The transparency of creoles - UvA-DARE

The transparency of creoles*
STERRE LEUFKENS
University of Amsterdam
Transparency is a one-to-one relation between linguistic units at all levels of
organization. It can be violated by several elements and operations, such as
apposition, cross-reference, grammatical relations, agreement, fusional
morphology, grammatical gender and expletives. A rigid parts-of-speech system is
non-transparent as well. Creoles have been argued to be more transparent than
older languages. This could be the result of the extreme contact situation in which
they emerged. Another possibility is that transparent features of creoles are
inherited from shared source languages. In this study, four creoles are compared
as to their transparency. They turn out to be transparent to a high degree, which
shows that transparency is indeed a characteristic of creoles. Especially formbased forms, i.e. elements and operations that have no semantic or pragmatic
counterpart or trigger, are infrequent in creoles. The creoles in the sample have
different source languages, so their transparency cannot be inherited from shared
source languages. It is argued that in an intensive contact situation, languages
need to be more transparent. All non-transparent features are stripped from the
language. Opacity can develop when a language grows older.
1. Introduction
Languages map meaning to form. A transparent one-to-one mapping seems the most logical
and efficient way to do this. However, natural language is not often transparent. The vast
majority of the languages of the world exhibit opaque structures (like grammatical gender and
negative concord), which do not conform to a one-to-one meaning-to-form relation.
Transparency in grammar is an advantage from a cognitive point of view. One-to-one
mappings are easier to learn, while the acquisition of opaque structures is slower. Consider for
instance the late acquisition of grammatical gender in Dutch, which is only fully mastered by
children around 8 (Blom et al. 2008). Apparently, transparency is the starting point of
language, while opacity is acquired later on.
This directionality does not only apply to language acquisition, but supposedly to
language genesis as well. Creole languages, relatively young languages, have been argued to
be more transparent than older languages (e.g. Seuren and Wekker 1986). Opaque phenomena
like grammatical gender are seldom attested in creoles (Trudgill 1999). Arguably, this is the
result of the intensive contact situation from which these languages resulted. In such a
situation, language must be fully efficient (i.e. transparent) to be effective. It is in growing
2
Sterre Leufkens
older that a language might acquire opaque features. Opacity then is ‘historical baggage’; it is
a non-functional extra.
However, not everyone believes in the transparency of creoles. McWhorter (1998,
2001) argues that simplicity, not transparency, is a typical creole feature. Others oppose to the
idea that the sudden emergence of creoles is responsible for their supposed special nature. It is
argued (cf. Arends et al. (eds.) 1995: 87-109) that properties of creoles are inherited from
superstrate or substrate languages. Well-studied creoles often share some source languages.
Among common ancestors are colonialist languages (French, Portuguese, English, Dutch) and
African languages (from the Kwa, Gbe and Fon families). Obviously, combinations of similar
languages result in similar creoles. The supposed transparency of creoles is then not a result
of the specific way in which creoles emerged, but a result of inheritance of shared ancestors.
This paper aims to find out whether creoles really are more transparent than older
languages. Four creoles will be studied on their degree of transparency. If they, as a group, are
more transparent than older languages, we can safely conclude that transparency is in fact a
characteristic of creoles. The hypothesis that a shared ancestor is the origin of this supposed
characteristic is put to the test by studying creoles with different source languages. If the
creoles turn out to be transparent, this cannot be the consequence of inheritance from a
common source language.
In the next section, transparency will be defined along the lines of the theoretical
framework of Functional Discourse Grammar. In Section 3, earlier work on the transparency
of creoles will be discussed. In Section 4 the methodology is explained and Section 5 presents
the results of the investigation of four creoles. In Section 6 conclusions are drawn.
2. What is transparency?
2.1. Definition
Transparency is the extreme situation in which there is a consistent one-to-one relation
between meaning and form (cf. Hengeveld forthcoming). In the framework of Functional
Discourse Grammar (henceforth FDG; Hengeveld and Mackenzie 2008), four levels of
organization are distinguished: pragmatics, semantics, morphosyntax and phonology.
Transparency applies to the interfaces between the pragmatic and semantic levels on the one
hand and the formal levels on the other hand. ‘Meaning’ is in this paper hence taken to
include semantic and pragmatic units, even though the term might not be perfectly suitable for
the latter.
Linguistic transparency can be violated in three ways: by redundancy, by domain
disintegration and by form-based form. Redundancy (Figure 1) comprises all relations
between one meaning unit and two or more formal units. One of the formal units could be left
out without loss of meaning and is hence redundant. Domain disintegration (Figure 2) occurs
when the integrity of formal units is violated. This happens for instance when two formal
units fuse, rendering a non-transparent many-to-one relation. Another kind of domain
The transparency of creoles 3
disintegration is discontinuity, where a formal unit is splitted, again resulting in non-parallel
alignment between two levels. The third type of violations of transparency is form-based form
(Figure 3), which includes all forms and formal operations that have no pragmatic or semantic
counterpart or trigger. For instance, expletive elements (e.g. it in it rains) are form-based
forms as they are elements that are there because of some formal rule, but do not mean or
refer to anything. Form-based forms can also be called non-functional, meaningless, or
syntactically autonomous forms.
Figure 1: Redundancy
Figure 2: Domain disintegration
Figure 3: Form-based form
Systematically combining these three types of non-transparency with the (interfaces between)
linguistic levels of organization, the result is a list of non-transparent phenomena. This list
will be given in Section 2.3, after an outline of the FDG model in Section 2.2.
Earlier definitions of transparency fail to capture crucial elements of the concept.
1
Seuren and Wekker (1986: 62) define what they call ‘semantic transparency’ as a one-to-one
relation between semantic units and morphemes. Grammars are argued to have three
strategies to maximize transparency: maximal uniformity of treatment of semantic categories
(uniformity principle), minimal reliance on rules or rule types that are highly language
particular (universality principle) and minimal processing (simplicity principle).
The uniformity principle is adopted under the current definition as well. Indeed it is
non-transparent if semantic categories are expressed differently due to morphosyntactic
influence (see Section 2.3.3). However, the universality principle is not adhered to in this
paper. Transparency is treated here as a property of rules or elements within a language. The
cross-linguistic particularity of a rule or element is not relevant for its transparency.
Furthermore, transparency is defined here as a property of languages. Language processing
(and its simplicity in a specific language) might be related to this property, but it does not
define it. Simplicity in Seuren en Wekker’s sense is therefore not part of the current definition
either (see Section 3 for a discussion of a different interpretation of simplicity).
Kihm (2000) defines transparency (‘optimality’ in his terms) as a consistent relation
between syntax and morphophonological form. I believe that transparency indeed applies to
such relations, but that in a complete study of transparency, interfaces between all levels of
language should be taken into account. A unified analysis of transparency should include
pragmatic, semantic, morphosyntactic and phonological units.
4
Sterre Leufkens
2.2. Functional Discourse Grammar
This section aims to
familiarize the reader with
the FDG model. The paper
should be understandable
without thorough knowledge
of
FDG,
but
some
background is helpful to
understand
certain
assumptions
and
interpretations.
Functional Discourse
Grammar (Hengeveld and
Mackenzie 2008, cf. Figure
4) is a top-down model,
which means that a linguistic
utterance is modelled as
starting as an intention to
communicate a message.
This message is formed in
the
non-linguistic
Conceptual Component. The
message then enters the
Grammatical
Component,
where it is formulated into
meaning units and encoded
into formal units. The
message
passes
through Figure 4: The FDG model of language (Hattnher and
different stages, which are Hengeveld 2007: 8).
modelled as levels. The output
of each level constitutes the input of the next level. The output of the Grammatical
Component is the input for the Output Component, where the message is put into sounds.
At the uppermost Interpersonal Level, pragmatic information (for instance reference or
the illocutionary force) is stored in the form of units called layers, namely the Move (M), the
Discourse Act (N), the Illocution (F), the speech-act Participants (P) (viz. Speaker (S) and
Addressee (A)), the Communicated Content (C), the Subact of Reference (R) and the Subact
of Ascription (T). These layers are hierarchically ordered, as illustrated in a simplified manner
in (1). The more to the left a unit stands, the higher it is in the hierarchy. Square brackets
stand around units that are not hierarchically ordered; units between them are equipollent. Not
all layers are necessarily represented in all utterances (an Act can for instance occur without a
The transparency of creoles 5
Communicated Content), and it is also possible that there are multiple instances of one unit
(e.g. several Referential Subacts in one Communicated Content).
(1)
M: A: [ F S A C: [T R] ]
Semantic information obtains at the Representational Level. At this level we again find
hierarchically ordered layers: the Propositional Content (p), the Episode (ep), the State-ofAffairs (e), the Configurational Property (f1), the Lexical Property (f2) and the Individual (x).
These layers represent semantic categories with specific properties. For instance, the
Propositional Content is a mental construct and can hence be modified by words like ‘maybe’.
A State-of-Affairs exists in time and space (and not mentally) and can be modified by among
other things time and place adverbials. The layers relate to each other as in (2). Again, units
can appear multiple times in a sentence, and some can be left out.
(2)
p: ep: e: f1: [ f2 x ]
Layers can be subject to modifiers (lexical elements) and operators (grammatical elements).
Formal units are located at the Morphosyntactic and Phonological Levels. The
Morphosyntactic Level contains Linguistic Expressions (Le), Clauses (Cl), Morphosyntactic
Phrases (Xp), Morphosyntactic Words (Xw) and Morphemes (Xm), ordered as in (3). The
latter three units can be of different types, where the type determines the first letter of the
abbreviation (e.g. Nw stands for a Morphosyntactic Word of the type ‘noun’). Units can
appear several or zero times in an utterance.
(3)
Le: Cl: Xp: Xw: Xm
Finally, the Phonological Level contains Utterances (U), Intonational Phrases (IP),
Phonological Phrases (PP), Phonological Words (PW), Feet (F) and Syllables (S), ordered as
in (4). They can again appear several or zero times in sentences. Furthermore, in some
languages there is no distinction between Phonological Phrases and Phonological Words.
(4)
U: IP: PP: PW: F: S
The square boxes at the left of Figure 4 contain the primitives, the building blocks, of each
level. Each language has these primitives, but the inventory is language-specific. So every
language has Lexemes, but each language has different Lexemes.
In the operation of Formulation, the non-linguistic message is converted to primitives
of the two highest levels (the Interpersonal Level and the Representational Level). A relevant
set of primitives here is that of Frames, which define which combinations of Lexemes are
possible. Lexemes are the second set of primitives at these levels; they are the basic semantic
units that FDG discerns. The upper two levels furthermore make use of Primary Operators.
6
Sterre Leufkens
Operators are elements that ‘operate on’, that is, do something to units of the respective level.
Primary operators can for instance assign semantic values to units at the Representational
Level.
In the operation of Encoding, primitives of the upper two levels are converted into
primitives of the lower two levels. Morphosyntactic Encoding converts meaning units to
morphosyntactic units; then Phonological Encoding converts these to phonological units. A
first set of primitives used in Encoding is that of Templates. These define how the units at the
Morphosyntactic and Phonological Levels are organized, for instance the ordering of multiple
Morphosyntactic Phrases in a Clause. At the Morphological Level, we furthermore find
Grammatical Morphemes. These are non-modifiable elements such as Auxiliaries and
Affixes. At this level we also find Secondary Operators, which can introduce irregular forms.
The Phonological Level has a set of primitives called Suppletive Forms: units of which the
form is morphologically unpredictable. Finally, Tertiary Operators can be found at this level,
which for instance introduce non-predictable intonation patterns.
Please recall that the Interpersonal and Representational Levels contain pragmatic and
semantic units (Lexemes, Frames and operators); in other words ‘units of meaning’. This is of
course a highly complicated concept. Meaning is in essence non-discrete (cf. Labov 1973)
and cannot be treated as discrete, countable units without consequences. FDG deals with this
problem by only defining a pragmatic or semantic primitive when this is relevant for the
description of a language. So an absolute tense operator and a relative tense operator are two
units of meaning in a language where they trigger different markers (Hengeveld and
Mackenzie 2008: 164).
The imperfection of this solution can be illustrated by means of
a word like head, which can denote a body part but also a person in charge. The word is a
polyseme: it has multiple related meanings. The different meanings might have different
distributions in some languages. The distinction is then linguistically relevant and there are
two units of meaning. However, it could also be argued that the meanings of head have a
metonymical relation to each other and that there is no clear boundary between them. Maybe
there is just one unit of meaning, with different readings in different contexts.
The problem of drawing boundaries between meanings falls outside the scope of this
paper (but cf. Aarts et al. (eds.) 2004 for an overview of work on fuzzy categorization) and
hence will not be dealt with any further. Suffice it for now to say that only linguistically
relevant meaning distinctions should be modelled in a linguistic theory, and that the theory of
FDG will do that for me in this paper. FDG at least provides some way of defining units of
meaning and thus makes it possible to define and study transparency.
2.3. List of non-transparent features
Transparency was defined as a one-to-one relation between units at the pragmatic and
semantic levels on the one hand and units at the formal levels on the other hand. Now that we
have seen which levels of organization and which units are distinguished in FDG, we can
specify which relations between levels are possible. Combined with the possible ways of
The transparency of creoles 7
violating transparency (redundancy, domain disintegration and form-based form), a list
emerges of possible non-transparent phenomena. The features in this list, printed in bold in
this section and summarized in Table 1 in Section 2.4, are grouped according to the interface
where they occur.
2.3.1. The interface between the Interpersonal and Morphosyntactic Level
At this interface, pragmatic meaning relates to morphosyntactic units. An example is the
relation between a Subact of Reference and a Noun Phrase.
No redundancy phenomena are attested at this interface, but domain disintegration can
appear in the form of discontinuity. If units belong together at the Interpersonal Level, it is
transparent when they are morphosyntactically contiguous. Discontinuity violates that
principle, as illustrated in example (5) from Dutch.
(5)
Heb je
die
man gezien
die
door
rood reed?
Have you
that
man seen
who through red
drove
‘Have you seen the man that drove through the red traffic light?’
At the Interpersonal Level, the underlined units ‘that man’ and ‘that drove through the red
light’ belong together as they form one Subact of Reference. However, the corresponding
morphosyntactic unit is splitted by the last word of the verb phrase, thus creating nontransparency.
A complication here is that to define discontinuity, one needs a theory of constituency
(Velasco, forthcoming). A constituent can only be discontinuous if it is a single unit in the
first place (e.g. the Noun Phrase in (5) is only discontinuous if you assume that a relative
clause together with its head is actually one unit). We therefore have to define what a
morphosyntactic unit is. FDG takes the position that units only belong together (i.e. form a
constituent together) if they stand next to each other (Velasco, forthcoming). Hence in (5),
‘that man’ and ‘who through red drove’ are two constituents. There is hence no discontinuity
at the Morphosyntactic Level as such (there are simply two units there), but there is a
discrepancy between the pragmatic unit and the corresponding morphosyntactic units. It is
that discrepancy that is non-transparent, and that I label discontinuity.
Note that it could also be argued that (5) is non-transparent at the interface between the
Representational and Morphosyntactic Level. The discrepancy would then exist between two
morphosyntactic units and one semantic unit (an Individual). However, in case of doubt I will
follow the top-down approach of FDG by assuming that the discrepancy lies on the highest
interface possible.
The morphosyntactic or phonological complexity of a constituent can be a trigger for
discontinuity. In many languages complex units are moved right, while simple units are
moved left. Such influence of formal complexity on morphosyntactic placement is a formbased formal operation and hence non-transparent in itself. Such complexity-based movement
8
Sterre Leufkens
is independent from discontinuity, as the two need not go together. It will be discussed at the
appropriate levels.
Discontinuity can also be introduced by raising. In example (6), we see a Referential
Subact at the Interpersonal Level, realised as an embedded clause at the Morphosyntactic
Level. The embedded clause has one argument, ‘the horses’. In (7), that argument is raised: it
is morphosyntactically an argument of the main clause (which is evident through its position
and the verbal agreement that it causes).
(6)
(7)
It seems that the horses are ill.
The horses seem ill.
The units at the Interpersonal Level and at the Morphosyntactic Level do not run parallel and
transparency is violated.
We have seen that discontinuity is non-transparent, as transparency predicts that
formal units always retain their integrity. Another way of violating domain integrity is fusion.
When the boundaries of two formal units disappear and the units integrate, a many-to-one
relation is the result. Fusional morphology is therefore non-transparent. Transparent
languages will have either isolating morphology, i.e. one-to-one relations between meanings
and words, or agglutinative morphology, i.e. one-to-one relations between meanings and
morphemes (Hengeveld 2004). Fusional morphology is located at this interface when a
pragmatic unit relates to fused formal units. It also appears at the interface between the
Representational and Morphosyntactic Level, when a semantic unit is involved.
There are two types of fusional morphology: cumulation and stem alternation
(Hengeveld 2007: 38). Cumulation refers to the expression of more than one grammatical
category in one morpheme. An example is the joint expression of person and number in
personal pronouns. The high amount of languages in which these categories are cumulated
could be a reason to believe that person and number are actually not separate categories at all.
However, since languages do exist that express these categories separately, we have to
2
assume that personal pronouns are in fact examples of cumulation.
Stem alternation appears when the form of a lexical stem is affected by the
expression of a grammatical category. A lexical stem is usually more robust than a
grammatical unit. Therefore the violation of a lexical unit is more ‘severe’ than the violation
of a grammatical unit. Stem alternation is hence more non-transparent than cumulation.
2.3.2. The interface between the Interpersonal and Phonological Level
At this interface, all relations between pragmatic meaning and phonological form are located.
A possible relation holds for instance between an Act (e.g. a question) and an Intonational
Phrase (e.g. a rising pitch contour).
In a transparent language, the groupings at the Interpersonal Level run parallel to the
groupings at the formal Levels. However, in many languages we find non-parallel alignment
The transparency of creoles 9
between the Interpersonal and the Phonological Level, which is a case of domain
disintegration. Consider example (8) from Dutch.
(8)
Ik wou
dat
hij
kwam
/kʋau
dɑti
kʋɑm/
I want.PST
COMP
he
come.PST
‘I wish he would come.’
(Hengeveld and Mackenzie 2008: 18)
At the Phonological Level, there are three units: /kʋau/ (‘I wish’), /dɑti/ (‘that he’) and /kʋɑm/
(‘would come’). These phonological units do not correspond to units at the Interpersonal
Level. The groupings of the Phonological Level are hence not transparently related to the
groupings of the Interpersonal Level.
2.3.3. The interface between the Representational and Morphosyntactic Level
At this interface we find relations between semantic units (for instance States-of-Affairs or
Individuals) and morphosyntactic units (for instance Clauses or Phrases).
A redundancy phenomenon at this interface is apposition, illustrated in example (9).
(9)
Manfred, my friend, is kind.
One Individual is realized twice in the Morphosyntactic Phrases ‘Manfred’ and ‘my friend’
(Hengeveld and Mackenzie 2008: 350). The same is true in example (10): there is a one-totwo relation between semantic and morphosyntactic units.
(10)
wa
sa
a-r
Ø-q’ə-s-žə-w-ʔa-aƔ-ś
you
me
3-ABS
3.ABS-AFF-1-again-2-say-PST-DECL
‘You said it to me.’
(Hengeveld and Mackenzie 2008: 350)
If one of the morphosyntactic units is not lexically realized, like the Kabardian affixes in
example (10), we speak of cross-reference, which is a type of apposition (Hengeveld and
Mackenzie 2008: 350). Cross-reference only applies when the lexically realized element (the
free pronouns in the Kabardian example) is optional. This is in FDG distinguished from
argument agreement, where both the lexically realized element and the predicate marker are
obligatory. Argument agreement is non-transparent as well, but at a different interface (cf.
Section 2.3.5).
Another instance of redundancy exists at this interface in the form of multiple
expression of semantic information, for instance of semantic class. In Bantu languages, the
semantic class of nouns is expressed on the nouns themselves as well as on adjectives,
10 Sterre Leufkens
demonstratives and verbs (e.g. Corbett 1991). This is non-transparent, as one semantic unit
has multiple morphosyntactic counterparts. Another example is multiple expression of
plurality, for instance in the English phrase three bananas where the numeral and the suffix
express overlapping information.
Yet another case of redundant expression of semantic information is negative concord.
One negation operator at the Representational Level correlates to multiple negation markers at
the Morphosyntactic Level, as in example (11) from Afrikaans.
(11)
Ek het
hom nie
gesien
I have him NEG seen
‘I haven’t seen him.’
(Zeijlstra 2004: 63)
nie.
NEG
Note that multiple expression of semantic information is not agreement. Agreement is the
copying of morphosyntactic material to other morphosyntactic units; a purely formal
3
operation where semantic information is completely irrelevant.
This distinction between redundant semantic expression and formal agreement is the
same as Booij (1993) makes in the domain of inflection. Booij (1993: 2-6) explains that some
inflectionally expressed categories (e.g. number in English) have a semantic correlate, so they
relate to the meaning of an inflected unit. He calls this inherent inflection. Other inflectional
categories (e.g. grammatical gender in Dutch) are purely morphosyntactically driven; the
inflectional forms do not carry any meaning but are just there because they are copied from or
triggered by some other morphosyntactic unit. This is called contextual inflection. We will
see later on that this distinction between meaning-based and form-based form is highly
relevant in the study of transparency in creoles.
Besides redundancy we also find domain disintegration at the interface between the
Representational Level and the Morphosyntactic Level. A violation of the integrity of formal
units and hence of non-transparency is discontinuity, already discussed in relation to the
interface between the Interpersonal and the Morphological Level. It is possible that not a
pragmatic unit, but a semantic unit correlates to two or more morphosyntactic units. An
example is the realization of the semantic negation operator in French by the circumfix ne
pas: one semantic unit corresponds to two morphosyntactic units. Note that if the circumfix
would be triggered by a pragmatic unit (e.g. in case of a topic marker), the discontinuity
applies at the higher interface between the Interpersonal Level and the Morphosyntactic
Level. The same is true for infixes: these can cause discontinuity at two interfaces.
As discussed above, fusional morphology is non-transparent since it results in manyto-one relations between meaning and form. Two types of fusional morphology are
cumulation and stem alternation, of which the latter is a more severe violation of
transparency. Cumulation at this interface occurs for instance in Spanish verbal suffixes,
where tense, mood and aspect operators are expressed in one morpheme. An example of stem
The transparency of creoles 11
alternation at this interface is found in English past tense verb forms like saw (see + past
tense).
2.3.4. The interface between the Representational and Phonological Level
A transparent relation at this interface involves a one-to-one relation between a semantic unit
(e.g. a State-of-Affairs) and a phonemic form (e.g. an Intonational Phrase).
If groupings at the Representational Level do not relate transparently to groupings at
the Morphosyntactic Level, we speak of non-parallel alignment, which was already
discussed above at the interface between the Interpersonal and the Morphosyntactic Level. An
example of non-parallel alignment at the Representational/Phonological interface is the
occurrence of a clitic expressing tense. The clitic forms one phonological unit with its host
(e.g. the predicate) at the Phonological Level. But at the Representational Level, the tense
operator scopes over more than this host (i.e. over the whole State-of-Affairs instead of only
the property expressed by the predicate). This is an example of domain disintegration, as the
integrity of two formal units is violated, resulting in a non-transparent relation.
2.3.5. Non-transparency at the Morphosyntactic Level
There are several elements and operations that exist at the Morphosyntactic Level, but do not
have any higher trigger or counterpart. A first example is expletive elements (or dummies):
units at the Morphosyntactic Level that do not relate to a unit at a higher level. Expletives are
often pronouns (e.g. it in it rains) or location adverbs (e.g. there in there is something wrong).
Travis (1984) establishes the following implicational hierarchy for expletives:
Argumentless passives and unaccusatives > expletives of displaced NPs >
4
expletives of displaced CPs > weather predicates
We can infer that if a language has expletives at all, they will be found with weather
predicates.
Another form-based form phenomenon is agreement. Hengeveld and Mackenzie
(2008: 350) define agreement as ‘a mechanism by which information properly pertaining to a
single element of the Clause is copied to one or more elements’. Agreement is hence a
completely formal procedure. There is no meaningful correlate or trigger, which distinguishes
agreement from redundancy phenomena listed above. Consider for instance subject-verb
agreement (argument agreement) in example (12) from French.
(12)
Nous chant-ons
3.PL sing-3.PL
‘We sing.’
12 Sterre Leufkens
The free subject pronoun and the person marker on the verb are both obligatory. The verb
marker is not a Referential Subact, as it is not able to refer on its own. The person and number
information in the marker is hence not coming from the Interpersonal Level, but copied from
the free pronoun. As this copying takes place at the Morphosyntactic Level, this is an instance
of purely formal (or syntactic) agreement, which is a non-transparent feature.
A related form-based form phenomenon is grammatical gender, the common name
for a semantically empty noun classification system (Corbett 1991). Grammatical gender only
becomes apparent through agreement, as the formal class of a noun is only relevant when
other forms adapt their form to it. Grammatical gender should not be confused with biological
gender, a meaning-based classification system that groups nouns according to for instance
5
animacy or sex distinctions.
Another form-based form phenomenon is tense copying, also known as sequence of
tenses or consecutio temporum. A tense operator from the main clause is copied at the
Morphosyntactic Level to the operator slot of the embedded clause (Hengeveld and
Mackenzie 2008: 367). The tense of the embedded predicate is hence not semantically
motivated. Example (13) comes from Amele, a language without a tense copying rule. The
verb in the main clause has a past tense operator; the embedded verb is in the future tense. In
English, a language with a tense copying rule, the main verb tense operator is copied to the
embedded verb. This results in the conditional tense of the embedded verb in the translation.
(13)
Naus uqa ege qila bele-q-an fo=ec
sisal-t-en
Naus he
I
today go-1.PL-FUT Q=NMLZ ask-1.SG/3.SG-REM.PST
‘Naus asked me whether we would go today.’ (litt.: whether we will go today)
(Hengeveld and Mackenzie 2008: 368)
Transparency predicts that organization at the lower two levels is based on and parallel to
organization at the higher two levels. Alignment should hence be based on pragmatic or
semantic functions to be transparent. This is the case in languages with interpersonal or
representational alignment (Hengeveld and Mackenzie 2008: 324). Interpersonal alignment
occurs when the pragmatic role of arguments (Topic or Focus) is relevant for their expression.
In languages with representational alignment, the semantic role of arguments (e.g. Actor,
Undergoer or Recipient) determines their expression. However, there are languages where
pragmatic and semantic roles are neutralized. This is called morphosyntactic alignment. In
those languages, there are syntactically relevant functions like Subject and Object, known as
grammatical relations. Morphosyntactic alignment is opaque, as syntactic functions are
form-based forms: their existence is not motivated by a meaningful unit or operation.
Consider for example the difference between Kharia (representational alignment) and
Dutch (morphosyntactic alignment). The intransitive clause in example (14) takes an Actor
argument, the intransitive clause in (15) takes an Undergoer. In Kharia, the semantic role of
the argument is relevant for the form of the predicate: if it is an Actor argument the predicate
The transparency of creoles 13
is in the active voice, if it is an Undergoer argument the predicate is in the middle voice
(Leufkens, forthcoming).
(14)
ho=kaɽ
biʔthuŋ=oʔ
that=SG.HUM
spit=A.PST
‘S/he spit.’
(Peterson 2006: 203)
(15)
ho=kaɽ
urumɖaʔ=ki
that=SG.HUM
sweat=M.PST
‘S/he sweated.’
(Peterson 2006: 204)
In the equivalent Dutch examples (16) and (17), the semantic role of the argument is
neutralized (the different form of the past tense suffixes is the result of phonological factors)
and we speak of a syntactic function Subject.
(16)
Zij
spuug-de
She
spit-PST
‘She spit.’
(17)
Zij
zweet-te
She
sweat-PST
‘She sweated.’
The now following non-transparent features are all instances of morphosyntactic influence on
formal operations. Firstly, consider function marking. Pragmatic function can be marked in
different ways, for instance by means of particles or word order. In some languages,
morphosyntactic information influences pragmatic function marking. The way that
morphosyntactic units are marked for a specific function is then different for pronouns and
complement clauses, or for nouns and verbs. This is non-transparent as a formal feature
triggers a formal adaptation.
A good example of a language that is transparent with respect to this feature is Kharia
(Peterson 2006). We see that the contrastive marker =ko applies to the pronoun in example
(18) and to the complement clause in example (19) in the exact same way. The
morphosyntactic nature of the clauses is irrelevant for pragmatic function marking.
(18)
u=je[ʔ]=ko
this=SG.NHUM=CNTR
‘But this is theirs.’
(Peterson 2006: 41)
ho=ki=yaʔ=ga
that=PL=GEN=FOC
heke
PRS.COP
14 Sterre Leufkens
(19)
oreʔj koŋtaŋ bui=na=ko
ho=ki=yaʔ
ox
cow
raise=INF=CNTR
that=PL=GEN
‘But raising oxen and cows was their custom.’
(Peterson 2006: 298)
dhatam
custom
aw=ki
COP=M.PST
Similar to the way that morphosyntactic information might determine pragmatic function
marking, in some languages morphosyntactic information influences semantic function
marking. The way that morphosyntactic units are marked for a specific function is then for
instance different for pronouns and complement clauses, or for nouns and verbs. English is
non-transparent in this respect, as pronouns are case-marked (he vs. him), but more complex
NP’s are not.
Similarly, in some languages we find influence of morphosyntactic information on
derivational and/or inflectional processes. To prevent any ambiguities around these often
differently interpreted terms: “[d]erivation is the morphological process that results in the
formation of new lexemes.” (Bauer 1983: 26-27). Inflection is then the morphological
process that does not result in the formation of new lexemes, but in new word-forms (Bauer
1983: 22).
Transparency predicts that derivation and inflection take the same form whatever the
Lexeme they apply to. A derivational or inflectional strategy should be applicable to units of
6
all categories and of all degrees of complexity. Examples of derivational strategies in
languages are (tonal) affixation, cliticization, reduplication, compounding and conversion (or
zero-derivation). Inflectional strategies are for instance (tonal) affixation, cliticization and
7
suppletion. Again consider Kharia, which predominantly makes use of the processes of
cliticization, conversion, reduplication and compounding (Peterson 2006). The
morphosyntactic nature of units is irrelevant: all these processes are applied to words of
different categories and different degrees of complexity.
Morphosyntactic category plays a role in another non-transparent feature. There are
languages, called flexible languages (Hengeveld 1992), where Lexemes are semantic units
without a specification of morphosyntactic category. However, in so called rigid languages,
all Lexemes have a specific morphosyntactic category. If in a rigid language a noun is used in
a predicate function, morphological adaptations are triggered, for instance the introduction of
a copula. Such adaptations are triggered by morphosyntactic information (i.e. word category),
which shows that this is a form-based formal operation. In a transparent language, we expect
that there is no influence of morphosyntactic category on the availability of units as
predicates or arguments.
Kharia is transparent with respect to this feature, as it is a flexible language (Peterson
8
2005). In Kharia, all semantic units can be used as predicates, whether denoting an event as
in (20), an individual as in (21), or a property as in (22). Furthermore, all semantic units are
available as arguments. Morphological category is hence irrelevant and there is no reason to
assume that Lexemes have a specified morphosyntactic category.
The transparency of creoles 15
(20)
lebu
ɖel=ki
man
come=M.PST
‘The man came.’
(Peterson 2006: 60)
(21)
bhagwan
lebu=ki
ro
ɖel=ki
God
man=M.PST and come=M.PST
‘God became man [=Jesus] and came [to earth].’
(Peterson 2006: 60)
(22)
betoʔɖ=siʔɖ=iɲ
hunger=PFV=1.SG
‘I am (=have become) hungry.’
(Peterson 2006: 161)
Flexible languages like Kharia are most transparent in morphosyntactic expression of
semantic units; rigid languages are least transparent. Most languages are not as flexible as
Kharia, but also not as rigid as English for instance is. Such languages have for instance a
clear noun-verb distinction, but no other morphosyntactic categories. Or there is conversion
from nouns to verbs (all nouns can be used as verbs without formal changes), but not the other
way around (verbs cannot be used as nouns). In general we can say that the more
morphosyntactically triggered adaptations a language requires, the more non-transparent the
language is.
Yet another non-transparent feature that has to do with morphosyntactic influence is
the way in which bound morphemes mark phrases in a language. At least two types of bound
morphemes are distinguished in the literature: clitics and affixes. I adopt the ideas of Zwicky
and Pullum (1983: 503), who characterize clitics as markers that are not selective as to their
hosts; they attach to any kind of morphosyntactic unit. Affixes are selective: they attach to
words of a specific morphosyntactic category and not to complex phrases. Now, if some
function in a language is marked by means of a bound morpheme, it is transparent when this
is always done by the same bound morpheme on whatever host it applies to. Hence, in a
transparent situation, phrases are marked by clitics. It is non-transparent when the head of the
phrase is marked by an affix, instead of by a bound morpheme that applies to the whole
phrase.
The last form-based form feature at the Morphosyntactic Level is influence of
morphosyntactic complexity on word order. In many languages, formally complex units
are moved to or located at the right of the sentence, while simple units go left. A formal
feature then determines another formal feature, which is non-transparent.
It can be difficult or even impossible to establish whether morphosyntactic complexity
or phonological complexity was the trigger for morphosyntactic movement. That the two are
not the same is proven by languages like Hawaiian, where words can be morphologically
16 Sterre Leufkens
complex but phonologically simple. In case of doubt, I will again follow the top-down
approach of FDG and assume that the highest level is responsible; hence that morphosyntactic
complexity precedes phonological complexity.
2.3.6. Non-transparency at the Phonological Level
In many languages, phonemes assimilate their form to the form of adjacent phonemes. This is
non-transparent, as it involves formal adaptation triggered by formal units. Phonological
assimilation can take various forms, for instance sandhi rules, nasalization, degemination,
deletion or insertion of phonemes, devoicing, diphthongization, vowel harmony and tone
perturbation.
Phonological assimilation can be sensitive to morphosyntactic boundaries. Consider
for instance nasalization in Kharia, which only takes place within morphemes and never
across morpheme boundaries. Of course, there is also phonological assimilation that is
insensitive to morpheme boundaries. This kind of assimilation results in stem alternation or
cumulation at the Morphosyntactic Level. Such phonological adaptations will therefore be
discussed under domain disintegration at the relevant interfaces.
One further form-based form is found here: influence of phonological weight on
word order. As discussed before, heavy phrases tend to be placed or moved right in
languages, light constituents are placed or moved left. Such movement will only be analysed
at this level when influence of morphosyntactic weight is excluded as a possible factor.
2.4. Transparency analysis
The non-transparent features established above are summarized in Table 1, which can be
employed to measure the degree of transparency of a specific language.
Table 1: List of non-transparent features
Interface
Interpersonal
Level /
Morphosyntactic
Level
Interpersonal
Level /
Phonological
Level
Redundancy
Domain
disintegration
- Discontinuity
(e.g. raising)
- Stem
alternation
- Cumulation
- Non-parallel
alignment
Form-based form
The transparency of creoles 17
Representational
Level /
Morphosyntactic
Level
Representational
Level /
Phonological
Level
Morphosyntactic
Level
Phonological
Level
- Apposition
- Multiple
expression of
semantic
information
(e.g. class,
plurality,
negation)
- Discontinuity
(e.g. French ne…
pas)
- Stem
alternation
- Cumulation
- Non-parallel
alignment
- Agreement
- Grammatical gender
- Tense copying
- Expletive elements
- Grammatical relations
- Morphosyntactic information
influences pragmatic / semantic
function marking
- Morphosyntactic information
influences derivational and/or
inflectional processes
- Morphosyntactic category
influences availability as
predicate / argument
- Head-marking through affixes
- Morphosyntactic complexity
influences word order
- Phonological assimilation
- Phonological weight
influences word order
The more frequent and pervasive a non-transparent property is in a language, the more nontransparent the language is as a whole. It would therefore be worthwhile to not only establish
whether a language has some property, but also to what extent this property occurs. The
degree of transparency of a language would then not just be a sum of properties, but a more
fine-grained measure. However, such an analysis would take a considerable amount of space,
since areas of grammar would have to be described completely. Therefore I will limit myself
to a discrete transparency analysis in this paper. A language will be considered non-
18 Sterre Leufkens
transparent with respect to some feature, when one or more instances of non-transparency are
attested in that area.
3. Creoles: the semantic transparency hypothesis
In their article on semantic transparency, Seuren and Wekker (1986) are among the firsts to
state that creole languages are more transparent than non-creole languages. They argue that
creoles are a typologically distinct group of languages, with transparency as its defining
characteristic. This characteristic is allegedly a direct consequence of the way in which
creoles emerged: through intensive language contact. This view on creoles has become to be
known as the semantic transparency hypothesis (see e.g. Muysken 1988).
An opponent of this hypothesis is McWhorter, who argues that although creoles can
be seen as a typologically distinct group of languages, their defining characteristic is not
9
transparency but simplicity (McWhorter 1998, 2001). According to McWhorter, one aspect
of simplicity is a low number of overt distinctions in a particular area of grammar
(McWhorter 2001: 135-137). McWhorter states that creoles often have a low number of overt
inflectional morphological distinctions, rendering them simpler (at least in the area of
inflectional morphology) than older languages.
The idea that creoles lack or have a simple morphology has gotten tremendous
critique, which I cannot discuss here for reasons of space (cf. Arends et al. 2005 for an
overview). One line of critique that should however be noted here is expressed by Aboh and
Smith (2009). The authors warn that simplicity in McWhorter’s sense only says something
about surface structures, not about the overall complexity of languages, which of course have
more structure than just surface structure. The authors furthermore stress the relevance of
studying interfaces between levels, because a low number of (overt or covert) distinctions in
some area of grammar could very well go hand-in-hand with a high number of (overt or
covert) distinctions in other areas.
McWhorter furthermore argues that creoles are not only simpler in their inflectional
morphology, but also in derivational morphology. Derivation in creoles is claimed by
McWhorter (1998: 796-799) to be more semantically regular than in non-creoles. This idea
has gotten serious critique as well, for instance by Braun & Plag (2003), who deny that
derivation in creoles is semantically regular. They point out that affixation, compounding,
reduplication and transposition are well-attested procedures in creoles. These processes all
(frequently, not marginally) produce semantically irregular, non-interpretable words. An
example is the Sranan word wasiwasi ‘wasp’, which is formed by reduplication of wasi
‘wash’ (Braun and Plag 2003: 96). The relation between the meanings is non-interpretable as
we cannot predict what wasiwasi means on the basis of wasi.
Semantic regularity and (semantic) transparency are treated as one and the same here.
The line of reasoning is this: if all formal units consistently represent semantic units, then we
should be able to infer exact meanings from formal units, as is the case in a word like airmail
(Bauer 1983: 19). If the form does not predict the meaning, the relation between them must be
The transparency of creoles 19
non-transparent. A word like blackmail (Bauer 1983: 19), or the earlier example wasiwasi, is
not interpretable from its morphemes and hence non-transparent.
But in my opinion, non-interpretability (semantic irregularity) is not necessarily non10
transparent. Someone who uses the word wasiwasi does not need to separate it in two units.
A speaker uses it as one formal unit related to one meaning. Of course, a speaker can notice
that it is a morphologically complex form, formed by reduplication of another word and that
the meaning of the new word cannot be predicted. But this is not necessary to understand and
use wasiwasi. Transparency is not violated at all. McWhorter’s claim that creoles are
semantically regular and Plag’s claim that they are not, are therefore both irrelevant to the
semantic transparency hypothesis.
The semantic transparency hypothesis does not only involve the idea that creoles are
more transparent than older languages. It also proposes a reason for this alleged characteristic,
namely the sudden emergence of creoles in a situation of extreme language contact. The
underlying idea is that language contact increases transparency, whereas opacity can only be
introduced during a longer period of language change. Creoles are young contact languages
and as such expected to be transparent.
This is a so called Universalist approach to creole genesis as it stresses the importance
of a universal tendency, rather than a language-specific one, for the formal properties of
creoles (Arends et al. (eds.) 1995: 11). Another famous Universalist approach, the bioprogram
theory (cf. Bickerton, 1988), states that creoles are similar because they are founded by
children with a universal innate language capacity (Arends et al. (eds.) 1995: 11). The
semantic transparency hypothesis, however, is a functionalist approach that does not presume
an innate universal language. The alleged similarity of creoles is accounted for by the idea
that underlying semantic structures are universal, and that in creoles these semantic structures
are represented directly in surface structure (Arends et al. (eds.) 1995: 11).
The Universalist component of the transparency hypothesis has been criticized as well.
It is argued in Superstratist and Substratist approaches (cf. Arends et al. (eds.) 1995: 87-109)
that creole languages inherited their properties from their source languages. As many wellstudied creoles have the same or similar source languages (English, Dutch, Portuguese and
French as superstrates and African languages as substrates), it is not surprising that creoles
show similarities. Bakker (2003) for instance argues that the common European superstrate
languages all happen to have a relatively poor inflectional morphology. This would explain
the alleged ‘simple’ morphology of creoles. The similarities among creoles, and their
morphological characteristics, are in this view not a result of the circumstances in which the
creoles emerged, but of inheritance from shared ancestors. The current study aims to falsify
the idea of shared inheritance.
4. Method and sample
In this study, a transparency analysis is executed on four creole languages. By checking the
languages on all non-transparent features in Table 1, their degree of transparency is measured.
20 Sterre Leufkens
It is expected that the creoles are transparent to such a high degree, that it distinguishes them
as a group from non-creole languages.
To exclude the possibility that transparency is inherited from shared source languages,
the sample includes creoles with typologically and genetically dissimilar source languages. If
the studied languages indeed turn out to be similar and transparent to a high degree, this
cannot be the result of similar and transparent super- or substrates, but has to be seen as a
creole characteristic.
The first creole language in the sample is Diu Indo-Portuguese. It is spoken in India
and its ancestors are Portuguese (Romance) and Gujarati (Indo-Aryan). The second language
is Nubi, spoken in Uganda, based on Egyptian or Sudanese Arabic (Semitic) and several
languages from the Kordofan and Nilo-Saharic families. Pichi, the third language in the
sample, is based on Krio, itself a creole based on English (Germanic) and several WestAfrican languages. Finally, the sample includes Sri Lanka Malay. This language is spoken in
India and is based on Sinhalese (Indo-Aryan), Malay (Austronesian) and Tamil (Dravidian),
and influenced by Dutch and English.
The reference grammars I consulted are all fairly recent and elaborate. I have consulted
the authors of these grammars for some additional information.
5. Results
11
5.1. Diu Indo-Portuguese
Diu Indo-Portuguese (Cardoso 2009; henceforth: DIP) is a language spoken in India, on the
isle of Diu. The small community on the island spoke Gujarati when Portuguese colonialists
arrived in 1535. The Portuguese took control over the island after a long period of battle, also
involving Turkish and Egyptian forces (Cardoso 2009: 70). This resulted in a contact
language that started to stabilize when the battles were over in 1554. African-born slaves and
the offspring of Eurasian couples mixed in this process as well, adding even more linguistic
varieties to Diu Indo-Portuguese.
5.1.1. Redundancy in DIP
As becomes clear in example (23), DIP has no cross-reference.
(23)
ɔm
larg-o
ped
man release-PST fart
‘The man let out a fart.’
(Cardoso 2009: 198)
However, apposition of two lexically realized elements is not prohibited, see example (24).
The transparency of creoles 21
(24)
ɔn
foy
dǝpǝy
nɔs, nɔs
Where
go.PST
then
1.PL 1.PL
‘Where did we go then, the two of us?’
(Cardoso 2009: 128)
doy?
two
DIP exhibits negative concord, as in example (25), but no other instances of multiple
expression of semantic information.
(25)
Nĩge
nã
apĩŋ-o
nobody
not
catch-PST
‘Nobody caught fish.’
(Cardoso 2009: 211)
pex
fish
5.1.2. Domain disintegration in DIP
There is discontinuity in Diu Indo-Portuguese, as in example (26).
(26)
es
tud
ɔn
foy
raprig ?
DEM PL
where
go.PST
girl
‘Where did the girls go?’ Litt: ‘Those where went girl?’
(Cardoso 2009: 177)
The Referential Subact corresponds to two non-contiguous morphosyntactic units (‘those’ and
‘girls’). Discontinuity does not occur in the form of raising, circumfixes or infixes. I found no
examples of discontinuity at the interface between the Representational and Morphosyntactic
Level.
DIP is a predominantly isolating language (Cardoso 2009: 107), but fusional
morphology occurs in the form of cumulation of person and number in personal pronouns.
The past form tiŋ of the auxiliary tǝ (‘IPFV’) can also been seen as an instance of cumulation.
Stem alternation is found in the verbal domain. Whereas most verbs have a regular
inflectional paradigm, some verbs inflect irregularly. The paradigm of the verb vay ‘to go’ for
instance is entirely suppletive (e.g. foy ‘go.PST’). Note that these suppletive forms in DIP are
all borrowed from Portuguese, a language with pervasive suppletion (Cardoso 2009: 288).
A Phonological Word in DIP has one accent on the last syllable (Cardoso 2009: 100).
Clauses have one intonational peak, usually assigned to a focal constituent (Cardoso 2009:
102). Phonological units and meaning units thus run parallel – I found no counter examples.
5.1.3. Form-based form in DIP
22 Sterre Leufkens
There is no agreement in DIP, nor any purely formal classification system such as
grammatical gender. Note that both Gujarati and Portuguese do have agreement and
grammatical gender (Cardoso 2009: 285-287).
DIP has no tense copying rule, which is demonstrated in example (27). The main
clause predicate is in past tense, but this is not copied to the subordinated predicate.
(27)
Yo
sab-iŋ
ki
el
tə
fɑl-a
1.SG know-PST
COMP 3.SG IPFV.NPST say-INF
‘I knew that he would say (litt.: is saying) “Liza” only.
(Cardoso 2009: 234)
Liza
mem
L.
EMPH
Cardoso (2009: 217) sums up various cases of constituent doubling in DIP. Most of these
doublings have semantic or pragmatic triggers (e.g. an emphasis operator) and do not violate
transparency. For instance ‘to me he gave it, to me’ means ‘he really gave it to me’ (Cardoso
2009: 219, translation mine). Such sentences are comparable to reduplication, the only
difference being that an entire constituent is reduplicated instead of a single morpheme or
word. However, sometimes a predicate is doubled for a morphosyntactic reason, making it a
form-based form phenomenon. Predicate doubling occurs when P, U or R arguments
(normally realised in post-verbal position) are topicalized. Topics occur in preverbal position,
but in a sentence with an A-argument, this position is already taken. The clash is resolved by
repeating the verb, resulting in an A-V1-P/U/R-V1 order. The reason for this doubled verb is a
morphosyntactic one, namely the clash of two templates. The verb copy is an expletive
element, as it is an element that is there because of some formal rule but does not carry any
meaning.
DIP exhibits representational (semantic) alignment. Cardoso (2009: 192) demonstrates
that arguments usually remain bare, but that Recipients are marked with dative case. Parguments are optionally dative-marked as well. There are hence semantically relevant
functions and no grammatical relations; DIP is transparent with respect to this feature.
Pragmatic function is in DIP predominantly marked by means of word order:
particularly prominent elements are expressed in preverbal position (Cardoso 2009: 212).
Contrastive function or emphasis is sometimes attained by means of constituent doubling
(Cardoso 2009: 218). Furthermore, there is the emphatic function marker mem, which applies
to all morphosyntactic units (Cardoso 2009: 220). As all these procedures apply to units of all
morphosyntactic categories and degrees of complexity, DIP is transparent in this respect.
The morphosyntactic nature of units sometimes influences marking of semantic
function in DIP. Dative case is marked by means of -a on pronominal arguments, but with -pe
on all other units (Cardoso 2009: 181).
There are a few derivational affixes in DIP (Cardoso 2009: 254), mostly inherited
from Portuguese, for instance the diminutive suffix –iŋ that applies to nouns. However, these
suffixes are unproductive in DIP. It is therefore questionable whether they really constitute
The transparency of creoles 23
instances of derivation. Compounding (Cardoso 2009: 257) is a productive strategy in DIP. It
is non-transparent, as it is only possible to compound nouns and adjectives or two nouns. The
morphosyntactic category of words is hence relevant for the derivational process of
compounding.
Person, number and aspect are in DIP usually expressed lexically, so not by means of
inflection. Tense is expressed by means of inflectional affixes on predicates. There are a few
verbs (the ones that were discussed under stem alternation) that do not take these affixes but
have their own suppletive paradigm, in which person and number are expressed as well. The
choice for affixation or suppletion is made per word – you have to memorize which verbs
have suppletive forms. This is non-transparent, as the use of affixation or suppletion cannot be
predicted by means of pragmatic or semantic information.
Semantic units are in DIP not always available as predicates or arguments. Only
Lexemes denoting Events can be used as predicates (Cardoso 2009: 110). All other semantic
units need morphological adaptations. There is hence evidence at least for a morphological
class of verbs, clearly distinct from nouns, which is non-transparent. The distinction between
adverbs and adjectives is less strong (Cardoso 2009: 113), so that DIP is not a completely
rigid language.
DIP is a predominantly isolating language (Cardoso 2009: 107). However, there are
some affixes in DIP, such as the diminutive suffix –iŋ mentioned above (Cardoso 2009: 254).
These affixes are not productive in DIP, which means, in my opinion, that they do not
constitute cases of non-transparency. The suffixes do not mark the head of a phrase: they are
part of Lexemes and cannot be applied to new Lexemes. There is no influence of
morphosyntactic category or complexity on bound morphemes – DIP is transparent with
respect to this feature.
Morphosyntactic placement in DIP is influenced by the morphosyntactic or
phonological weight of constituents (Hugo Cardoso, personal communication). Possession for
instance is expressed by means of the construction dǝ + possessor. If the possessor is
expressed by means of a pronoun, it is placed before the noun. If the possessor is a noun, it is
usually expressed post-nominally. This already indicates that light elements are placed more
to the left, while heavy units are realised more to the right. More evidence for this is the fact
that nominal possessors can sometimes precede the possessed noun, but only if they are very
light, as in example (28).
(28)
dǝ
tete
kaz
of
T.
house
‘Tete’s house.’
(Hugo Cardoso, personal communication)
The weight of the (possessor) phrase is hence of influence on morphosyntactic placement.
On the Phonological Level, there is some assimilation of phonemes to surrounding
phonemes: vowels that are adjacent to a velar nasal consonant are nasalized, though this is not
24 Sterre Leufkens
obligatory (Cardoso 2009: 93). This so called nasality spread is highly restricted and
infrequent (Cardoso, personal communication). No examples were found of nasality spread
across morpheme boundaries, but it is a theoretical possibility. However, it is not plausible
that nasality spread is a process that leads to cumulation or stem alternation. I will therefore
treat it here as an example of phonological assimilation restricted to morphemes.
5.2. Nubi
Nubi is an Arabic-based creole spoken in Uganda. Wellens (2003, chapter 7) argues that it is
probably based on a Central Arabic trade pidgin based on Egyptian Arabic and Sudanese
Arabic, which must have existed before 1820. Later on there was intensive contact with WestSudanese Arabic.
5.2.1. Redundancy in Nubi
Nubi seems to exhibit cross-reference, as in example (29).
(29)
Anas
kun kweisin
people.PL
be-Ø good.PL
‘People are good.’
(Wellens 2003: 181)
However, it is not really the case that we are dealing with two expressions of one individual
here. The predicate (kweisin) is only marked for plural, not for person. This can hence be
interpreted as a case of double marking of plurality, and not of cross-reference. Nubi does
allow for apposition of two lexically realized elements.
We saw that in Nubi, plurality can be marked more than once. It is optionally
expressed on the noun, on the adjective and on the demonstrative, as in example (30).
(30)
Wele
wadin
Boy.PL other.PL.
‘these other boys’
(Wellens 2003: 98)
dolde
DEM.PROX.PL
Nubi also (marginally) allows negative concord (Wellens 2003: 183).
5.2.2. Domain disintegration in Nubi
Nubi does not exhibit any form of discontinuity: there are no circumfixes or infixes and I
found no instances of raising.
The transparency of creoles 25
Cumulation occurs in Nubi in personal pronouns. Stem alternation is also attested, as
suppletion and ablaut appear in the forming of plurals of nouns and adjectives. An example is
the adjective ‘big’: SG kebir, PL kubar (Wellens 2003: 76). These forms are inherited from
Arabic.
There are phonological adaptations in Nubi that are not sensitive to morpheme
boundaries, which can lead to cumulation and stem alternation. Vowels can be deleted in
between two consonants of the same place of articulation or in word-final position (Wellens
2003: 44) or when preceding another vowel. Stressed vowels can fuse with another vowel to
form a diphthong (Wellens 2003: 46). Furthermore, there is vowel harmony (Wellens 2003:
47). This means that vowels tend to assimilate their height in the neighbourhood of other
vowels as for instance in silu ‘take’ that may become sulu. The verbal prefixes gi- and bioptionally assimilate their height too (bu-sulu), which clearly shows that vowel assimilation
can cross morpheme boundaries. Consonants can acquire or lose voice when preceding voiced
or voiceless consonants or pauses (Wellens 2003: 49). Nasal consonants adapt their
articulation place to following consonants (Wellens 2003: 49). Consonants are sometimes
deleted when they stand in between two vowels (Wellens 2003: 50). While geminates are
usually retained word-internally, there can be degemination at word boundaries (Wellens
2003: 51). All these assimilation processes weaken or ignore the boundaries between formal
units and hence give rise to fusion. Both grammatical and lexical stems can undergo these
adaptations, so they can cause cumulation and stem alternation in Nubi.
A Phonological Word in Nubi gets one accent on one of the last three syllables
(Wellens 2003: 42). I found no examples of Phonological Words that could not be related
transparently to higher level units. I found no information on larger phonological units (such
as phrasal or sentential intonation) so I do not know whether phrasing at the Phonological
Level runs parallel with phrasing at the Interpersonal and Representational Levels.
5.2.3. Form-based form in Nubi
Nubi does not exhibit agreement or grammatical gender (Wellens 2003: 61). Nubi has no
tense copying rule (Wellens 2003: 191). I also found no expletive elements in Nubi.
It is difficult, if not impossible, to establish the alignment type of Nubi, as arguments
are most often zero-marked. Only arguments with the semantic role of Recipient are always
introduced by the particle na (Wellens 2003: 180), as in example (31).
(31)
umon
kutu sum
na
3.PL
put
poison
to
‘They put poison for Mohammed.’
(Wellens 2003: 180)
Mohamadi
M.
Semantic role is relevant for the expression of Recipients, so in that domain there is
representational alignment in Nubi. However, the consulted grammars give insufficient
26 Sterre Leufkens
information on Actor and Undergoer arguments. I have not been able to find out whether
these roles are neutralized and we should speak of Morphosyntactic or another type of
alignment in that domain. I therefore leave open whether Nubi is transparent with respect to
this property.
Nubi has several focus marking particles (e.g. ya), which may apply to any sentence
constituent (Wellens 2003: 172). Topicalization is achieved by means of fronting, leftdislocation, cleft constructions, or some other change in word order (Wellens 2003: 184-189).
Morphosyntactic information is irrelevant to the application of these operations, which is
transparent.
Only the semantic function of Recipient is marked explicitly in Nubi, by the particle
na. As this particle can occur with any type of morphosyntactic unit, semantic function
marking is transparent in Nubi.
Stress shift is often used for derivational purposes, for instance to nominalize verbs
(Wellens 2003: 133). There is also a lot of conversion (zero-derivation) in Nubi, adjectives for
instance can be used as nouns and as adverbials without a change in form (Wellens 2003: 76).
The morphosyntactic category of Lexemes determines which derivation strategy has to be
used. This is non-transparent as morphosyntactic category influences derivation.
In Nubi, person is expressed pronominally, so not by means of inflection. Tense, mood
and aspect are also marked lexically. Plurality is marked inflectionally, by means of vowel
mutation (Wellens 2003: 56) or affixation. Which strategy is applied (vowel mutation or
affixation) differs per Lexeme. This is non-transparent, as the meaning of a noun or a verb
does not predict or correlate with the inflectional strategy that is followed. Nubi is hence
opaque when it comes to inflection.
Events, Properties and object-like units (e.g. nadede raha ‘this is peace’, Wellens
2003: 72) can all appear as predicates without formal adaptations. Argument function can be
taken by Properties (mo kweis ‘with goodness’, Wellens 2003: 76) and by Individuals. Events,
however, are always realized as verbs, which have to undergo formal adaptations (e.g. stress
shift) to be able to function as argument (Wellens 2003: 133). Nubi is hence not completely
transparent with respect to flexibility.
Nubi is a predominantly isolating language. There are, however, some bound
morphemes in Nubi, all of them affixes. There are nominal suffixes marking plurality and
verbal prefixes marking aspectual meaning. If have not found any examples of the marking of
phrases larger than only a noun or only a verb, so I cannot establish whether in Nubi only the
head of a phrase is marked, or the whole phrase.
A pronominally expressed Recipient is placed before other heavier arguments, while
this would be the reverse if the units were of equal weight (Wellens 2003: 180). That means
that ordering can be determined by morphosyntactic weight, which is non-transparent.
All phonological assimilations in Nubi described above under stem alternation and
cumulation, apply across morphemes but also word-internally. A voiced consonant can for
instance be devoiced when preceding a voiceless consonant (Wellens 2009: 49). There is
hence form-based form of this type in Nubi.
The transparency of creoles 27
5.3. Pichi
Pichi is spoken on the island of Bioko, off the coast of Equatorial Guinea. It is based on Krio,
itself a creole language with English as its lexifier, brought along from the mainland. Other
West-African languages (some of them pidgins) can be seen as substrate languages. From the
arrival in 1858 of Spanish colonialists, Pichi has been influenced by Spanish.
5.3.1. Redundancy in Pichi
Apposition of two lexically realized elements is allowed in Pichi, as in example (32), but
cross-reference does not occur.
(32)
Djunais
wèt
Bòyé dɛn
fɛt
dè
D.
with B.
3.PL IPFV fight
‘Djunais and Boye they are fighting.’
(Yakpo 2009: 407)
Pichi exhibits negative concord, as in example (33).
(33)
[…]
è
no
gɛt
no
3.SG not
get
not
‘He has no problem whatsoever.’
(Yakpo 2009: 215)
problema
problem
This is the only type of double expression of semantic information attested.
5.3.2. Domain disintegration in Pichi
There are no circumfixes or infixes in Pichi, but raising is allowed, see example (34).
(34)
Bòyé fiba
se
è
B.
seem QUOT
3.SG
‘Boye seems to have money.’
(Yakpo 2009: 421)
g ɛt
mɔní
get
money
Actually, this is not a prototypical example of raising, as we do not know whether Bòyé is
really an argument of the predicate ‘seem’ –there is no formal adaptation. This might be a
case of pure displacement, and it is unclear whether that is raising. However, if this is
displacement and not raising, the construction is still discontinuous, as Bòyé belongs to the
28 Sterre Leufkens
Referential Subact at the Interpersonal Level, but is not located there at the Morphosyntactic
Level.
Pichi is predominantly isolating. There is a clear instance of cumulation: personal
pronouns combine person, number and case in one form (Yakpo 2009: 179). Furthermore,
there can be reduction of grammatical morphemes, especially reduction of consonant clusters.
Word-final consonants are for instance deleted if the next word starts with a consonant with
which it could form a cluster. Consonants in word-final position are devoiced. Nasal
consonants sometimes assimilate their articulation place to the following consonant and
vowels sometimes assimilate their height to vowels in surrounding syllables (Yakpo 2009: 5658). Only grammatical morphemes are assimilated, while phonemes of content words are in
Pichi fully articulated (Yakpo 2009: 58). We are hence dealing with processes leading to
cumulation, located at the interface between the Interpersonal and Morphosyntactic Level,
and at the interface between the Representational and Morphosyntactic Level. I found no
examples of stem alternation in Pichi.
Alignment of the lower two levels and the upper two levels does not always run
parallel (Kofi Yakpo, personal communication). Pichi has an object clitic =àn that attaches to
a predicate, resulting in one phonological unit. But the predicate-object combination cannot
be seen as a single pragmatic or semantic unit. This is hence an example of non-parallel
alignment.
5.3.3. Form-based form in Pichi
There is no agreement in Pichi, nor is there grammatical gender (Yakpo 2009: 153). There is
also no tense copying rule (Yakpo 2009: 211).
Pichi has quite a list of verbs that take expletive elements as arguments (Yakpo 2009:
417), for instance weather predicates (fɔl ‘to rain’), copula elements (fiba ‘to seem’) and
evaluative verbs (gud ‘to be good’). These verbs occur with a syntactic argument that has no
semantic counterpart.
Pichi has no grammatical relations. In fact, nearly all arguments are zero-marked. Only
personal pronouns get function marking: first person A-arguments are for instance expressed
by means of à, U-, P- and R-arguments by means of mi. A P-argument can be expressed by
means of the clitic =àn, but other arguments cannot (Yakpo 2009: 395-397). These are
actually the only indications that we are dealing with representational alignment here. But
since there is no evidence that semantic roles are neutralized, we have no reason to believe
that there are grammatical relations in Pichi.
The pragmatic function of focus can be expressed in Pichi by means of suprasegmental
phonology. There are also several focus-assigning particles. Thirdly, cleft constructions exist
that assign focus (Yakpo 2009: 282-298). Topic function is usually expressed by means of
left-dislocation, optionally combined with the use of the particle naw (Yakpo 2009: 303-306).
All strategies apply to all morphosyntactic units. Morphosyntactic information is hence
entirely irrelevant for pragmatic function marking in Pichi, which is transparent.
The transparency of creoles 29
Semantic function is only expressed on personal pronouns. This is non-transparent, as
the complexity of morphosyntactic units determines morphosyntactic expression.
The derivational process of compounding in Pichi is transparent, as units of different
categories and different degrees of morphosyntactic complexity can be compounded (Yakpo
2009: 135). The derivational process of repetition that Yakpo (2009: 142) distinguishes is
transparent as well, because it applies to all morphosyntactic units. However, there are also
derivational processes that only apply to specific hosts and are therefore non-transparent.
Conversion for instance applies to some units but not to all (Yakpo 2009: 130); verbs can for
instance be conversed to nouns, but not the other way around. The process of reduplication
only applies to verbs (Yakpo 2009: 142). Morphosyntactic category influences derivation
here.
Grammatical information is in Pichi often expressed analytically, so not inflectionally
(Yakpo 2009: 124). There are some marginal other strategies, such as the use of bound
morphemes (affixes and clitics) and suppletion. Some of these strategies are non-transparent.
For instance the use of tonal affixation is restricted to pronouns (Yakpo 2009: 125). The
morphosyntactic complexity of units hence determines the inflectional strategy followed.
Pichi is relatively flexible, as many Lexemes in Pichi can be used as predicates and as
arguments without morphological adaptations. Events and Properties can be used as
predicates (see e.g. Yakpo 2009: 410). Yakpo’s dynamic verbs (Lexemes denoting Events)
can be used as arguments without a change in form. Inchoative-stative verbs (Lexemes
denoting Properties) can be used as arguments as well (Yakpo 2009: 130). So Events and
Properties can be predicates and arguments, but not everything is so flexible. Only few
Morphosyntactic Nouns can occur as predicates without change in form; most Nouns need a
copula-construction to be able to predicate (Yakpo 2009: 35, 130). There is hence some
evidence for morphological classes of nouns and verbs, which causes non-transparency in
Pichi.
As Pichi is an isolating language, there are only few bound morphemes. One is the
suffix –wan, which turns adjectives into adverbials. It is not fully productive anymore in
Pichi, as it only applies to some monosyllabic adjectives (Yakpo 2009: 127). Therefore, I
cannot say whether it marks phrasal heads only, or phrases as a whole. Pichi also has clitics,
for instance us= which is a question marker. This proclitic usually takes simple nouns as its
host, but in us=kayn moto ‘Q=kind car’, meaning ‘which car’, we see that it actually scopes
over a phrase and not only over the noun. This is evidence that in Pichi, there is no headmarking by affixes but instead phrase-marking by clitics.
Morphosyntactic or phonological weight influences morphosyntactic placement in
Pichi. In expressing possession, a possessive pronoun is expressed prenominally (in bɔyfrɛn
‘her boyfriend’, Takpo 2009: 181), while heavier possessors are expressed postnominally
(e.g. pikin fɔ mi anti ‘child of my aunt’, Yakpo 2009: 173). This is non-transparent as formal
information determines formal placement.
30 Sterre Leufkens
Phonological adaptation processes occur in Pichi within words and across words.
These processes are instances of form-based form, as the place or manner of articulation of
some phonemes trigger formal adaptations in other phonemes.
5.4. Sri Lanka Malay
Sri Lanka Malay is spoken at Sri Lanka. People at this island spoke Tamil and Sinhalese,
when the Dutch brought Malayan people to the island in the 17th century. The structure of the
lexifier Malay changed radically under the influence of Tamil and Sinhalese (Nordhoff, 2009:
805).
5.4.1. Redundancy in SLM
Apposition of two lexical elements is allowed in SLM, as illustrated in example (35).
(35)
Mr Sebastian aada, se
aada, kitham duuva
arà-oomong.
Mr S.
exist 1.SG exist 1.PL
two
NPST-speak
‘Mr. Sebastian is here, I am here, the two of us are talking.’
(Nordhoff, forthcoming)
SLM does not exhibit cross-reference.
Multiple expression of semantic information occurs in SLM, as plurality is marked
twice in phrases with a numeral, as in example (36).
(36)
Kandi=ka
hathu
Kandy=LOC INDEF
thiga-pulu
three-ty
riibu=kee
mlaayu
thousand=SIMIL Malay
pada
PL
arà-duuduk
NPST-exist.ANIM
‘There are 30,000 Malays in Kandy.’
(Nordhoff 2009: 243)
Negative concord is not allowed (Nordhoff 2009: 671-674).
5.4.2. Domain disintegration in SLM
12
There are no circumfixes , infixes, nor instances of raising in SLM. Discontinuity is hence
not attested in any form (Nordhoff, forthcoming).
There is some cumulation in SLM as person, number and case are cumulated in single
forms (Nordhoff 2009: 225). Moreover, there is assimilation of phonemes to adjacent
phonemes at the Phonological Level. This assimilation ignores morpheme boundaries, as
stem-final nasal consonants tend to assimilate their place of articulation to the first consonant
The transparency of creoles 31
of a bound morpheme (Nordhoff, forthcoming). Degemination between stem-final and affixinitial consonants is also allowed. Both processes occur with grammatical and lexical stems,
so that this can lead to cumulation as well as stem alternation. Both processes can be located
at the interfaces between the two higher levels and the Morphosyntactic Level.
Nordhoff (forthcoming) distinguishes at the Phonological Level between
Presuppositive Phrases (LH tone) and Assertive Phrases (L tone). Referential Subacts
correlate transparently to Presuppositive Phrases and predicates relate to Assertive Phrases
(Nordhoff forthcoming). The mapping between Phonological phrasing and Interpersonal and
Representational phrasing is hence transparent. I found no counter examples.
5.4.3. Form-based form in SLM
There is no agreement in SLM (Nordhoff 2009: 495) and no formal classification system such
as grammatical gender (Nordhoff, forthcoming). SLM has no tense copying rule (Nordhoff,
forthcoming). We can be sure that SLM has no expletive elements, as it does not use a
dummy in weather predicates, illustrated in example (37).
(37)
Arà-uujang
NPST-rain
‘It is raining.’
(Nordhoff 2009: 504)
SLM has representational alignment, as semantic roles are expressed by means of casemarking clitics (Nordhoff, forthcoming). There are no grammatical relations.
Pragmatic function in SLM is marked by means of two information structure clitics
(=jo and =jona, Nordhoff 2009: 378), which assign contrastive focus. Topical constituents
are expressed sentence-initially (Nordhoff 2009: 688). Both strategies apply to all kinds of
morphosyntactic units and are hence transparent.
Semantic function marking is transparent as well in SLM, as the morphosyntactic
nature of units is never relevant for the use of the case marking enclitics of SLM (Nordhoff,
forthcoming).
Morphosyntactic information is sometimes relevant for derivational processes, as there
are two affixes (nominalizer –àn and causativizer –king) that cannot attach to pronouns,
clauses or deictics (Nordhoff, forthcoming).
SLM uses two inflectional strategies: affixation (Nordhoff 2009: 405) and cliticization.
Tense is for instance expressed by means of prefixes on predicates, while plurality is
expressed by means of enclitics. Some affixes only occur in subordinated clauses, indicating
that morphosyntactic information is relevant to inflectional processes.
Events and Properties can be used as predicates (Nordhoff 2009: 602). Some
Morphosyntactic Nouns can be predicates as well, but others need a copula-construction
(Nordhoff 2009: 472). Likewise, not all semantic units can be arguments in SLM: Events
32 Sterre Leufkens
need the nominalizer =àn to function as arguments. This means that SLM is not completely
transparent with respect to flexibility.
SLM is predominantly isolating, but nonetheless has several bound morphemes. These
bound morphemes do not mark the head of the phrase, but the phrase as a whole. For instance
the dative-marking clitic =nang can attach to pronouns, simple nouns, but also to a whole
clause (Nordhoff, forthcoming). It always attaches to the last word of a phrase. SLM is hence
transparent with respect to this property.
There are non-transparent phonological assimilation rules in SLM. The case marking
clitics =pe and =nang both undergo a change of form if they are attached to singular pronouns
(Nordhoff, forthcoming). The form of the host hence matters for the expression of these
clitics. Other phonological assimilation processes, described above under domain
disintegration, are form-based form phenomena as well.
There can be influence of formal complexity on morphosyntactic placement in SLM. It
is for instance possible that a heavy constituent moves to sentence-final position, as in
example (38).
(38)
Se=ppe
1.SG=POSS
oorang
man
thuuva
old
pada anà-biilang [kithang pada
PL
PST-say
1.PL
PL
Malaysia=dering
anà-dhaathang
katha]
Malaysia=ABL
PST-come
QUOT
‘My elders said that we had come from Malaysia.’
(Nordhoff, forthcoming)
5.5. Combined results
Combining the transparency analyses of the four languages, we get Table 2. A +-sign in this
table means that the non-transparent feature is present in the language. A –-sign means that
the language does not exhibit the feature, hence that the language is transparent in that
respect. A question mark means that I have not been able to find sufficient information.
Table 2: Combined transparency analyses of four creoles.
Type
Redundancy
Domain
Feature
Apposition
(cross-reference)
Double expression of semantic
information
(negative concord)
(plurality)
(semantic class)
Discontinuity (IL/ML, RL/ML)
DIP
+
+
Nubi
+
+
Pichi
+
+
SLM
+
+
+
+
+
+
-
+
+
+
-
The transparency of creoles 33
disintegration
Form-based
form
(raising)
(circumfixes, infixes)
Stem alternation (IL/ML, RL/ML)
Cumulation (IL/ML, RL/ML)
Non-parallel alignment (IL/PL,
RL/PL)
Agreement
+
+
-
+
+
?
+
+
+
+
+
-
-
-
-
-
Grammatical gender
Tense copying
Expletive elements
Grammatical relations
+
-
?
+
-
-
Morphosyntactic information
influences pragmatic function
marking
Morphosyntactic information
influences semantic function marking
Morphosyntactic information
influences derivational processes
Morphosyntactic information
influences inflectional processes
Morphosyntactic information
influences availability as predicate/
argument
Head-marking through affixes
Morphosyntactic / phonological
weight influences word order
Phonological assimilation
-
-
-
-
+
-
+
-
+
+
+
+
+
+
+
+
+
+
+
+
+
?
+
+
+
+
+
+
+
All creoles exhibit redundant structures, mainly apposition of two lexically realized elements
and negative concord. In this respect, the creoles cannot be said to be very transparent.
Domain disintegration is attested in all creoles as well, but to a lesser extent. Even
though all languages tested are predominantly isolating, there are instances of cumulation in
all of them. In all studied languages personal pronouns exist that combine the grammatical
categories of person and number. DIP exhibits stem alternation, although it should be said that
the non-transparent verb forms are all directly inherited from lexifier Portuguese. It is not the
case that DIP has fusional morphological rules; in fact the units are borrowed as integer nontransparent forms. Nubi exhibits stem alternation, but like in DIP, the suppletive forms are
34 Sterre Leufkens
inherited, in this case from Arabic. DIP and Pichi exhibit discontinuity. The only clear case of
non-parallel alignment is found in Pichi.
The creoles studied all show some form-based form phenomena. Firstly, all creoles
show phonological assimilation processes and influence of morphosyntactic complexity on
word order. Both these features are related to an extra-linguistic factor: ease of articulation
and cognitive processing. Apparently then, ease of articulation and processing overrule
transparency. Note that this is in line with the ideas of Seuren & Wekker (1986), who believe
that simplicity (which they define as minimal processing) is maximized in creoles.
More opacity of the form-based form type is attested in the domain of availability of
semantic units as predicate and argument. A transparent language is fully flexible, but the
studied creoles are not. However, the creoles are more flexible than non-creole languages.
Crucially, there are fewer restrictions on the expression of semantic units than in the
superstrate languages, such as Portuguese and English. The creoles also have fewer headmarking affixes than their source languages do, and pragmatic and semantic function marking
are less often influenced by morphosyntactic information. In all creoles, we find a large
amount of conversion (zero-derivation), which indicates that morphosyntactic category is not
as important in the creoles as it is in the source languages. Inflection is non-transparent in all
languages, but note that it is again not very pervasive in the creoles. Only few grammatical
categories are expressed by means of inflection; the creoles share a preference for lexical
marking.
The most striking result of the study is the absence of the remaining form-based form
phenomena from all tested creoles. None of the creoles exhibits agreement or grammatical
gender. Moreover, none of the languages in the sample has a tense copying rule. Grammatical
13
relations are absent from all creoles as well.
Arends et al. (2005) find that several Romance-based creoles exhibit (in terms of Booij
1993) inherent (meaning-based) inflection, but not contextual (form-based) inflection. The
results of the current study are hence in line with the findings of Arends et al.; there is
reluctance to form-based form in creoles. The only counter example here is the presence of
expletives in Pichi and DIP.
The results clearly show that the transparency of creoles cannot be accounted for by
means of inheritance. All creoles exhibit transparent features that were not present in the
source languages. Grammatical gender is for instance present in Gujarati and Portuguese, but
not in DIP. English has grammatical relations, but Pichi has not. Portuguese is strongly
fusional, but DIP is (despite a few stem alternating morphemes) mainly isolating. There are of
course clear cases of sub- and superstrate influence, but these cannot account for the creoles’
transparency.
5.6. Towards a hierarchy of opacity
We have seen that creole languages are indeed quite transparent. This transparency cannot be
an inherited feature, since the creoles are much more transparent than their source languages
The transparency of creoles 35
are. Arguably, non-transparent elements and operations are stripped from languages in a
situation of intensive language contact. Communication in such a situation needs to be as
efficient, i.e. as functional as possible. Especially non-functional, form-based forms are not
helpful in communication and hence are not taken over in the new language. An emerging
language starts out transparently, while opacity might develop later on after a process of
language change.
Likely, the order in which opacity is developed in a language is not random. The
results of the current study already suggest a possible ordering in which opaque features
develop in languages. Apparently, redundancy is the first type of opacity to appear in a creole.
Domain disintegration is acquired later on and in the next stadium of the development of a
language, form-based form phenomena start appearing. The results of the current study
indicate that rigidity in morphosyntactic categories, and certain restrictions on derivational
and inflectional processes are the first form-based form phenomena to enter a language.
Consecutively, expletives appear. In a later stage, grammatical relations, agreement, tense
copying and grammatical gender emerge.
We have now established a first implicational hierarchy: redundancy > domain
disintegration > form-based form. But the results also contain patterns within the three opacity
categories. I will now discuss these patterns separately, along with parts of Table 2 in which I
have changed the order of features and languages. Counter examples to stipulated hierarchies
are delineated in these tables. Note that the ranking of opacity of the languages differs per
category – Nubi is for instance most transparent when it comes to form-based form, but most
opaque in the domain of redundancy. This proves that the categories do not strictly follow up
on each other – it is very well possible that form-based form starts appearing in a language
that has acquired only few instances of domain disintegration.
Table 3: Hierarchy of redundancy features
Type
Redundancy
Feature
Apposition
(cross-reference)
Double expression of semantic
information
(negative concord)
(plurality)
(semantic class)
DIP
+
+
Pichi
+
+
SLM
+
+
Nubi
+
+
+
-
+
-
+
-
+
+
-
Within the redundancy category (Table 3), we see that all languages allow apposition of two
lexicalized elements, but none of the languages shows cross-reference.
Multiple expression of semantic information is attested in all creoles as well. DIP,
Pichi and Nubi have negative concord, while SLM and Nubi show double marking of
plurality. None of the languages in the sample has a semantic classification system, so there
36 Sterre Leufkens
can be no multiple expression of semantic class. I do not think that multiple expression of one
category of semantic information implies the double expression of another – the categories
seem to be unrelated. I therefore will not stipulate an internal hierarchy here.
Table 4: Hierarchy of domain disintegration features
Type
Feature
Domain
Cumulation (IL/ML, RL/ML)
disintegration
Stem alternation (IL/ML, RL/ML)
Discontinuity (IL/ML, RL/ML)
(raising)
(circumfixes, infixes)
Non-parallel alignment (IL/PL,
RL/PL)
Nubi
+
SLM
+
DIP
+
Pichi
+
+
?
+
-
+
+
-
+
+
+
Within the category of domain disintegration (Table 4), cumulation occurs most often,
followed by stem alternation. This is in accord with expectations, as stem alternation is a
more severe violation of transparency. Discontinuity is less frequent than stem alternation and
apparently acquired later than fusional morphology. Within discontinuity, circumfixes and
infixes are not attested – raising is arguably higher in a discontinuity hierarchy. Non-parallel
alignment appears after discontinuity. The three categories hence appear in the order given in
(39), in which I left out the subcategories.
(39)
Fusional morphology > discontinuity > non-parallel alignment
Pichi is the most opaque language in the sample when it comes to domain disintegration. The
fact that stem alternation is not attested in Pichi is therefore unexpected. It could be argued
that there is no single hierarchy as in (39), but that there are in fact two separate hierarchies,
as in (40).
(40)
Stem alternation
^
> discontinuity > non-parallel alignment
Cumulation
This two-dimensional hierarchy captures that discontinuity implies fusional morphology but
not necessarily the subcategory of stem alternation.
Table 5: Hierarchy of form-based form features
Type
Form-based
form
Feature
Morphosyntactic / phonological
weight influences word order
Nubi
+
SLM
+
Pichi
+
DIP
+
The transparency of creoles 37
Phonological assimilation
Morphosyntactic information
influences availability as predicate/
argument
Morphosyntactic information
influences derivational processes
Morphosyntactic information
influences inflectional processes
Morphosyntactic information
influences semantic function marking
Expletive elements
Morphosyntactic information
influences pragmatic function
marking
Grammatical relations
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
-
-
+
+
-
-
+
-
+
-
?
-
-
-
Agreement
Grammatical gender
Tense copying
Head-marking through affixes
?
-
-
-
A hierarchy of form-based form phenomena appears in Table 6. Influence of formal
complexity on word order is found in all languages, just as phonological assimilation. As
discussed above, these are probably very common processes in the languages of the world, as
they rely on cognitive and articulatory principles. Morphosyntactic restrictions on the
availability of units as predicates and arguments, and morphosyntactic influence on
derivational and inflectional processes, are more common than formal influence on semantic
function marking. Pragmatic function marking is never influenced by morphosyntactic
information. This is apparently the most robust transparent trait in languages in the domain of
form-based influence on formal operations.
Expletive elements are attested in Pichi and DIP. No other form-based forms were
found in the creoles. Apparently, these are the most severely non-transparent properties of the
list: they are the strongest violations of transparency and appear in languages only after all
other features have appeared.
5.7. Towards a hierarchy of interfaces
Patterns can also be found in the occurrence of non-transparency on the different interfaces.
Look for instance at the occurrence of fusional morphology (Table 6).
38 Sterre Leufkens
Table 6: Fusional morphology on IL/ML and RL/ML
Interface
RL/ML
IL/ML
RL/ML
IL/ML
Feature
Cumulation
Cumulation
Stem alternation
Stem alternation
Pichi
+
+
-
DIP
+
+
+
-
Nubi
+
+
+
-
SLM
+
+
+
+
We already saw that cumulation precedes stem alternation. Interestingly, stem alternation at
the interface between the Representational and Morphosyntactic Level precedes stem
alternation at the higher interface. Apparently, fusional morphology develops bottom-up.
The reverse is true for discontinuity and non-parallel alignment (Table 7). They appear
at the higher interface first, and then on the lower interface.
Table 7: Discontinuity on IL/ML and RL/ML
Interface
IL/ML
RL/ML
IL/PL
RL/PL
Feature
Discontinuity
Discontinuity
Non-parallel alignment
Non-parallel alignment
Nubi
-
DIP
+
?
?
Pichi
+
-
SLM
+
-
This, however, might be a consequence of the decision to analyze discontinuity and nonparallel alignment on the highest interface possible. Moreover, the features are attested so
infrequently in the sample that it is not justified to draw conclusions here. Further research is
needed to investigate whether opacity develops bottom-up or top-down.
No new interface-specific hierarchies can be distinguished at the
pragmatic/morphosyntactic and the semantic/morphosyntactic interfaces. Patterns at those
interfaces have already been discussed in section 5.6.
6. Conclusions
In this study, both components of the transparency hypothesis are corroborated. Firstly,
creoles are indeed more transparent than non-creole languages. The creoles typically lack
form-based form phenomena such as grammatical relations, agreement, grammatical gender
and tense copying. Morphosyntactic influence on derivation, inflection and availability for
predicate and argument function is less pervasive in the creoles than in their source languages.
The same is true for domain disintegration. Only redundancy phenomena are well represented
in creoles.
The second component of the semantic transparency hypothesis states that creoles are
transparent as a result of their sudden emergence in a contact situation. An alternative account
is that creoles are similar because their source languages are similar. The transparent nature of
The transparency of creoles 39
creoles would then be an inherited feature. The study shows that the latter account cannot be
right, as the creoles exhibit transparent features that were not present in the source languages.
The transparent nature of creoles can only be a result of the circumstances in which they
emerged.
Apart from corroborating the hypothesis, the paper has provided a concrete and
specific definition of the concept of transparency, where earlier definitions were sometimes
vague and not testable. The new definition enables linguists to investigate transparency in a
precise and scientifically valid manner.
The paper furthermore suggests directions for further research on transparency. A
hierarchy of opacity was proposed, itself containing several smaller hierarchies that represent
the order in which opacity is acquired by originally transparent languages. Future research
should point out whether this ordering of non-transparent features is corroborated by
linguistic facts in other languages.
References
Aarts, Bas, David Denison, Evelien Keizer & Gergana Popova (eds.). 2004. Fuzzy grammar:
A reader. Oxford: Oxford University Press.
Aboh, Enoch & Norval Smith. 2009. Simplicity, simplification, complexity and
complexification. Where have the interfaces gone? In Enoch Aboh & Norval Smith (eds.),
Complex processes in new languages, 1-25. Amsterdam: Benjamins.
Arends, Jacques, Pieter Muysken & Norval Smith (eds.). 1995. Pidgins and creoles: An
introduction. Amsterdam: Benjamins.
Arends, Jacques, Josje Verhagen, Eva van Lier, Suzanne Dikker & Hugo Cardoso. 2005. On
the presence versus absence of morphological marking in four Romance-based creoles.
Bakker, Peter. 2003. Pidgin inflectional morphology and its implications for creole
morphology. In Geert Booij & Jaap van Marle (eds.), Yearbook of morphology, 3-33.
Dordrecht: Kluwer.
Bauer, Laurie. 1983. English word-formation. Cambridge: Cambridge University Press.
Bickerton, Derek. 1988. Creole languages and the bioprogram. In Frederik Newemeyer (ed.),
Linguistics: The Cambridge survey,276-284. Cambridge: Cambridge University Press.
Blom, Elma, Daniela Polišenská & Fred Weerman. 2008. Articles, adjectives and age of
onset: the acquisition of Dutch grammatical gender. Second language research 24. 297331.
Booij, Geert. 1993. Against split morphology. In Geert Booij & Jaap van Marle (eds.),
Yearbook of Morphology, 27-49. Dordrecht: Kluwer.
Braun, Maria & Ingo Plag. 2003. How transparent is creole morphology? A study of Early
Sranan word-formation. In Geert Booij & Jaap van Marle (eds.), Yearbook of morphology,
81-104. Dordrecht: Kluwer.
40 Sterre Leufkens
Cardoso, Hugo. 2009. The Indo-Portuguese language of Diu. Amsterdam: University of
Amsterdam dissertation.
Corbett, Greville. 1991. Gender. Cambridge: Cambridge University Press.
Evans, Nicholas & Toshiki Osada. 2005. Mundari: The myth of a language without word
classes. Linguistic typology 9. 351-390.
Hattnher, Marize & Kees Hengeveld. 2007. Advances in Functional Discourse Grammar:
Introduction. Alfa – Revista de Lingüística 51(2) (Advances in Functional Discourse
Grammar). 7-10.
Hengeveld, Kees. 1992. Non-verbal predication: Theory, typology, diachrony (Functional
Grammar Series 15). Berlin and New York: Mouton de Gruyter.
Hengeveld, Kees. 2004. Morphological types in Functional Discourse Grammar. Web papers
on Functional Discourse Grammar 79. 1-10.
Hengeveld, Kees. 2007. Parts-of-speech systems and morphological types. ACLC Working
Papers, 2(1). 31-48.
Hengeveld,
Kees.
Forthcoming.
Transparent
languages
in
Discourse Grammar. In Kees Hengeveld (ed.), Transparent
Special issue of Linguistics in Amsterdam.
Functional
languages.
Hengeveld, Kees and Lachlan Mackenzie. 2008. Functional Discourse Grammar. A
typologically-based theory of language structure. New York: Oxford University Press.
Kihm, Alain. 2000. Are creole languages “perfect” languages? In John H. McWhorter (ed.),
Language change and language contact, 163-199. Amsterdam: Benjamins.
Kihm, Alain. 2005. Noun Class, Gender and the Lexicon-Syntax-Morphology Interfaces. In
G. Cinque and R. Kayne (eds), The Oxford Handbook of Comparative Syntax 40, 459512. Oxford, Oxford University Press.
Labov, William. 1973. The boundaries of words and their meanings. In C.-J. Bailey and R.W.
Shuy (eds.), New ways of analyzing variation in English, 340-373. Washington DC:
Georgetown University Press.
Leufkens, Sterre. Forthcoming. Kharia. In Kees
languages. Special issue of Linguistics in Amsterdam.
Hengeveld
(ed.),
Transparent
McWhorter, John. 1998. Identifying the creole prototype: Vindicating a typological class.
Language 74. 788-818.
McWhorter, John. 2001. The world’s simplest grammars are creole grammars. Linguistic
typology 5(2-3). 125-166.
Muysken, Pieter. 1988. Are creoles a special type of language? In Frederik Newmeyer (ed.),
Linguistics: The Cambridge survey, 285-301. Cambridge: Cambridge University Press.
Nordhoff, Sebastian. 2009. A grammar of Upcountry Sri Lanka Malay. Amsterdam:
University of Amsterdam dissertation.
Nordhoff, Sebastian. Forthcoming. Sri Lanka Malay. In Kees Hengeveld (ed.).
Transparent languages. Special issue of Linguistics in Amsterdam.
The transparency of creoles 41
Peterson, John. 2005. ‘There’s a grain of truth in every “myth”, or, Why the discussion of
lexical classes in Mundari isn’t quite over yet’. Linguistic typology 9(3). 391- 441.
Peterson, John. 2006. Kharia: A South-Munda Language. Volume I: Grammatical analysis.
Osnabrück:
Universität
Osnabrück
dissertation.
http://www.southasiabibliography.de/Bibliography/Austroasiatic/Munda/Kharia/Kharia__
A_South_Munda_Language/kharia__a_south_munda_language.html
(accessed
29
September 2009).
Plag, Ingo. 2001. The nature of derivational morphology in creoles and non-creoles. Journal
of pidgin and creole languages 16(1), 153-160.
Seuren, Pieter & Herman Wekker. 1986. Semantic transparency as a factor in creole genesis.
In Pieter Muysken & Norval J. Smith (eds.), Substrata versus universals in creole genesis,
57-70. Amsterdam: Benjamins.
Travis, Lisa. 1984. Parameters and Effects of Word Order Variation. Cambridge: MIT
dissertation.
Trudgill, Peter. 1999. Language contact and the function of linguistic gender. Poznan studies
in contemporary linguistics 35.133–152.
Yakpo, Kofi. 2009. A grammar of Pichi. Nijmegen: Radboud University of Nijmegen
dissertation.
Velasco, Daniel. Forthcoming. Discontinuity and displacement in a functional theory of
grammar.
Wellens, Ineke. 2003. An Arabic creole in Africa. The Nubi language of Uganda. Nijmegen:
Catholic University of Nijmegen dissertation.
Zeijlstra, Hedde. 2004. Sentential negation and negative concord. Utrecht: University of
Utrecht dissertation.
Zwicky, Arnold M. & Geoffrey K. Pullum. 1983. Cliticization vs. inflection: English n’t.
Language 59 (3). 502-513.
*
I am indebted to Kees Hengeveld and to all members of the FDG research group at the University of
Amsterdam for their valuable feedback on earlier versions of this paper. I would also like to thank
Norval Smith for introducing me to the ideas and literature on transparency of creoles, as well as Hugo
Cardoso, Sebastian Nordhoff, Kofi Yakpo and Ineke Wellens for their help with analyzing the
languages of their expertise. Contact information: [email protected], Dept. of Theoretical
Linguistics,
Spuistraat
st
nd
120,
Abbreviations: 1, 2, 3=1 , 2 , 3
rd
1012
VT
Amsterdam.
person, A=active voice, ABS=absolute case, AFF=affix,
ANIM=animate, CNTR=contrastive, COMP=complementizer, COP=copula, DECL=declarative,
DEF=definite DEM=demonstrative, EMPH=emphatic, EXS=existential, FOC=focus, FUT=future,
GEN=genitive,
HUM=human,
IL=Interpersonal
Level,
INDEF=indefinite,
INF=infinitive,
IPFV=imperfective, IRR=irrealis, LOC=locative, M=middle voice, ML=Morphosyntactic Level,
NEG=negation,
NHUM=non-human,
NMLZ=nominalizer,
NPST=non-past,
PL=plural,
42 Sterre Leufkens
PFV=perfective, PL=Phonological Level, POSS=possessive, PROX=proximate, PRS=present,
PST=past,
Q=question
marker,
QUOT=quotative
marker,
REM.PST=remote
past,
RL=Representational Level, SG=singular, SIMIL=similative.
1
Transparency should, in my opinion, deal with pragmatic meaning as well as semantic meaning. I
therefore prefer to speak of ‘transparency’ rather than ‘semantic transparency’. However, where I
quote or discuss literature I will use the authors’ terminology.
2
This is the reason that personal pronouns in this paper are glossed with dots between the number and
person categories, e.g. 3.PL instead of 3PL.
3
A purely formal version of negative concord, which I will call negation agreement, does exist. It is
for instance found in Nunggubuyu (Hengeveld & Mackenzie 2008: 351), where a clausal negation
operator is copied at the Morphosyntactic Level to the heads of all phrases in its scope.
4
Travis includes ‘ > referential NPs’ at this end of her hierarchy, but I do not agree with her that
Referential NPs are expletive elements. Therefore, I leave them out of this discussion.
5
Interestingly, the presence of semantic class in languages correlates with agglutinative morphology
(Kihm 2005). Languages with grammatical classification systems (grammatical gender) more often
have fusional morphology. This is precisely the kind of correlation that transparency predicts.
6
Often, grammatical information is only expressed on certain morphosyntactic units (e.g. in English
plurality is marked on nouns and verbs, but not on adjectives). This does not violate transparency, as
transparency does not predict that information is expressed on all morphosyntactic units. It is opaque,
however, when a strategy for inflectional marking is influenced by morphosyntactic information (e.g.
when nouns are marked for plural by means of affixation, while verbs are marked by means of
suppletion).
7
Note that suppletion is always non-transparent because it involves fusional morphology. Apart from
that, the inflectional strategy of suppletion can be non-transparent when it is applicable to some
morphosyntactic units but not to others.
8
But see Evans and Osada (2005) for a plea against the existence of flexibility, and an alternative
analysis of Mundari, a language related to Kharia.
9
Despite McWhorter’s rejections of the transparency hypothesis, many researchers have interpreted
transparency and simplicity as the same or in any case overlapping notions. I will therefore discuss
McWhorter’s ideas on simplicity here.
10
Plag (2001: 2) actually recognizes this when he states: “Bauer’s examples of transparent vs. opaque
formations […] are airmail and blackmail, respectively. The crucial problem, however, is whether a
word such as blackmail should be considered morphologically complex at all.”
11
The non-transparent features were ordered according to interface in Section 2.3. However, to
improve the readability of the results section, I decided to discuss the non-transparent features of each
language in the order redundancies – domain disintegration – form-based form.
12
Actually, there are two nominalizers pàr-…-an and ka-…-an (Nordhoff (2009: 305). These,
however, are not productive and considered to be allomorphs of –àn. I hence analyze –àn as the real
morpheme.
The transparency of creoles 43
13
The absence of tense copying in the creoles could be explained by the infrequence of that feature in
languages of the world. The chance is quite high that, with every sample of four languages, not one of
them will exhibit tense copying. However, especially the absence of agreement and grammatical
relations cannot be explained in this way, since those are very frequent phenomena.