns. Class 3, 5, and 7
are large, and semantically they appear to be highly heterogeneous (thus, referred
to in grammars as «miscellaneous» classes). The non count part of class 6 is not as
large, but the connection between semantics and the «portion» of terms that a
language grammatically treat as uncountable, is an interesting one. The classes
that I decided to leave out of the analyses then were:
- the «human» class 1 (including exclusively human terms), and the small sub
class 1a with a majority of human and animal terms (often animals with human
connotations in one way or another),
- the large prefix-less class 9 which sometimes is referred to as the «animal class»
as a lot of animal terms belong here, but which also seems to attract a lot of
borrowings,
- class 11 which in Setswana is small and in the process of disappearing; an
increasing number of class 11 nouns shift to class 5,14 and finally
-class 14 which exhibits (although not exclusively) a lot of abstract concepts.
Class 9 is the only one of these classes that represents as great a challenge as the
classes that I chose to work with. It is a large class, and apart from the numerous
animal terms and various borrowings it also includes a variety of other terms.
Ideally I should therefor have included it, but after working on the other classes, I
soon realised that including class 9 as well would be too big a task for a thesis of
this kind.
In the analyses, I have left out the few animal terms and most abstract
terms in my sample. In order to integrate these terms in a proper manner, I
would have to compare them to the «animal class» 9 and the «abstract class» 14.
When I started to enter the nouns into the database, I had only vague ideas
of how to categorise them. Each noun stem was entered together with
information of class affiliation (singular and plural prefixes), dictionary
definition, source of derivation in cases of derived nouns, and finally very
general semantic classifications, like «concrete object», «human», «plant», etc.

14 It w ould of course be interesting to investigate the ongoing replacement of class 11 nouns into class

5, but it would require a study on its own.

-25-
After that first rough sorting, and after comparing lists of nouns from the
different classes with one another, I slowly became aware of similarities of
various kinds between larger or smaller portions of nouns within each class. For
each new possibly relevant semantic category, I had to return to the data base and
add or replace tags. Reorganising the nouns according to new categories often led
to further suggestions, which made it necessary to add or remove still more tags
in the database.
Contini-Morava (1994:8) points out an important problem with tagging
semantic categories onto nouns in this way, when she states: "In order to create a
database, one has to anticipate which classificatory categories will be useful before
entering the data, in a way guessing at the very analysis that the tool is intended
to help discover."
There is of course no certain way to ensure that I am not only thrusting my
own categories on the data. But it is possible at least to minimise the problem: By
comparing my suggested analyses with other works on Bantu noun class
semantics, and with comparable data from other languages or other parts of
language, it should be possible to avoid the most fundamental
misrepresentations. The aim of a study of this kind will at any rate be a
suggestion for a semantic description, not a presentation of the «true and only»
analysis.

-26-
4.2. Class 3
In this section (4.2) I will suggest that the class 3 network centres around the
prototype 'tree', and that the network of different schemas all related (directly or
indirectly) to the prototype by extensions of varying types, result in the two class
schemas: 'living' and 'long'.
Both schemas reflect categorisation principles that are relevant in many
other languages, namely: degree of animacy and configuration.

First, however, I will present briefly some other suggestions for the semantics of
class 3 in various Bantu languages.

Earlier treatments of the semantics in Bantu class 3

Cole (1955:75) in his standard Setswana grammar gives the following description
of class 3: "Nouns of this class are miscellaneous and impersonal. They include
names of parts of the body, tools and instruments, animals, a few special terms
applied to domestic animals, natural phenomena, and most names of trees and
plants".
In treating the semantics of class 3 in various Bantu languages, most (if not
all) authors have recognised the fact that almost all names for trees, and a lot of
other plants, are found in this class. More or less different generalisations over
this class' members all seem to be compatible with this fact.
Torrend (1891:80) concludes his treatment of Bantu class 3 in this way: "[ ... ]
it seems pretty evident that the predominant notion in this class is that of
«objects which are light, move, change, grow, produce, or, in general which
contain some principle of life and production, a notion intimately connected with
that of 'power of growing up' like a tree»."
Meinhof's (1906:6) concluding description is: "Belebtes, aber nicht
personliches" (living, but not human).
Within Denny and Creider's (1986:220-1) binary system, Proto Bantu class 3
refers to units of 'extended solid figure' (within the larger group of
configurational count nouns). They define extended as "characterized by relative
length in one dimension at the expense of the other two, or less often, by relative
length in two dimensions at the expense of a third". They also discuss the
possibility of a development of 'kinds' from the configurational classes, and then
suggest that class 3 contains plants as far as 'kinds' are concerned.
Batibo (1987:10) claims that ancestral Bantu speakers categorised class 3
nouns as objects "planted" in space or time.

-27-
 Spitulnik (1987:56-7) uses the label 'potent/living' for class 3 in ChiBemba,
and she argues that 'tree' is the prototypical member of this class.
Contini-Morava (1994:10-12) characterises class 3 nouns in Kiswahili as
referring to "entities with vitality (neither human nor prototypically animal}",
and she claims that plants/trees are the most central subcategory.

Proposal for a schematic network of Setswana class 3

My own suggestion for Setswana is (not surprisingly) that class 3 members centre
around the prototypical notion 'tree', and that it is possible to extract from the
network, the abstract class schemas 'living', and 'long'.
I will present graphical representations of the extension relationships and
the instantiation relationships before I go on to describe each schema in the
network. (The numbers in brackets within each «box» representing a schema,
correspond to the numbers in the text dealing with the different schemas.)
For graphical reasons I have chosen to present the instantiation
relationships «belonging» to each class schema ('long' and 'living') in two figures.
They should of course be read as integrated parts of the same network, but
graphical constraints make it difficult to present a readable version with both
shemas integrated into the same figure.

The extension relationships in class 3 can be represented as m the following

figure:

lively (11) {12)

human •11·····11•· «pro~sses
activity L..as
_ thin
_ g_s>>....J•III······· .. ···•·······················•··•······..······················
1
olher

~~~~gs
(8)

·!!······················..
body- (7)
coverings
made of
c::J i
1
ee ................u•.
~
L.::_r. . . . . . . . ll··
objects (4)

~~t::
skin (10) olher (9) •Ill·············· skin ·•11··········..······ ··········II•· long, (3) long objects
related •11·······..·· skin
~;:.:.....-....J ~. :. wooden ·············II• (not (5)
states objects ~ ~ objects wooden)
(6) ; ;
long
bod ,11.......:
l l .
•...................... 11,, obJects
(13)
yparts connected
to ground

figure 4

-28-
Adding the instantiation relationships based on the class schema 'living', this
figure is produced:

living

lively (ll)
human
activity

(2) objects (4)

other (8) ...................11•· madeof
body- plant
coverings material

(3) long objects

skin (10) other ............. 11 1 (not (5)
related •11........... skin
wooden)
states objects

long i,, ..................... b" (13)

bodyparts 111• o Jects
connected
to ground

figure 5

-29-
The next figure shows the instantiations based on the class schema ' long'.
Although only few schemas instantiate this schema, it entails a large portion of
the class 3 concepts in this network (between 45 and 66 % depending on how large
portion of the schema 'plant' that can be considered 'long').

long

lively (11) (12)

human •11·····11•· «processes
activity as things>> ·•ll······································································
objects (4)
other (8)
body· ·········II•· madeof
plant
coverings •II························ body- (7) material
coverings
made of
skin (10) other (9) •III·.. ······· · ·· · L.:s~kin
::..:...___.
related •II··········· skin ob'ects
states objects
(6) l
l
long
bodyparts •II·······'
L....................111, objects (13)
connected
to ground

figure 6

(1) The prototype: trees

In my noun sample 37% of the nouns in this class refer to trees and other plants
(if we leave out pure abstracts and animals). Tree names, as the ones below,
constitute the largest subpart within this group (42%).

mokoso a white thorn tree; (Acacia albida) 15

mooka a species of thorn tree; (Acacia karoo)
moruku d u tambuti tree; Spirostachus africana
mo setlha a species of acacia tree; (Peltophorum africanum)
mo s u camel-thorn; Acacia giraffe

As described for other Bantu languages, trees constitute a highly productive

semantic category within class 3; almost invariably new loan words referring to
trees are incorporated into this class. This is rather remarkable considering the

15 Unless something else is stated, the English translations are those provided by the dictionary

(Matumo 1993)

- 30-
fact that other loan words tend to fall into «default classes» usually lacking a
noun prefix (class 9 in Setswana), or to connect with the phonologically most
relevant class based on similarity between the class prefix and the loan word's first
syllable.
I see 'tree' as the prototypical schema for class 3, as it forms the basis for
metaphorical or metonymical extensions to most other concepts in the class. The
class schemas that can be extracted from the network of extensions, reflecting the
connections between the prototype and the rest of the other groups in the
network, are: ' long' and 'living' (with the further elaboration 'material with
«live» origin').
It is a peculiar fact, however, that the most frequently used generic term for
tree is setlhare , a class 7 noun. One would perhaps expect it to be a class 3 noun,
as are almost all the names for specific trees. However, in most other Bantu
languages ' tree' is a class 3 noun (e.g. Kiswahili: mti , Zulu: umuthi), and
Setswana also have at least one class 3 term for tree, although it is not frequently
used (more).l 6

(2) Other plants

As exemplified below, class 3 also contains a lot of other plants (bushes, grass,
creepers, herbs etc.), as well as some parts of plants.

mmodula the flower on a corn stalk; the outer and loose husk that
clings to corn
mofikapitse a species of creeping plant
mogarugaru a species of river grass
mogokang a hooked thorn bush that grows in sandy places
moragangaka a species of herb, used for healing

l6 Allowing oneself to speculate, one could perhaps suggest that there is a possible link between
set/hare and class 7, based on another meaning of that noun: 'medicine'. As class 7 exhibits as an
abstract schema 'participation in action', with 'instruments' («intermediary in action») as the
prototype, the «action aspect» of trees (as «instruments of healing»: medicine) could perhaps
link it to that class.

-31-
In western scientific terms, trees are often treated as a subgroup of 'plant'. There
is, however, no reason to thrust such an 'expert category' 17 on a semantic network
that is supposed to reflect the generalisations of «ordinary» language users, - not
of skilled botanists. 18
Within the semantic network of class 3, I therefore find it reasonable to
treat plants other than trees, as extensions from the category 'tree' on the basis of
their many similarities with trees. Plants, as well as trees, instantiate the schema
'material with «live» origin', which again instantiates the even more abstract
class schema 'living'.
Many plants will also clearly instantiate the class schema ' long'. Most plants
have long parts, but it would require a thourough study of Setswana flora to
decide accurately which plants that one would perceive as «long gestalts)). In the
n etwork I have reflected this uncertainty with a questionmark on the
instantiation arrow from the class schema 'long' to the schema 'plant'.

(3)Long objects made of wood

These objects can also be regarded as semantic extensions from the prototype: The
semantic relation between 'trees' and other 'long objects' is a metaphorical one,
whereas the relation between 'trees' and other 'wooden objects' is synecdochical.
Both relations are at work in connecting 'long, wooden objects' to the prototype of
class 3. The semantic connection between this group of concepts and the prototype
produces the two schemas 'long' and 'material with «live)) origin'.
The following are examples of concepts belonging to this group:

morapa a horizontal bar on two upright poles to hang blankets over

mosetshe a peg
mo sogo le a wooden beam
mots he a pestle; a stamping stick

17 Categories representing scientific models, or 'expert models' as described in Ungerer & Schmid
(1996:52-5).

18 Both English and Norwegian d ictionaries list two interpretations of 'plant': one «scientific»
w hich includes trees ("any living organism that typically synthesises its food from inorganic
substances, possesses cellulose cell walls, responds slowly and often permanently to a stimulus,
lacks specialised sense organs and nervous system, and has no p owers of locomotion"), and one
which excludes them ("such an organism that is green, terrestrial, and smaller than a shrub or
tree; a herb"). (My italics.) (Hanks, P. (ed.) Collins English Dictionary 1986:1173)

-32-
motswaiso a firebrand; a torch; a partially burnt stump; a piece of charred
wood; a stick from yesterday's fire

At least one language (Fulfulde, another Niger-Kongo language) has reserved a

noun class exclusively for this conceptual group (the 'ngal-class') (Endresen 1993).

(4) Objects made of other plant material

The same type of semantic extension as that from 'trees' to 'wooden objects' in
general (synecdoche), can be found from other 'plants' to 'objects made of plant
material'.
As one can see from the examples below, this schema is a generalisation
over a group of rather varied concepts. They do, however, share the schema
'material with «live» origin', which they also share with both the group of
'plants' and with the prototype 'tree'.

mhikwana branch in gate to show that people are out

mmitsa charm that attracts customers; a charm that attracts men; (a
prostitute) 19
morurelo a pot used for cooking medicines 20
motlaagana a temporary house; a house made of bushes; a tent; a
tabernacle
mutlwane a snare; a trap with a beam21

19 Such charms would usually consist of plant material. The relation between the primary meaning

and the term 'prostitute' could be one of metonymy; referring to the prostitute by using the word
for a charm that she might wear to attract men. It is, however, equally possible that the
charm and the prostitute independently have got their terms from the verb go bitsa (to call, cry
out, draw attention to), which is the verbal basis for the derived noun mmitsa.
20 This might not immediately look like a «member» of this conceptual group. It can, however, be

argued that it is connected to the rest of the group by a metonymic relation «from content to
container». The pot is associated with it's normal content, medicine, and medicines are
produced from plant material.
21 It may be made either of plant material, or of skin material (Prof. S.A. Swanepoel p.c.). In both

cases it will be linked to the network and instantiate the abstract schema 'material with
«live» origin'.

-33-
 (5) Other long objects or phenomena (not wooden)
In addition to the long wooden objects, there are, as we can see below, some other
long objects and phenomena of varying types in my noun sample.
Extending the network to entail these concepts, we see that we are leaving
the «domain of trees and plants)).22 Hence, the similarity judgements are of a
slightly more abstract kind.
I consider these concepts extensions from the group of ' long wooden
objects', as they instantiate the class schema 'long', but are not connected to the
'living' schema.

moeme erect pointing, or erect thing; a name sometimes given to a

modimako, or tall pole with black and white stripes, and
crowned with civet cat tail that stands in the centre of the
Bogwera camp at the second initiation23
mothaladi a stripe in colours; a streak between clouds; a scrap of porridge
left in a pot
mot halo line
motontonyane thin long bucket
mot shet srudi ma lane; traffic lane

The metaphor «a tree is a body»

So far we have been concerned with schemas whose similarities to the prototype
is rather obvious (except perhaps the 'long objects' that are not made of wood or
plant material).
There is another «cluster)) of schemas, all related to the human body in
different ways. The connection of these schemas to the prototype needs some
comments.
Many languages exhibit a conceptual connection between trees and the
human body, for instance by using similar terms to refer to either a part of a body
or a part of a tree. In English the term 'trunk' can refer to the human torso as well
as to the 'woody stem' of a tree. In Setswana thupa refers either to a small twig, or
to the penis, matlhare usually means leaves, but can also mean body-covering

22 'domain' in the sense u sed in Langacker 1987.

23 This noun comes close to being «maximally schematic»; it doesn't necessarily refer to anything

specific, but to any upright or erect thing. Such a thing is by necessity 'long'. The noun is a
nominalisation, based on the verb go ema - to stand.

-34-
 (although not in every day use). One can also in some languages (including
Setswana), refer to different parts of a tree by using body part terminology; tlhogo
ya setlhare - the head of a tree, molala wa setlhare - the neck of a tree, legetlha la
setlhare - the shoulder of a tree, mpa ya setlhare - the stomach of a tree, leota la
setlhare - the leg of a tree. It is not difficult to grasp such meanings, even for
speakers of languages that don't make use of such a metaphorical connection
between parts of a tree and parts of a body.
I suggest that the reason for the intuitive understanding of such terms
must be that humans easily construe the similarities between bodies and trees;
such a metaphor is felt to be «natural». But it is not only a cluster of characteristics
that makes us perceive this similarity. Rather, the whole 'gestalt' (the «perceived
whole»; cf. Ungerer & Schmid 1996:33-7) of a tree is perceived as similar to that of
a body. Both trees and bodies are long, erect objects that are organically alive, that
have an outer cover (bark or skin), and that have different parts branching off
from a central «trunk».
Keeping class 3 out of the discussion for a moment, it seems reasonable on
a general basis to assume that the starting point for extension in the metaphorical
connection between trees and bodies is bodies. In accordance with assumptions
about the «bodily basis of meaning» (cf. Lakoff 1987, Johnson 1987) one can argue
that it is «natural» for humans to bring primary bodily experience into their
conceptualisation. Hence, seeing a tree as a body, could be a more basic (in the
sense: more usual, as well as developmental prior) direction of extension, than
seeing a body as a tree. As an example of this direction of comparison we have
already seen that Setswana employ body part terms on trees.
I have already introduced several reasons to suggest that the «class
prototype» in class 3 is 'tree'. The metaphorical connection between trees and
bodies are clearly present in the semantic network of class 3, and with the central
position of trees in class 3, I find it reasonable to propose that the direction of the
extension relationship here is from trees to bodies, and through that relationship,
further on to other concepts linked in different ways to bodies.
All arguments (productivity, number of terms, saliency) suggest that class 3
is a «tree-class» and not a «body-class».
The class 3 concepts that show a semantic connection to the human body,
can be grouped as follows: 'long body parts', 'body-coverings made of skin', 'other
body-coverings', 'other skin made objects', and sicknesses or special 'states of the
skin' of human bodies. The term for 'body' itself - mmele - is also a member of
this class.

-35 -
 Concerning the graphical representations of the networks, we see now that
the extension arrows can be qualitatively different, depending on whether or not
the schemas connected by extension belong to the same semantic domain or not,
and in cases of extension from one domain to another: on how qualitatively
different the two involved domains are.

(6) Long body parts

Long body parts of course instantiate the class schema 'long' that they share with
the prototype 'tree'. They also instantiate the schema 'material with «live»
origin', which is an elaboration of the more generalised class schema 'living'. The
long body parts represent metaphorical extensions from the prototype on the basis
of conceived similarities between trees and the human body.
The following are examples from my noun sample:

mmopo the bridge of the nose; the muzzle of an animal

moeledi a navel cord
mokwatla the back; the spine; a backbone
moletse a leg
mono a thumb

(7) Body-coverings made of skin

The next extension from the prototype is also based on the metaphorical
relationship between trees and bodies; trees have bark as their outer covering,
while bodies have skin. And a body-covering is easily conceived metaphorically
as the «skin of the bearer», that is: as a part of the body. A second connection to
'body' is established by a synecdochical «part for whole» relation type, as these
objects are made of skin (skin is part of a body).
The group instantiates 'material with «live» origin', that is a further
elaboration of the class schema 'living'.
The following skin-made body-coverings were found in my noun sample:

moitlamo girdle; waistband; belt; sash

mothikga ornamental skin clothes
motlokolo goat-skin apron, for girls and women worn over their
buttocks
motseto traditional leather strip worn by men through loins and
buttock; napkin

-36-
 From this group two other small (in numbers) extensions can be drawn:

(8) Other body-coverings

This small group consisting of the three nouns listed below, resembles the
previous one; the only difference is the lack of synecdochical relation to 'body'
through 'skin'. But body-coverings are still metaphorically connected to 'body',
even if not instantiating the schema 'material with «live» origin' itself, as the
former group does. Rather it shares the local schema - 'body covering' - with that
group. The local schema can be represented like the one below and integrated
into the larger network figure:24

other
body-
coverings

figure 7

mofitshana a plain iron ring worn on the wrist

moitshom elo armour; protective clothing
motlopo a tortoise-shell; 25 (clitoris)

(9) Other objects made of skin

Here it is the synecdochical relation to 'body' through 'skin' that motivates the
connection to the network, and the group instantiates the schema 'material with
«live» origin'.
The following are examples of this group of concepts:

mofalo a small skin water-bag

mogogorwana a dummy calf made of skin, stuffed with grass; a dummy calf
that is licked by a lactating cow whose own calf has died

24 To emphasis that this figure is to be read as an integrated part of the greater class n etwork, I

have copied the relevant part (concerning the exten sion relationship). The lay-out of the
figure therefore appears a bit strange. In the following I will present examples all local
schemas in this way.
25 This is reasonably interpreted as m etaphorically related to skin.

- 37 -
mokobolwane calfskin
mophanyan e chapped skin
moropane a timbrel; tambourin26

(10) Skin related states (or sicknesses)

Metonymy based on spatial nearness is the basis for this further extension from
skin made objects. This group (see the examples below) doesn't instantiate the
schema 'material with «live» origin', but connects to two of its instantiations
('body-coverings made of skin' and 'other objects made of skin') by a 'local
schema': 'related to skin'.

mmokwana measels
mobipo a weal; a swelling in which there is no fluid; or pus under the
skin
mofufutso perspiration; sweat
mothopa small boil
motshwa an itchy eruption

figure 8

(11) Lively human activities

The following nouns represent another group of concepts that seems rather
uniform, and that corresponds to Spitulnik's concept 'lively human activities' (I have
adopted her term.):

mmino dance
moduduetso applause; ululating, of a woman in a dance; yodelling, of a
girl, or woman

26 This is actually made of both skin and plant material.

-38-
mogolokwane ululation
moopelwane the shouting, ululations and singing of a victorious army,
choir, etc.
moteka a merry noise; a feast

This schema represents a metaphorical extension from the class prototype into
the «domain of human activities». The group connects to the prototype by
instantiating the class schema 'living'.
We see now that the class schema 'living' exhibits a very high degree of
abstractness: In this case 'living' refers to that «portion» of human activities that
are particularly lively.

(12) «Processes as things»

The abstract schema 'living', as we have seen, encompass entities or phenomena
that are associated with «live» origin, but also less concrete notions (cf. 'lively
human activity').
Nouns that refer to a process constitute another group of «not concrete»
concepts that I suggest is connected to the network. Many of these nouns are, as I
have pointed out in the examples below, derived from verbs.

modiga an end of a dispute, or discourse [go diga - to cast down; let

fall]
mogano a breaking up, or cessation of rain; time for resuming work
after rain
mokgwasa rustling; actual sound, as of something moving through twigs
and foliage; (rumour )
mongatego a fainting fit [go ngatego - to faint]
moo no a snore; a grumbling; moan [go ana- to moan; groan; snore]

This group of notions is suggested by Contini-Morava (1994:13) in her proposal

for a class 3 network, and I share her opinion that, in describing a process as a
thing, these nouns «fit well with the other liminal entities in this class, that fall
somewhere between animate and inanimate». By describing a process or an
event, they point not to a «dead object», but rather to an entity filled with
movement, sound, or other characteristics of ' living' things. Thus, they
instantiate the class schema 'living'.

-39 -
 (13) Objects connected to ground
Spitulnik (1987:58) argues that in ChiBemba there is a connection between trees
(that are linked to the ground by their roots) and other things that are dug in the
ground. My data (see the examples) also suggest a metaphorical extension from
trees to such a group, connected by the following local schema: 'in ground'.

figure 9

mogobo di a straight burrow

molala-thakadu an antbear hole, or lair
motlhobodika a large vent of an antheap
mot 1hodi the eye of a fountain
mot lobo a pit, or quarry; a place where clay for pottery, or sebilo, is dug;
a storage place

It is possible that most nouns in this group in addition instantiate the class schema
'long' as well, reflecting yet another connection to the class network. (I have,
however, not integrated this possibility into the graphical representation of the
network.)

Concluding remarks to the class 3 network

The semantics of class 3 doesn't seem to represent the great confusion that one
might be led to believe, by those who see the Bantu noun classes as formal classes
only. As we have seen, my suggested network produces 'tree' as the prototype.
Both Spitulnik (1987) and Contini-Morava (1994) also explicitly give trees a
prototype position within class 3 (for ChiBemba and Kiswahili respectively).27
They label their most abstract generalizations: "potent/living" (Spitulnik) and
"entities with vitality (neither human nor prototypically animal)" (Contini-

27 Contini-Morava uses the label 'most central subcategory', a notion that I interpret as 'prototype'.

-40-
 Morava). I consider their suggestions compatible with my own most abstract
schema 'living'. The «old bantuists», Torrend (1891) and Meinhof (1906), also
appreciate the 'living' (- but «not personified») character of the members of this
class. Denny and Creider's (1986) contribution differs, as they focus on the
configurational aspect of this class' members: 'extended'. Their «class 3 meaning»
resemblet; my schema 'long' , but they defocus the possible 'living' character of the
class.28 However, by drawing attention to the configurational aspect, they have
extracted an important property of a substantial amount of class 3 nouns
(including the prototype). With the exception of Contini-Morava (1994), all the
others either miss this point completely, or only vaguely hint towards it.29 Batibo
(1987) offers yet another interpretation of the semantics in this class, but even that
focuses on an important aspect of trees: that they are "planted".
It is my impression that the semantics in class 3 is a lot easier to reveal than
some of the other so called «miscellaneous» classes. This is, I think, due to two
circumstances. Firstly, it seems that there is less variation between different
Bantu languages regarding the selection of nouns that the class contains. 30
Secondly, I assume that it is of importance that the prototype in this class is
exceptionally strong; trees constitute a large and very salient and homogeneous
group, and is about the only group of nouns that is almost completely productive
in incorporating every new term introduced to the language.
Accordingly, we see that those who have investigated the semantics of class
3 (myself included), bring forward semantic descriptions that are largely
compatible.
Rather than revealing any «new» semantic groups, my own contribution
to the understanding of class 3 is: establishing a network that explicitly suggests
meaning relations between different groups of notions and commenting on the
nature of these relations, as well as positing the abstract generalisations (schemas)
that bind the parts of the network together.

28 They only very loosely mention that class 3 "may constitute a class for plants on the kind side of

the system" (Denny and Creider 1986:223).

29 As Torrend (1891:79) does, when he mentions "such tools or artificial objects as remind one of the
form of a tree [... ]" (My italic.)
30 This impression base on the data used as illustrations by the authors that I refer to.

-41-
4.3. Class 5
In this section (4.3) I propose that the semantics of class 5 exhibits a network of
extensions related to a possible prototype fruit ', leading to the two internally
unrelated class schemas: 'non-individuability' and 'roundish'.
That implies that as far as configuration is concerned, the majority of concepts in
class 5 oppose the majority of class 3 concepts.

Earlier treatments of the semantics of Bantu class 5

Class 5 in Setswana is one of the semantically least transparent classes. Cole
(1955:79) again makes use of the term «miscellaneous» for a description of its
content. He treats class 5 and 6 together (class 6 consists partly of the plural
counterparts of class 5 nouns, but also of a great number of uncountable nouns),
and comes forward with the following list of concepts as the content of the classes
5 /6: "names of parts of the body, animals, plants, natural phenomena, some
collective nouns, and a number of personal nouns". All these concepts, with the
exception of collective nouns (which clearly belong only to class 6), are listed
again in his treatment of several other classes.
As the following short presentation will show, the search for a core
meaning of class 5 has led to rather different suggestions.
Torrend (1891:91-2) in his treatment of class 5 centres on the notions of
"unproductive", "hard, or bare, or flat". The rest of his treatment of class 5 is
rather fragmentary; he merely lists selected nouns, or groups of nouns, from the
class, leaving the impression that those nouns in some way or another are more
central to the class than the not mentioned ones. He, however, makes no attempt
to interlink them. The groups of nouns that he emphasises are still, as we will
see, present in class 5 in contemporary Setswana: fruits, the sun/ the day, and
"words and distinct sounds".
Meinhof (1906:9) treats class 5 as the singular counterpart of class 6, which
he sees as "urspriinglich dualisch" (originally dualic).
Denny & Creider (1986:220-1) argues that Proto Bantu class 5 belongs to the
group of configurational classes. Class 5 is described as consisting of "non-
extended, solid figure units". According to Denny & Creider 'non-extended' may

-42-
be "negatively characterised as not extended31 , and is positively characterised as
rounded, protruded, humped, bunched, etc." They add, however, that it is
possible that class 5 additionally incorporates a few nouns on the basis of 'kind'
rather than 'configuration'. When focusing on 'kind', they suggest that class 5 is
the class for fruits, or in a more general sense: "typical products taken for human
use from plants without regard to their configuration".
Batibo (1987:13) argues that class 5 consists of objects "usually existing in
pairs or identical forms".
Spitulnik (1987:62) treats class 5 and 6 together. She claims that in
ChiBemba both the classes 5/6 (the singular nouns in class 5 and their plural
counterparts in class 6) and the non-count class 6 nouns "center around the
related notions of 'non-individuability', 'collectivity' and 'abundance' [... ].
Although their semantics differ only minimally, these two classes have been
given separate designations because there are a sizeable number of nouns which
are non-count and occur exclusively in Cl. 6. Some singular forms of Cl. 5 are
back-formations of Cl. 6 mass nouns, but typically the count nouns of Cl. 5 I 6
denote fruits, paired items (especially paired body parts), and constituents of
groups."
Contini-Morava (in an updated version of her work found on Internet,
1996) argues that the most productive semantic category within class 5 consists of
fruit terms, and that fruits constitute the basis for several semantic extensions,
particularly concerning shape and size. She states that the majority of fruits are 3-
dimensional, round, large 32 , and that it is possible to conceive of them either as
solid objects or as containers. On this basis she suggests the following semantic
extensions: protrusions, swellings, lumpy substances, 3-dimensional containers,
and hollow spaces. She then proposes a more abstract extension: from
growth/ swelling and containment to large things in general. (In Kiswahili, class 5
can be used to derive augmentatives.) As a 'superschema' for the class, Contini-
Morava suggests 'plant offspring', based on the two instantiations 'fruit' and

31 They define "extended" as: "characterised by relative length in one dimension at the expense of

the other two, or less often, by relative length in two dimensions at the expense of the third".
(Denny and Creider 1986:220)
32 She admits that treating fruits as large objects might seem counterintuitive, but claims that it can

be justified, because "[f]ruits are «large» in relation to their earlier stages of growth, and it is
when they become large that they are of most value to humans". (Contini-Morava on Internet,
1996)

-43-
'leaves'. The extensions from 'fruit' have already been mentioned, and the
extensions from 'leaves' are: 'curved flat objects', 'things with broad parts', and
'curved outlines'.
It seems then, that the semantics of Bantu class 5 is a lot more complex
than that of class 3. A possible explanation could be that class 5 exhibits greater
variability in semantic content between the different Bantu languages (for
historical or other reasons). It is, however, equally possible that this apparent
variability rather demonstrates a semantic complexity, and that most authors by
their preconceived notions about core meanings, have been led to pick out one
single aspect among several possible ones.
The different semantic core meanings proposed above, can be summarised
as: dual, non-individual, unproductive, non-extended, and the most frequently
mentioned group of items that belong to class 5 (in almost all Bantu languages),
is: fruit.

Proposal for a schematic network of class 5

My own suggestion is that the semantic network of class 5 produces both a class
schema for 'non-individuated' (with «duality» as one further elaboration) and a
class schema representing the configuration 'roundish' . 'Fruit', but also a lot of
other concepts can be characterised as being instantiations of both these schemas.
By 'non-individuated' objects I refer to:
1) objects that normally occur as 'one of a «multitude»' . A multitude is here
defined as a group of three or more items (typically very many) of the same 'kind'
and of more or less the same shape.
2) objects that normally occur as 'one of a pair'.

'Roundish' objects in the sense that I will use here, need not be perfectly globular or
circular. They contrast with long objects by being of relatively similar lengths in all
relevant dimensions.
The abstract concept 'non-individuated' is clearly more difficult to handle
than a configurational concept like 'roundish' . 'Non-individuability' depends, to
an even greater degree, on the viewer's subjective conceptualisation or «focal
attention» (cf. Langacker 1987).
A third aspect of the class 5 semantics is that the majority of concrete,
touchable objects consist of «deader material» than their more «live»
counterparts in class 3. This might constitute a second striking opposition
between these two classes. (The first opposition is based upon the configurational

-44-
contrast 'long' vs. 'roundish'.) I have, however, not integrated this «non-live»
aspect in the network figure, the main reason being of a practical nature: I would
have to split many of the «schema-boxes» further, leading to a network figure
that would be hard to comprehend. That does not imply that I do not appreciate
the possibly relevant status of an abstract schema: 'non-live objects'. I will,
however, suggest that since the proposed class prototype does not reflect the
notion of 'non-live objects' it is probably not as central as the two class schemas
that I have integrated in the proposal for a class 5 network.
The network of extensions that I suggest for class 5, could be graphically
represented as follows:

single (6) non-indiv.

r on-extend.
roundish ( ) ............................. ,. ~inguistiS,
11 Items
. . 2

c:J(1.... . . .
body parts •I11..................... . ed (5) ~10
!Paired (8) pa1r non-md1v.
roundish •111 ..................................................... roundish roundish
, ,............ body parts •ll ............... 11, •non-fruit>> .....................IJ• non-indiv.
~domments paired 11 fruit (non-ro und)
non-round) objects (4)
body parts . ~ll) •11.......... ..........HI• roundish
non-md1v. ('Ind'IV.) (9
humans objects (3) ...................Jio roundish
enclosures I
locations

(12)
t' negative immature ethnic
nega IVe •o(l ............ humans ·•fii ..·OI• humans ·•11 ....11•· groups
states

figure 10

-45-
Again the instantiation relationships are shown in two different figures to avoid a
graphical representation that is virtually impossible to read. They are, of course, to
be read as different parts of one unified representation.
First, I show the instantiations based on the schema for 'non-individuated' :

non-
individuated

single (6)
roundish
.---~:;,
(8,..,
) body parts ·•IJ .................... paired (2
non-indiv.
·•11.............................. .. ................. roundish roundish
paired (7) ,11J ............ body parts
u,
.,.._..._""" ................ oil• <non-fruit» ..................... non-indiv.
(non-round)
non-round)
objects 4
body parts ____. ..........u, roundish
(indiv.) CY
objects (3) .................oil• roundish,-
enclosures1
locations

(12) .
negative immature ethnic
•II.......... .. negative ·•11 .. ·11•• humans ••11 ..·11•• groups
states humans

figure 11

- 46 -
The second representation show the instantiations of the schema 'roundish':

roundish
(non-
extended)

locations

(12) .
negative negative immature ethnic
states
•11········.... humans •lll.. ·lll• humans •IU.. ·III• groups

figure 12

(1) Fruit: a possible prototype for class 5

As a prototype, 'fruit' is not as clear as the prototype in class 3. Actually, the main
reason for suggesting that it is possible to treat it as a prototype at all, is that
evidence from other Bantu languages (including the reconstructed Proto-Bantu),
indicates that fruit names have always been in class 5, and that there is a regular
relationship between trees/plants in class 3 and fruits in class 5: fruit names will
often have the same stem as the trees or plants that they grow on, the only
difference being the noun class prefix: morula is the class 3 noun for a «morula-
tree)), and lerula is the class 5 noun for a «morula-fruib).
In Setswana all the indigenous fruits, as those listed below, have class 5
names. That indicates a former productivity. Hence, it is possible that the group of
fruits has been more «saliently presenh) in class 5 at an earlier stage of history.

lega pu water-melon
leg omane a species of pumpkin

- 47-
lekwele an edible tuber; a table potato
lephu tse pumpkin
lerotse a melon; a melon that is not of the sweet species

In contemporary Setswana class 9 has absorbed most of the «modern» fruits

imported to the area. The names of these imported fruits have been treated like
many other loan words. They have been placed in the prefix-less class (9), where
they keep the original first syllable (or only modified in accordance with Setswana
phonotactics) without adding a class prefix. This gives apole for apple, panana for
banana, pere for pear etc. It is, however, possible to add the class 5 prefix to at least
two of these «new» fruit names, namely apple (leapole) and banana (lepanana L
thus indicating quite recent productivity. This fact together with the fact that almost
all Bantu languages assign fruits to class 5, strengthens the impression of this class
as the «fruit class», despite the fact that the fruits in class 9 might outnumber the
ones in class 5 in contemporary Setswana.
I include tubers, bulbs and other fruit-like plant products in the 'fruit' term
(the biological, as well as the English or Norwegian definition of fruit is of course
irrelevant here).
'Fruit' is the starting point for several extensions. The similarity judgements
between 'fruit' and the other extensions (that will be treated below) produces the
two class schemas: 'non-individuated' and 'roundish'. 'Fruit' instantiate both of
these schemas.
That does not imply of course that all fruits by necessity are round. Rather, it
refers to a prototypical characteristic of fruits; the prototypical fruit is reasonably
considered 'round', as are in fact the majority of the group as such.

(2) Other non-individuated, roundish objects («non-fruit»)

Class 5 also entails a lot of other roundish objects that normally occur together
with several others of the same kind and of more or less the same shape. As such,
they instantiate the same two class schemas that the prototype does. H ence, one
could argue that the prototype should include this group as welt establishing a
relationship like:

-48-
 non-indiv.
roundish
objects

non-indiv.
roundish •llt···········m•· roundish
fruit <non-fruit>>

figure 13

-with the most abstract schema established as the class prototype from which the
other extensions are drawn. However, it does not seem reasonable to suggest as
prototype a concept that is as abstract as this, and that exhibits such a highly
heterogeneous internal structure as the examples below show. Therefore, I have
rather treated this group as an extension from the prototype 'fruit', sharing with it
the two class schemas 'roundish' and 'non-individuated', but differing as far as
«live» vs. «dead» material is concerned (in that most of these concepts are
«deader» than fruits).

legwarane a stone, tooth, etc., for rubbing skins

lekwakwa a sharp-edged stone; a rough stone of volcanic formation
lelepa a difficult or intricate knot; a puzzle; a riddle; a mystery33
lengope a lump of porridge when badly stirred
lep hoto a knot in a stick, or piece of timber; a place in a branch, or reed
where a twig has been cut off

(3) Other roundish (but individuated) objects

As the following examples show, class 5 also includes some concepts referring to
roundish objects that can h ardly be seen as instantiations of the schema 'one of a
«multitude»'.

lefiswana little pot

lekuka a skin sack; a milk sack, made of skin (nowadays it may be made
out of other materials). N.B. it is used for processing fresh milk
into sour milk or madila

33 The knot is the primary meaning. The other m eanings are metaphorical extensions.

- 49-
lesiba the outer skin of a goat's bowel, and the musical instrument
made from it-34
letsatsi the sun; a day; heat caused by the sun
letshego a trivet, or three-legged stand for a pot, etc.: a stone etc. for a pot,
or kettle to stand on to be boiled

These concepts relate to the prototype through extensions, and represent

instantiations of the class schema 'round'.

(4) Other non-individuated (but «non-round») objects

This group of concepts represents yet another extension from the prototype. It
represents an instantiation of the schema 'one of a «multitude»', and thereby also
the class schema 'non-individuability'. The following are examples from this
group:

lekala department35
lerapo bone
letlepu mirth; plentitude of food, or any other material possessions;
abundance, especially of food
letlhomaganyane a small seam in a kaross36
leudi a speck; a shred, as of a blanket

(5) Paired, roundish parts of the body

Paired (or dual) objects represent a special case of 'non-individuability'. It is when
we are confronted with these concepts, that we see that the schema 'one of a
«multitude»' is not general or abstract enough to constitute the class schema (as
two hardly can be considered a «multitude»). The need therefore arises for
establishing the more abstract class schema: 'non-individuated'.

34 The secondary meaning representing a case of synecdoche.

35 This meaning represents a metaphorical extension from the more basic meaning 'branch' (not

given in the dictionary). 'Non-individuability' is a relevant aspect both of the primary

meaning and of this extension, as 'department' is meaningful only on the «background» that
there exist several.
36 A seam consists of many small identical stitches.

-50 -
 Most body parts in class 5, as the ones below, are easily described as
instantiations of both the schemas 'paired' and 'roundish'. They represent yet
another extension from the prototypical group.

Iera go a buttock
lerete testis; scrotum
letheka waist; loin
letswafo a lung
letswele a fist; a small compact group of people37

The following three small (in numbers) groups of concepts are easily related to the
group of 'paired, roundish body parts' by extension:

(6) Single, roundish body parts

Only one body part from my class 5 sample letlaparapelo - 'pericardium', is not
paired. Its configuration, however, is clearly 'roundish'.

(7) Paired, «non-round» body parts

A couple of other body parts from my sample can hardly be considered 'roundish',
but do represent instantiations of the schema 'one of a pair':

lefuka a pinion; a wing; a feather

lesufu a thighbone or upper leg bone of an animal or human being;
the upper arm bone
letlhakore flank; side

(8) Paired, roundish body-adornments

This extension differs from the two above: It shares with the group of 'paired,
roundish body parts' all its instantiation relationships, but differs by representing
concrete objects attached to the body, rather than being parts of the body itself. As
such, the following concepts represent metonymic relationships based on «spatial
nearness» to the 'paired, roundish body parts'.

3 7 The fist is the primary meaning. The secondary meaning is an example of a metaphorical
extension.

-51-
lenyena an ear-ring
les eka a bangle or bracelet; formerly a twisted metal ring worn on the
arm and below the knee; a leg ornament; a coiled spring

(9) Roundish enclosures or locations

These notions arc indirectly linked to the prototype through extensions. I have
treated them as extensions from the group of roundish, but individuated objects,
which itself is an extension from the prototype. The roundish enclosures and
locations instantiate the schema 'round'. The entities that these concepts refer to,
differ from the groups of entities treated above, in the way human beings deal with
them: They are not three-dimensional, «manipulable» objects that we can hold in
our hands, but either two-dimensional locations or three-dimensional enclosures.
The roundish shape could be either concrete or imagined. The following are
examples from this group of concepts:

ledutela a pond or pool in the veld; a waterhole in the veld

lefatshe a country; a land; a state; the earth; the world
legotlhe universe 38
lelapa a low wall or enclosure in a home; a home; a family39
letsha a pan, or shallow lake; a vlei, a saltpan

It is, however, possible that one should connect these concepts to the schema for
'non-individuability' as well, on the basis that they could be conceived as
«containing» non-individuated entities; either an unspecified number of human
beings (often a «multitude"), or the mass entity, water. In that case one should
perhaps rather treat this group as a direct extension from the prototype.

(10) Non-extended (in the domain of time), non-individuated «linguistic items»

Formal grammatical terms and some other linguistic phenomena occur in class 5,
as for example the following:

38 Probably perceived as a kind of roundish, huge enclosure (three-dimensional) or location (two-

dimensional).
39 The enclosing wall is the primary meaning. When referring to a home or a family lelapa is a
metonym based on «spatial nearness».

-52-
lebitso a name; a nomenclature
lediri a verb
lentswe a voice; a word; the sound of a voice
leselo noun
letlhaodi adjective

The «new» formal terms are probably analogies based on the earlier existing
vocabulary concerning parts of language or sounds. The striking difference between
these language related phenomena and the language related phenomena in other
classes, is that the class 5 group mostly consists of characterisations of single words
or sounds (of a short duration), whereas concepts referring to more enduring
language events are found in other classes; e.g. poielO - 'a tale' (class 9), morapelo -
'a prayer' (class 3), tlhamane -'a fairy tale' (class 9).
The link between this group of concepts and the other concepts in class 5 is
based on a metaphorical extension from the configurational opposition 'roundish'
vs. 'long' belonging to the three dimensional domain, to an opposition in the
domain of time: short duration («non-extended» - or metaphorically: 'roundish' -
in the domain of time) vs. endurance («extended» - or ' long' - in the domain of
time). That makes this group an instantiation of the schema 'roundish'.
In addition these single sounds and word-types normally occur together with
a «multitude» of other similar parts of language. This makes them much more
related to 'non-individuated' concrete objects, than could possibly be said about the
more enduring language phenomena like stories, prayers etc. In other words: this
group can be perceived as still another instantiation of the class schema: 'non-
individuated'.
To conclude: I have treated this group of concepts as a metaphorical
extension from the 'non-individuated roundish objects' «that are not fruits»; both
groups instantiate the two class schemas, but they differ in 'semantic domain'.

(11) Non-individuated human beings

At first sight one gets the impression that most human beings in class 5 are marked
by negativity (lefelepa - a cheat; a rogue) or immaturity (lesokgola - a child just able
to run about). However, the names of many ethnic groups (except the Batswana
themselves and a few others) are also found in this class; it is not uncommon to
integrate «new» groups that have not had a Setswana term, into this class rather
than into «the human class 1». This works even for distant and unrelated groups
that the Batswana hardly could have been in a position to develop a negative

-53-
attitude towards.4° I thus find it more likely that, rather than plain negativity, the
notion of 'non-individuability' is the origin of the class 5 human terms: One could
argue that there is a human tendency to perceive people from other ethnic groups,
people with negative or scary qualities and the not fully grown and mature in a de-
individuated manner. The following are examples of «class 5 persons»:

Iefelepa a cheat; a rogue

Iekgoba a slave; a member of a slave family
Lekhalate a coloured; a person of mixed blood in Southern Africa41
lesokgola a child just able to run about
letagwa a drunkard

In the network figure I have represented 'ethnic groups', 'immature humans',

and 'negative humans ' as three further elaborations of the schema 'non-
individuated human beings'.

(12) Negative states

A group of relatively homogeneous abstract concepts (homogeneous in the sense
of centring around a notion of «negativity» or «undesirability») constitutes the
last group of concepts that I have integrated into the class 5 n etwork. (See the
examples.)

Iebekebeke starvation
Iehuma poverty; great sorrow; damage; loss
Ieitlhapelo riotous mirth42
lenyora thirst
lesotlO contempt; mockery

Although I have established 'non-individuability' as the generalisation over the

humans in class 5, we have also seen that «negativity» easily suggests itself as
part of the immediate associations for many of these terms. That is, of course, not

40 Dr. Joseph Tsonope p.c.

41 This is a borrowing from English, with the class 5 prefix attached to it.
42 This notion is considered negative because it is associated to being drunk (Prof. S.A. Swanepoel
p .c.).

-54-
a contradiction: The tendency to treat some humans in a «de-individuated»
manner, easily leads to negative associations. This negative aspect could provide
a further extension to negative states that humans might experience, leading to
the local schema: 'negativity'.
The 'local schema' connecting «negatively marked human beings» and the
negative states can be represented as follows (and should, of course, be «read
into» the schematised class network):

negativity

negative negative
states ·•II················ humans

figure 14

Concluding discussion of the proposed network for class 5:

My proposal for a semantic network of class 5 in Setswana actually incorporates
several of the notions that have been suggested as the semantic core of class 5 in
various Bantu languages and by various authors. Torrend's (1891) notion
"unproductive" seems to focus on the «non-live» character of a lot of this class'
m embers as opposed to many class 3 members (proposed here as a possible
additional abstract schema entailing many class 5 concepts).
Meinhof (1906) argues that duality was the original concept of both class 5
and 6. He goes to great lengths in trying to combine the notion of duality with the
indisputable fact that fruits are class 5 concepts. As mentioned in chapter 2, he states
that "Friichte zerfallen entweder von selbst in zwei Teile oder werden vom
Menschen, da er zwei Hiinde hat, regelmaf5ig in zwei Teile zerlegt".43 (My italic.)
Within my proposed network of interrelated concepts, nothing suggests such a
direct link between paired objects and fruit. Rather, the two groups of concepts are
connected by sharing the rather abstract class schema 'non-individuated'.
Batibo's (1987) statement that the original Bantu class 5 concept was "existing
in pairs or identical forms", is also compatible with part of my suggested network; it

43 In English translation: "Fruits either divide into two parts by them selves, or are regularly

divided into two parts by humans, because they have two hands."

- 55 -
coincides with the schemas 'one of a pair' and 'one of a «multitude»', which
together compose the more abstract schema 'non-individuated'.
Although Denny and Creider (1986) feel the need to mention fruit
specifically, and thereby suggesting that class 5 might exhibit an aspect of «kind
semantics», they emphasise the configurational aspects of this class, giving the label
"non-extended, solid figure units" to its members. As opposed to Meinhof and
Batibo, who focus on 'duality' and 'non-individuability' (as I interpret them),
Denny and Creider focus on what I have treated as another abstract schema within
class 5: the configurational concept 'roundish'.
A possible explanation to this diversity could be that the different suggestions
only reflect a situation where class 5 has developed in different directions in
different Bantu languages. However, Meinhof (1906), Batibo (1987) and Denny and
Creider (1986) all base their analyses on Proto Bantu material. So they have, I
believe, «seen different things» within the same material, because they were
restricted by the belief that there would be one single core meaning to detect. In my
view, all of them point to groups of concepts from which one can extract important
aspects of the semantics of contemporary Setswana class 5, but only to parts of it. As
those parts are basically the same as the different abstract schemas in my suggested
network, I consider this a support to my analysis.
On the basis that both class 5 and the non-count part of class 6 centre around
the same notions of non-indiv iduability, collectivity and abundance, Spitulnik
(1987) gives the two classes a common treatment. (Meinhof 1906 also treats the two
classes together, but based on the notion of duality.) I share her opinion that non-
count nouns (which include physical masses and entities that to some extent is
conceived in a «mass-like manner») indeed have something in common with
'non-individuated' entities.

- 56 -
 Within the framework of my analysis this relationship could be represented
as follows:

amassed

homo-
geneity

one of one of
a pair m any

CLASS 6 CLASS 5

figure 15

Contini-Morava (on Internet 1996) points to the fact that in Kiswahili a lot of large
things are included in class 5, and that "a noun stem normally associated with any
of the other classes, if put in Class 5, aquires the connotation of large size". She
suggests that it is the category 'fruit' that motivates the extension to large things, on
the basis of associations of growth/ swelling and containment, and that fruits can be
conceived as «large» in relation to earlier stages of growth.
Concerning Setswana, there is absolutely nothing in my class 5 data that
could suggest an extension to large things. On the contrary, a large portion of the
class 5 concepts refer to small objects. (Small in the sense of easily manipulable, -
in other words: small as compared to the human body.) As a matter of fact, I
believe that a third class schema 'small' possibly should be integrated into the
class 5 network together with the other two. I have, however, chosen not to do so,
as I see the other two class schemas 'non-individuated' and 'roundish' as more
fundamental for the understanding of the semantics of class 5.
It is of course not a problem that Kiswahili includes in class 5 a lot of
concepts referring to large objects, while Setswana doesn' t. But it represents a
problem that Contini-Morava (1996) argues that fruits are conceived as «large»,
while I would argue that they are conceived as «small». We have no reason to

-57-
believe that speakers of these two closely related languages conceive fruits that
differently.
Comparing Contini-Morava's examples with my own data, I find that
except for the augmentative use of Kiswahili class 5, the types of concepts
included in class 5 in the two languages to a large extent are of the same character.
It seems therefore as if the only reason for suggesting that fruits are «large», is the
augmentative use of class 5 in Kiswahili. If that is the case, such a solution seems
forced on the data. I would rather suggest that 'large' might be an independent
class schema ('superschema') in Kiswahili class 5, not necessarily connected to the
prototype 'fruit'.

-58-
4.4. Class 6
In section 4.4 I will suggest that the non count nouns in class 6 have the schema
'liquids' as their prototype. The different extensions from the prototype reflect
one (or both) of the characteristics of 'liquids ': 'no inherent boundary' and
'internal homogeneity'. These characteristics are represented as the two class
schemas in the schematic network of class 6.

Class 6 includes regular plural formations of roots occurring in the singular

classes 5 and 14. In addition, the class exhibits a substantial number of class 6
nouns that don't have singular counterparts. It is the latter group of non count
nouns that is the basis for the semantic analysis in this section. I will only briefly
comment on the relation between the non count class 6 nouns and the count
noun plurals of class 6.

Earlier treatments of the semantics of Bantu class 6

Cole (1955:81-2) divides these class 6 nouns that "are normally used in the plural
only" in four categories: 1) "Basic collective nouns [... ] mainly signifying liquid
and semi-liquid substances." 2) "Collective nouns formed from stems normally
appearing in other classes, [... ] and signifying large numbers or aggregations of
animals." 3) "Abstract or semi-abstract nouns." 4) "Nouns indicating times or
seasons: Owing to their temporal significance, these function mainly as adverbs
of time."
Torrend (1891:97-100) connects the "sub-class MA [class 6] without singular
with the class BU-MA [classes 14/6] ". He states that "The substantives which
have no other classifier than MA- are principally those of fluids or quasi-fluids,
or again of things which melt naturally." He also mentions that "In several
languages, principally in Chwana [Setswana], MA- is often used to express great
number."
Meinhof (1906:8-9) claims that class 6 originally contained: 1) "doppelt vor-
handene Dinge". The other semantic sub-groups he finds are described like this:
2) "Aus dem dualischen Sinn hat sich ein kollektiver entwickelt." 3) /IDie Fliis-
sigkeiten und ahnliche Stoffe gehen nach der ma- Klasse in der Regel ohne

-59-
Singular dazu." 4) "Aus der kollektiven Bedeutung ist die abstrakte ab-
zuleiten."44
Denny and Creider (1986:218, 224-5) initially separate count nouns from
mass nouns. They claim that: "Mass nouns constitute a separate subsystem
which utilises some of the morphemes of the main (count noun) system, but
which has sufficient semantic distinctness to justify its separate treatment." The
class 6 mass nouns are held to be liquids and viscous substances. Class 6 nouns
are also found in Denny and Creider's count noun system, on the configurational
side, where they are supposed to represent collections of "non-extended solid
figure" units. Most of these nouns would merely be the plural counterparts of
class 5 nouns which are described as (single) units of "non-extended solid figure".
But they also recognise the fact that class 6 "is often used to mark collective
plurals, i.e., those where the units cohere together".
Batibo (1987:15) suggests "usually amassed" as the class 6 noun «label»:
"Nouns in this category denote objects which are usually amassed together so
that they are conceived as one entity. Most of such objects belong to non count
nouns, especially liquids."
As mentioned in section 4.3, Spitulnik (1987:62) treats class 5 and 6
together, claiming that "their semantics differ only minimally". However she
states that "[f]orms in class 6 which are not plurals of class 5 nouns (i.e. the non-
count Cl. 6) denote masses, or augmented or collective versions of entities
usually denoted by nouns of other classes."

All these treatments of this sub-group of class 6 seem to point to the centrality of
liquids. 45 The collective meaning and the abstracts are also recognised by some.
The only suggestion that clearly sets itself apart from the others, is Meinhof's
(1906) "doppelt vorhandene Dinge". My data certainly don't support the inclusion
of such a notion in the non count part of class 6. On the contrary, I share the other
authors' opinions (as I interpret them), that liquid masses are very central to
«non count class 6».

44 In English translation: 1) «doubled things» 2) From the dualic sense a collective has derived. 3)
Liquids and similar substances fall into the rna-class, usually without a singular. 4) From the
collective meaning the abstract is derived.
45 Contini-Morava (1994 and 1996) does not include class 6 in her analyses of Kiswahili noun classes.

-60-
Proposal for a schematic network of class 6
The semantic network that I will propose in the following, includes the schema
'liquids' as prototype. 'Internal homogeneity' and 'no inherent boundary' are
suggested as the two class schemas. These are abstract generalisations that, in
addition to explicating the physical masses themselves, reflect important aspects
of more peripheral groups that I will suggest are related to the prototype by
extensions.
The extension relationships that I will suggest for class 6 may be
represented as follows:
other (Z) (3) (7)
physical ............. Ill•· collective ••11.......... . ordinary
(1) ..................111 masses plurals plurals
liquids
....__ __. .......11• unspecified collectiv~6)
concrete .........u•
events
objects (4)

temporary •tll ......l(lo qua1..

Ities
states

figure 16

The instantiation relationships are showed in the following two figures:

no
inherent
boundary

(7)
•11~ ........... ordinary
............... " 111' masses plurals
s tates ·•11.............. liquids
6
.....__ __. .......11• unspect'ft' ed collective( )
concrete ......... 111•
events
objects (4)

temporary ..
•ttl ......ur•· qua1tties
states

figure 17

-61-
 internal

(3) (7)
, 11~........... ordinary
plurals
masses
states •1(1............. liquids
..__ __. .......u.. unspecl"f"led collective(6)
concrete .......... JI•·
objects (4) events

temporary ..
••II ......JI•• qua 11ties
states

figure 18

(1) The prototype: liquids

Liquids are perhaps the most typical example of things languages tend to treat as
uncountable concepts, or 'mass concepts'. According to Langacker (1987:203-7 and
1991:18-19), mass terms reflect construed unboundedness and homogeneity.
Unboundedness refers in his terms to the non-existence of "a limit (internal to
the scope of predication) to the set of interconnected entities that constitute a
region" (Langacker 1991:585). Homogeneity refers to absence of internal
variability (literally or construed), and is intimately linked to the notion of
indefinite expansibility and contractibility; any subpart of a mass (however small
or large) counts as an instance of the category.
Liquids are among the clearest examples of mass phenomena in the
domain of three-dimensional space (which contains concrete, physical
substances). They are inherently unbounded in a literal sense. That does of course
not suggest that we never meet clearly delimited portions of liquid. But only
containers can provide liquids with clearly observable spatial limits. Milk is
considered milk, whether it takes the shape of a glass it is in, is spread out on the
surface of a floor, is in the cow's udder etc.
The other central aspect of liquids is their internal homogeneity. We can
split them up or merge them into any size, without having to use a new term;
any portion of milk- from a particular drop of milk to the generic use of the term

-62-
(«milk is healthy») - is correctly referred to as «milk)). The reason that liquids
tend to be treated like this in languages is that they don't display any observable
internal structure.
As we will see, the other groups of concepts in class 6 deviate from the
liquids largely in terms of degree of real display of unboundedness or
homogeneity. The prototypical status of the liquids is reasonably assumed exactly
because of this special status; as entities that exhibit maximal literally perceived
unboundedness and homogeneity. Thus, they instantiate the two class schemas
'no inherent boundary' and 'internal homogeneity' in a maximally concrete way.
The following are examples of liquids from the class 6 noun sample:46

madila thick milk; sour milk

makubelo the last milk in a sack which only comes out after shaking
makgaritlha the dregs of beer, traditionally reserved for older men
maphoko froth on ground caused by rain water
masi milk

(2) Other physical masses

This group of concepts refer to «powdery)) or «grainp) substances. The nouns
listed below are examples of such concepts.

mabududu dry dust

makhafe fringes
mareremela clots in a gravy; curds of sour milk, etc.
matsanko wounds
ma tshetshetho bran from sifted meal; husks etc., from sifted grain

They differ from 'liquids' mainly in that it is possible (to a greater or lesser extent)
to conceive of the substances in question as more individuated than liquids. In

46 That liquids are considered the prototype of non count class 6, does not imply that all liquids are

found in this class. The singular class 14 (which is the most typical «abstract class») also
entails a number of liquids and other physical masses like bojalwa (alcoholic drink) and
bogobe (porridge). These nouns can pluralize, using the class 6 prefix. When they do, they
naturally don't refer to «several instances» of a certain mass/liquid, but rather to different
types or kinds: Majalwa and magobe refer to different types of drinks or porridge, in the same
way that «wines» in English refer to different types of wine.

-63-
other words: Our sensory apparatus allows us to perceive the internal
constituents of these «objects» to a greater degree than in dealing with liquids.
We can (although scarcely) see and talk about «a speck of dust» whereas we
cannot observe any «speck of water».47 On the other hand, it would make as
little sense to talk about thirteen or one thousand «dusts» as it would to talk
about thirteen or one thousand «waters». We simply don't interact with the
substance called dust in such a way that the exact number of specks matters to us
(except, of course, in rare cases when we wish to draw attention to one particular
speck). The fact that humans tend to perceive dust as a mass is reflected in the fact
that this little speck of dust neither in English nor in Setswana can be referred to
as «One dust».
Thus, I see these concepts as extensions from the prototype, instantiating
the same two class schemas ('internal homogeneity' and 'no inherent boundary')
that the liquids do. On a scale of progressively greater individuation (as
illustrated in Langacker (1987:205) with the example water> dust> sand >grass >
tile > cattle > timber ) these concepts come very close to the «less individuated»
end of this scale.

(3) Collective plurals

From 'physical mass' I will suggest that there exist in Setswana a semantic
extension to 'collective plurals'. With some nouns that have «normal» singular
and plural variants in other classes ( e.g. nku -sheep, sg. class 9, dinku -sheep, pl.
class 10), one has the possibility to «use» the class 6 prefix for the purpose of
profiling collectivity (manku - «collectivity /mass» of sheep, class 6). These
concepts do not share the characteristics of the physical masses treated above in
any literal sense. The extension is metaphorical. It can be argued that a collective
plural reflects a mass construal (Langacker 1987:203-5). It reflects that the speaker
has «imposed» a kind of homogeneity and unboundedness upon the objects in
focus. As such, even the collective plurals instantiate the two class schemas
'internal homogeneity' and 'no inherent boundary'.
The listed nouns below show some examples of this «collective use» of
terms. The typical candidates for «collectivising» are animals and plants,

47 A drop of water clearly does not have the same status as a speck of sand: Whereas a speck can be
argued to be the smallest, inherently bounded entity that dust is composed of, the same can not
be said about a drop of water.

-64-
although some other objects and phenomena also can be treated in this way (as
we see in the case of magosi).

magosi kings; kingdoms; chiefs; chieftains

makolane a thicket of palm trees
manku many sheep; flocks of sheep
maroana a thicket of mimosa trees of a certain species48
mabi a thicket of a certain species of thistle 49

I would argue that it is neither the mere fact that these terms are "signifying large
numbers or aggregations (Cole 1955:82), nor that they refer to "animals of the
II

same kind that are confined to the same place (Ntsime and Kruger 1989:67), that
II

leads to the use of a collective term. It is rather the speaker's mental construal (the
individuals are «de-focused» and the group as an undifferentiated mass is
highlighted) that determines the choice of a collective rather than an ordinary
plural. A conceptualiser can, in other words, ignore the fact that it is easy to identify
the internal constituents of the collective plurals, and that these constituents
usually are not precisely identical. It is, however, not surprising that «great
numbers» and «confinement to the same place» are the two situations that usually
are given as explanations for the use of the collective plural in Setswana; a few
widely dispersed sheep are, of course, not as easily conceived as a mass, as a big
group of individuals that stay so close together that one hardly can make up their
individual outlines from a distance.

(4) Unspecified concrete objects

The same «imposed» mass construal must be the reason for trea ting the
examples below as non count nouns. Some of them sound familiar to an English
or Norwegian speaker (malao - spices), whereas others do not represent entities
that English or Norwegian treat as non count terms (in fact some are not treated
as lexical items at all): maadimo - borrowed things, mantsa - loose hanging
things.

48 The dictionary lists "a species of mimosa trees" as the meaning. Setswana speakers that I have

discussed this with, claim that this is the collective term for an otherwise class 3 noun.
49 This is probably also a case of dictionary error, when the dictionary gives the meaning: "a

species of thistle". In this case, however, it is not unlikely that the collective plural would be
the most frequently used term for the thistles, as they u sually grow in thickets.

-65 -
maadim6 borrowed things; borrowing or a loan; borrowings or loans
malao spices
mantsa loose hanging things
manot6 strays, as of locusts, cattle, etc.
masomosomo many; crowds

What is going on here is basically the same as for the collective plurals, the
difference being a matter of degree. The requisite abstraction is more extreme in
these cases. A comparable (and well used) English example is a term like
furniture. The actual entities referred to by such a very schematic term may be
highly diverse. Nonetheless, it is possible to construe those entities in a
homogeneous way, because of their similar fundamental character, not of
configuration, but of function .
Because of construed homogeneity and unboundedness this group of
concepts is reasonably seen as yet another metaphorical extension from the
categories of physical mass, instantiating the class schemas 'internal
homogeneity' and 'no inherent boundary'.

(5) States
So far I have delt with concrete objects (touchable objects existing in three-
dimensional space), but class 6 also entails a substantial number of less concrete
concepts. Langacker (1987:207) suggests that "abstract nouns - like their more
concrete counterparts - designate regions that can be either bounded or
unbounded", and that this is reflected in a distinction between count and non-
count nouns. It is a fact that abstract nouns, just like the concrete ones, can be
divided into count nouns (like the English: (a) complaint, (an) argument, (a) yell,
etc.) and non count nouns (despair, hope, yelling, etc.)
The abstract non count nouns in my class 6 sample all refer to different
states. Jackson (1990:10) divides the group of states further into temporary states
and qualities .50 He defines a 'quality' as a "more-or-less permanent characteristic
of someone or something, while a 'state' is a less permanent type of situation".
Although differing in degree of permanency, both temporary states and qualities
are conceived not as single, momentary (or rather: «bounded») episodes, but as

50 His focus here is on verbs and situation types, but I find the definitions applicable to my data as

well, although I only use part of it; he actually divides 'states' into four subgroups: quality,
temporary state, private state, and stance. Gackson 1990:10-5)

-66-
conditions that last for some time. The start and end points of these conditions
are not in focus. In other words: temporary states and qualities don't exhibit
inherent boundaries with respect to time. In that way they resemble the concrete
masses; the difference only concerns which domain the lack of bounding refers to
(three-dimensional space vs. time).
One can also reasonably suggest that humans construe temporary states and
qualities as being homogeneous; we think of them as representing «single,
coherent types of conditions» and remove the possibility of internal variation
and complexity from focus.
Thus, the 'states' seem to represent a metaphorical extension from liquids
(and the other physical masses), related to them by the class schemas 'no inherent
boundary' and 'internal homogeneity'.

Temporary states
The following temporary states are from the class 6 noun sample:

makgwere a state of confusion in the mind; problems; difficulties

mahupuhupu state of being swollen and bruised from a beating; being bitten,
as by bedbugs
mangetengete state of being dented, or pierced all over
manyerenyere state of being very wet, very muddy
matlhatlharara state of being untidy; being in a dishevelled and unkempt
state

Qualities
The following qualities are from the class 6 noun sample:

maatla strength; power

mafatswa persuasiveness
makgakga rudeness; impertinence
maona immaturity
matletsetletse great fullness; abundance, as provisions

(6) «Collective events»

The last group of non count concepts in the semantic network of class 6 I have
called 'collective events'. These events, I will suggest, are construed in more or
less the same way that the 'unspecified concrete objects' are. The «internal

-67 -
 diversity» is de-focused, and «imposed homogeneity» is highlighted: Each event,
as matshediso - condolences, (or any other from the example below) of course
consists of several distinguishable smaller «action episodes», like the actual
utterances, the way of behaving in such a situation, etc. But that is not what is
focused upon. Rather, these notions refer to the whole «set of action episodes» as
if they consisted of a «homogeneous substance». In referring to matshediso one is
not concerned about the internal structure of this event. What matters is that the
whole «set» of behaviours altogether is taken to be an instance of condolences. In
addition, I would suggest that the start and end points of these events are not
focused.
Thus, 'collective events' represent metaphorical extensions into the more
abstract «domain of actions» from the 'unspecified concrete objects' belonging to
the spatial domain. They instantiate the class schemas 'internal homogeneity'
and 'no inherent boundary'.

madume greetings
maiphitlha-
phitlhwane evening recreation; hide and seek
mapekepeke flashes of light; sparks
magawegawe disputes; uproars
matshediso condolences

Although it is not evident in my proposed schematic n etwork, there are clearly

some similarities between this group of concepts and the 'states'. Both groups
refer to «situation types». Hence, they produce a common abstract schema that
one could label 'situation types'.
For graphical reasons I have not integrated that into the network, as it is
not of major importance for the general understanding of the semantics of class
6.

(7) Count noun class 6 plurals (ordinary plurals)

Apart from including non count nouns, class 6 is one of the plural classes in
Setswana: Noun stems with class 5 and class 14 prefixes (both these classes are
singular classes) occur with the class 6 prefix when they are used as plural nouns.
Historically class 6 is a Proto-Bantu merge of two different classes, one referring to
plurals and the other to masses (see for example Greenberg 1963:31). The class of

-68 -
masses seems to be the origin of the non count part of today's class 6, while the
rest consists of regular plural forms.
Although of different origin, a possible semantic link between the two
groups of nouns can be established on the background of our construal of plural
phenomena in general. Langacker (1991:76) argues that ordinary plurals belong to
the mass noun category: "the profile of the plural predication is a region that is
unbounded within its scope in the domain of instantiation". What distinguishes
a plural from a non count mass noun for Langacker (1991:77-8) is that a plural
"profiles a mass that we can think of as being formed by replicating indefinitely
many times a discrete entity that we are accustomed to dealing with individually
[... ]" This «replicability» is either not present at all (as in madi - blood) or de-
emphasised in different degrees dependent on the perceivable «particulate
nature» of the entity in question (ranging from examples like matshetshetho -
bran from sifted meal, to the 'collective plurals'). Accordingly, he refers to plurals
as 'rep licate mass', while a non count mass noun is referred to as 'non-replicate
mass'.
The 'replicate' and the 'non-replicate' masses thus share the class schema
'no inherent boundary', while they differ in (conceived) internal structure.
The network that I have suggested for class 6 is based primarily on the non
count nouns. If I were to present the complete class 6 network, the suggested
network would become only a relatively small part of the larger network
including all the different other schemas representing the plural versions of the
class 5 schemas (of which we have seen that a majority instantiate a 'non-
individuability' schema) and the class 14 schemas (which typically would include
abstracts and liquids and viscous substances). Integrating the schema 'ordinary
plurals' as one among the other parts of the n etwork as I have done in this
section, then, is an extreme simplification, perhaps even an indefensible one. Yet
I have chosen to do so in order to show that there might be an associative link
between these plurals and that part of a «full network» that represents the non
count nouns.
The schema in the class 6 network that is most directly linked to the
ordinary plurals, are the 'collective plurals'. It would clearly be unreasonable to
claim that the large group of ordinary plurals are extensions from the collective
plurals. An extension relationship will have to go in the opposite direction,
starting off somewhere in the larger «full network» of class 6. That is what I have
tried to indicate when I, in the figure representing class 6, have represented the

-69-
relationship between the ordinary plurals and the rest of the non count class 6
network.

The mass term in linguistics

In linguistic literature the terms 'non count noun' and 'mass noun' tend to be
equated. The definition of 'mass' given in Crystal (1991:212) illustrates the point:
"A term used in the grammatical classification of nouns; opposed to count." It is
easy to establish the non count nouns in a given language; they simply don' t
occur marked for 'number'. Intuitively it seems rather likely that the grammatical
term 'mass' in addition points towards certain semantic qualities. Accordingly,
Crystal continues: "The term refers to those nouns which the speaker treats as
continuous entities, having no natural bounds (contrasting with the separable
'countable' quality of count nouns)." However, faced with the fact that different
languages don' t treat all possible candidates for a «mass description» in a similar
manner, he states: "but the distinction is not made on semantic grounds alone;
the contrasting pattern of co-occurrence with determiners, quantifiers, etc., is the
main evidence, e.g. *an anger vs. some anger shows this to be a mass noun.
There is no logical reason why nouns should be count or mass: a concept may be
countable in one language, but mass in another, as in the case of information,
which is mass in English, but countable in French (des informations). "
Many linguists have used the apparent non-systematicity between the
ascribing of count vs. non count nouns to apparently similar concepts, as a
«proof» of the arbitrariness of grammar. I would rather interpret this as an effect
of the grammar's sensitivity to human conceptualisation. Wierzbicka (1985:316)
gives some striking examples as to what extent we are able to change our
viewpoint concerning which entities that possibly could be perceived as mass (or
non count) entities. She refers to examples from literature, among them the
mother termite who is concerned about her child and utters the following:
"Johnny is very choosy about his food. He will eat book, but he won't touch
shelf." Here «book» and «shelf» are treated as mass concepts. Such shifts in
conceptualisation are indeed possible: the sentence has not become
incomprehensible or «ungrammatical»; it rather helps us to see «book» and
«shelf» as «foodstuff», on a par with «porridge». Hence, it does not seem unlikely
that different languages place the boundary between countable and uncountable
differently, and that one single language can draw a «countability boundary»
between seemingly similar concepts like «oats» and «wheat». Or in Wierzbicka's
own words: "Obviously, there are different possible ways of conceptualising the

-70-
world, there are different possible cognitive perspectives, and a termite's
perspective may differ from that of an Englishman. But the different concep-
tualisations have their own logic, which grammar reflects and illuminates."
On the other hand: some objects and phenomena are more readily accepted
as mass terms than others. To be able to accept «book» or «shelf» as mass terms,
we have to take the termite's viewpoint. For human beings it is a lot more
natural to treat a book as a self-contained clearly bounded entity that cannot be
«portioned out» in small parts that still count as «book». In other words: «book»
is a clear candidate for the label 'countable'. The most likely candidates for mass
descriptions will be objects with no clear-cut boundaries, and which can be
divided arbitrarily and still be named with the same term. Such concepts will
tend to be treated as mass terms in most languages. But there are also a
substantial number of concepts that can be conceived of as easily in a «countable
manner» as in a «mass-like manner» (as for example Crystal's (1991:212) example
information vs. des informations). Such phenomena will tend to be treated
differently in different languages.
What is not completely predictable is not by necessity arbitrary: Although
we see that not exactly the same concepts are treated as mass concepts in Setswana
and English, that does not imply a lack of semantic motivation.(cf. Lakoff 1987).
On the contrary, we have seen that variability of human construal can lead to
different «solutions» in cases where it is not all that obvious whether it is
reasonable to treat something as a mass or not (madume vs. a greeting/ greetings).
The greatest bulk of nouns in the two languages will, however, be confined to the
same «countability category» for obvious reasons: Only a weird language would
conventionalise number marking on terms referring to liquids and not to those
referring to clearly delimited concrete objects.

Concluding remarks to the proposed class 6 network

Explicit referral to a notion of 'mass construal' in my analysis, is perhaps the most
important contribution to the understanding of the non count part of class 6.
Without the understanding that a language user has the ability to construe a
situation, object or phenomenon in alternate ways, it is difficult to see the
semantic connection between as apparently different objects as masi (milk) and
mabodi (the collective plural of podi - goat).
The non count class 6 nouns in Setswana shows us which concepts the
speakers of this language have treated as «mass-like». We see that some of these
concepts are treated in the same manner in English, while others are not.

- 71 -
 Liquids, collective plurals, and to a certain extent abstracts, have been put
forward as relevant groups in the other works with which I have compared my
analysis. With the exception of Spitulnik (1987) no explicit attempts have been
made neither to connect these three groups of concepts semantically, nor to
connect other perhaps less obvious groups to these most typical ones.
Lastly, there seems to be a tendency that reduplication of noun roots is used
a lot more frequently in «non count class 6» than in any other class. If that is the
case, it might represent an interesting indication on a connection between the
noun form and the semantic content; it surely would make sense if for example
states that reflect the result of repeated actions (like mangetengete - state of being
dented, or pierced all over) rather regularly were represented by reduplicated
roots. Unfortunately though, I have not been able to dig into this possibly
interesting area.

-72-
4.5. Class 7
In the following I will suggest that the class 7 semantics is based upon the
prototypical schema 'instrument', in the sense of «intermediary in actions». We
shall see that the notion of action becomes important; almost all extensions from
the prototype instantiate the class schema 'participation in action'. The network
represents a «metonymic chaining» where almost each schema mirror one of the
participants in a prototypical «action chain».

Earlier treatments of Bantu class 7 semantics

Cole (1955:84-6) again uses the label "miscellaneous" when he characterises the
semantics of this class. According to him, the class "includes names of parts of the
body, tools and instruments, animals, languages and characteristics, and a
number of personal nouns". He divides the personal nouns in two categories:
those that refer to "people having a mental or physical defect, or some habit,
custom or other peculiarity of manner or appearance which characterises them as
being abnormal" and those that refer to "people who are experts or have some
special ability or desirable characteristic". Cole also points to the fact that a lot of
nouns in this class are derived from verb stems.
Torrend (1891:112) sketches four groups of notions belonging to class 7: 1)
the names of languages, 2) the word for "a thing, and some substantives in which
this word is understood", 3) "the words for any sort of limited break, or cut, on
land or water, or on a body", 4) "short" things, or rather, as he explains it further:
things that are "relatively thick in one part and small in another, or halved, or
protruding with a thick basis and to a comparatively small height etc.".
Meinhof (1906:12-3) states that in class 7 "Sitte, Gebrauch, Werkzeug"
(custom, use, tool) are meanings that are easily recognised. Comparing with
Torrend's notions, "Sitte" (custom) includes languages, as the Bantu languages do
not distinguish between language and the manners, ways or customs of the
people under consideration. Meinhof continues by claiming that "a us dem
Begriff «Sitte, Art und Weise» entwickelt sich die Verwendung des Prafixes zur
Bezeichnung von Abstrakten". 51 He then maintains that the class 7 prefix
expresses the notion "Ding, Sache" (thing), but does not, as Torrend does, refer to
the fact that even the generic term 'thing' belongs to this class. The use of the

51 In English translation: "from the notions «custom and manner», the use of the prefixes to signify

abstracts has developed".

- 73 -
 "Sachenprafix" («thing-prefix») to refer to human beings implies, according to
Meinhof, that such persons are being treated "verachtlich" (contemptuously).
In Denny and Creider's (1986:223) classificatory system, class 7 belongs to the
noun classes that can be classified on the basis of "kinds". They suggest that the
primary meaning in class 7 is "instrumental artifact", something that would
include Meinhof's "Werkzeug" and perhaps also represent one specification of
Meinhof's notion "Gebrauch". Denny and Creider also argue that it is "a fairly
natural extension from 'used object' to 'despised object' ", and that it appears that
this additional meaning had been established in Proto Bantu for class 7. If this is
applied to human beings, it resembles Meinhof's argument that the «thing-
prefix» adds a mark of contempt to the persons that the term is applied to,
although Meinhof focuses on thing, whereas Denny and Creider focus on use.
Batibo (1987:17) argues that the original notions in class 7 "denoted objects
with marked or special characteristics. Such characteristics could be deformity,
artificiality, eventfulness or speciality." Deformity, or "lack of something" as he
explicates it, may be seen as connected to the notions of contempt that both
Meinhof and Denny and Creider touch upon. And artificiality (Batibo himself
adds: "man-made") is reasonably interpreted more or less along the same lines as
Denny and Creider's notion "instrumental artifact". Meinhof's "Werkzeug"
could probably also be included in the concept 'artificiality'.
Spitulnik (1987:64-7) starts her treatment of class 7 by claiming that this
class often is "considered a kind of residual class in Bantu since it contains the
widest variety of 'inanimate' nouns, including the generic term for 'thing' ". She
goes on to demonstrate that many ChiBemba class 7 nouns are generic nouns or
refer to artifacts and parts of things. As we have seen, both ' thing' and artifacts
have been extracted from class 7 by other authors also. According to Spitulnik the
class 7 prefix in ChiBemba in addition functions to derive augmentatives of
nouns from other classes. It is also productive in deriving nouns from verbs, and
Spitulnik claims that when they do, the derivation seems to imply a notion of
"discrete, bounded thing". She also, as many others, recognises that numerous
nouns referring to names of various Bantu languages are found in this class. The
human beings in class 7, she refers to as "those who are unsocialized or
completely outside the social world". Contempt is often a result of such
situations, cf. Meinhof and Denny and Creider.
Contini-Morava (1994:13) refers to Denny and Creider's statement that the
"primary meaning [of Proto-Bantu Class 7] is instrumental artifact". She agrees on
the assumption that 'primary meaning' is interpreted as 'prototypical meaning'.

-74-
Further, she is adding to the notion of instrumental artifact (or "utilitarian
objects" as she calls it) the specification: "small enough to hold in the hand",
because, as she puts it: "this applies to the majority of terms for instrumental
artifacts in Class 7, and it provides a motivation for the major semantic extension
within the class, to 'small entities in general', not all of which are instrumental
artifacts". In her proposal for a semantic network, Contini-Morava includes the
following: small artifacts, small animals, immature beings, people with physical
defect/lack, pieces/parts of things, shortened things, pointed things / parts, part of
substance, similarity /manner, small body parts, ailments associated with body
parts and concrete object associated with verb (product, implement, etc.).

As we can see above, there are, despite of different viewpoints, some rather
conspicuous similarities between the quoted authors: Instruments/ artifacts, the
generic term 'thing', languages/customs as well as negatively marked human
beings all seem to be among the notions that most of these different authors
extract from their data.

Proposal for a schematic network of class 7

My own suggestion for a semantic network is built around 'instruments' as the
prototype. That is hardly surprising, considering the other contributors'
considerable attention to the same phenomenon. However, my focus will not be
on the «man-made» aspect of these objects, cf. «instrumental artifact». Rather, I
have come to the conclusion that the central point concerning these objects is
their instrumental functions; that they participate in specific manners related to
different actions. This is then the background for different semantic extensions
from the prototype to other groups of concepts within the network.
The whole network of extension relationships in class 7 seems to represent
a parallel to a «chain of actions», or what Langacker (1991 :285-6) refers to as "the
canonical event model" which "represents the normal observation of a
prototypical action". Of course, «action chains» or «event models» are usually
thought of as belonging to grammatical structure involving verbs, and the
semantic role of a noun is suggested from a broader context. Nevertheless, a
comparison with typical elements (or participants) in an action chain has been
clarifying for the understanding of the semantics of class 7 nouns: We shall see in
the following that the extension relationships in class 7 in a way mirror the
relations between typical participants in action chains; the different noun
schemas together constitute a metonymic chaining based on the different

-75-
relations between typical «participants» in actions: instrument, agent, manner,
place, and affected.

The network of extension relationships that I have suggested for class 7, can be
represented as follows:

manners \.<:J . (6)

objects
of acting affected
(languages) •II•································· instruments .............................. HJ, by action
(mte:- (1)
med1ary 13\
in actions) ............. n•• places ,_,
associated
~ith actions

(5)
other negatively positively
negative •11·..··•••·•· acting •111........1111 acting
persons persons persons

figure 19

The network including the class schema can be represented as follows:

participant
in action

other (S) negatively positively

negative •11•.......... acting 11'"'"'"11•· acting
1

persons persons persons

figure 20

-76-
(1) The prototype: instruments
Langacker (1991:285) refers to the 'role archetype' instrument as "a physical object
manipulated by an agent to affect a patient; it serves as an intermediary in the
transmission of energy" .52 The «setting» for an instrument is in other words an
active (typically human) performer of an action, using the instrument with the
purpose of affecting a passive (typically inanimate) object.
In dealing with noun semantics we are, of course, no longer speaking about
semantic roles in the usual sense. Rather, we are focusing on what kind of
concepts certain nouns refer to. In this case we are faced with a group of concepts
prototypically referring to physical objects manipulated by an agent to affect a
patient. Or in other words: instruments function as «intermediary in actions».
We see that this prototype definition equates with Langacker's definition of the
'role archetype' instrument above.
In the following, we will see that these nouns that instantiate the
'instrument' schema, together with other schemas in the network instantiate an
abstract schema that we may call 'participant in action'.
As the great majority of nouns referring to concrete, non-animate objects
in class 7 refer to different types of instruments, and as we will see that these
instruments serve as the source for several other semantic extensions in the
schematic network of class 7, I have chosen to represent the instruments as the
class prototype.
As the examples below shows, many of the nouns referring to instruments
are derived from verbs; verbs referring to the action in which the instrument is
used as a means. I suggest that this fact represents an additional indication that
these nouns probably are construed as intimately linked to the action that the
verbal origin refers to.

52 Discussing semantic roles, Langacker (1991:285) argues that they represent pre-linguistic
conceptions and refers to them as role archetypes in order to "call attention to their primal
status and non-linguistic origin".
Included in the definition of instrument are two other roles: The role archetype agent: "a
person who volitionally initiates physical activity resulting, thorough physical contact, in
the transfer of energy to an external object". And the role archetype patient: "an inanimate
object that absorbs the energy transmitted via externally initiated physical contact and
thereby undergoes an internal change of state". Included in the notion of archetypes is of course
a conception of prototype, and Langacker's definitions of semantic roles clearly show that they
are based on prototypical characteristics.

-77-
segelelo a vessel for drawing water [go gelela -to draw water/fluid]
segopa a dustpan; a scraper [go gopa -to remove by scraping; sweep
up]
sekampane a bow
sekgampane a kind of musical instrument
sepalangwa vehicle [go palama - to ride, palangwa - is being ridden]

(2) Manners of acting, characteristics (including languages)

This schema represents a generalisation over nouns that indicate the manner or
way in which some action is being undertaken. The nouns in question can also be
used as manner adverbs. 53
Many of these nouns belong to the small group of noun roots that exhibit
«multiple class membership»: In Bantu languages some roots can take two or
more prefixes, and a change of prefix results in a change of meaning. In the
examples below, I have, to illustrate the point, given an alternative form (the
same root attatched to a different prefix) with a different meaning for each item.
Among various types of characteristics, I suggest that names for languages
should be included in this group. The reason is that these terms not only refer to
languages in the strictest sense, but also to the characteristics of the people who
speak the language; their manners, customs and culture. In my noun sample
only two names of languages are represented. (Actually only one of them refer to
a language by the usual term, the other having status as a kind of nickname.)
However, names for languages are assigned to this class on a productive basis, not
only in Setswana, but in most Bantu languages. 54
I suggest that the «manner concepts» represent metonymic extensions
from the prototype 'instruments'. The «manner concepts» instantiate the class
schema 'participant in action'.

sebele personality; reality; a person himself; bodily form [mmele -

body]

53 In the standard grammar of Setswana, Cole (1955:364) calls attention to the adverbial use of class
7 nouns "indicating manner, fashion or characteristics".

54 Thus the term for Norwegian is made by adding the class 7 prefix to the (English) name for the
country of the speakers, giving Senorway as the result. Senorway then refers to the language of
the Norwegians, as well as the manners and characteristics of the Norwegians.

-78-
segomo behaving like cattle [kgomo -a cow, bull]
segosi royal; pertaining to royalty [kgosi - a chief, king]
sejatlhapi English [go ja : to eat, tlhapi : fish] 55 [Mojatlhapi - an
Englishman]
Sekololo language of the Bakololo; in the custom, or tradition of the
Bakololo [Mokololo - a «kololo-person»]

(3) Places associated with actions

The examples below refer to places that typically are associated with certain
actions. Thus, they share a «connection to action » with the prototypical group:
instruments. In that way they can be seen as metonymic extensions from the
prototype.
The group represents an instantiation of the class schema 'participant in
action', although its «participation» is clearly less central than what is the case for
the prototypical group; the place for an action constitutes a «background» rather
than a true participant in an action event.

sebothelo a resting place for cattle [go botha - to repose; rest]

sebatla an open valley, or a treeless place, or plain; an open space in
the veld, or a level place, hence a spot that is suitable for
holding unregulated meetings; a kind of Hyde Park Corner
sefero a path between hedges56
setsha a plot; a plot of ground for building, or other purposes; estate

(4) Persons (agents) with extreme manners of acting

The personal terms that this schema is a generalisation over, refer to human
beings who are characterised by «extreme manners of acting»: Either they are
«experts» who use their skills in an exceptionally good manner, or their manners
are exceptionally fierce or at least intensively negative, cf. the examples below.
Often, these nouns are derived from verbs referring to the type of action these
persons are engaged in.
It is an interesting fact that several of the human terms in class 7 have their
less intensified versions in class 1: A sejadi is an «expert planter», whereas a

55 literally, «the fish eaters' language».

56 where people walk.

-79-
mojadi is just any ordinary planter, and a mogwanti (bully) is not as bad as a
segwanti .
This schema represents a metonymic extension from the prototype schema
'instrument', as it focuses on another «participant» in the action chain; namely
the agent who is performing some act(s) in an exceptional way either by means of
a concrete (spatial) instrument or by means of metaphorically understood
«instruments» like a certain personal skill or ability. Thus, this schema
instantiate the class schema 'participant in action'.

I have chosen to represent the «positively acting persons» and the «negatively
acting persons» as two different instantiations of 'persons with extreme manners
of acting'. In the network figure I have represented the relationship between
these two groups with a double extention arrow, indicating that I have no reason
to present one group as more central than the other:

Positively acting persons:

sebui orator; speaker [go bua- to talk; speak]
seemana a mighty one, hence manna wa seemana , a man of war. Also:
mightiness
segwaba a person who can accomplish or contrive to do difficult things
sejadi planter (go jala -to sow]
seopedi a renown, or famous singer; an expert singer. Also: musical
instrument (go opela - to sing]

Negatively acting persons:

seaka a harlot; a prostitute
segwanti a bully [go gwanta - to be wilful; conduct oneself in a defiant,
or impudent manner; ... ]
seitaodi an unruly individual; a lawless individual; one who is not
law-abiding [go itaola - to be lawless; be delinquent; govern
oneself]
serukutlhi a rebel; a troublemaker [go rukutlha - to rebel; brawl; riot]
seswapelo a sloven; a dead hearted one; desperado [go swa - to die, pelo -
heart]

-80-
(5) Other negatively marked persons
It can be argued that the schema 'negatively acting persons' is responsible for a
further metonymic extension to persons with negative characteristics in general,
but who lack the strong focus on action, as the following nouns exemplify:

sel€ma a fool
semumu a dumb person
serothola a foolish, careless person; one not able to do anything
properly
setete a spoilt child
setlatla a fool; a silly person

The connection between this group of cocepts and the rest of the network can be
represented with a local schema 'negativity', that this group shares with the
'negatively acting persons':

negativity

other negatively
negative ·•II·········· acting
persons persons

figure 21

(6) Objects affected by action

This schema reflects a generalisation over concepts that refer to objects which are
involved in actions or events, but in a clearly passive way. Jackson (1990:24)
defines the semantic role 'affected' as "an object that is directly involved in a state
or event but does not play an active part in it".
The following examples show that these nouns refer either to objects that
typically have been affected by extensive use (with the result that they have
become worn down), or to objects that typically are affected by the action/ event
that is referred to with the verb that they are derived from.

sefolotsana an aborted foetus of an animal; a weakling [go folotsa - to

abort]
segogou an old and worthless skin; (an old skinny person)

-81-
sejo food; a type of food [go ja -to eat]
seketana a rag
sekgwene a stump

This schema thus represent yet another example of a metonymic extension

from the prototype 'instruments' based on shared participation in a «chain of
actions/ events». It shares with the prototype the class schema 'participant in
action'.

Concluding remarks concerning the suggested class 7 network

My suggestion for a class 7 network represents a rather radical break with former
treatments of the class 7 semantics.
That 'instruments' in my proposed network employ the position of class
prototype, is hardly controversial though. It is clearly compatible with most other
works on Bantu class 7 (cf. the quoted authors in the beginning of this chapter).
However, whereas most of these other works focus on «thing», «use» and «man-
made» in dealing with the instruments, I have directed the focus towards
participant in action.
An instrument can easily be seen from different angles, or rather: our
"focal adjustment" may differ (cf. Langacker 1987:489). 57 Hence, a cup - senwelO -
may be seen as a «thing», a «dead object», a «man-made object», a «utilitarian
object», etc. It can also, as I will suggest is the most relevant viewpoint concerning
its class 7 membership, be seen as an object that an agent uses in the act of
drinking (go nwa -to drink, the applied form: nwela). Supporting my suggestion
is the fact that a majority of instrument terms are derived from verbs. 58
Focusing on the fact that instruments take part in actions provides us with
the motivations for the semantic extensions to other schemas in the class 7
network. With the exception of one schema, I have suggested that all schemas in
the network of class 7 instantiate the abstract class schema 'participant in action'. I
have not come across any treatment that gives the action aspect such a central

57 "variation in how a situation is conceived, particularly variation pertaining to selection,

perspective and abstraction" (Langacker 1987:489).
58 An investigation of degree of saliency of the verbal origin of these nouns (for mother tongue
speakers) would be valuable, as a high degree of saliency could represent a further support to
my analyses. Unfortunately though, it has not been possible to include that within the limits
of this thesis.

-82-
place within the semantics of class 7. Rather, it has been common to focus on the
«dead», «thing-like» nature of the majority of class 7 concepts. While
appreciating this tendency, my interpretation of the Setswana data don't support
the centrality of such notions. It is possible that a «full» semantic network,
explicitly labelling all relevant generalisations found in the class, would have to
integrate a schema 'inanimate' (or 'dead object' /'thing'), but that does not seem to
be the generalisation that is best suited for capturing the semantics of class 7 in
general.
Both Spitulnik (1987:64) and Contini-Morava (1994:13) appreciate the fact
that class 7 exhibits a highly productive process of deriving nouns from verbs.
Since they don't put any weight on the notion of action as a possibly important
aspect of class 7, however, they don't interpret this productive process as an
indication that verbal origin might be central to the understanding of the class 7
semantics in the sense that it points towards the centrality of actions . Rather,
Spitulnik (1987:65) continues the «tradition» of interpreting class 7 as the «thing
class», when she suggests "At the most general level, the type of semantic change
accompanying Cl. 7 (ici-) prefixation to a verbal infinitive or verb stem can be
glossed as 'thing' or 'thing of' ". She even suggests a re-interpretation of what has
often been referred to as the «manner of» derivational function of the class 7
prefix, claiming that "[ ...] it seems here that our characterization of 'thing (of)' is
just as adequate. Thus, iciBembd denotes that which is most significantly of the
Bemba, e.g. their way of life, language, etc." I would argue that Spitulnik's
characterisation is not as adequate; according to my view there is no indication
either in Bantu languages or culture that suggest this «thing-interpretation» of
languages or other kinds of manner concepts.
After appreciating the prototypical status of 'utilitarian objects small
enough to hold in hand', Contini-Morava (1994:13) makes a point of «smallness»,
suggesting that '"small enough to hold in the hand' [... ] provides a motivation for
the major semantic extension within this class, to 'small entities in general' [...]".
Again, it is possible that 'small' is a generalisation over a large number of class 7
concepts, and that it could be included in a «full» network of the class. But I have
not seen any indication of its centrality in my data. It is possible, even, that it is
not the size itself that is relevant, but rather the «manipulability» of objects that
are used in actions. In that case, relatively small size would follow rather than be
an independent semantic factor.
Concerning the large amount of nouns derived from verbs, Contini-
Morava (1994:13) introduces the schema 'concrete objects associated with verbs

-83-
 (product, implement, etc.)'. In my view that does not seem to be a very natural
generalisation over a group of concepts; the direct referral to verbs makes it
sound more like an example of «expert categorisation».
It is also my impression that she structures her suggestion for a network in
an unfortunate way. Contini-Morava (1994:11) explains that the topmost category
in a chart, what Langacker calls a 'superschema', is "a maximally abstract category
that holds together the various subcategories". The 'superschema' is in other
words identical to what I have called 'class schemas' in my networks. In her class
7 chart (reproduced in my chapter 2) the topmost category is 'utilitarian objects
small enough to hold in hand', but that category is referred to as the prototype in
the text (Contini-Morava, 1994:13)! And as no other group is represented as an
instantiation (indicated in her charts with solid lines) of the topmost category, it
clearly cannot have the status of a 'superschema'.59
In sum, we have seen that class 7 has provided us with the greatest extent
of variability between different analyses. In my view, focusing on the «thing»
aspect might have been a blind alley, but there remain even more thorough
investigations before one can draw very clear conclusions about the matter.
Nevertheless, I have suggested that the aspect 'participant in action' is more
relevant to the class 7 semantics than the «thing» aspect, and further work could
be done to evaluate such a stand. Experiments aiming at testing the saliency of
the verb base of many class 7 concepts could be one method of further
investigation.

59 An additional weakness is that by using lines instead of arrows in the charts, she does not clearly

indicate the direction of the semantic relationships, unless the lines are to be read
systematically from top to bottom.

-84-
4.6. Concluding remarks
The semantic networks that I have suggested in this chapter, have turned out to
be in accordance with «typologically natural systems» (as explicated in Corbett
1991):

- Many languages exhibit gender distinctions based on configuration. That is

reflected in the configurational contrast between class 3 (long) and 5 (roundish).

- Degree of animacy is also reflected in many languages in various ways. In my

suggested networks we see that animacy or the degree to which something can be
perceived as consisting of «live» material is an additional relevant factor in
characterising class 3 and 5. Of course it is relevant to the meaning of the
«human» class 1 as well. Class 1 exhibits the highest degree of animacy; human.
A large portion of class 3 nouns refers to entities which, although neither human
nor animal, represent «live» material to a certain extent (plants, objects made of
plant material or skin) or can be associated with particularly lively events. The
majority of class 5 and class 7 nouns consists of clearly less «live» entities. 60

- The internal structure of class 7 is based upon concepts that we usually find
reflected in other parts of language (connected to syntax), namely concepts
referring to different participants (or semantic roles) in an «action-chain».
Although usually not thought of as a part of nominal semantics, it is possible to
find other languages which also seem to «utilize» this «action-chain» perspective
in dealing with nouns.
In Fula the (primarily verbal) infix -ir- connects the three concept-groups
instrument, manner, and place. Arnott (1970:348-353) introduces the term
"modal" and describes how "[it] is used with reference both to the manner in
which something is done, and to the means or instruments employed. In fact the
modal and instrumental meanings merge into one another [... ]." Further, he
describes how the locative "is identical in shape with the modal exten sion", and
then how both the modal and locative extension occur in nouns as well; the
modal when the noun is "referring to an instrument", the locative when the
noun is "referring to a place associated with an action".

60 As mentioned earlier, class 9 is sometimes referred to as the «animal class». As such, class 9

would also belong in an «animacy hierarchy ».

- 85 -
 In Hausa "various types of nouns may be formed from verbs by employing
a ma- prefix with or without an -[ii] suffix". The types are as follows: 1) "Nouns
signifying the agent or doer of an action", 2) "[n]ouns signifying a place where the
activity indicated by the original verb is performed", and 3) "[n]ouns signifying a
tool or implement involved in the action indicated by the original verb". (Kraft
and Kirk-Greene 1979:194-96)
In sum, Fula treats together (with the same infix) 'instrument', 'manner',
and 'place'. Hausa connects 'instrument', 'place', and 'agent' . Setswana seems to
integrate all these concepts (and add one more) into the noun class 7:
'instrument', 'manner', 'place', 'agent', and 'patient' ( represented by the schema
'objects affected by action').

- Even the 'non-individuated' schema in class 5 seems related to phenomena

known from other languages: Some languages mark morphologically single
entities of something that normally is treated with a collective term (thus
indicating a 'mass construal'): In Russian the 'singulative suffix' -in is used in this
manner. Izjum (raisins) and soloma (straw) are collective terms. It is, however,
possible to draw attention to a single raisin and a single straw by adding the
singulative suffix: izjumina and solomina. 61 Arabic exhibits the same
phenomenon: formally tuffa:~ (apples) and $adaf (shells) are singular nouns, but
they are used with a collective meaning. In order to speak about one single (or a
definite number of) apple(s) or shell(s) one has to add the feminine suffix -a(t) ,
resulting in tuffa:}Jgjjj and $adafaf..11_62
In these cases the «marked» (and probably least frequent) terms are those
that refer to single entities. In Setswana, the schema 'non-individuated' seems to
suggest that the «unmarked>> case is referring to the plural versions of these
entities. A frequency study of the use of singular terms (class 5) as opposed to
their plural versions (class 6), would be useful. On a hypothesis that the plural
forms of the nouns that are connected to the 'non-individuability' schema are
«semantically unmarked», one would expect the plurals to be significantly more
frequent than their singular counterparts.

- Finally: the distinction between countable and uncountable notions is of course

a very common one.

61 Tore Nesset p.c.

62 Elie Dib Wardini p.c. (The (t) in the suffix is pronounced only in certain surroundings.)

-86-
The analyses that I have presented in this chapter do not account for all the data
from my noun sample. They do, however, give an account of the great majority
of the data. Leaving out the few animals and purely abstract notions (with the
exception of abstracts in class 5), my analyses provide explanations of between
85% and 69% of the data belonging to the four classes. {The figures are as follows:
class 3: 85%, class 5: 76%, class 6: 83%, and class 7: 69%.)
It seems reasonable that the figures are not higher. First of all, my data only
represent a relatively small part of the nouns in the classes under consideration.
It is therefore not impossible that some generalisations (representing local
schemas, or even more important schemas) have escaped me. It is also not
unlikely that one will never be able to account for absolutely every concept; it is
probably practically impossible to reveal every association that is relevant in such
large networks of similarities.
Further studies might, however, throw some more light on the semantics
of these noun classes. In addition to analysing a larger data sample, there are
other fields of inquiry that probably also would be useful for an increased
understanding:

1) Thorough analyses of that portion of the nouns that exhibits «multiple class
membership» (nouns that can change class prefix, and thereby change meaning)
would be useful. I am aware of the fact that a few such studies have been carried
out, but these studies seem to take it for granted that it is only the «multiple class
nouns» that can be studied from a semantic point of view (as in Rakgokong 1986).
Therefore these studies have not searched for possible semantic connections
between this limited portion of nouns and the numerous other nouns.

2) Systematic analyses of borrowings from the same point of view (comparing

borrowings with proposed semantic networks for the different classes) would also
be important.
Borrowings are normally treated in the following ways:
- The new noun may by and large retain its shape, only modified according to
Setswana phonotactics, and become a member of the prefixless class 9 (phone >
founo ). Class 9 includes the majority of borrowings in Setswana.63
- If the foreign noun's first syllable (after phonotactic modification) resembles one
of the existing noun class prefixes, the noun may be assigned to the class of that

63 Cf. for example the extensive list of borrowings in Janson and Tsonope (1991:137-151).

-87-
particular prefix (starch > setatShe (class 7 with the prefix se-)). This is, however,
not necessarily the case. Sementshise (sandwich) is a class 9 noun; the plural of
this term is disementshise (class 10).
- In some cases where phonology would suggest one class, the noun is assigned to
another class according to semantic principles. One example is that ' tree' terms on
a regular basis are assigned to class 3 (by attachment of the class 3 prefix)
independent of the phonological form of the original term. There are other
examples as well, although they might not be numerous: the class 7 noun
sekopisi (meaning photocopier) from English copy/copies suggests that the
language users in this case have assigned the noun to a class according to
semantic and not phonological principles (as a photocopier is clearly an
instrument).
It would be useful to investigate whether, and to what degree, a
concurrence of semantic and phonological motivation (cf. sekurufu - from
English screw) would lead to increased probability for assigning a noun to a
certain class, as compared to cases where only phonology seems to be responsible
for the assignment.

3) A study of the relationship between original verbal meaning and the noun
meaning in the cases of nouns derived from verbs could also possibly yield useful
information.

4} Integrating extensive cultural knowledge into the semantic analyses would

clearly also represent another fruitful point of view (and is absolutely necessary
for a complete analysis).

5) Finally, a broad and systematic comparison with data from other language
families could also be used as an additional means of evaluation concerning
hypotheses about the semantics of Bantu noun classes.

-88-
Chapter 5. Psycholinguistic testing of the hypotheses

5.1. Introductory remarks

As a supplement to the analyses based on the material from the noun database, I
have conducted a psycholinguistic experiment in the form of a nonsense word
test presented for 78 mother tongue speakers of Setswana. The aim of this
experiment was to investigate whether or not (or to what degree) mother tongue
speakers would have semantic associations on noun class prefixes connected to
nonsense stems.
Asking a mother tongue speaker of Setswana what the «noun class
meanings» are, or what the different noun prefixes «mean», does not yield much
useful information. On a conscious level it seems that - apart from the rare cases
of semantically transparent classes like 1/2 - the speakers themselves have little
feeling that the classes may be semantically based. Such a situation could at first
sight seem to support a claim that the Bantu noun classes are largely formal. But
in my view there is no reason to be content with such a conclusion. Denny and
Creider (1986:230) state that:

The cognitive salience of classificatory systems is in general not great (in the
sense that speakers are often unaware of the bases in terms of which
classifiers are chosen), but a weak kind of saliency [... ] may exist and result in
the proper assignment of new vocabulary items to appropriate classes.

Although the speakers cannot give any explicit description of the semantics of the
noun classes, one should thus not rule out the possibility that one might reveal
parts of an unconscious knowledge, when faced with a task that involves
spontaneous associations, rather than carefully weighed statements. Experiments
that seek for such unconscious knowledge can be designed (c.f. Corbett 1991:89):

The most obvious type of experiment to elucidate gender assignment is to

present the native speaker with unfamiliar nouns (unfamiliar either because
they are rare or because they are artificial nouns invented by the linguist) and
to give a test, [... ], to see which gender they are assigned to.

As an example of successful testing Corbett (1991:92) refers to a nonsense noun

test64 investigating the supposedly straightforward phonological assignment rules
of Qafar genders:

64 constructed by R.J.Hayward. His results were given to Corbett by personal communication.

-89-
 The informant was able to produce the answer rapidly and in every case the
gender assigned was that which is predicted by the assignment rules.
Interestingly, he was very puzzled as to how he had done it, though he was
well educated. This suggests that even simple assignment rules do not operate
at a conscious level.
(My italics.)

If speakers operate at an unconscious level even when dealing with rather

«straightforward» systems, we can definitely not expect them to be conscious
about less salient systems.
Corbett (1991:7-69) divides gender systems into semantic and formal
systems. The formal systems are described as either phonological (as the Qafar
example mentioned above) or as morphological as the Bantu nominal classes
(genders). Concerning phonological systems, he is rather optimistic with regards
to the possibility of obtaining useful information by experiments. The following
quotation shows that his optimism does not apply to the morphological systems:

For investigating morphological systems the problems involved in u sing

invented forms are much worse; it is difficult to give the n ecessary
information on the paradigm of an invented noun in a natural way. A variant
of this approach was suggested by Greenberg (1962:173), namely to ask Bantu
sp eakers who know European languages to assign genders to nouns from these
languages, even though they have not been borrowed - an artificial borrowing
experiment; results obtained by Richardson (1967:378-80) proved
' tantalizingly enigmatic'.
(Corbett 1991:90)

Richardson's experiment was conducted with one informant only, and his
treatment of the results implicitly indicates that the number of artificial
borrowings that his informant had to deal with, was very low (Richardson 1967:
379-80). He states that "While some English words denoting abstract qualities or
ideas, [... ], were duly accommodated in the appropriate Bemba "abstract" class
(buu- 14a), others were scattered over three classes (zero 9a, ama- 6, buu- 14a)
[...]". He then goes into details considering the assignment of the English word
'vowel'. His informant first assigned this noun to the classes 1a/ 2a, but ended up
alternating between the classes 9a /6 and 5/ 6. In my opinion his conclusion that
" [... ] the results proved as tantalizingly enigmatic as the situation they were
intended to illuminate" can not be used as evidence for the impossibility of
obtaining useful results from artificial borrowing experiments (or for that sake:
nonsense word tests) considering his extremely sparse data. Until more
convincing results are presented, I choose to regard tests of this type a possible

-90-
 means for investigating unconscious associations that might shed light on the
semantics of Bantu noun classes.
I suspect that the pessimism towards experiments of this type, stems partly
from a widespread belief that there is not much semantics to detect in these
systems, except for the traces of an earlier semantic system (cf. chapter 2 in this
thesis), and partly from a more general scepticism towards experimenting as such
(cf. for example Kiparsky and Menn (1977:64) who refer to a "«strangeness effect»
which causes subjects' performance to deteriorate relative to their normal
speech"). Expecting to find positive results of experiments is, of course, dependent
upon believing that there is a semantic system that is still «alive)).
Within the framework of cognitive grammar, one assumption is that
speakers use their cognitive capacities to evaluate similarities of varying kinds,
and to generalise over smaller and greater numbers of linguistic items, thereby
«creating)) increasingly more abstract schemas (c.f. Langacker 1987). This is
considered a general process. Hence, a useful hypothesis that should be
investigated, is that speakers of classifier languages like Setswana, generalise over
large numbers of nouns, so that more or less abstract schemas emerge within
each class, enabling the speakers to make use of (usually) unconscious
associations when evaluating a «new)) (borrowed or nonsense) term.
Most ' class schemas' (in the sense of the term that I have used in my
analyses) represent highly abstract notions. As such they do not necessarily reflect
the immediately most salient characteristic of a given concept. It should therefore
come as no surprise that linguistic «knowledge)) concerning these very abstract
schemas is of a less conscious nature than the few salient «class schemas)) (as the
class 1/2 schema 'human being'). It is hardly speculative to claim that «being
humam) probably is a very salient characteristic of 'man' (monna), 'woman'
(mosadi), 'visitor' (moeti), 'farmer' (molemi) etc., whereas it is much less obvious
what a speaker would perceive as the salient characteristic of lerOtse (a melon of
a certain species): It may be the colour, the taste, the quality of being foodstuff etc.
In addition to this, however, it is possible for the speaker to compare and make
«similarity judgements)) between great numbers of items on a higher level of
abstraction. This may then lead to the establishment of abstract schemas (as
'roundish', 'non-individuability' etc.), even if usually not on a conscious level.
On this background one could presume that the noun prefixes (or more
precisely: the concordia! elements together with the noun prefix) should, if not
«triggen), then at least favour certain meaning associations. In spite of the
discouraging remarks of linguists like Richardson (1967) and Corbett (1991), I

- 91-
have conducted a nonsense word test with the aim of revealing unconscious
associations between certain class prefixes and a few selected meanings that
would be compatible with certain schemas in my suggested semantic class
networks.
In the following I will first comment briefly on a small pilot test, before I
present the construction and layout of the final tests, as well as their practical
execution. Then I will present and discuss the test results of each test. Lastly, I will
sketch the more general conclusions that one might draw from the test results.

5.2. The pilot test

Before I started working with the noun classes, I carried out a three month study
of general Setswana grammar at the University of Potchefstroom, South Africa.
During that stay, I constructed and carried out two small nonsense word tests; one
with 26 second and third year students and the other with 19 master students. All
were mother tongue speakers of Setswana. The aims of the pilot test were:
1) to find out whether the subjects would be willing to respond to such tests, and
understand how to respond according to my instructions, and
2) to investigate to what extent the answers would support my (at the time) few
guesses about «class meanings».
The most important task was to explore whether a test based on nonsense noun
roots would be feasible at all. Concerning the second point, I did not expect to find
many useful answers, as the tests had been designed without the necessary
hypotheses.
Two types of test items were constructed:
Type A): test items where «a full nonsense noun» (that is: a «real» prefix plus a
nonsense root) was presented together with 5 alternative meanings. The subjects
were asked to pick out one meaning for each nonsense noun.
Type B): test items where a nonsense root was presented without a prefix, and
with one fixed meaning. The subjects were asked to add the prefix that they
considered most suitable.
The nonsense nouns were constructed according to Setswana phono-
tactics65, and were carefully examined by a fluent speaker (although not a

65 as it is described in Cole 1955.

-92-
Motswana)66, to assure that they were not phonologically too close to any existing
terms.
The meanings were given in English to avoid interference from concordia!
elements in Setswana phrases. (The students have a good mastery of English, so I
did not consider the use of English a major problem.)
An overhead projector was used to present the test items, one at a time,
and the subjects were instructed to answer in writing.
The first test was presented for one group of 21 second year students and
one group of 5 third year students. The test consisted of 10 nonsense nouns. The
first five items were of «type A»: The students were asked to choose among 5
alternative meanings for each nonsense noun. One of the meanings I suspected,
on the background of Coles' grammar, to be easily associated with a certain class
prefix. The other four meanings were supposed to be less good examples from the
class. In the following example from the test, my hypothesis was that the prefix
se- more easily would be associated with a kind of instrument, than with any of
the other meanings:67

Example from the test (type A):

sera tala A. a kind of dish (food)
B. a kind of vehicle
C. a body part
D. a person who is performing a certain skill
E. a kind of instrument

The last five items were of «type B»: The students were asked to fill in a prefix (or
none, as would be the case with class 9). Each meaning was assumed to be very
typical for one specific class, so that one could expect most subjects to choose the
prefix from that certain class.

Example from the test (type B):

tsuro a type of bush

66 Prof. S.A. Swanepoel.

67 As soon as I, on a later stage, started to systematise a relatively large noun sample, I realised
that at least the meanings Band D could equally easily be combined with the prefix se-.

-93-
The next test was presented for 19 master students. It was designed in a similar
manner, although modified according to the experiences of the first test. This
time my Setswana professor, who's mastery of Setswana is close to mother
tongue, assisted me in making up one very likely meaning and four, to his
intuition, very unlikely meanings for each of the 6 nonsense nouns in the first
part of the test (type A). I presented both the singular and the plural form of the
nonsense nouns, because it is not possible to distinguish between the singular
prefixes of class 1 and 3 (mo- ). In the second part (type B), I chose to combine all 6
roots with an animal term, with the aim of searching for possible associations of a
class prefixes to different types of animals (domestic, wild, small, big, different
configuration etc.) This was done without any preconceived idea of possible
associations.68 I instructed the subjects to supply the plural prefix together with
the chosen singular prefix, to avoid the problems of the mo- prefix (which is
either a class 1 or a class 3 prefix) and the prefix-less class 9. Using «N-» for class 9
would clearly not be a good solution; there would always be doubts about the
interpretation of a «blank», it could represent a «non-answer», but it could also
mean that class 9 had been chosen, but that the subject had not seen the point of
using «N-» (which would be very academic an «unnatural») in the answer.
I also instructed the subjects to respond even if they found all alternatives
equally weird. In the first test the students had been given the opportunity not to
answer, and that led to relatively many «blanks».
Finally I presented a type A test that I divided in two; one with the
meanings given in English and the other with the meanings in Setswana. I
presented the English version for one half of the master students and the
Setswana version for the other half. I wanted to compare the answers, looking for
possible indications that the Setswana phrases as predicted would «invite» the
subjects to choose meanings according to similarities between the prefixes of the
nonsense words and some prefixes or concordia! elements used in the Setswana
meanings.

The most important conclusion of the pilot test was that conducting a test with
this design, is indeed possible. The subjects did not find it difficult to understand
how to respond, although they found the nonsense nouns weird. Except for the

68 As mentioned in chapter 1, I eventually left out animals from my analyses of the classes that I
treat, realising that this would be an impossible task without including an analysis of class 9
(sometimes referred to as the «animal class))).

-94-
 most obvious cases, like choosing 'person' for a word beginning with mo-, or
choosing mo- for the meaning 'person' (with le- and se- as second and third
choice), I (in accordance with my expectation) didn't get many clear answers.
However, the fact that they actually did respond as expected to the few «clear»
cases like the combining of 'person' with mo- (or alternatively le- or se-) indicated
that they had not been choosing blindly; they had tried to perform the test
seriously.
I concluded that it would be possible to conduct a test with this design.
However, a lot of things would have to be changed.
First of all, I could not of course make the test without having worked out
rather clear hypotheses for the types of meanings associated with the different
classes.
Secondly, while I would still present as one of the choices, the meaning or
the prefix that I expected to be a good candidate for the full nonsense noun or the
nonsense root respectively, I would have to be equally careful when selecting the
other four alternatives. They should be good representatives for their own
classes, instead of only vaguely being «not good candidates» for the class that I
was investigating, as they were in the pilot test.
Thirdly, I could not expect to obtain useful answers unless only one
parameter was changed at a time. In the pilot test I did not consequently keep to
the same choices of meaning when investigating each prefix. Using one set of
meanings when investigating one class prefix and another set of meanings (partly
overlapping or totally different) when investigating another class prefix, makes it
virtually impossible to evaluate the results.
Fourthly, I would have to test each combination of prefix and meaning
several times, while only changing the nonsense root. If not, I would never be
sure whether the test subjects had reacted to some association of the root rather
than of the prefix.
Finally, I should restrict both the choice of meanings and prefixes to the 5
classes that I had selected for my thesis.

5.3. Construction of the final tests

The tests were restricted to dealing with the classes 3, 5, 6, and 7. However, I have
also included the class 1 prefix. On the assumption that class 1 (human terms
only) is completely semantically transparent to the speakers, one could expect that
the subjects at least would show a clear preference for combining the class 1 prefix

-95-
with meanings referring to human beings. Failing to do so, would indicate that
the subjects were answering completely blindly, without even trying to use their
linguistic intuition. Hence, if I found no strong correlation between the class 1
prefix and meanings referring to humans, I could probably conclude that all test
results were worthless, because the subjects had not taken the task seriously.

Test types
The final tests also consisted of nonsense items of the two types that I described in
the preceding section, but with some modification of type B. The subjects were
not given a free choice of prefix; this time they were presented with a meaning,
and asked to choose among 5 alternative forms, that had a common nonsense
root, but different prefixes (the prefixes of the classes 1, 3, 5, 6 and 7).
In this way it became clearer what the alternatives were, and it was the only
way to make sure that the subjects kept to the prefixes that I wanted them to
choose among.
Example:
a soft-drink setlhorape
matlhorape
letlhorape
motlhorape (metlhorape)
motlhorape (batlhorape)

The type A consisted, as in the pilot test, of a full nonsense noun (constructed of a
«real» prefix plus a nonsense root) presented with five alternative meanings.

The reasons for keeping largely to the design of the pilot test was that:
1) such a test had proved successful in the sense that it seemed easy to make the
test subjects perform the task,
2) performing the test was not very time consuming,
3) because the test can be formed as a written «cross off test», it is possible to test
several individuals simultaneously, and
4) a «cross off» system means that systematising the results is possible without a
lot of time consuming interpretations of more or less vague answers.

The advantage of such a test is then that it is possible to test a high number of
subjects. Due to more or less diffuse answers from the pilot test, and the rather
pessimistic comments in the literature as to the possibility of revealing a Bantu

-96-
class semantics, as well as problems connected to the «unnaturalness» of such
testing, I did not expect to find only clear and unambiguous answers. I did,
however, hope that above a possible «mass» of rather chaotic answers, there
would be reflected some systematicity that could be used in evaluating my
suggestions for some parts of the semantic networks. To be able to draw any
conclusions at all, I felt that I needed a substantial number of subjects. Therefore I
chose to give priority to forming a test that easily could be carried out with many
subjects, rather than forming a test with a great variety of tasks that would imply
time consuming interaction with every subject.
I constructed 8 tests with 90 single test items altogether, and organised
them in a Part 1 and Part 2. The two parts were to be conducted on different
occasions; responding to 90 items is clearly too much for a single session.
Test 1 (with altogether 20 test items) focus on constructed meanings that
would instantiate the prototype schema of class 1, 3, 5, 6, and 7. Test 7 and 8 (with
20 test items in each) focus on constructed meanings that would instantiate less
central schemas in the 5 classes.
The other tests were designed to investigate:
-the postulated prototypical character of the class networks (test 2, 3, and 4- with
2 test items for each test),
- the fact that human terms can be found in different classes (test 5, with 16 test
items), and
- how the subjects would react to constructed meanings that «invited to two
different foci» (test 6, with 8 test items).
Of these tests I will only present and discuss test 5 here. The design of the 4
other small tests (2-4 and 6) for various reasons turned out to be failures. Hence, I
have chosen to present these tests and their results in the appendix, together with
comments on why I think that they were unsuitable.

The nonsense roots

The nonsense roots were constructed according to Setswana phonotactics (cf. Cole
1955), and controlled and adjusted by a mother tongue speaker (Simon
Chilembo), to ensure that the phonological distance to other «real» items was
large enough, and that the roots did not sound foreign or unnecessarily odd.
In my noun data base I had tagged the noun roots for both cv structure and
end vowels. 51% were of a cvcv type, 25% were of a cvcvcv type, 17% were of a
cvcvcvcv type, and the rest were small groups of different shapes. I constructed

-97-
the nonsense roots in accordance with these percentages of cv types, to make
them appear as typical as possible.
I chose the end vowels of the nonsense roots equally carefully; using the
actual distribution of end vowels within each noun class (from my sample) as
patterns for the nonsense roots' end vowels.

5.4. Data collection

The tests were conducted with 78 mother tongue speakers in Gaborone, the
capital of Botswana. The majority of test subjects were secondary school students
between 16 and 19 years old; 44 from Gaborone Secondary School (GSS) and 25
from Maru A Pula. I also tested 9 adults between 22 and 53 years old, and of
varying professions (a teacher, a receptionist, a housemaid, a gardener, a librarian,
two secretaries and two university lecturers).
The GSS students were tested in five groups; four small groups of five
students, and one large group of 24 students. With the exception of 2 students, all
the GSS students took part in both the Part 1 and the Part 2 tests. The Maru A
Pula students were collectively tested in their class. They were only taking part in
the Part 1 test. The grown-ups were tested individually. 7 of the 9 took part in
both the Part 1 and the Part 2 tests.
When introducing the subjects for the test, I emphasised that I was asking
for their immediate feeling about the nonsense words, and that there were no
«right» or «wrong» answers to the test items. They were urged to give an answer
to all items, even when they did not feel that one alternative was any better than
the others. They were also instructed that they were not allowed to make more
than one choice for each test item. Finally, they were instructed to start with the
first item and proceed without looking back at the earlier answers, and that there
was no need for them to use a long time on each item, as it was their first
immediate response I was interested in. As I was always present myself, I could
check that they were following my instructions about the practical proceedings of
the tests.
The test items were, as explained elsewhere, randomly distributed in the
test, but they were ordered with respect of type A and type B. Half of the tests
started with all the type A items, and the other half with all the type B items. This
was primarily done to avoid that the same test items would suffer from the
subjects' poorer concentration towards the end of the test, but it turned out to be a
useful barrier towards a possible temptation to look at the responses of the
neighboring student. I distributed the two variants of the test so that no student

-98-
sat next to another who started out with identical organisation of test items,
something that made it virtually impossible to compare one's own answers to
the answers of other subjects.
The subjects performed the tests by marking their choices with a text
marker (or by ticking off the chosen item with a pen).

5.5. Presentation of tests and results

In the following I will present the tests 1, 5, 7, and 8 together with the results, and
a discussion of these. A more general discussion is presented as a section of its
own in the end of this chapter.
The results of the tests were incorporated into a database, resulting in 5860
records. The database has made it possible to combine different elements in such
a way that I have been able to check possible differences between different groups
of subjects (based on age, which school they attend, etc.). I have also been able to
check for example if the nonsense roots (without regard to a prefix) have led to
preferrals of certain meanings. As I have found no systematic correlations of
these kinds, I have not included any such figures in the presentations below.
It is important to note that in the final test paper, all the test items were
randomly distributed to ensure that it was impossible to find patterns that could
influence the answers. 69 If, on the contrary, I had grouped together the five
nonsense nouns based on the same root, but with different class prefixes and with
the same five meaning alternatives, it would be possible for some subjects to
answer on the clear assumption that they were to make a one-to-one link
between nonsense forms and meanings. It would then have been impossible to
judge when an answer had been given on the basis of genuine intuition, and
when a subject had used a kind of «filling-in-the-gaps strategy»; combining some
prefixes and meanings according to whether or not they had already been «used».
The intermingling of test items, and the instruction that the subjects were to start
with the first item, and then proceed without «jumping back or forth» (they were
observed while performing the test), should have eliminated the possibilities of
such «filling-in-the-gaps strategies».

69 See the final test paper in the appendix.

-99-
Part1
5.5.1. Test 1
The first test aims at investigating the prototypical meanings of the five classes (1,
3, 5, 6 and 7). Since each meaning was constructed as a good example of one of the
prototype schemas, my hypothesis was that more subjects than what could result
from purely random choices, would choose a connection between a meaning
«belonging to» a certain class schema and the prefix appropriate for that class.
The test meanings were formed on the basis of my hypotheses of the class
prototypes at the time of constructing the tests. The only difference between those
hypotheses and my final suggestion concerning class prototypes, is that the
notion of 'non-individuability' has been added as a relevant aspect of the class 5
prototype. That notion has therefore not been reflected in the tests.

The test meanings of the first test were as follows:

'a person' (class 1, schema: 'human')

'a tree that grows in Europe' (class 3, schema: 'tree')
'a small, round ball' (class 5, schema: 'roundish')
'a soft-drink' (class 6, schema: 'liquid')
'a tool that is used for making soap' (class 7, schema: 'instrument')

Each class prefix (here: the class 7 prefix se- ) was tested with two different
nonsense roots, as in:

serutsa a small, round ball

a tree that grows in Europe
a soft-drink
a tool that is used for making soap
a person

setsifo a person
a small, round ball
a soft-drink
a tree that grows in Europe
a tool that is used for making soap

-100-
Each corresponding meaning (here: a «class 7 meaning») was also tested twice,
with two other nonsense roots, as in:

a tool that is used for making soap maboroleta

leboroleta
seboroleta
moboroleta (babOroleta)
moboroleta (meboroleta)

a tool that is used for making soap motlhorape (batlhorape)

letlhorape
rna tlhorape
motlhorape (metlhorape)
setlhorape

In other words: the combination of a certain class prefix with a certain test
meaning (here: the class 7 prefix se- and a meaning that is compatible with the
schema 'instrument') was tested four times. This procedure was carried out for all
the five classes, resulting in 20 single test items only to test these few presumed
prototypical meanings. Forming the tests in this manner, meant that I could only
test very few of my suggested schemas. In spite of this obvious limitation, I chose
to form the tests like this, rather than risking not being able to draw any
conclusions, because I could not control if the subjects had reacted more to
associations of the root rather than of the prefix.

Results
I have presented the results in three tables: Table 2 shows the responses in terms
of chosen meanings when the subjects were confronted with «full nonsense
nouns» (real prefixes plus nonsense roots). The columns show how many times
each meaning was chosen, when the subjects were presented with nonsense
items attached to the prefixes that are shown in the upper row. The numbers are
based on the responses of the two nonsense roots -rutsa and -tsifo . Table 3 shows
the responses in terms of chosen forms (the only difference being the prefixes)
when confronted with a certain meaning. The columns show how often each
nonsense form (different real prefixes attached to the same nonsense root) was
preferred, when the subjects were presented with the meanings that are shown in

- 101 -
the upper row. Here the numbers are based on the two nonsense roots -boreleta
and -tlhorape . Table 4 adds up the numbers from the two preceding tables.
The shaded cells mark the highest scores, whereas the cells marked with
double lines show which combination of prefix and meaning that I expected to be
chosen most frequently.
The last row gives the results of a significance test conducted by Grete
Usterud Fenstad.70 Only the results of the expected most frequent choices (the
numbers in the double lined cells) were tested for significance. If the p-value is
smaller than 0.01, the number of responses is significant in the sense that the
result can not be attributed to chance (statistically random distribution of
answers). (See appendix 3 for an explanation of the mathematical method used.)
It will become clear below, that the results of the significance test is
valuable only to a certain degree. The test was carried out based on a «zero
hypothesis» that the subjects had responded by pure guessing (which would have
led to a statistically level spreading of answers over the alternative choices).
Accordingly, the results of the significance test is suitable to answer the general
question of whether or not one can claim that the subjects have made use of
(possibly unconscious) semantic associations when dealing with the test items. In
some cases, however, it seems that the subjects have chosen to disregard the class
1 and class 6 prefix/ meaning. On the assumption that they have made use of
semantic associations, that is not surprising: It would be a reasonable hypothesis
that both class 1 members (humans) and class 6 members (non count concepts)
differ from the other class' members in semantic saliency (exhibiting a higher
degree of saliency). Actually some of the test results below seem to support such a
hypothesis. In some cases, then, when the subjects have more or less completely
ruled out class 1 and 6, and the choice between the other three alternatives has
resulted in a very level spreading of answers, the significance test will still give a
positive result (i.e. the p-value will be smaller than 0.01) if the expected choice
were one of the other classes (not 1 or 6). Therefore one has to exercise caution
when dealing with the results of the significance test.

70 Senior Lecturer, Department of Mathematics, University of Oslo.

- 102-
The results of test 1 is presented in the two tables below:

I 1 (mo- ba-) 3 (mo- me-) 5 (le-) 6 (rna-) 7 (se-) total :

person I 116 (76%) I 8 (5%) 33 (22%) 33 (22%) 22 (14%) 212
tree 9 (6%) 46 (30%) 22 (15%) 25 (16%) 17 (11%) 119

ball 9 (6%) 11 (7%) I 31 (21"/o) 12 (8%) 24 (16%) 87

drink 7 (5%) 46 (30%) 26 (17%) 50 (33%) 29 (19%) 158

tool 12 (8%) 42 (27%) 39 (26%) 32 (21%) 61 (40%) 187

total: 153 153 151 152 153

p-value : p == O p = 0.0009 p = 0.435 p== 0 p==O
Table 2. From form to meaning, test 1.

I person tree b a ll drink tool to t al:

1 (mo- ba-) I 125 (81%) I 3 (2%) 3 (2%) 5 (3%) 1 (1%) 137

3 (mo- me-) 12 (8%) I 64 (41%) I 18 (12%) 37 (24%) 28 (18%) 159

5 (le-) 13 (8%) 20 (13%) I 83 (54%) I 34 (22%) 51 (33%) 201

6 (rna-) 0 3 (2%) 6 (4%) 5 (3%) 12 (8%) 26

7 (se-) 5 (3%) 66 (42%) 44 (29%) 73 (47%) I 62 (40%) I 250

total: 155 156 154 154 154

p-valu e: p == 0 p == 0 p ==0 p == 1.0 p ==0

Table 3. From meaning to form, test 1.

In general I expected the answers to the form-to-mean ing test items (Table 2) to be
more levelled out, as compared to the m eanin g-to-form items (Table 3). In the
mean ing-to-form tasks I h ave already forced upon the subjects a con nection
between a certain nonsense root and a certain meaning, thereby leaving only the
prefixes for them to ch oose amon g. This, I believe, should put more focus on the
relation between the p refixes and th e meanings, than what is possible in the
form-to-meaning tasks. When a nonsense root p lus a prefix is to be connected to
one of several alternative meanings, the risk that it is the combin ation of the root
and a certain meaning that is focused, is clearly present.

-103 -
The class 1 prefix mo- (ba-) and the meaning 'a person'
Table 2 and 3 above show a significant tendency to connect the meaning
'person' 71 and the class 1 prefix. Given the exceptionally homogeneous semantic
content of class 1, this was of course expected. The whole point of integrating the
class 1 prefix and a "class 1 meaning" into the test, was merely to control whether
or not the subjects had understood the task and were willing to respond seriously.
No significant correlation between the class 1 prefix and the meaning 'person'
would have indicated that the subjects had selected their answers by mere chance.
It can, however, be seen that they must have used their intuition (or for that sake:
their knowledge, as most Batswana would be able to state that a noun with the
prefix mo- (+ pl.:ba-) refers to a human being, and that the most typical prefix
used in nouns referring to a human being is mo- (+ pl.:ba-)) . Responding
according to the predictions to the only obvious case, indicates that the subjects
really have tried to deal with the whole task in a serious manner.

The class 3 prefix mo- (me- ) and the meaning 'a tree that grows in Europe'
From form to meaning:
Table 2 shows that a significant 30% of the subjects select 'tree' when they are
confronted with a nonsense noun with the class 3 prefix.
Although the number of subjects selecting 'tree' is significant when taking
all response possibilities into account, ' tree' does not represent a clear first choice.
Both 'tree' and 'drink' are selected by 30%, and 'tool' by 27%.
The choice of 'tool' is by no means as surprising as the choice of 'drink'. It is very
difficult to conceive of a tool without a configuration. Since class 3 can be
characterised partly as a «configurational class» (for long items), it is virtually
impossible to guard against subjects reacting on the basis of configuration rather
than on the basis of function.
The 'drink' response, however, is puzzling. As we shall see below, the
expected connection between the meaning 'soft-drink' and the class 6 prefix
turned out to be clearly problematic. Anticipating the course of events, my
interpretation is that 'soft-drink' probably has been an unfortunate test item; the
subjects might not have imagined a liquid (as I wanted them to), but rather
focused on the usual appearance of soft-drinks, namely tins and bottles or the
like. It is also possible that 'soft-drink' is favoured simply because it refers to

71 In the following I will use the same abbreviated versions of the «test m eanings» as I u se in the

tables.

- 104-
«stuff» that often has strange names. The other choices of meaning are better
candidates for «normal» Setswana names; hence harder to connect to the
unfamiliar nonsense terms.
The relation between the meanings 'tree' and 'ball' is of particular interest.
While 'tree' receives the already mentioned 30% of the answers, 'ball' receives
only 7%. In my analysis I claim that one of the oppositions between typical class 3
members and typical class 5 members is configurational ('long' and 'roundish').
Accordingly, they should be rather easily kept apart in the test results, given that
the test items invite the speakers to «use» configurational associations (conscious
or unconscious). The test responses suggest such a «barrier» between the two
classes.

From meaning to form:

Table 3 shows a similar asymmetrical response between the class 3 prefix (chosen
by a significant 41%, and the class 5 prefix (13%).
The only unexpected response in Table 3 is represented by the class 7 prefix.
That this prefix is chosen by 42% of the subjects can not be a result of accidental
choices; in fact slightly more subjects choose the class 7 prefix than the class 3
prefix which have been shown to be significant. The class 3 and the class 7
prefixes together represent 83% of the answers. The only possible reason for the
high score of the class 7 prefix that I can think of, is that the subjects have been
influenced by a «setlhare -effect». As will be remembered from chapter 4, the
generic term for tree is not a class 3 noun, but the class 7 noun setlhare. I think
that this might have led the subjects to choose according to the phonology of the
generic term, overriding the fact that virtually all names of trees are class 3
nouns. An incident that supports this assumption, is that one of the subjects
actually wrote setlhare next to the test item setlhorape . The root tlhorape is too
close to the real root tlhare to be a good test item.

The class 5 prefix le- and the meaning 'a small, round ball'72
From form to meaning:
When the subjects are presented with a nonsense form with the class 5 prefix
(Table 2), they do not clearly favour any of the meanings, nor do they rule out

72 At the time of constructing the tests, I had not yet come to the conclusion that 'non-
individuability' also is «present» in the prototypical schema. Accordingly, the test meaning
does not reflect that notion.

-105-
any. The choices are rather level; ranging from 15% ('tree') to 26% ('tool'). The
number of subjects selecting the expected 'ball' (21%) is not significant.

From meaning to form :

When, on the contrary, they are presented with the meaning 'ball' (Table 3), they
strongly favour the class 5 prefix le- (54%), a result that is significant. Further in
Table 3 we see that the class 7 prefix makes the second choice, with it's 29% . I had
reasons to expect this to be the second choice, as 'ball' could be a rather good
instantiation of the class 7 schema 'paricipation in action'. The class 1 and the
class 6 prefixes are practically ruled out, with 2% and 4% respectively, while the
class 3 prefix gets 12 %. Again we see that the difference between the choice of the
class 3 and the class 5 prefix is striking; the class 5 prefix is chosen 4 1/ 2 times as
often as the class 3 prefix. The result meets the expectation that a semantic item
like 'a small, round ball' more easily will be associated with the class 5 prefix [e-
than with the class 3 prefix rna- (me).
The assumption that the form-to-meaning test items lead to a higher rate
of accidental answers than the meaning-to-form items do, is clearly born out in
testing the combination of 'ball' and the class 5 prefix, as the difference between
the answers in Table 2 and Table 3 is striking.

The class 6 prefix ma- and the meaning 'a soft-drink'

From form to meaning:
Confronted with the form-to-meaning test items (Table 2), the subjects show a
clear tendency to favour 'drink' (33% which represents significance).
Concerning the connection between the class 6 prefix and the other
meanings, I suspect that the subjects have interpreted the class 6 prefix as the
plural - and not the non-count- prefix. The plural class 6 prefix is used regularly
for singular class 5 items, but in addition it serves as a collective plural for nouns
from other classes. Thus, «class 5 persons» and «class 5 tools» would take the
class 6 prefix in the plural. Names for trees, as class 3 notions, are usually given
the class 4 prefix (me- ) in the plural, but referring to thickets of trees the class 6
prefix may be used.

From meaning to form:

The picture is drastically altered when focusing on the meaning-to-form items
(Table 3). When responding to 'drink', the subjects for whatever reason practically
rule out the class 6 prefix (it is selected by only 3%, which of course is not

-106-
significant). It is, however, worth noting that the class 6 prefix gets extremely low
scores whatever «test meaning» that is in focus. The scores range from 0 to 8%,
indicating that the subjects did not feel any of the meanings to be good «class 6
candidates». That was, of course, expected for the meanings 'person', 'tree', 'ball'
and 'tool', but not for the meaning 'soft-drink'. I believe that this response
indicates rather strongly that 'soft-drink' has been an unfortunate item. I might
not have been able to direct the subjects' attention towards the liquid itself, but
rather towards the usual appearance of soft-drinks: the tins and bottles that
usually contain the drink.
The subjects show a marked preference for the class 7 prefix (47%). The
reason remains unclear, except that the subjects could have been influenced by
terms like seno tsididi (cold drink) or seela (liquid). Both seno and seela are
nominalisations of verbs, consisting of the class 7 prefix se- plus the verbal roots
(phonologically modified in the case of seno ) for 'drink' and 'flow'.

The class 7 prefix se- and the meaning 'a tool that is used for making soap'
Except for the unproblematic relationship between the class 1 prefix and the
meaning 'person', it is only the combination of the class 7 prefix and the meaning
'tool' that is chosen most frequently both in the form-to-meaning and meaning-
to-form tests.

From form to meanzng:

A significant 40% of the subjects favour 'tool', when they have to deal with a
nonsense noun with the class 7 prefix. The second choice is given by 19%; less
than half of those that choose 'tool'. Except from 'tool' as the significant first
choice, Table 2 shows a rather level spreading of the other alternatives.

From meaning to form:

Again, a significant 40% of the subjects favour the class 7 prefix se- when they are
confronted with the meaning ' tool'. This time it is the class 5 prefix that
constitutes the second choice, with 33% of the answers.
The alternative choices in Table 3 are less level than those in Table 2. That
both the class 3 and the class 5 prefixes are chosen by quite a number of subjects, is
rather natural, when we consider what I have pointed out earlier about tools and
configuration. But it remains counterintuitive that the class 5 prefix score so
much higher than the class 3 prefix; I would rather have expected the opposite, as
I conceive the typical shape of a tool to be long rather than roundish.

-107-
The most striking response pattern for the whole of Test 1 is obtained by adding
up the numbers for all chosen combinations of prefixes and meanings, including
both the form-to-meaning tasks and the meaning-to-form tasks. That has been
done in Table 4. Each cell in the table represents the correlation between a certain
meaning and a certain noun class (represented by the prefixes). The cells with
double lines mark the expected combinations, and the shaded cells mark the
combinations that were chosen most frequently. The pattern that emerges show
that the overall tendency is to favour the expected combinations of prefixes and
meanings.

person/cl.l person/ cl.3 person/ cl.S person/ cl.6 person/ el.7

241 20 46 33 27
tree / eLl tree / cl.3 tree / el.S tree / cl.6 tree / cl.7
12 110 42 28 83

ball/cl.l ba ll/ cl.3 ball/cl.S ball/ cl.6 ball/cl.7

12 29 114 18 68

drink/ cl.l drink/ el.3 drink/ cl.S drink/ el.6 drink/cl.7

12 83 60 55 102
tool / eLl tool/cl.3 tool/cl.S tool/cl.6 tool/cl.7
13 70 90 44 123
Table 4. Correlations between meaning and class, test 1.

5.5.2. Test 5
In this test I was looking at the distribution of human terms over the classes 1, 5
and 7. The test includes one «neutral» meaning ('a person who is working in a
car factory') expected to be treated as a class 1 term (with mo- ), one «expert» kind
of meaning ('a person who is an expert in a special way of dancing'), that I, at the
time of constructing the tests, thought was the «typical class 7 person» (with the
prefix se- ). The combination of the class 5 prefix (le- ) and a certain type of
meaning represented a problem, because I had seen, not one, but three typical
groups: the negative, the small/immature and many ethnic groups. I chose to
represent all of them in order to see if any one was chosen with significantly
higher frequency, or if they would be levelled out ('a very stupid person', 'a very
small child', ' a person from a foreign nation'.) The class prefix attached to a
nonsense root was tested twice with two different nonsense roots.

-108-
Example (test type A):
seatepha a person who is working in a car factory
a very stupid person
a person who is an expert in a special way of dancing
a person from a foreign nation
a very small child

Each of the five meanings were also tested with two different nonsense roots. The
choice of nonsense form was this time limited to the three items that would be
compatible with the forms of class 1, class 5 and class 7, as in the following
example:

a very stupid person sethephima

lethephima
mothephima (bathephima)

Results
From form to meaning:

1 (mo- ba-) 5 (le-) 7 (se-) total:

working 1 22 (14%) 12 (8%) 9 (6%) 43

s mall 9 (6%) 16 (10%) 14 (9%) 39

stupid 42 (27%) 71 (46%) 81 (53%) 194
foreign 51 (33%) 33 (21%) 11 (7%) 95

expert 29 (9%) 22 (14%) 39 (25%) 90

total: 153 154 154

p-value: p = 0.959 p=O p = 0.049
Table 5: From form to meaning, test 5.

working-> a person who is working in a car factory

stupid-> a very stupid person
expert-> a person who is an expert in a special way of dancing
foreign-> a person from a foreign nation
small -> a very small child

-109-
Presenting the subjects with class 1 nonsense terms did not lead to the expected
result; the apparently «neutral» meaning 'a person who is working in a car
factory' was selected by only 14% which is not significant. Instead many preferred
'a person from a foreign nation' (33%) and 'a very stupid person' (27%).
Of all the three meanings equally expected for forms with the class 5 prefix,
it is 'a very stupid person' that is most frequently chosen (46%).
When the subjects had to deal with nonsense forms with the class 7 prefix,
only 25% (which is not significant) chose the expected meaning ' a person who is
an expert in a special way of dancing'. The first choice 'a very stupid person' got
53%. Although I originally had expected a majority of the subjects to choose
'expert', both answers are in accordance with my present analysis of the class 7
network. Why so many more subjects have preferred 'stupid' to 'expert', I am not
able to explain, though, unless it is easier to imagine naming fools with strange
names, while one perhaps would expect a reminiscence of go bina (to dance) in a
form meaning 'a person who is an expert in a special way of dancing'.

Looking at the whole table, it seems that many subjects have avoided the two
notions 'a person who is working in a car factory' (the expected «class 1
meaning») and 'a very small child'. It is perhaps more difficult to connect such
«everyday language» notions to strange nonsense items, than it is to connect the
other three more «marked » meanings to the nonsense words. If this is the case,
the subjects are left with the meanings 'stupid', 'foreign' and 'expert' from which
to choose. 'Foreign' is probably the least «marked» of these three meanings, hence
the best candidate for class 1.

From meaning to form:

working small stupid foreign exp ert total :

1 (mo- ba-) I 96 (62%) I 74 (48%) 21 (13%) 118 (77%) 35 (23%) 344

5 (le-) 34 (22%) I 55 (36%) I 34 (22%) I 33 (21%) I 9 (6%) 165

7 (se-) 24 (16%) 25 (16%) 101 (65%) 3 (2%) I 111 (72%l I 264

total: 154 154 156 154 155

p-value: p~o p = 0.265 p = 0.999 p = 0.999 p=O
Table 6: From meaning to form, test 5.

-110-
 Table 6 above shows that for each meaning one prefix is more or less clearly
preferred, although in three of the five cases it is not the prefix that I originally
had expected as a first choice. The pattern that emerges is one where the subjects
tend to distribute the different meanings between class 1 and class 7. Of course, as
class 1 can be regarded as the «default class» for human beings, it can hardly be
surprising that it is the class 1 prefix that is chosen most frequently (altogether it
is chosen 344 times, while the class 7 prefix is chosen 264 times and the class 5
prefix 165 times). The interesting point though, is that for the two meanings 'very
stupid' and 'expert' most subjects seem to «agree» that class 1 is not suitable; 65%
treat 'very stupid' as a class 7 noun, and 72% treat 'expert' as a class 7 noun. (The
other possibilities are given from 6 to 23% of the answers.)
The only originally expected choices that have been selected by a significant
number of subjects, are the class 1 prefix for the meaning 'working' and the class 7
prefix for the meaning 'expert'.
As I do not any longer consider small size or immaturity as such to be class
5 schemas (to the extent that human beings with such characteristics are found in
class 5, it is rather because of their «non-individuated» character), I am not
surprised that more subjects prefer this notion to be associated with class 1 than
with class 5, especially because the meaning itself is not «marked».
Even if there is no «rule» that foreigners or other ethnic groups are given
class 5 names, it is so usual that I find it rather surprising that there is not a more
level distribution between choosing the class 1 prefix (77%) and the class 5 prefix
(21%).
If I am right that 'non-individuability' represents a class schema that
connects most class 5 personal terms, it makes great sense that the subjects more
or less have rejected the class 5 prefix which represents a kind of de-focusing on
the individual and rather distributed their choices between the «default class»
and the more «action related class»
There is a tendency in the language that some «class 5 persons» are
moving to class 1. 73 Tt wou]d have been interesting to investigate whether or not
a group of very old subjects would have made more use of the class 5 prefix, than
the mostly very young subjects that I have used, did.

73 I do not know the extent of this phenomenon, and it falls outside the scope of this thesis.

-111-
Part 2
The Part 2 was also presented to the majority of the subjects. It consisted of two
more tests (7 and 8) of the same type as the first test in Part 1, but this time aiming
at investigating less central schemas. The idea was that this second test were to be
presented for the same subjects, if the first test from part 1 showed any
interpretable patterns. If, on the contrary, the answers to this first test based on the
most central or prototypical notions were totally unintelligible, there would be no
sense in presenting the last part, since that was constructed on the basis of less
central groups of notions.
As each of the five constructed meanings in these tests, as well as each of
the five prefixes attached to a nonsense root, were tested twice, each of these two
tests consisted of 20 test items.

5.5.3. Test 7
In this test I let the following meanings be representatives of schemas within the
five classes:
'a person who is working with boats' (class 1, schema: 'human being')
'a long thin line' (class 3, schema: 'long objects/phenomena')
'the round part of an animal's back-leg' (class 5, schema: 'paired, roundish body
parts')
'the kind of things that are kept in a box' (class 6, schema: 'unspecified concrete
objects')
'a thing that a chief holds in his hand during a religious ceremony' (class 7,
the original schema was: 'other artifacts' i.e. «non instrumental», which is
no longer relevant in my suggestion for a class 7 network. The meaning
would be an example of 'instrument' in my present suggestion for
network.)

As in test 1, I tested the combination of a certain class prefix and a certain

meaning (compatible with one of the schemas in the semantic network of the
class in question) four times; twice with the test type A and twice with the test
type B. Consider the following example of testing the combination of the class 3
prefix mo- and the schema 'long objects/phenomena':

motsero (metserO) -a thing that a chief holds in his hand during a religious
ceremony
- the kind of things that are kept in a box

-112-
 - a person w ho is working with boats
- a long thin line
-the round part of an animal's back-leg

mokorite (mekorite) - a thing that a chief holds in his hand during a religious
ceremony
-a person who is working with boats
- a long thin line
-the kind of things that are kept in a box
- the round part of an animal's back-leg

a long, thin line -moforobeta ( meforobeta)

- maforobeta
- seforobeta
- leforobeta
- moforobeta (baforobeta)

a long, thin line - sefeo

- mofeo (mefeOJ
- lefeo
- mafeo
- mofeo (bafeo)

Results
From form to meaning:

1 (mo- ba-) 3 (mo- me-) 5 (le-) 6 (rna-) 7 (se-) total:

person I 81 (84%) I 14 (14%) 13 (13%) 12 (12%) 15 (15%) 135

line 1 (1%) I 20 (20%) I 31 (32%) 9 (9%) 14 (14%) 75

round leg 6 (6%) 20 (20%) 24 (24%) 8 (8%) 18 (19%) 76

things (box) 7 (7%) 21 (21%) 14 (14%) I 59 (60%) I 8 (8%) 109

rel. thing 2 (2%) 23 (23%) 16 (16%) 10 (10%) I 42 (43%) I 93

total: 97 98 98 98 97
p-value: p=O p =0.460 p =0.133 p=O P "'O
Table 7: From form to meaning, test 7.

-113-
person-> a person who is working with boats
line-> a long thin line
round leg-> the round part of an animal's back-leg
things (box) -> the kind of things that are kept in a box
rel. thing - > a thing that a chief holds in his hand during a religious ceremony

In Table 7 one can see that 3 of the 5 expected choices have been selected by a
significant number of subjects: As much as 84% selected 'a person who is working
with boats' when dealing with nonsense items with the class 1 prefix.
The class 6 prefix also seems to lead to a strong favouring of the expected
meaning. A significant 60% choose the expected 'the kind of things that are kept
in a box'. Unfortunately though, this class 6 meaning is clearly a mishap; the use
of the English plural makes it impossible to draw any conclusions whatsoever, as
the subjects are probably led to react to the class 6 prefix as the plural and not as
the non count prefix.
That leaves us with only one significant answer apart from the obvious
case of class 1; the selection of 'a thing that a chief holds in his hand during a
religious ceremony' when the class 7 prefix is used. 43% has chosen that
alternative, which is more than twice as many as those who chose the second
alternative.
The answers concerning class 3 and class 5 are too level to represent any
useful information.

From meaning to form:

person line round leg things (box) rel. thing total:

1 (mo- ba-) I 68 (69%) I 1 4 3 0 76

3 (mo- me-) 3 I 43 (44%) I 26 38 23 133

5 (le-) 13 33 I 31 (32%) I 4 30 111

6 (rna-) 3 4 2 '"' 1 52

7 (se-) 11 17 34 10 44 (45%) 116

t o ta l : 98 98 97 97 98
p-value: p =O p =O p =0.0016 p =O p =O
Table 8: From meaning to form, test 7.

- 114 -
When the subjects respond to the meanings (Table 8), the responses are not quite
as level. There is a tendency that the expected prefixes have been chosen.
The meaning 'person' as usual leads to a significant number of subjects
choosing the class 1 prefix (69%).
Comparing the class 3 and class 5 meanings ('line' and 'round leg'), we see
that the combination of meaning and class prefix show a tendency towards the
expected choices, although the differences are not striking. The significance test
based on a zero hypothesis of random spreading of answers on all five
alternatives, however, show that both in the case of the class 3 meaning and the
class 5 meaning, the expected prefixes have been chosen by a significant number
of subjects, 44% and 32% respectively. The answers clearly show that we have to
reject the zero hypothesis, but concerning the class 5 meaning we are not left with
a clear preference of the expected choice, since it is mostly the rejecting of the class
1 and the class 6 alternative that leads to result that the expected choices are
chosen by a significant number of subjects. The choice between the remaining
three options are in fact quite level.
The results of the class 6 meaning do not, for reasons mentioned in the
preceding section, give any useful information.
The class 7 meaning shows a preference for the expected choice; a
significant 45% of the subjects select the class 7 prefix.

-115-
The complete correlation of meanings and classes are presented in Table 9. Again
we see a tendency towards favouring the expected choices:

person/ cl.l line/eLl round /cl.l in box/ cl.l relig. I cl.l

149 15 17 15 15
person/ cl.3 line/cl.3 round/cl.3 in box/cl.3 relig. I cl.3
4 63 57 47 37
person/ cl.S line/cl.S round/cl.S in box/cl.S relig. I ci.S
19 53 55 12 48
person/ cl.6 line/ cl.6 round/ cl.6 in box/cL6 relig. / cl.6
10 25 16 101 9

person/ cl.7 line/cl.7 round/ cl.7 in box/ cl.7 relig ./cl.7

13 40 50 20 86
Table 9. Correlations between meaning and class, test 7.74

5.5.4. Test 8
In this test, the following meanings representing various schemas, were tested:

'a person who is behaving well' (class 1, schema: 'human being')

'a dress that is made of skin' (class 3, schema; 'body-coverings made of skin')
'a big, round meeting-room' (class 5, schema: 'roundish enclosures/ locations')
'the state of being a little bit tired' (class 6, schema: 'states')
'a worn-out dress' (class 7, schema: 'objects affected by action')

The example shows the testing of class 5:

lekesele - a big, round meeting-room

- a worn-out dress
- a dress that is made of skin
- a person who is behaving well
-the state of being a little bit tired

74 Person= a person who is working with boats. Line = a long, thin line. Round= the round part of an

animal's back-leg. In box = the kind of things that are kept in a box. Relig. = a thing that a
chief holds in his hand during a religious ceremony.

-116-
 lekathe - a dress that is made of skin
- a big, round meeting-room
- a worn-out dress
- the state of being a little bit tired
- a person who is behaving well

a big, round meeting-room - makgotepha

- lekgotepha
- sekgotepha
- mokgotepha (bakgotepha)
- mokgotepha (mekgotepha)

a big, round meeting-room - lekugOthebo

- mokugathebo (bakugathebo)
- makugOthebo
- sekugOthebo
- mokugathebo (mekugathebo)

Results
From form to meaning:

1 (mo- ba-) 3 (mo-me-) 5 (le-) 6 (rna-) 7 (se-) total:

'person I 82 (84%) I 8 (8%) 17 (17%) 8 (8%) 8 (8%) 123
skin-dress 5 (5%) I 47 (48%) I 25 (26%) 27 (28%) 26 (27%) 130
room 1 (1%) 13 (13%) I 17 (17%) I 22 (22%) 14 (15%) 67
state 4 (4%) 4 (4%) 13 (13%) 14 (14 0 17 (18%) 52
womdress 6 (6%) 26 (27%) 26 (27%) 27 (28%) I 3o (32%) 115
total : 98 98 98 98 95
p-value: p:=:O p:=:O p = 0.744 p = 0.921 p = 0.002
Table 10: From form to meaning, test 8.

person -> a person who is behaving well

skin-dress -> a dress that is made of skin
room-> a big, round meeting-room
state -> the state of being a little bit tired
womdress-> a worn-out dress

-117-
The next bulk of test meanings has led to less clear answers. The only potentially
interesting answers are represented by the responses concerning the two
meanings involving 'dress'. The problem with these constructed meanings is that
one might not be testing the notions 'skin' and 'worn-out' (which are the reasons
for claiming that the meanings could fit with class 3 and 7 respectively), but
rather the shared notion ' dress'. (The Setswana term for a dress is mosese (class
3).) However, in Table 10 above we see that the distribution of answers con-
cerning these two meanings show a significant tendency (48%) towards choosing
'skin dress' as the class 3 meaning (as expected), and also a significant tendency
(32%) towards choosing 'worn-out dress' as the class 7 meaning, (as ex-
pected).These facts indicate that apart from reacting upon the 'dress' part of the
meaning, the subjects probably have been influenced by the «modifying
element».
The expected choices concerning the class 5 and the class 6 prefixes have
not been selected by a significant number of subjects; the meaning 'round room'
(class 5) was chosen by only 17%, and 'state' (class 6) was chosen by only 14%.
The m eaning 'person' was as usual chosen by a significantly high number
of subject (84%) when dealing with items with the class 1 prefix.

From meaning to form:

I person skin-dress room s tate worn dress total:

1 (mo- ba-) I 53 (54%) I 4 (4%) 2 (2%) 3 (3%) 3 (3%) 65

3 (mo- me-) 10 (10%) I 54 (55%) I 35 (36%) 6 (6%) 43 (44%) 148

5 (le-) 15 (15%) 15 (15%) 32 (::\::\0/n' 45 (46%) 31 (32%) 138

6 (rna-) 3 (3%) 4 (4%) 3 (3%) 16 (16%) 6 (6%) 32

7 (se-) 17 (17%) 21 (21%) 26 (27%) 28 (29%) I 15 (15%) I 107

total: 98 98 98 98 98
p-value: p =O p =O p = 0.0009 p =0.818 p = 0.877
Table 11: From meaning to form, test 8.

When the meanings are the starting point (Table 11), the picture changes slightly.
Now, we see an indication that 'dress' has led to the choice of the class 3 prefix in
both variants ('made of skin' or 'worn-out'). A significant 55% of the subjects
choose the class 3 prefix as the response to the meaning 'skin dress', but only an
insignificant 15% chose the class 7 prefix when reacting on 'worn dress'.

-118-
 The usual significant connection between 'person' and the class 1 prefix
show up again; 54% choose that, although is also possible for personal nouns to
have the class 5 and 7 prefix (and in some rare cases even the class 3 prefix).
The number of class 5 prefix responses to 'round meeting-room' (33%) are
significant as far as the distribution of answers over all 5 alternatives is
concerned, but leaving the class 1 and class 6 option aside gives a different picture:
The ch oices between the remaining 3 alternatives are very level.
Lastly, I do not have any suggestions for the tendency to prefer the class 5
prefix for a 'state', rather than choosing the expected class 6 prefix. Only
insignificant 16% chooses the class 6 prefix.

The complete correlation of meanings and classes are presented in Table 12. This
time there is no strong tendency towards favouring the expected choices.

person/d. I skin-d. /eLl room/eLl state/eLl wornd./el.l

135 12 19 11 11

person/ cl.3 skin-d./cl.3 room/cl.3 s tate/ el.3 wornd./cl.3

15 101 60 33 69
person/ cl.S skin-d./cl.S room/cl.S state/cl.S worn d./cl.S
16 28 49 67 45
person/ cl.6 skin-d. /cl.6 room/cl.6 state / cl.6 worn d. /cl.6
7 8 16 30 23

person/ cl.7 skin-d./ el.7 room/ cl.7 state/ cl.7 wornd./cl.7

23 47 52 55 45
Table 12. Correlations between meaning and class, test 8.75

The results of these tests (7 and 8) confirm my prediction that the less central
notions would be more difficult to test in this way, than the apparently
prototypical notions in test 1.

75 Person = a peson who is behaving well. Skin-d. = a dress that is made of skin. Room = a big, round
meeting-room. State = the state of being a litle bit tired. Worn d. = a worn-out dress.

-119-
5.6. Discussion
Here I will attempt to outline what we can conclude from the test results.
We have seen that the tendency towards choosing the expected alternatives
in some of the tests is large enough to conclude that at least we have to rule out
the possibility that the Setswana noun classes constitute a purely formal system.
This, however, is hardly controversial. A common stand, also among those who
tend to play down the role of semantics in the Bantu noun classes, is that even
within mainly formal systems there is usually still a kind of semantic «core».
Corbett (1991:8), who treats the Bantu genders (noun classes) as 'formal
morphological systems', still claim that: "In a sense all gender systems are
semantic in that there is always a semantic core to the assignment system."
It would, then, be more interesting if my test results could lead to any
conclusions about what kind of semantic structures that the noun classes
represents, as well as the position or importance of the semantics within these
systems. As we shall see, these questions can hardly be fully answered by the test
results. The results are, however, compatible with representing each noun class
as a system of several abstract (generalised) schemas where some are more central
than others. To a certain degree, they even indicate that this probably is a better
way of understanding the noun class semantics, than any of the other possible
positions that I will discuss below. The test results also represent a good starting
point for further investigation, as they clearly indicate that it is possible to use
methods like this in revealing the semantics of the noun classes.

Are the noun classes semantically or formally based?

If the classes were exclusively formally based, one could predict that the test
results would exhibit a statistically random distribution of answers. However, the
significance-testing of the obtained results, shows in a convincing way that this is
not the case: The results of most tests clearly represent a high degree of non-
random distribution of answers. Accordingly, we have to conclude that to a
certain degree the noun classes are semantically based.
It is interesting to note that all subjects whom I asked about their feelings
about the tests (after the test sessions), answered that they had no intuition about
the alternative choices. One person said she knew that tree names begin with mo-
(the class 3 prefix), but that all the other items were equally strange. All others
whom I asked (on a very informal basis) said that none of the alternative choices
were any better than the others. The test results, however, reveal that their

- 120 -
responses must have been based on systematic associations, at least to a certain
degree.

What is the nature of the «class semantics»? Is it better understood in terms of

complex bundles of properties (kinds), or in terms of relatively simple, abstract
categories?
This complex question focuses on the character of the semantic group(sf6 that
constitute the semantics of each class.
The implicit claim that many Bantu grammarians make is that the
semantics (if any) is describable in terms of «natural classes of entities» (examples
of such entities would be 'plants', 'body parts', 'tools', etc.). Cole (1955) is not an
exception when he begins the treatment of each noun class with lists of «natural
kinds» that are found in the class. Such a treatment results in an extremely high
degree of overlapping groups. Comparing the classes 3/ 4, 5/ 6, and 7/ 8, we find
that very few of his semantic groups are belonging to only one class (or one pair
consisting of a singular and a plural class)?7
The overlapping groups are shown in the table below (the numbers refer to
the noun class labels):
3 4 5 6 7 8
'parts of the body'
'tools and instruments'
'animals' ~~~3~E~3
'natural phenomena'
'personal nouns' ~~~~ii~E~
'plants'
Table 13.

We are then left with 'a few terms applied to domestic animals' (class 3/ 4), 'trees'
(class 3/ 4), 'some collective nouns' (class 5/6), and 'languages and characteristics'
(class 7 / 8) as groups of entities that are uniquely bound to one pair of classes
(singular and plural) only. In restricting oneself to dealing with semantic groups
only on this level (as many bantuists at least implicitly have done), this situation

76 Leaving aside for a moment the discussion of whether each noun class can be described with one or

several groups of notions

77 In d ealing with class 5, he includes the non count class 6 nouns, as he doesn 't distinguish between

the regular plurals in class 6 - plurals of class 5 nouns- and the non count nouns of class 6.

-121-
 gives a good indication as to why it has been so difficult to come up with sensible
suggestions for «class semantics>>.
The meanings in my tests have been formed on the assumption that the
semantics of the classes only in part reflect such a «kind semantics». As
generalisations over large number of single concepts have been taken to be the
basic starting point for my analyses, there was no implicit restriction that these
generalisations necessarily would lead to a resulting pattern of «kinds» only.
However, some test meanings are compatible with the «kind» type of meaning; 'a
person', 'a tree that grows in Europe', ' a tool that is used for making soap' , etc.
Other test meanings, on the other hand, are not easily conceived as being
based on «natural kinds», as they would simply be too few or too specific to form
natural semantic groups: Is it possible to imagine a natural semantic system that
makes use of «kinds» like 'balls' (where 'a smalt round ball' would belong),
'lines' ('a long, thin line'), 'belongings of chiefs' or 'things used in religious
ceremonies' ('a thing that a chief holds in his hand during a religious ceremony'),
and even distinguishes between the «kinds» 'clothes made of skin material' ('a
dress that is made of skin') and 'clothes that are worn out' ('a worn out dress')?
If the test results had shown significant answers only when meanings that
can easily can be claimed to represent «kinds» were involved, that would have
been an indication that I have chosen an unfortunate starting point; namely the
assumption that the semantics of the classes to a large extent reflect more abstract
concepts. However, this is clearly not the case: Several meanings that are less
compatible with a «kinds understanding» were combined with the expected class
prefix by a significant number of subjects (for example: 'a small, round ball' led a
significant number of subjects to choose the class 5 prefix, and the class 3 prefix
led a significant number of subjects to choose the meaning 'a dress that is made of
skin', whereas the class 7 prefix led a significant number of subjects to choose the
meaning 'a worn out dress'). Such meanings I would suggest, reflect abstract
categories based on properties like 'round', 'material with «live» origin', 'affected
by action', etc. that hardly can be perceived as «kinds». The significant results of
several such meanings would then have to be rendered inexplicable if the
semantics of the classes were to be explained only in terms of the traditional
«kinds».
On the contrary, we find that the test findings are compatible with semantic
patterns that reflect both «kinds» and more abstract properties.

-122-
Does any semantic category within each class exhibit a more central status than
the others?
Assuming now that the semantics of each noun class has to be described in terms
of several semantic categories, it is possible to conceive of two different situations
concerning the status of each different category: Either there is no fundamental
difference between the categories' positions within the classes, or some categories
are more central.
Comparing the results from test 1 (with test meanings based on supposedly
central categories/schemas) with the results from the tests 7 and 8 (based on less
central categories/schemas), is the only way that the test results as such possibly
can shed some light on this issue.
If the conceptual status of the different semantic groups were equal, one
should not expect that systematically more answers to test 1 than to the tests 7 and
8 were significant.
In fact, we see that 8 of 10 (80%) answers to the test 1 items are significant
(that is: a significant number of subjects have chosen the expected alternative),
whereas 14 of 20 (70%) answers to the tests 7 an 8 are significant.
At least one can say that even with the reservation that it might be difficult
to be conclusive about this issue (based on the few comparisons of items that this
test makes possible), the test results are compatible with an assumption that some
categories are more central than others. I even feel justified in suggesting that the
results are less compatible with the opposite assumption; that no categories are
more central than others. The material is, however, far too small to draw any
strong conclusions.

Should the semantics of the noun classes be describes in terms of related or

unrelated groups of concepts?
There is nothing in the test results that can be used as support for either of these
suggestions. On the other hand, there is nothing that precludes any of them
either. The hypothesis that my analysis is based upon- that the semantics of the
noun classes can be described in terms of semantic networks (which of course
presupposes related groups of concepts) - can, in other words, neither be
confirmed nor disconfirmed by the findings of these tests.
This vagueness is, however, not due to the actual responses: Neither
unambiguous results in favour of all the expected choices, nor the opposite result
(none or very few expected results) could shed any light on the issue of
interrelatedness.

-123-
 Other types of tests would have to be used in attempting to reveal possible
interrelatedness between concepts in a network. An example of a test type that
possibly could be used is: 'triades' (where one presents the subjects with groups of
three words, and the task is to remove the one that «does not belong to the
group»).

Concluding remarks
My analysis in chapter 4 implies that I suggest that:
1) The classes are semantically definable.
2) The class semantics is based both on «kinds» and on more abstract concepts.
3) Each class can be defined in terms of several groups of concepts.
4) These groups of concepts do not have the same status within the class (some
are more central than others).
5) The different groups of concepts within each class are related to one another in
a «network structure».

Now we have seen that the test results seem to support some of these claims to a
greater or lesser degree. Other results are for varying reasons useless. But no
results are directly incompatible with any of the above claims.
In addition we can say that the test results are, if not always equally
convincing, at least promising regarding the prospects of formulating new tests
that might yield more and better information concerning noun class semantics.
In the type of test that test 1, 7 and 8 represent, each expected combination
of prefix and meaning was tested four times, each with a different nonsense root
(twice from form to meaning and twice from meaning to form). This was done to
ensure that associations based on the root would not influence the results in one
specific direction, and to have a sufficient number of responses concerning each
combination to make it possible for consistent patterns to emerge, even with a
rather high degree of random choices. But these tests still suffer from an obvious
weakness; only one meaning was tested in each test. In some cases we have been
left with a suspicion that a test meaning has been unfortunate (as 'soft-drink' in
test 1), but with no possibility to compare such an item with other very similar
ones (as others involving a liquid). It is clearly possible to form a test according to
such a demand. The practical drawback is that the number of test items will
explode. Nevertheless, I don't think there exists any shortcuts; testing of this type
will necessarily have to be extensive. I believe that in the long run it will be more

-124-
 fruitful to test small parts thoroughly. Choosing to test more concepts briefly will
only result in vague and confusing results.
In sum, test 1, 5, 7 and 8 are all well suited for such more extensive testing.
It would have been interesting to present the tests 2, 3 and 4 («prototypicality
tests» presented in the appendix) for subjects who had not been presented with
test 1 where the same expected first choices occur. That could have given an
indication on whether or not I am right in suspecting that the test design itself
has been partly responsible for the low rates of the expected choices. However, it
is possible that other test types are better suited for the issue of prototypicality.
Test 6 («double profiling», also presented in the appendix) is too
complicated; looking back, I believe it was premature to begin with such rather
advanced tests before good procedures had been found for the simpler types.
In general, it could possibly be rewarding to try to construct more test
meanings from the same «semantic fields», but formulated in a way that,
according to hypothesis of class meanings, would place them in different classes
(like long body parts contra roundish body parts, linguistic entities of short contra
long duration etc.)
What I have achieved by performing the tests, is showing that even if the
subjects themselves state that they do not have any intuitions about which
answers to choose, it is rather clear that they have been influenced by
unconscious associations between certain class prefixes and certain meanings, and
that to a certain degree some results support my analyses, while none are
incompatible with the analyses.
The most important problem that remains unsolved, however, is the lack
of good suggestions as to how one could go about testing hypothesis about
semantic networks and relations between concepts.

-125-
-126-
Chapter 6. Conclusion

The aim of this thesis has been to investigate whether or not it would be possible
to establish a plausible semantic characterisation of some of the largest and most
heterogeneous noun classes in Setswana. I think that my analyses of the classes 3,
5, 6, and 7 has shown that this is indeed possible.
It is actually surprising that so little extensive research has been done in
this field earlier. I suspect that this, at least partly, is due to the explicit or implicit
viewpoint that categories (linguistic as well as non-linguistic) have to be defined
in terms of «sets of necessary and sufficient features» shared by all category
members.
Accepting the possibility that categories might be structured in different
ways has cleared the way for suggesting semantic characterisations of the noun
classes in terms of semantic networks. Cognitive Grammar (as outlined for
example in Langacker 1987 and 1991, Lakoff 1987) has provided me with
«theoretical tools» that have proved useful. It has enabled me to account for
numerous semantic connections between different subgroups of nouns within
each class, as well as made it possible to extract abstract generalisations ('class
schemas' or 'superschemas') that characterise the classes as such . We have seen
that some classes can be characterised by one such generalisation, whereas others
may be characterised by two or more such abstract generalisations (o r
'superschemas') that don't need to be semantically interconnected.
That these most abstract generalisations within the semantic networks
have turned out to be in accordance with «typologically natural systems»
represents an indication that they really reflect a main characterisation of the
semantics of these classes.
Much remains to be done concerning the psychological status of networks
like the ones presented in this thesis, however. As a preliminary attempt, I
conducted the psycholinguistic test described in chapter 5. Although not giving
fully convincing evidence that all my proposed semantic subgroups within the
classes represent cognitive units, the test results clearly show that the subjects in
general have employed semantic associations in dealing with the test items (cf.
the results of the significance test). Comparing the test results with the subjects'
common statement (after performing the test) that they experienced no intuitive
feeling that any of the choices were better than others, indicates that the semantic
associations that the subjects undoubtedly have made use of, must be of a rather
subconscious nature. It is possible though, that speakers differ with regard to

-127-
degree of conscious intuition about such associations. One mother tongue
speaker on an occasion explicitly expressed to me the feeling that class 5 nouns
refers to «deader» things than class 3 nouns do (a statement that is in accordance
with my analyses of class 3 and class 5). In the theoretical framework of this
thesis, that would indicate that, for him, the superschema (class schema) 'living'
has cognitive salience.
The «naturalness» of the proposed superschemas together with the results
of the psycholinguistic tests suggests that the answer to the question posed in the
title of this thesis, is closer to yes than no. The question was: do the Setswana
noun classes represent conceptual categories marked by grammar?

-128-
REFERENCES
Arnott, D. W. 1970. The Nominal and Verbal Systems of Fula. Oxford: Oxford
University Press.
Batibo, H. M. 1987. The Bantu Ancestors' Vision of the World. In: Journal of
Linguistics and Language in Education. Vol. 2.
Bleek, W. H. I. 1862. A comparative Grammar of South African Languages.
London and Cape Town.
Cole, D. T. 1955. An Introduction to Tswana Grammar. Cape Town: Longman
Penguin Southern Africa.
Corbett, G. 1991. Gender. (Cambridge Textbooks in Linguistics.) Cambridge:
Cambridge University press.
Crystal, D. 1991 [1980]. A Dictionary of Linguistics and Phonetics. Oxford: Basil
Blackwell Ltd.
Contini-Morava, E. 1994 and 1996. Noun Classification in Swahili: a Cognitive-
Semantic Analysis using a Computer Database. Forthcoming. The 1996
version presently available on Internet.
Croft, W. 1990. Typology and Universals. (Cambridge Textbooks in Linguistics.)
Cambridge: Cambridge University press.
Denny, J. P. and Creider, C. A. 1986. The Semantics of Noun Classes in Proto-
Bantu. In: Craig, C. (ed.) 1986. Noun Classes and Categorization.
Amsterdam / Philadelphia: John Benjamins Publishing Company.
Dixon, R. M. W. 1982. Where Have All the Adjectives Gone? and other essays in
Semantics and Syntax. Berlin, New York, Amsterdam: Mouton Publishers.
Endresen, R. T. 1993. Semantisk karakteristikk av nominalklassene i Fulfulde.
Oslo: Universitetet i Oslo.
Greenberg, J. H. 1963. The Languages of Africa. Bloomington: Indiana University.
Hanks, P. (ed.) 1986.Collins Dictionary of the English Language. London and
Glasgow: William Collins Sons & Co.
Jackson, H. 1990. Grammar and Meaning. A Semantic Approach to English
Grammar. London and New York: Longman.
Janson, T. and Tsonope, J. 1991. Birth of a National Language. The history of
Setswana. Gaborone: Heinemann Botswana.
Johnson, M. 1987. The Body in the Mind. The Bodily Basis of Meaning,
Imagination, and Reason. Chicago and London: The University of Chicago
Press.

-129-
Kiparsky, P. and Menn, L. 1977. On the Acquisition of Phonology. In: Macnamara,
J. (ed.). Language Learning and Thought. New York, San Fransisco,
London: Academic Press.
Kraft, G. H . and Kirk-Greene, A. H . M. 1979 [1973]. Hausa. A complete working
course. New York: David McKay Company Inc.
Lakoff, G. 1987. Women, Fire, and Dangerous Things . Chicago and London:
University of Chicago Press.
Lakoff, G. and Johnson M. 1980. Metaphors We Live By. Chicago and London:
University of Chicago Press.
Langacker, R. W. 1987. Foundations of Cognitive Grammar. Vol. I. Theoretical
Prerequisites. Stanford, California: Stanford University Press.
Langacker, R. W. 1991. Foundations of Cognitive Grammar. Vol. II. Descriptive
Applications. Stanford, California: Stanford University Press.
Matumo, Z. I. (comp.) 1993. Setswana English Setswana Dictionary. Gaborone,
Botswana: Macmillan Botswana Publishing Company.
Meinhof, C. 1899. Grundriss einer Lautlehre der Bantusprachen. Berlin: Dietrich
Reimer.
Meinhof, C. 1906. Grundzuge einer vergleichenden Grammatik der Bantu-
sprachen. Berlin: Dietrich Reimer.
Ntsime, J. M. and Kruger, C. J. H. 1989. E antswe letseleng. Setlhopha sa 10.
Pretoria: Via Afrika Limited.
Rakgokong, P. M. 1986. Nominal shift (of original nouns and deverbatives) in
Setswana : a revaluation. M. A. thesis. Potchefstroom University.
Richardson, I. 1967. Linguistic Evolution and Bantu Noun Class Systems. In: La
Classification Nominale dans les Langues Negro-Africaines. Aix-en-
Provence: Editions du Centre National de la Reserche Scientifique.
Rosch, E. 1973. Natural Categories. In: Cognitive psychology 7.
Rudzka-Ostyn, B. (ed.) 1988. Topics in Cognitive Linguistics. In: Koerner, E. F. K.
(gen. ed.), Amsterdam studies in the theory and history of linguistic
science. Amsterdam and Philadelphia: John Benjamins Publishing
Company.
Spitulnik, D. A. 1987. Semantic Superstructuring and Infrastructuring: Nominal
Class Struggle in ChiBemba. In: Studies in African Grammatical Systems.
Bloomington, Indiana: Indiana University Linguistic Club.
Taylor, J. R. 1989. Linguistic Categorization. Prototypes in Linguistic Theory.
New York: Oxford University Press.

-130-
Torrend, S. J. J. 1891. A Comparative Grammar of the South-African Bantu
Languages. London: Kegan PauC Trench, Triibner & Co.
Tsohatzidis, S. L. (ed.) 1990. Meanings and Prototypes. Studies in linguistic
categorization. London and New York: Routledge.
Ungerer, F. and Schmid, H.-J. 1996. An Introduction to Cognitive Linguistics.
London and New York: Longman.
Wierzbicka, A. 1985. Oats and Wheat: the Fallacy of Arbitrariness. In: Haiman, J.
(ed.) !conicity in Syntax. Amsterdam: John Benjamins.
Wierzbicka, A. 1990. 'Prototypes save': on the uses and abuses of the notion of
'prototype' in linguistics and related fields. In: Tsohatzidis, S. L. (ed.),
Meanings and Prototypes. Studies in linguistic categorization. London and
New York: Routledge.
Wittgenstein, L. 1958 [1953]. Philosophical Investigations. (Translated by G. E. M.
Anscombe.) Oxford: Basil Blackwell.

-131-
- 132 -
 APPENDICES
Appendix I. Noun lists
This appendix contains all the class 3, 5, 6, and 7 nouns from my noun sample.
They are grouped according to the semantic categories that are elaborated in my
analyses, but I have not provided explanatory remarks here. Hence, the reason for
assigning a noun to a certain category might not always be obvious (particularly
not to non-Setswana speakers).

Class 3
(1) Trees
mhapu (mofapu) a species of hardy thorn trees
mhota a species of evergreen tree; the raintree
mmasa a tree with tanning bark
mmil6 a species of wild fruit tree; Vangueria infauta
moan a a baobab tree
modumela a species of acacia tree
modutu a species of white-wood tree; Celtis africana
mofata (mhata) a species of tree which grows in heavy sand; lonchocarpus
nelsii
mofeie fig
mogorogoro green monkey orange
mokana Corchorus kirkii, a species of tree
mokhibe a species of tree
mokokwane a species of tree
mokoso a white thorn tree; (Acacia albida)
moku Acacia robusta; a thorn tree with long, straight, small-hooked
horns
montSantsa Bauhinia petersiana, a species of tree
mooka a species of thorn tree; (Acacia karoo)
morolwane a syringa tree
morukudu tambuti tree; Spirostachus africana
moruthure cat thorn; common hook thorn; water thorn; Fabaceae
morutlhatshana a species of thorn tree that grows in dry regions; Acacia caffra
mosetlha a species of acacia tree; (Peltophorum africanum)
mosilapele Rhus lancea; the karee tree
mosu camel-thorn; Acacia giraffe

-133-
motsnaba a species of wild fig; sycamore fig
moupaphiri a species of karee tree; Maerva angolensis, or Flacourtia-indica

(2) Other plants

mhatolantswa a species of bush; Bequaertidendron magalismontanum
mmabi a species of wild cotton bush
mmidi mealies; maize
mmodula the flower on a corn stalk; the outer and loose husk that
clings to corn
mofikapitse a species of creeping plant
mogabaditsana a species of plant that bears edible berries
mogamane Kalahary food berry; Fabaceae
mogarugaru a species of river grass
mogogo a bush with tender stems, edible black fruit and sweet to the
taste
mogokang a hooked thorn bush that grows in sandy places
mogolabosigo datura; a Stramonium plant
mogwang a shoot, or blade just emerging above the ground; the blade of
an ear of corn; the leaf of a reed
mojao poison leaf; Dichapetalaceae
mokapana same as mokawa, Citrullus neudinianus ; Herero cucumber; a
species of creeper, edible, used medicinally as a purgative and
as poison
mokgopa aloe
mokhure Ricinus communis ; plant bearing thorny pods, favourite
food for goats
moko marrow; kernel of fruit, or pip
mokome a present of berries, or bulbs
mokwalakwala spineless monkey orange
molodi the innermost bark of a thorn tree, used for making twine; a
cord; fibre
monamo the rank, or runner of a creeping plant
monyelenyel€ peeling plane (Achiaaceae ); a crocus; Ochna sp.
moragangaka a species of herb, used for healing
morokolo small sour plum; Olacaceae
morotologa sour plum
mositlwane a thistle-like plant; a species of aspargus

-134-
mosukujane Lippisa jaranica ; the leaves are infused and used as a drink
for fevers; fever-tree; wild tea, used as a cure; wild mint
mothata include species of Cordia ovalis; Pappea capensis
motlhabo a shoot; a sprout in growing grain
motlhagala species of mistletoe; infusion as tea
motlhonwane species of bush, or Maytenus tenuispina
motlhuje black root; blue bush; monkey plum; poison peach
motsotsojane cross berry; Grewia retinervis; Tilianceae
motswalakgoro a bush bearing edible berries; Waltheria indica
mouta mould
moyoant live-long bush

(3) Long objects made of wood

mogoma a plough; a hoe
mokasa a block of wood; a thick stump of wood
mokgomilo a hook
mon t eo a staff for beating sandals; (buttermilk)
mopaela a switch; a stick for beating with
mophakalego a cross-beam; a bar; a top cross-piece
morapa a horizontal bar on two upright poles to hang blankets over
mosetshe a peg
mostpun a beam; a cross-beam; a pole in a house for hanging karosses,
etc.
mosogole a wooden beam
motimela a charred piece of wood
mots he a pestle; a stamping stick
motswaiso a firebrand; a torch; a partially burnt stump; a piece of charred
wood; a stick from yesterday's fire
motswi an arrow

(4) Objects made of other plant material

mhikwana branch in gate to show that people are out
mmitsa charm that attracts customers; a charm that attracts men; (a
prostitute)
morurelo a pot used for cooking medicines
moswang the contents of an animal's stomach

- 135 -
motlaagana a temporary house; a house made of bushes; a tent; a
tabernacle
mutlwane a snare; a trap with a beam

(5) Other long objects or phenomena (not wooden)

moe me erect pointing, or erect thing; a name sometimes given to a
modimako, or tall pole with black and white stripes, and
crowned with civet cat tail that stands in the centre of the
Bogwera camp at the second initiation
mokgwabo the tracks of a drove of cattle, or a crowd of people
mokOlo a descending slope, a steep bank, a steep descent
mokwakwa state of clay loam during dry season
moraladi a stripe; a streak
moru ti a shade; a shadow
moseja used mainly as an ADV.S., to mean on the other side; across,
the land adjoining, or on the side of a river
mothaladi a stripe in colours; a streak between clouds; a scrap of porridge
left in a pot
mot halo line
motlha a time; a season; an era; date
motontonyane thin long bucket
motsekedi a sling
motshegare daytime; ranging from late morning till late afternoon
mot{hetsnedima lane; traffic lane
motshisa a stripe
mouwelo echo

(6) Long body parts

mmopo the bridge of the nose; the muzzle of an animal
moeledi a navel cord
mokukutho a bone at the tail of the animal; coccyx
mokwatla the back; the spine; a backbone
moletse a leg
mongobutsa a bone in an animal's leg above the shin
mono a thumb
motwane leg of a calf.

-136-
(7) Body-coverings made of skin
moitlamo girdle; waistband; belt; sash
mothikga ornamental skin clothes
motlokolo goat-skin apron, for girls and women worn over their
buttocks
motseto traditional leather strip worn by men through loins and
buttock; napkin
(8) Other body-coverings
mofitshana a plain iron ring worn on the wrist
moitshom elo armour; protective clothing
motlopo a tortoise-shell; (clitoris)

(9) Other objects made of skin

mofalo a small skin water-bag
mogogorwana a dummy calf made of skin, stuffed with grass; a dummy calf
that is licked by a lactating cow whose own calf has died
mokobolwane calfskin
mophanyan e chapped skin
moropane a timbrel; tambourin

(10) Skin related states (or sicknesses)

mmokwana measels
mobipo a weal; a swelling in which there is no fluid; or pus under the
skin
mofufutso perspiration; sweat
mogote Sg.: heat; fever. (Pl.: megote . refers to strong drinks)
mothopa small boil
motshwa an itchy eruption

(11) Lively human activities

mmino dance
moduduetso applause; ululating, of a woman in a dance; yodelling, of a
girl, or woman
mogolokwan e ululation
mokgobo a reproach; abuse; reviling
moopelwane the shouting, ululations and singing of a victorious army,
choir, etc.

- 137-
morapelo a prayer; a supplication; a prayer meeting
morupo a circumcision
mosate a seat of government; a capital city
moteka a merry noise; a feast
motse a village; a home; a homestead

(12) «Processes as things»

modiga an end of a dispute, or discourse
(go diga - to cast down; let fall)
mogano a breaking up, or cessation of rain; time for resuming work
after rain
mokgwasa rustling; actual sound, as of something moving through twigs
and foliage; (rumour )
mongatego a fainting fit
(go ngatego - to faint)
moo no a snore; a grumbling; moan
(go ona - to moan; groan; snore)

(13) Objects connected to ground

mogobodi a straight burrow
molala-thakadu an antbear hole, or lair
motlhobodika a large vent of an antheap
motlhodi the eye of a fountain
mot lobo a pit, or quarry; a place where clay for pottery, or sebilo, is dug;
a storage place

Remaining nouns
The following nouns are not easily fit into the suggested semantic network of
class 3:
mfama half
mogadikego a stomach ache
mogontlho act of taking advantage of another
mokea (used mainly as a REL. S., that which i crooked, or uneven)
mokgweleo a burden; a heavy load
mokokomalo botulism
molafo rabies
molao a law; a rule; a regulation

-138-
mole bela used mainly as a ADV. OFT., to express early afternoon;
excessive heat and glare of the sun; just after midday; when
the sun gets into your eyes
mopalo the state of being famished; malnutrition
mope leta spelling
mophetsoso testis
moputso reward; prize
morafe a tribe; a nation
moripo a slope; the bottom of a hill; a dam constructed at the bottom
of a hill
mo s imana used for anus
mos oko very thick unfermented porridge
Mosupologo Monday
mota (mmoto) the part below the anus
motshomilo water running from the nose
motshotelo dry dung dust
mot so used principally with numbers more than ten; a unit
motswatswa used mainly as a REL. S., and as an ID:, a snow-white thing;
spotless whiteness

Animals
mmoulo a mule
moduolo a large, blackish hare
mokgerwa a lean, starved dog; a mongrel
mokwidinyan e Namaqua dove; Oena capensis
moditswe a species of black lizard
mo gofu a big rhinoceros
mogokong young locusts in the stage prior to flying, and when they have
changed in colour from black to red
mogol egwe a species of small bird
mogotomoduan e a poisonous grub that covers itself in grass, fatal to animals
that swallow it
mogwelegwele a jay
mokabaoane klipspringer
morubisi an owl, believed to be used by witches at night
mos welesw ele a bird, found in thickets, that makes a noise like that of
chopping

- 139-
motlhakane a locust ready for eating
motoutwane a species of moth very destructive to clothing

Class 5
(1) Fruit
Iegapu water-melon
Iegelegele unripe fruit; (lekgela is more frequently used)
legomane a species of pumpkin
Iekwele an edible tuber; a table potato
lemi the fibrous part of a pumpkin, melon, etc.
leris6 a species of large edible bulb
lerotse a melon; a melon that is not of the sweet species
Ierula the fruit of the cider tree Anacardiaceae
letseme an immature corn, or sweet reed

(2) Other non-individuated, roundish objects («non-fruits»)

leb6te a bruise; blood blister in the flesh
legwarane a stone, tooth, etc., for rubbing skins
lehura a piece of fat
lek6k6ma a side-burden on a pack-ox; the impedimenta of a travveler
carried on his back; the rough points and swellings on
smooth surfaces; impediments on what would otherwise be a
fine, smooth blisterless piece of wood
lekwakwa a sharp-edged stone; a rough stone of volcanic formation
lelepa a difficult or intricate knot; a puzzle; a riddle; a mystery
lengope a lump of porridge when badly stirred
lenyaphiri limestone; a block of limestone with holes in it
lephoto a knot in a stick, or piece of timber; a place in a branch, or reed
where a twig has been cut off
lerophi a blister; small boil; corn
lesope a ruin; a deserted house, or ruins of a kraal

(3) Other roundish (but individuated) objects

lefiswana little pot
lekuka a skin sack; a milk sack, made of skin (nowadays it may be made
out of other materials). N.B. it is used for processing fresh milk
into sour milk or madila

-140-
lesiba the outer skin of a goat's bowel, and the musical instrument
made from it
letsatsi the sun; a day; heat caused by the sun
letshego a trivet, or three-legged stand for a pot, etc.: a stone etc. for a pot,
or kettle to stand on to be boiled

(4) Other non-individuated («but non-round») objects

lekala department
leketlo a chip of anything; a bit of anything broken off
lephata department
lerapo bone
lesapo a bone
letlepu mirth; plentitude of food, or any other material possessions;
abundance, especially of food
letlhomaganyane a small seam in a kaross
leudi a speck; a shred, as of a blanket

(5) Paired, roundish parts of the body

lebotswana the top of the shin bone

legofi palm of hand; handful; applause
legope the shoulder blade
legwele ankle
le hinini a gum; the place in a gum where a tooth is missing
lekgwafo a lung
lelefu a tonsil; an uvula
lelokololo a joint of a finger, wrist or ankle
lepara the upper inner part of a man's thigh, or that of a male
animal; space between legs; not used of a female; a wide step
lerago a buttock
lerete testis; scrotum
leth eka waist; loin
letlhafu the muscle or lump of flesh in the calf of the leg, and in the
arm between elbow and wrist
letlhap ono retina
letswafo a lung
letswele a fist; a small compact group of people

-141-
(6) Single, roundish body parts
letlaparapelo pericardium

(7) Paired, «non-round» body parts

lefuka a pinion; a wing; a feather
lesufu a thighbone or upper leg bone of an animal or human being;
the upper arm bone
letlhakore flank; side

(8) Paired, roundish body-adornments

lenyena an ear-ring
leseka a bangle or bracelet; formerly a twisted metal ring worn on the
arm and below the knee; a leg ornament; a coiled spring

(9) Roundish enclosures or locations

lebatla an open space; a public assembly; a gathering of people at the
kgotla
ledutela a pond or pool in the veld; a waterhole in the veld
lefatsh e a country; a land; a state; the earth; the world
Iegae a home; a dwelling place
legotlhe universe
lehalahala hall
lekgelebe a bank or dam for holding a small quantity of water
lekoma a mud or brick wall
lelapa a low wall or enclosure in a home; a home; a family
lethudi a compartment under the eaves of a house; stoep under the
eaves of a hut or house
letsna a pan, or shallow lake; a vlei, a saltpan

(10) Non-extended (in the domain of time), non-individuated, «linguistic items»

leaka (maaka is commonly used); a lie
lebitso a name; a nomenclature
lediri a verb
leemedi pronoun
lentswe a voice; a word; the sound of a voice
lesalao a load shouting
leselo noun

-142-
lesupi demonstrative
letlhalosi adverb
letlhaodi adjective
letsibosi interjection; an exclamation

(11) Non-individuated human beings

lebelete a wanton; a prostitute
lefelepa a cheat; a rogue
lefetwa one passed by; an old maid
lekau a youth; a young man
lekgadi an individual of the Manthatise clan
lekgarebe a virgin; an unmarried lady or recently married young
woman
lekgoba a slave; a member of a slave family
Lekhalate a coloured; a person of mixed blood in Southern Africa
lekorotlo a round-shouldered person
lentaatshwene a valiant man; one who has slain a man in battle; one who
has killed a ravenous beast
lepempet lh wane a small boy just big enough to herd kids of goat
lesafe an albino
lesetedi an individual of the Griqua race
lesokgo la a child just able to run about
leswebelega a hired soldier; a mercenary
letagwa a drunkard
letswalanoga a deceiver; a scamp

(12) Negative states

lebekebeke starvation
lebo go suffering; agony
lefaratlhatlha confusion; a confused state of affairs; a state of confusion
lehuma poverty; great sorrow; damage; loss
leitlhapelO riotous mirth
lenyora thirst
lepono nakedness
leso death
le sotlo contempt; mockery

-143 -
REST
leanetema agenda
Iebadi scar
Iebududu dry, hard ground
lediana a hindrance
legakgala the very early dawn; the first streak of light; a caterpillar with
shiny silvery dots along the body and which feeds on mosu
or umbrella thorn and on mogonono
legale blade, or cutting edge of a knife, chisel, etc.
legare a razor blade
lejeleputswa gold reef
lekarapa a mine recruit; helmet made of metal or some hard protective
material
Iekeisen e location
lekhurane rash
lemao needle; injection
lenga kare the grapple plant
leotlana a chief's officer; a town policeman
Ierol e powder; dust
lesaba a great secret meating
Ies ire a veil for the face
leso mepedi dozen
leswagadi a plant used in tanning
lesware footsoreness, either in persons or animals; rash, particularly
that which appears on a recently shaved skin
letetsa mucus
Ietlabula the time of the first ripe fruits; summer time; season
preceding harvest time
letlalokwalo parchment
letlatlana dusk; eventide; early evening twilight; the gloaming
letlhatso vomiting
Ietlhogotshweu a species of bush with white flowers; an old man
Ietlhwa snow
Ietlotlobo a snake skin; the scale of a fish; a piece of skin pulled off after
a scald, etc.
letshogadi jackboot
Ietshoroma ague; malaria; rigor; shiver; fever; influenza

- 144 -
letsoku red ochre, used with fat as an unguent for the body
letsopa a clay for making pots, etc.

Animals
lebolobolo a puff adder
lehututu ground hornbill
lekaba a pack-ox
lekanyane a wild dog
lemphorwana a young bird; a nestling
leowang a crested crane, red and white in colour
lerane plover
lesaiboko a very big goat; an angora goat
letlonkana a black wolf
letototo manded mongoose
letsutsuropu a species of bird
lewang a stork
legou a species of wild goose
leio a crab; a crabfish
lengodu a female blue tick (likely)
lenong a vulture; a black vulture
lephu sa a cow that has ceased to give milk
lesogo a partridge, included the red-winged one

Class 6
(1) liquids
madila thick milk; sour milk
makub elo the last milk in a sack which only comes out after shaking
makgaritlha the dregs of beer, traditionally reserved for older men
maphoko froth on ground caused by rain water
masi milk

(2) Other physical masses

mabududu dry dust
makhafe fringes
marerem ela clots in a gravy; curds of sour milk, etc.
marorodi embroidery
matsanko wounds

- 145-
matshetshetho bran from sifted meal; husks etc., from sifted grain

(3) Collective plurals

magoma a species of burr weed
magosz kings; kingdoms; chiefs; chieftains
makolane a thicket of palm trees
manku many sheep; flocks of sheep
marakangwaga ground plants with edible red fruits
maroana a thicket of mimosa trees of a certain species
mabi a thicket of a certain species of thistle

(4) Unspecified concrete objects

maadimo borrowed things; borrowing or a loan; borrowings or loans
mafalo scrapings
malao spiCes
man o t 6 strays, as of locusts, cattle, etc.
mantsa loose hanging things
masomosomo many; crowds
ma teng entrails of animals; contents; mysteries; deep things
matlhakola a remnant
matshwititshwiti large numbers
matswaketswake a jumble, or confused mess; a hodgepodge of food
maungobolekane canned fruit

(5) States
Temporary states
mahupuhupu state of being swollen and bruised from a beating; being bitten,
as by bedbugs
makgwere a state of confusion in the mind; problems; difficulties
maleka equal; state in which two opponents, or opposing sides are
evenly balanced
malobolobo state of being drenched, or saturated
mangetengete state of being dented, or pierced all over
manyerenyere state of being very wet, very muddy
matlhatlharara state of being untidy; being in a dishevelled and unkempt
state

-146-
Qualities
maatla strength; power
mafatswa persuasiveness
magagarapa shape of an animal's horns that point sidewards and then
turn forward and in
maikaelelO determination; intention, or aim; intentions; purpose(s)
makgakga rudeness; impertinence
makgeme that which is watery; action of always flowing or running, as
running eyes
manyama auspiciousness; luck
maona immaturity
maswato laziness
matletsetletse great fullness; abundance, as provisions

(6) «Collective events»

mad u me greetings
maiphitlha-
phitlhwane evening recreation; hide and seek
mapekepeke flashes of light; sparks
magawegawe disputes; uproars
matshediso condolences

REST
mabapi concerned with
madiso herding or grazing; pasturage
magamugamu a watery swelling; a juicy morsel
magwata roughness of a surface; blisters on a surface
mahube the blush of dawn; the streaks of colour and light at dawn;
(joint of meat)
maits oketso crooked ways
makalaba large eyes
makgafa a honeycomb
menangwane details of a subject that is under discussion; ideas
manyediana mumps
marinepe the border of a garden
maruru winter clouds with cold, but no rain
matikiri quarry

-147-
 Class 7
Instruments
sealo a skin, etc. spread out for lying on; a sleeping mat
(go ala - spread out on the ground; make bedding ready for
use)
sea no a totem; an object of veneration
(go ana - tell; relate; venerate; treat as sacred)
sebaga a bead; a string of beads; a necklace
sebediso leaven; yeast
sebetapoane a superior kind of necklace
(go beta - choke)
sebetsa a dangerous weapon; any instrument of punishment
(go betsa - hit)
seboleledi a mechanical device for recording speed
(go bolelela - tell somebody)
seburuburu a whirling toy; anything whirled to keep opponents away
sedibeletso hand vessel placed to catch dripping liquid in
(go dibeletsa - cause to be caught or trapped)
sediriso appliance
(go dirisa - use)
sedulo a seat (same as setilo)
sefawa shrapnel; a missile
sefemo a weapon of defence
(go fema - avoid blows; ward off; parry)
sefikantswe a cairn; a gravestone; a heap of stones piled up for some
purpose. According to tradition, passersby entering a town
usually contributed to the pile to ensure good luck
segafa a bead
segajo a stone engraving, or engraving on stone
(go gaba - hollow out; scoop out; hew out; cut out, as an
engraver)
segelelo a vessel for drawing water
(go gelela - draw water I fluid)
segopa a dustpan; a scraper
(go gopa - to remove by scraping; sweep up)
seikaego a thing to lean against
(ikaega - rest upon or against; depend upon)

-148-
seiphemelo that whith which one defends oneself
(go iphemela - defend oneself)
sekampan e a bow
sekano a seal; a cement used by bees, etc.; a wax
(go kana - seal; cement; stop holes with cement)
sekei yoke-pin
sekgampane a kind of musical instrument
sekgele a banner; a flag; a kind of umbrella made of black ostrich
feathers; the wool of a black sheep wound round a stick and
planted as a goal; a goal; a prize
sekgopo a tool for gathering or collecting things
(go kgopa - trip up)
sekhobe a shoe made of home-made leather
sekhurumelo lid, or a cover
(go khurumela - close)
sekokolo a harrow for breaking clods of earth after ploughing
sekopisi photocopier
sekorolo scroll
sekotsekara scotch cart
sekurufu a screw
selaga a snare; a trap
selefera silver; silver coins, or money
seiekanyi gauge
(go lekanya - measure)
seiepe an axe
selogo a weaving machine; a handloom
(go loga - plait; weave)
seme a whip
senwelo a drinking vessel; a cup
(go nwa - drink, nwela - sink; drink for)
seo lomank wane a small type of anthill
sepaga mo a ladder; any mode of transport, such as bicycle, a car, etc.
(go pagama - climb; mount; ride)
sepalangwa vehicle
(go palama - to ride, palangwa - is being ridden)
sepa pied i a battering ram device for pushg or striking with great force
sephemelO-sehubeng a defence for the breast; a breast plate

- 149-
 (go femela - protect, sehuba - chest)
sephimolo a duster; a towel
(go phimola - wipe; wipe off, or out)
sephoti treshing machine
(go phota - tresh)
sepomo an implement for dividing something; scissors; shears; a
pruning hook
(go poma - divide; cut off; snuff a candle; prune)
sera fa a receptacle for unthreshed corn made of poles; a pole
platform on which garden watchers stand
serapo an oar
serua a bed chamber pot; a night pot
sesiramatlho goggles
(go sira - cover, matlho - eyes)
set ek u a special winnowing basket
seteo a lash; a whip
sethato a charm
setilobanka settee
setlhabamokgosi siren
(go tlhaba mokgosi - make a noice)
setlhaga a nest; a bird's nest
setlhatlhosi lift; escalator; cage for taking passengers or goods from one
floor to another
(go tlha tlhosa - lift up)
setshamekiso toy
(go tshameka - play)
setsholetsi crowbar
(go tsholetsa - lift up)
setshwantso a picture; an image; a parable; film; bioscope
setsidima a bell
(go tsidima - to tinkle, or ring, as a bell)
setswalo a door; a shutter
(go tswala - shut)
sibi a cake of dry dung

-150-
(2) Manners of acting, characteristics (including languages)
sebele personality; reality; a person himself; bodily form (mmele -
body)
segomo behaving like cattle (kgomo - a cow, bull)
segosi royal; pertaining to royalty (kgosi - a chief, king)
sejatlhapi English (go ja : to eat, tlhapi : fish) (Mojatlhapi - an
Englishman)
Sekololo language of the Bakololo; in the custom, or tradition of the
Bakololo (Mokololo - a «kololo-person»)
semolao legal; legally
senoka sideways
serodumo disloyalty; instability of affection
setlhanka servile; pertaining to servitude

(3) Places associated with actions

sebaa a small bare place on the ground
(go baya -put down)
sebothelo a resting place for cattle
(go botha - to repose; rest)
sebatla an open valley, or a treeless place, or plain; an open space in
the veld, or a level place, hence a spot that is suitable for
holding unregulated meetings; a kind of Hyde Park Corner
sefero a path between hedges
setsha a plot; a plot of ground for building, or other purposes; estate

(4) Persons (agents) with extreme manners of acting

Positively acting persons
sebui orator; speaker
(go bua - to talk; speak)
seemana a mighty one, hence monna wa seemana , a man of war. Also:
mightiness
segwaba a person who can accomplish or contrive to do difficult things
sejadi planter
(go jala -to sow)
sekgapho a fast walker, or quick traveller
seopedi a renown, or famous singer; an expert singer. Also: musical
instrument (go opela - to sing)

-151-
Negatively acting persons
seaka a harlot; a prostitute
segagapi an extortioner
seganana naughty individual
segokga a hefty fellow; a tough fellow
segwanti a bully
(go gwanta - to be wilful; conduct oneself in a defiant, or
impudent manner; ... )
seitaodi an unruly individual; a lawless individual; one who is not
law-abiding
(go itaola - to be lawless; be delinquent; govern oneself)
seitsamaisi a superior, self-possessed, self-sufficient person; one noted for
something; one who regulates his own movements
sephaphathegi a very troublesome person or animal
sera an enemy troop of soldiers; an enemy
seruku tlhi a rebel; a troublemaker
(go rukutlha - to rebel; brawl; riot)
seswapelo a sloven; a dead hearted one; desperado
(go swa - to die, pelo - heart)
setlhodi aspy

(5) Other negatively marked persons

sea ramo an unfortunate fellow; a sufferer
selema a fool
semumu a dumb person
serothola a foolish, careless person; one not able to do anything
properly
setete a spoilt child
setlatla a fool; a silly person
setsururu a careless person; an untalented person

(6)Objects affected by action

seem era an infirmity of long standing; an old weakness, or sore
sefolotsana an aborted foetus of an animal; a weakling
(go folotsa - to abort)
segogou an old and worthless skin; (an old skinny person)
sefudu a falling of the womb; a prolapsus of the walls of the vagina

-152-
 (juduga - remove one's abode; depart; migrate to)
sejo food; a type of food
(go ja - to eat)
sekabetla a piece of cardboard, etc.; a portion of food, etc
(go kabetla - to cut off pieces)
seketana a rag
sekgwene a stump
sekwatla dry and shrivelled up (of skins, berries, etc.), used mainly as a
REL.S. (also: work-gang; roving work-gang; group of people)
selalelo an evening meal; a supper; the sacrament of the last supper;
(lalela - sup; take the evening meal)
serekolodi a ransom; a thing given as a ransom
(go rekolola - buy back; redeem; ransom)
serwala food provided by the friends of a girl at her betrothal
(go rwala - carry)
sesa a stump (of a tree)
seso salt that has lost its flavour
sesupo a sign; evidence; a testemony; testemonial
(go supa - to point out/ at)
sethunya a flower; a blossom; a gun
(go thunya - to explode; burst; blossom)

REST
sebeelo a pledge
seboka congress; group of people gathered for a specific purpose
(go boka -chase away)
sebopelo a woman's womb
sedikwadiki circular in shape; measureless in extent; numberless
(go dika - go around something)
seetlaetlana a cherished one; a much-cared-for child
sefako hail
sefane surname
sefara the dividing of a river into two streams, or a road into two
paths
sefoka a strong aroma, either good or bad; a smell
(go foka -blow, as the wind)
segakgamatso a wonder

-153-
segateledi nightmare
segofi [ARCH.] the palm of the hand
segopolo a thought
(go gopola - think; meditate; remember)
segwete an edible root shaped like a carot; beetroot
seidi giddiness
seimana nightmare
sekaka a desert; any impediment, or hindrance to progress; a
waterless country
(go kaka - be drawn two ways; be engrossed in, hindered)
sekate trouble, or difficulty; a state of danger, or difficulty
sekgala a long distance, a long way off
sekgokgono an elbow
sekgwamolelo volcano
(go kgwa molelo - spit out fire)
sekhutlontsi polygon
sekidi a patch; an island; a large piece of meat, etc.
sekwalo a title; a writing; a heading
(go kwala -write)
selaole species of bush; Lantana rugosa ; brandy bush; violet (colour)
selete a deep hole in the ground; a deep pit
selopo an elephant's trunk
semangmanyane a little bit of anything
senana area above pubis; mons pubis
senganga thick or sticky porridge; clay which sticks to the wheels;
hooked thorns of Acacia mellifera
(go nganga - nag)
sengwatha a lion's share of something
sennanne such-and-such a thing, etc.
senono a stalk of sweet reed at the start of th ear, which when chewed
serves to tie the bundle of reeds
senyane species of grass with white heads, used for making hats
sea a thing (contraction of selO )
sepitlaganyane DIM. of sepitla , a tight corner; a critical position
sere the butter
seretologa sour plum (same as moretologa )

-154 -
seriri hair that is overgrown; hair; a slaughtered animal at a
marriage feast
seropo a small excrescence on the body; a small swelling on an
animal's forehead caused by a cut made in order to
distinguish ownership
seroa Pagodus , a herbaceous perennial with small tuberous
rootstock
se rokole a bud
seroronya deep mud
sesetlana a storm; a hurricane; a strong rushing wind; cold weather
sesikalaba a large live coal; a fire-brand
sesole military (the army)
sethoboloko noon
setlaa the jaw; the jaw bone (more common:letlhaa)
setlhokolane small round wild beans of the blue gurri bush (also:
motlhakolane )
setlog olo a nephew; a niece; a man's sister's child; a woman's husband's
sister's child; a grandchild
se tlolO an odd number; a number over and above a fixed one
(go tlola - skip)
setlwasetlwane a species of thistle that bears red berries
seto to a corpse
setsala an animal's womb
(go tsala - beget; bring forth)
se tshi Acacia hebecleda ; a species of woody plant
setshwaelo an act of consecration or dedication; a hen, goat, sheep or cow
given to a child in the hope that it will multiply
(go tshwaya - brand an animal; mark in any way as a sign of
ownership; mark a script and award marks)
setsubaba a spot of any colour, but white on an animal
setulo luck; chance

Animals
sekwakwalala a cricket; the cicada beetle; screeching beetle
selealee a tarantula spider
seokomana a huge animal
serolo-botlhoko a bush buck

-155 -
serunyi a mole
setlhong a hedgehog
sebokwana small worm
segongwane a young locust, just able to fly
segotshane a species of falcon
sengalaga cattle with big horns
sepirinka an ass

-156-
Appendix II. Tests, part 1 and part 2

NONSENSE WORD TEST

Name:

Age:

Sex:

Occupation:

Examples with nonsense words:

motlwapo (metlwapo) a person who is fond of children

an alcoholic drink
a kitchen utensil
a long, black stick
the round tail of a special kind of rabbit

a kitchen utensil mothoketho (methoketho)

sethoketho
lethoketho
mothoketho (bathoketho)
mathoketho

CHOOSE ONLY ONE MEANING/NONSENSE WORD FOR EACH BOX !!

-157-
morutsa (barutsa) a small, round ball
a soft-drink
a person
a tool that is used for making soap
a tree that grows in Europe

mothethi (methethi) a tree that grows in Europe

a very noisy and happy party
a dress that is made of skin
a box that is made of wood
a shelter made of leaves and branches
a long, thin line
a bush that grows in Norway
a hole in the ground

setsifo a person
a small, round ball
a soft-drink
a tree that grows in Europe
a tool that is used for making soap

leatepha a person who is an expert in a special way of dancing

a very small child
a very stupid person
a person from a foreign nation
a person who is working in a car factory

-158-
lerutsa a tree that grows in Europe
a small, round ball
a person
a tool that is used for making soap
a soft-drink

sekigo the language of a European nation

a worn-out dress
a tool that is used for making soap
a thing that a chief holds in his hand during a religious ceremony
the nest that the snow-mouse builds

matsifo a soft-drink
a tree that grows in Europe
a small, round ball
a person
a tool that is used for making soap

serutsa a small, round ball

a tree that grows in Europe
a soft-drink
a tool that is used for making soap
a person

-159-
letitle a big, round meeting-room
a small, round ball
the round part of an animal's back-leg
a rude word
a small, round fruit
an ugly person
the bad feeling you get after doing something wrong

moatepha (baatepha) a very stupid person

a person who is an expert in a special way of dancing
a person who is working in a car factory
a very small child
a person from a foreign nation

marutsa a soft-drink
a person
a tool that is used for making soap
a tree that grows in Europe
a small, round ball

mosumalelo (mesumalelo) a dress that is made of skin

a bush that grows in Norway
a tree that grows in Europe
a box that is made of wood
a long, thin line
a shelter made of leaves and branches
a hole in the ground
a very noisy and happy party

-160-
motarala (batarala) a very small child
a person who is an expert in a special way of dancing
a very stupid person
a person who is working in a car factory
a person from a foreign nation

seatepha a person who is working in a car factory

a very stupid person
a person who is an expert in a special way of dancing
a person from a foreign nation
a very small child

motsifo (metsifo) a soft-drink

a small, round ball
a person
a tree that grows in Europe
a tool that is used for making soap

sepotsime the nest that the snow-mouse builds

the language of a European nation
a worn-out dress
a thing that a chief holds in his hand during a religious ceremony
a tool that is used for making soap

morutsa (merutsa) a person

a soft-drink
a tree that grows in Europe
a small, round ball
a tool that is used for making soap

-161-
letophu a small, round ball
the bad feeling you get after doing something wrong
a big, round meeting-room
the round part of an animal's back-leg
a rude word
a small, round fruit
an ugly person

motsifo (batsifo) a tool that is used for making soap

a person
a small, round ball
a soft-drink
a tree that grows in Europe

letarala a very stupid person

a person from a foreign nation
a very small child
a person who is working in a car factory
a person who is an expert in a special way of dancing

letsifo a tree that grows in Europe

a tool that is used for making soap
a soft-drink
a small, round ball
a person

-162-
setarala a person from a foreign nation
a person who is working in a car factory
a person who is an expert in a special way of dancing
a very small child
a very stupid person

a person setlhorape
motlhorape (metlhorape)
motlhorape (batlhorape)
matlhorape
letlhorape

a small, round ball leboreleta

seboreleta
moboreleta (baboreleta)
maboreleta
moboreleta (meboreleta)

a person from a foreign nation mothephima (bathephima)

lethephima
sethephima

a person who is working in a car factory lethephima

mothephima (bathephima)
sethephima

-163-
a tree that grows in Europe letlhorape
setlhorape
motlhorape (batlhorape)
matlhorape
motlhorape (metlhorape)

a small, round ball that is made of wood lephubi

mophubi (mephubi)
maphubi
sephubi
rnophubi (baphubi)

a very small child mothephima (bathephima)

sethephima
lethephima

a tree that looks like a ball sephubi

lephubi
mophubi (baphubi)
mophubi (mephubi)
maphubi

a person who is an expert in a special way of dancing letikga

setikga
motikga (batikga)

-164-
a small, round ball matlhorape
motlhorape (batlhorape)
letlhorape
setlhorape
motlhorape (metlhorape)

a tree that grows in Europe mabor€deta

leboreleta
moboreleta (meboreleta)
seboreleta
moboreleta (baboreleta)

a very stupid person sethephima

lethephima
mothephima (bathephima)

a small, round ball that is made of wood lephogalaro

maphogalaro
mophogalaro (mephogalaro)
sephogalaro
mophogalaro (baphogalaro)

a liquid that comes from trees lephogalaro

maphogalaro
mophogalaro (bap hogalaro)
sephogalaro
mophogalaro (mephogalaro)

-165-
a person who is working in a car factory setikga
letikga
motikga (batikga)

a tool that is used for making soap motlhorape (batlhorape)

letlhorape
matlhorape
motlhorape (metlhorape)
setlhorape

a tool used for making a special liquid sephubi

lephubi
mophubi (baphubi)
maphubi
mophubi (mephubi)

a very stupid person motikga (batikga)

setikga
letikga

a soft-drink setlhorape
matlhorape
letlhorape
motlhorape (metlhorape)
motlhorape (batlhorape)

-166-
a liquid that comes from trees maphubi
lephubi
mophubi (baphubi)
mophubi (mephubi)
sephubi

a person moboreleta (meboreleta)

seboreleta
leboreleta
moboreleta (baboreleta)
maboroleta

a tool used for making a special liquid mophogalaro (mephogalaro)

mophogalaro (baphogalaro)
maphogalaro
lephogalaro
sephogalaro

a very small child motikga (batikga)

letikga
setikga

a tool that is used for making soap maboreleta

leboreleta
seboreleta
moboreleta (baboreleta)
moboreleta (meboreleta)

-167-
a soft-drink moboreleta (baboreleta)
seboreleta
maboreleta
leboreleta
rnoboreleta (rneboreleta)

a tree that looks like a ball sephogalaro

maphogalaro
mophogalaro (mephogalaro)
mophogalaro (baphogalaro)
lephogalaro

a person from a foreign nation letikga

motikga (batikga)
setikga

a person who is an expert in a special way of dancing

sethephirna
rnothephirna (bathephima)
lethephirna

-168-
TEST, PART 2 Name:

a person who is working with boats maforobeta

moforobeta (baforobeta)
leforobeta
seforobeta
moforobeta (meforobeta)

the round part of an animal's back-leg mofeo (bafeo)

lefeo
mafeo
sefeo
mofeo (mefeo)

a dress that is made of skin sekgotepha

mokgotepha (bakgotepha)
lekgotepha
mokgotepha (mekgotepha)
makgotepha

a worn-out dress mokugothebo (bakugothebo)

lekugothebo
mokugothebo (mekugothebo)
makugothebo
sekugothebo

the state of being a little bit tired mokugothebo (mekugothebo)

makugothebo
lekugothebo
mokugothebo (bakugothebo)
sekugothebo

-169-
a thing that a chief holds in his hand during a religious ceremony
lefeo
sefeo
mafeo
mofeo (bafeo)
mofeo (mefeo)

a long, thin line moforobeta (meforobeta)

maforobeta
seforobeta
leforobeta
moforobeta (baforobeta)

a worn-out dress makgotepha

lekgotepha
mokgotepha (mekgotepha)
mokgotepha (bakgotepha)
sekgotepha

a person who is behaving well sekugothebo

mokugothebo (mekugothebo)
lekugothebo
mokugothebo (bakugothebo)
makugothebo

a big, round meeting-room makgotepha

lekgotepha
sekgotepha
mokgotepha (bakgotepha)
mokgotepha (mekgotepha)

-170-
the kind of things that are kept in a box seforobeta
maforobeta
moforobeta (meforobeta)
leforobeta
moforobeta (baforobeta)

a person who is working with boats mafeo

mofeo (bafeo)
sefeo
mofeo (mefeo)
Iefeo

a long, thin line sefeo

mofeo (mefeo)
lefeo
mafeo
mofeo (bafeo)

the state of being a little bit tired mokgotepha (mekgotepha)

sekgotepha
makgotepha
lekgotepha
mokgotepha (bakgotepha)

a thing that a chief holds in his hand during a religious ceremony

moforobeta (meforobeta)
seforobeta
maforobeta
moforobeta (baforobeta)
leforobeta

-171-
a person who is behaving well lekgotepha
mokgotepha (mekgotepha)
mokgotepha (bakgotepha)
sekgotepha
makgotepha

a dress that is made of skin makug6theb6

mokug6theb6 (mekugothebo)
sekug6theb6
lekug6theb6
mokug6theb6 (bakugothebo)

the round part of an animal's back-leg leforobeta

seforobeta
maforobeta
moforobeta (baforobeta)
moforobeta (meforobeta)

the kind of things that are kept in a box mofeo (mefeo)

sefeo
mafeo
mofeo (bafeo)
lefeo

a big, round meeting-room lekug6theb6

mokug6theb6 (bakugothebo)
makug6theb6
sekugothebo
mokug6theb6 (mekugothebo)

-172 -
setser6 a person who is working with boats
the kind of things that are kept in a box
the round part of an animal's back-leg
a long, thin line
a thing that a chief holds in his hand during a religious ceremony

makathe a worn-out dress

a big, round meeting-room
a person who is behaving well
the state of being a little bit tired
a dress that is made of skin

mokesele (mekesele) a dress that is made of skin

a worn-out dress
a person who is behaving well
a big, round meeting-room
the state of being a little bit tired

mokorite (mekorite)
a thing that a chief holds in his hand during a religious ceremony
a person who is working with boats
a long, thin line
the kind of things that are kept in a box
the round part of an animal' s back-leg

letser6 a long, thin line

the kind of things that are kept in a box
a thing that a chief holds in his hand during a religious ceremony
a person who is working with boats
the round part of an animal's back-leg

-173-
lekorite the kind of things that are kept in a box
a thing that a chief holds in his hand during a religious ceremony
a long, thin line
the round part of an animal's back-leg
a person who is working with boats

mokorite (bakorite)
a person who is working with boats
the round part of an animal's back-leg
the kind of things that are kept in a box
a thing that a chief holds in his hand during a religious ceremony
a long, thin line

sekathe the state of being a little bit tired

a big, round meeting-room
a dress that is made of skin
a worn-out dress
a person who is behaving well

makesele a person who is behaving well

the state of being a little bit tired
a worn-out dress
a big, round meeting-room
a dress that is made of skin

mokathe (bakathe) a person who is behaving well

a dress that is made of skin
a worn-out dress
the state of being a little bit tired
a big, round meeting-room

-174-
matsero the kind of things that are kept in a box
a thing that a chief holds in his hand during a religious ceremony
the round part of an animal's back-leg
a long, thin line
a person who is working with boats

lekesele a big, round meeting-room

a worn-out dress
a dress that is made of skin
a person who is behaving well
the state of being a little bit tired

mokathe (mekathe) the state of being a little bit tired

a worn-out dress
a dress that is made of skin
a big, round meeting-room
a person who is behaving well

makorite the round part of an animal's back-leg

the kind of things that are kept in a box
a thing that a chief holds in his hand during a religious ceremony
a person who is working with boats
a long, thin line

mokesele (bakesele) the state of being a little bit tired

a dress that is made of skin
a worn-out dress
a person who is behaving well
a big, round meeting-room

-175-
motsero (metsero)
a thing that a chief holds in his hand during a religious ceremony
the kind of things that are kept in a box
a person who is working with boats
a long, thin line
the round part of an animal's back-leg

sekesele a dress that is made of skin

a worn-out dress
a person who is behaving well
the state of being a little bit tired
a big, round meeting-room

sekorite a thing that a chief holds in his hand during a religious ceremony
the round part of an animal's back-leg
the kind of things that are kept in a box
a long, thin line
a person who is working with boats

motsero (batsero)
the round part of an animal's back-leg
a person who is working with boats
the kind of things that are kept in a box
a long, thin line
a thing that a chief holds in his hand during a religious ceremony

-176-
lekathe a dress that is made of skin
a big, round meeting-room
a worn-out dress
the state of being a little bit tired
a person who is behaving well

-177-
-178-
Appendix III. Statistical Model and Method

Statistical Model and Method

by
Grete Usterud Fenstad
Senior Lecturer
Department of Mathematics
University of Oslo

An individual is given k alternative answers to a certain question. We denote the

probability of correct answer by p. If the individual has no preference but picks
one of the possible answers at random, the probability of correct answer is 1 I k,
otherwise we believe p > 1 I k.
If several individuals (n) are asked the same question with the same k
alternative answers, it is reasonable to conclude with p > 1 I k if sufficiently many
(y) of them answer correct, i.e. if the test statistic

is large.
There is strong evidence for p > 1 I k if the probability of the observed
value or even more extreme values of the test statistic is small when p = 1 I k
(the so-called P-value or the significance probability).
We calculate the P-value for each question and conclude with p > 1 I k
when the P-value is less than 0.01, having a probability of at most 0.01 for wrong
conclusion for each question.

-179-
-180-
Appendix IV. Description and results of the tests 2-4 and 6

Test 2, 3 and 4
In these three tests my aim was to investigate the suggested prototypical structure
of the classes 3, 5 and 7. For each class I presented a range of meanings, each
compatible with one specific «low level schema» within the same class. The aim
was to see whether or not the subjects would favour the prototypical meaning,
and how the distribution of choices between all the other meanings would come
out. For each class I had two test items each with different roots. In this test only
the test type A would make sense.

Example:
mothethi (methethi) a dress that is made of skin
a bush that grows in Norway
a tree that grows in Europe
a box that is made of wood
a long, thin line
a shelter made of leaves and branches
a hole in the ground
a very noisy and happy party

Results
The most surprising result is the systematically low scores of the meanings that
are closest to what I have suggested as class prototypes, namely 'a tree that grows
in Europe', 'a small, round ball' and 'a tool that is used for making soap'. I am
particularly puzzled by the extremely low scores of 'tree', as trees seem to
constitute an exceptionally clear case of a prototype (c.f. section 4.2).
Characteristically, this was the only test item that made one of the subjects
comment, after performing the tests, that she did not have any feelings about any
of the items, except that trees begin with mo- .
Looking back, I can suggest two explanations: Firstly, there is a possibility
that the subjects have avoided the meanings that occur frequently in the test (all
these examples of «prototypical meanings» occur 14 times throughout the test,
while the rest of the meanings in these «prototype tests» occur only twice).
Selecting the same meaning again and again might intuitively seem «strange», as
long as there are other reasonable alternatives. Accordingly, these tests should

-181-
not have been mingled with the rest. Secondly, I do not believe that this test
design is well suited for prototypicality testing.

Test 2 (concerning class 3) -results:

40
36
34
35
30
24
25
tree-> a tree that grows in Europe
20 party-> a very noisy and happy party
15 skin-> a dress that is made of skin
10 wood-> a box that is made of wood
10
leaves -> a shelter made of leaves and branches
5 line-> a long, thin line
0 bush-> a bush that grows in Norway
tree bush wood skin line ground party leaves ground-> a hole in the ground

Test 3 (concerning class 5) - results:

35 34
35
30 room -> a big, round meeting-room
25 23 ball-> a small, round ball
20
leg-> the round part of an animal's back-leg
15 12 word-> a rude word
fruit-> a small, round fruit
10
person -> an ugly p erson
5
0
.. . )) feeling-> the bad feeling you get after doing
fruit ball leg room word person feeling something wrong

Test 4 (concerning class 7)- results:

49 48
50
45
40 36
35
30
25
20 16
language-> the language of a European nation
15 worn-out-> a worn-out dress
10
4
tool-> a tool that is used for making soap
5
reLthing -> a thing that a chief holds in his hand
0
tool rei. thing nest language worn- during a religious seremony
out nest-> the nest that the snow-mouse build

-182-
5.5.4: Test 6
This test was intended to invite to a shift in focal attention; two elements that
could easily be associated with different classes are included in each meaning. I
wanted to find out: 1) if this would lead the subjects to prefer the forms that were
compatible with one of the two classes that each element intended to be an
example of, and 2) whether the distribution between those two forms would
exhibit some systematicity.

The constructed meanings were:

'a small round ball [class 5] that is made of wood [class 3]',
'a tree [class 3] that looks like a ball [class 5]',
'a liquid [class 6] that comes from trees [class 3]', and
'a tool [class 7] used for making a special liquid [class 6]'.

Each meaning was tested twice with two different nonsense roots, and only with
the test type A.

Example:
a liquid that comes from trees map hub i
lephubi
mophubi (baphubD
mophubi (mephubD
maphubi

Results
It is difficult to be conclusive about these results, except that there seems to be a
certain tendency towards favouring the first and probably most salient part of the
meaning. The «modifying element» does not seem to influence the choices
much.

-183-
Class 5 + 3 Class 3 + 5
(a small, round ball that is made of wood): (a tree that looks like a ball):

70 80 76
70
70
60 55
60 56
50 50
40 40
30 30
20 20
10 10 11!11
o ~~~~~~~~~~L··~·~
· ~ 0
1 mo- 3mo- 51e- 6ma- 7se- 1 mo- 3mo- 5 1e- 6 ma- 7se -
ba- / me- / ba- / me-

Class 6 + 3 Class 7 + 6
(a liquid that comes from trees): (a tool used for making a special liquid):

60 54 70 66

50 60
42
40 50 43
30 40
30 23
30
20
20
10 10
0
... .. 0
..
1mo- 3mo- Sle- 6ma- 7se- 1mo- 3mo- Sle- 6ma- 7se-
/ ba- / me- / ba- / me-

-184-