Available online at www.sciencedirect.

com

Journal of Pragmatics 40 (2008) 2003–2026

www.elsevier.com/locate/pragma

The acquisition order of coherence relations:

On cognitive complexity in discourse
Wilbert Spooren a,*, Ted Sanders b
a
Department of Language and Communication, VU University Amsterdam, De Boelelaan 1105,
NL-1081 HV Amsterdam, The Netherlands
b
Utrecht institute of Linguistics – UiL OTS, Trans 10, NL-3512 JK Utrecht, The Netherlands
Received 14 July 2006; received in revised form 14 February 2008; accepted 23 April 2008

Abstract
This article presents an analysis of the acquisition order of coherence relations between discourse
segments. The basis is a cognitive theory of coherence relations (Sanders et al., 1992) that makes predictions
about the order in which the relations and their linguistic expressions are acquired, because they show an
increasing cognitive complexity. The child language literature lends support to two distinctions in the
theory, Basic Operation (causal versus additive) and Polarity (positive versus negative). In two studies,
additional data were collected to test the validity of two other distinctions, Source of Coherence and Order of
the Coherence Relation.
In the first study, children described a picture or conversed freely with the investigator. Both distinctions
turn out to be necessary to account for the acquisition patterns. In the second study, the children’s proficiency
in dealing with negative causal relations was investigated. The two studies use different research designs.
The first is a study of relatively naturalistic, only partially structured elicitation of extended stretches of
speech produced by children, the other is an experiment on the understanding and production of coherence
relations in short sequences of statements relying on nonsense words that lack a conventional semantic
content. The two procedures tap very different kinds of communicative skills and linguistic as well as
conceptual knowledge. The combination of these two studies allows us to draw valid conclusions about the
acquisition of the various coherence relations. The data support the claim that cognitively complex
coherence relations show up later than cognitively simple relations.
# 2008 Elsevier B.V. All rights reserved.

Keywords: Child language; Discourse coherence; Cognition; Dutch

* Corresponding author at: Department of Language and Communication, VU University Amsterdam, De Boelelaan
1105, NL-1081 HV Amsterdam, The Netherlands. Tel.: +31 20 598 6572; fax: +31 20 598 6500.
E-mail address: [email protected] (W. Spooren).

0378-2166/$ – see front matter # 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.pragma.2008.04.021
2004 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026

1. Introduction

Coherence is an essential trait of well-formed discourse. Between the segments of a coherent

text, relations like Contrast and Cause–Consequence must be inferred. These relations play a
central role in processes of text comprehension, memory, and reproduction. Consider fragment
(1) as an example.1

(1)
(a) De juf is heel aardig
‘The teacher [female] is very nice’
(b) want bijna altijd zegt ze: ‘‘Goed zo, Deborah’’
‘because she nearly always says: ‘‘Well done, Deborah’’
(c) en dat is altijd lief
‘and that’s always sweet’
(d) een meester is heel anders
‘a teacher [male] is very different’
(e) want die ben ik niet gewend
‘because I’m not used to him/them’
(Deborah 7;3)

The example shows several things. For instance, the units in this fragment are linked through
coherence relations: between clauses (b) and (c) an Evaluation relation holds; between (a) and
[(b, c)] and between (d) and (e) a Claim–Argument relation obtains; the relation between [(a–c)]
and [(d, e)] is that of Contrast. The example also shows that coherence relations need not be
marked explicitly: The Contrast relation connecting [(d, e)] to its context has to be inferred by the
reader, without the help of explicit linguistic cues like connectives (for example and, but, and so).
In our analysis of this example we use coherence relations and connectives to characterize the
discourse connectedness of the fragment. Traditionally, linguists have approached this problem
by looking at overt linguistic elements and structures. In their seminal Cohesion in English,
Halliday and Hasan (1976) describe text connectedness in terms of explicit clues that make a text
a text, such as reference, substitution, ellipsis and conjunction. Cohesion occurs ‘‘when the
interpretation of some element in the discourse is dependent on that of another’’ (Halliday and
Hasan, 1976:4).
Over the last decennia, the dominant view has come to be that the connectedness of discourse
is a characteristic of the mental representation of the discourse rather than of the discourse itself
(see Sanders and Pander Maat, 2006, for a recent overview). According to most cognitive
scientists, linguists and psycholinguists, understanding discourse means that readers construct a
mental representation of the information in the text (Graesser et al., 1997; Gernsbacher and
Givón, 1995; Garnham and Oakhill, 1992). The term coherence is generally used to refer to this
type of conceptual connectedness. Discourse coheres in several ways. Generally speaking, at
least two types can be identified:

1
Unless indicated otherwise, our examples come from a corpus of child language data collected for Study 1. Notational
conventions: utterances were divided into units; a unit is a main clause, an adverbial clause or an appositive relative
clause. We have discarded pauses etc. as these seem irrelevant to our analyses. After each fragment, we give the name of
the informant and his/her age in years;months.
 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026 2005

1. Referential coherence: smaller linguistic units (often nominal groups) may relate to
the same mental referent;
2. Relational coherence: A discourse representation shows coherence, because the
representation of the discourse segments (most often conceived of as clauses) are
connected by coherence relations like CAUSE–CONSEQUENCE

Although there is a principled difference between the cohesion and the coherence approach to
discourse, it is important to realize that coherence phenomena may be of a cognitive nature,
but that their reconstruction is often based on linguistic signals in the text itself. Both
coherence phenomena under consideration – referential and relational coherence – have clear
linguistic indicators that can be taken as processing instructions. For referential coherence
these are devices such as pronouns and demonstratives, and for relational coherence these are
connectives and (other) lexical markers of relations, such as cue phrases and signaling
phrases.
In this article we will focus on relational coherence. Coherence relations (for the term, see
Hobbs, 1979; for their analysis in a formal semantic context, see Asher and Lascarides, 1998) are
commonly regarded as ‘‘the cornerstone of comprehension’’ (Graesser et al., 2003:82). Examples
are relations such as CAUSE–CONSEQUENCE, LIST, and PROBLEM–SOLUTION. These
relations hold locally, that is, between adjacent clauses, but also between larger discourse units,
from groups of clauses up to whole sections, paragraphs and even chapters (Sanders, 1997; see
Rhetorical Structure Theory: Mann and Thompson, 1988; Taboada and Mann, 2006). Our view
on discourse coherence difference, then, somewhat from Hickmann’s (2003), who characterizes
cohesion as ‘‘typically defined on the local level of adjacent clauses’’, whereas coherence is to be
found ‘‘on the more global level of plot organization’’ (p. 93).
There have been many studies on the acquisition of connected discourse over the past four
decades. Bloom et al. (1980) summarizes a number of findings in the following ordering of the
acquisition of relations:

Additive < Temporal < Causal < Adversative

Connectives display the following pattern:

and < and then < because < so < but

Of these, and is first used for additive relations, and later also for temporal, causal and adversative
relations. The other connectives are used more specifically: and then for temporal relations,
because and so for causal relations, and but for adversative relations.2
In this article we present an explanation for this order, in terms of a theory of coherence
relations that is based on the cognitive complexity of general conceptual relations that underlie
coherence relations. The theory has been formulated as a classification of coherence relations
(Sanders et al., 1992, 1993). In our classification, coherence relations are treated as cognitive
entities, as elements of a cognitive construct, the discourse representation. Below, we present
an overview of the four parameters underlying the classification (see Appendix A for
the complete classification) and some of the findings in the literature that relate to the
distinctions in the classification. Then we will report two empirical studies that were carried

2
We disregard the ‘place holding’ function of and and confine ourselves to clause-combining functions of the
connectives discussed (Schiffrin, 1986).
2006 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026

out in order to investigate the predictions about the order of acquisition following from the
classification.

1.1. An overview of the classification and how it fits acquisition data

For our classification we use four basic relational concepts. They are described in the
following sections. For an extensive description, see Sanders et al. (1992, 1993). We also provide
an overview of relevant findings in acquisition research.

Basic Operation: additive versus causal relations

The Basic Operation refers to the intuition that discourse segments can either be connected
strongly (in causal relations) or weakly (in additive relations). In additive relations there is a
relation of conjunction between the two segments (2). In causal relations the Basic Operation is
implicational as well as conjunctional (3).

(2) Additive relation (List)

Zij waren aan het vissen
en hij was aan het zwemmen.
‘They were fishing
and he was swimming’
(Hanneke 7;8)
(3) Causal relation (Result)
Daar leunt een dikke man tegen een paal
en daardoor vallen de blokken van de paal om.
‘There is a fat man leaning against a post.
and because of that the building blocks of the post fall down’
(Daan 11;11)

The results from both naturalistic (e.g., Bloom et al., 1980; Eisenberg, 1980) and experimental
studies (Piaget, 1924/1969; Katz and Brent, 1968) of first language acquisition show that additive
relations are acquired before causal relations, and that additive connectives are acquired before
causal connectives.

Polarity: positive versus negative relations

In a positive relation, the segments are linked directly. In a negative relation the link involves a
negation of one of the segments. For example, in (5) the link between the segments relies on the
implication relation ‘‘if mothers like something, their children like it as well’’, which is negated
in the second segment.

(4) Positive relation (Explanation)

Soms lees ik wel wat
omdat ik wil weten wat er in het boek staat.
‘Sometimes I read something,
because I want to know what is in the book’
(Bram 7;2)
 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026 2007

(5) Negative relation (Concession)

En mama vindt dat leuk.
Maar ik vind dat helemaal niet leuk.
‘And mummy likes that.
but I don’t like that at all’
(Deborah 7;3)

The observational studies of Bates (1976), Bloom et al. (1980) and Eisenberg (1980) as well as
the experimental studies of, for example, Wing and Scholnick (1981) clearly show that positive
relations are acquired before negative relations.

Source of Coherence: content, epistemic and speech act relations3

In a content relation, the link involves a real world connection: The relation involves the
propositional or ideational content of the related segments (example (6)); in an epistemic relation
the connection involves the level of reasoning and concluding (example (7)). In a speech act
relation, the connection is at the level of illocutions (example (8)).

(6) Content relation (Consequence–Cause)

Hier is een boot gezonken
omdat ze tegen een boei zijn gevaren.
‘Here a ship has sunk
because they sailed against a buoy.’
(Hanneke 12;0)
(7) Epistemic relation (Evidence)
Dat zijn aapmensen of mensen van vroeger.
‘Those are apemen or ancient people’
Want wij zijn in het prehistorisch huis geweest
‘Because we have been in the prehistoric house’
En daar hebben ze ook van die vellen.
‘And there they also have those hides’
(Bram 7;2)
(8) Speech act relation
Nieuws vind ik niet leuk.
‘News I don’t find nice’ [=I don’t particluarly like the news [on television]]
Want daar vind ik niks aan.
‘Because I don’t like it at all’
(Jordy 6;5)

There are two reasons why conclusions about the relevance of Source of Coherence for
acquisition data are not straightforward. First, most acquisition studies in the literature use
different categories, which makes it hard to interpret the results directly in terms of Source of

3
In our original proposal we distinguish between semantic and pragmatic relations, the latter subsuming both epistemic
and speech act relations. For reasons of terminological uniformity we adopt in this article Sweetser’s (1990) terminology.
See Sanders and Spooren (2001) for an argument that our bifurcation maps on Sweetser’s trifurcation in this way.
2008 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026

Coherence. Second, for the categories that can be related to the content-epistemic-speech act
distinction, some studies (Peterson, 1986; Bloom and Capatides, 1987; Bates, 1976) seem to
show that content relations are acquired earlier, whereas other studies (Kyratzis et al., 1990;
Piaget, 1923/1971; Corrigan, 1975) suggest that non-content relations (especially speech act
relations) are acquired earlier.

Order of the segments: basic order or non-basic order

Relations are presented in basic order if the information in the first segment corresponds with
the antecedent of the Basic Operation; relations are in non-basic order if the information in the
first segment corresponds with the consequent of the Basic Operation. For instance in (9) the
Basic Operation is a causal one ‘‘someone playing beautifully can cause people to cry’’. The first
segment relates to the first element in this causal link, the second segment relates to the second
element. In example (10) the same Basic Operation is used, but now the links to the causal
connection are reversed: The first segment relates to the second part of the causal link (the
consequent), the second segment relates to the antecedent.

(9) Basic order relation (Cause–Consequence)

Een man speelt viool
‘A man plays the violin’
en iedereen begint te huilen [. . .].
‘and everybody starts to cry [. . .].’
(Bas 12;0)
(10) Non-basic order relation (Consequence–Cause)
Die mensen huilen allemaal
‘Those people are all crying’
omdat die vioolspeler zo mooi speelt.
‘because that violin player plays so beautifully.’
(Jory 7;4)

Order of the Segments does not distinguish between different types of additive relations, since
additive relations are logically (although not pragmatically) symmetric.
Prior research shows that in specific, experimental, situations, children make more errors
identifying non-basic order relations than basic order relations (Bebout et al., 1980). This
suggests that basic order relations are acquired earlier. However, in more natural conditions
(e.g., observational studies like Hood and Bloom, 1979; French and Nelson, 1985; McCabe
and Peterson, 1985) such a difference is not found. This suggests that there is a confounding
with research methodology. Source of Coherence seems to be another relevant factor
accounting for this difference between observational and experimental studies: experimental
studies have focused on content relations, whereas non-content relations (especially speech
act relations) are more frequent in observational studies. A final relevant factor for the
difficulties children have with non-basic order relations in experimental studies may be the
cognitive load on the participants: Peterson and McCabe (1987) have shown that the children
performed better after having carried out the task a number of times or when they could use a
picture as a memory aid. In short, this survey shows that research results with respect to order
of segments are inconclusive.
 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026 2009

1.2. How a theory of coherence relations can account for these data

We want to suggest that the differences in emergence reported in the literature can be
accounted for in a systematic way. In our analysis, some relations are considered more complex
than others as they involve more complex concepts. For instance, Concession relations are more
complex than List relations because they involve negation on top of additivity, and Cause–
Consequence relations are more complex than List-relations because they involve causality in
addition to additivity. From the cognitive claims attached to the analysis, it follows that such
complexity patterns show up in the order in which coherence relations are acquired. That is, one
might expect List relations to be acquired before Contrast relations and before Cause–
Consequence relations.
Similarly, epistemic relations are complex, as they require abstract reasoning, usually
exploiting relations that exist between real world events. The conclusion in (11) that this man is a
cowboy exploits the real world relationship that cowboys use lassos to catch horses.4

(11) Dat is een cowboy

‘That’s a cowboy,’
die wil paarden vangen
‘who wants to catch horses’
want hij heeft een lasso.
‘because he has a lasso’
(Deborah 7;3)

As to the complexity of content versus speech act relations, Kyratzis et al. (1990) argue that
speech act relations are least complex.
With respect to the fourth distinction, between basic and non-basic relations, basic order
relations are less complex than non-basic order relations, at least for the content domain: events
connected by a Cause–Consequence relation iconically reflect the order of the events in the real
world.
These considerations lead to two types of predictions:
Hypothesis 1. Weak prediction:
Developmental trends cannot be described without the distinctions made by the classification.
Hypothesis 2. Strong prediction:
Relations that are analyzed as simple are acquired before relations that are analyzed as
complex.
In order to test these hypotheses we conducted two studies, in which the relevance of the basic
concepts was central. The first is a corpus study of Dutch, in which we compared coherence
relations produced by children more or less spontaneously in a partially structured setting. The
corpus study was inconclusive with respect to the acquisition of the most complex relations,
namely the negative causal ones. Therefore, we conducted an experiment to test children’s
capacity to use these. In sum, we have used two different research designs. The first is a study of
relatively naturalistic, only partially structured elicitation of extended stretches of speech

4
Analyses of epistemic relations in the mental space framework, like Dancygier and Sweetser (2000) and Verhagen
(2000), inevitably portray such relations as more complex than their non-epistemic counterparts.
2010 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026

produced by children, the other is an experiment on the understanding and production of

coherence relations in short sequences of statements relying on nonsense words that lack a
conventional semantic content. The two procedures tap very different kinds of communicative
skills and linguistic as well as conceptual knowledge. The combination of these two studies
allows us to draw valid conclusions about the acquisition of the various coherence relations.

2. Study 1: children’s use of coherence relations while describing and conversing

Our first study is a corpus analytic study under experimental control, using Dutch children as
informants. The children were given two tasks, both aiming at the elicitation of complex
utterances: a description task (biasing for content relations) and a conversation task (biasing for
epistemic and speech act relations). The research was carried out with elementary school children
from different age groups: grade 1 (age 6–7) and grade 6 (age 11–12). This cross-sectional design
allows inferences on the developmental patterns in the children.
The general research question is to what extent the classification explains for differential
patterns in the acquisition data, in accordance with the weak and the strong prediction. The weak
prediction is that differences in use of coherence relations between the two age groups can be
described in terms of the primitives of the classification. According to the strong prediction,
coherence relations that are classified as cognitively simply are acquired before complex
coherence relations. More specifically this leads to the following four research questions.

1. Is there any evidence for the strong prediction based on the classification of relations,
namely that complex relations are acquired later than simple relations?

The observations of Bates (1976), Bloom et al. (1980), and Eisenberg (1980) and the
experimental data of Piaget (1924/1969), Katz and Brent (1968), and Scott (1984) support the
distinctions based on the concepts Basic Operation and Polarity. They also suggest that less
complex relations are acquired earlier than complex relations. Therefore, we expected younger
children to use less negative and causal relations than older children.

2. Is there any reason to assume a difference in degree of complexity between content

and other relations?

Different researchers arrive at different conclusions concerning the degree of difficulty of

content versus other relations. Diachronically, epistemic and speech act relations are derived
from content relations (cf. Sweetser, 1990; Tabor and Traugott, 1998). On that basis it is tempting
to assume that non-content relations are more complex than content relations (following Bates,
1976). But ontogeny need not replicate phylogeny (Traugott, 1997). The most sophisticated
picture is presented by Kyratzis et al. (1990), who claim that speech act relations are acquired
before content relations, which in turn are acquired before epistemic relations. However, these
findings are based on data collected in a context that is heavily biased toward non-content (more
specifically, speech act) relations. An unanswered question is whether the same goes in a context
biased for content relations. To answer this question we will compare the amount of content and
non-content (epistemic and speech act) relations produced by the two age groups.

3. Is the Source of Coherence determined by the context (text type) in which the utterance
was produced?
 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026 2011

The seemingly contradictory results with respect to Source of Coherence may be attributable
to a difference in context in which the data were collected. We have argued elsewhere (Sanders,
1997) that text type and the type of coherence relations found in a discourse correlate strongly.
Informative and descriptive texts show large numbers of content relations, whereas epistemic and
speech act relations dominate argumentative texts. This being so, the type of context in which
data are collected seems a major factor in the type of relations produced by the children. We
expected to find more content relations in the descriptive task, and more non-content (epistemic
and speech act) relations in the conversation task.

4. Do we need a separate basic concept Order of the Segments, apart from Iconicity?

This question pertains to the discussion whether the Iconicity principle suffices to account for the
ordering patterns in the data or whether we need an additional principle, Order of the Segments. The
Iconicity principle (or Order of Mention principle, Clark and Clark, 1977:358) states that iconic
ordering is more easy than non-iconic ordering, because it reflects the order of the events ‘in the real
world’. As Iconicity applies to events only, it is restricted to content relations. The crucial question
is whether non-content (epistemic and speech act) relations also show preferences for specific
orderings. If so, we need an additional principle Order of Segments, apart from Iconicity.

2.1. Method

2.1.1. Materials
Each child was given two tasks: a description task and a conversation task. In the description task
(which biased for content relations) children had to describe a number of ‘Where’s Waldo?’
(Handford, 1991) pictures, that show large numbers of causally related events, such as a man
walking under a ladder, whistling and unaware that on the ladder a window cleaner has just dropped
a bucket of water. In the conversation task (bias for epistemic and speech act relations) the children
had to formulate an opinion concerning a number of controversial topics, such as vetoed TV
programs, wearing brand clothing at school, the amount of the children’s pocket money, etc.

2.1.2. Participants
The investigation was carried out with six children from grade 1 and six children from grade 6
of an elementary school in a medium-sized town in the south of the Netherlands. The choice of
these age groups was inspired by choices made in related research: Younger children do not
produce wide varieties of coherence relations, and as to the upper boundary of grade 6, Piaget
(1924/1969) claims that at age 11–12, children begin to use even the most complex relations.
Indeed, Evers-Vermeul (2005) finds hardly any use of these negative causal relations in her study
of young children (roughly 1;6–5 years of age), and if they show up, they appear after positive
causal relations emerge.
The two age groups in our study consisted of equal numbers of boys and girls. The reading
level of the children varied equally in each age group. The mean age of the children in the
younger age group was 7;2 (min. 6;7, max. 7;8). The mean age of the children in the older age
group was 12;1 (min. 11;11, max. 12;7). All children had Dutch as their first language.

2.1.3. Procedure
Each child was interviewed individually in a separate room of the elementary school they
attended. Each child was given the two tasks, each task lasting circa 10 min. In the description
2012 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026

task, the children were given a picture from a Where’s Waldo book, along with the following
instruction: ‘‘In this picture all sorts of things are happening. Can you tell me what is happening
and how that has come about?’’ In order to prevent a bias from this formulation for specific
orderings of descriptions, the formulation of the instruction was varied with ‘‘Tell me how this
has come about, what is happening here?’’ In the conversation task, the children were stimulated
to give their opinion on a number of topics and to motivate their opinion. The responses of the
researcher were restricted to presenting the instruction, and to back-channel behavior.

2.1.4. Data analysis

The children’s utterances were recorded on a cassette recorder and transcribed. The unit of
transcription and analysis was the clause: main sentences, adverbial subordinative sentences and
appositive relative clauses were taken as minimal units. Only uninterrupted utterances were taken
into account.
Subsequently, the transcripts were analyzed by two judges, on the basis of the set of coherence
relations specified in Mann and Thompson’s (1988) Rhetorical Structure Theory (RST). This
model of analysis was chosen because of its descriptive width. Of each utterance it was decided
between which clauses an RST-relation existed. Then this relation was analyzed in terms of the
four basic concepts of Sanders et al. (1992). Appendix B gives an example of a protocol
fragment. Examples (2–11) above are taken from the corpus and illustrate the categorization.
Appendix C shows further criteria for distinguishing between content, epistemic and speech act
relations.
Following Bloom et al. (1980) we have used two strategies to guarantee the reliability of the
analysis. The coders have made use of all the contextual information available to interpret the
relations between the utterances. Furthermore we have the principle of cumulative semantic
complexity: the relations have been interpreted as specifically as possible. In terms of the
classification this means that additive or temporal relations which can have a causal
interpretation, are coded as causal relations.
Note that by following this procedure we are studying the occurrence of coherence relations as
well as their linguistic marking. It can be asked whether we study emergence or acquisition. By
looking at the use of the relations, it follows that our prime focus is emergence. However, we also
investigate acquisition because we study whether coherence relations are correctly used, that is,
we pay attention to the (mis)match between connective and coherence relation. Occasionally the
children’s use of a connective is deviant. An example is (12).

Example (12)
[Wat doe je het liefste als je uit school bent en waarom?]
[What do you like to do most when you are out of school and why?]
1 Nou ik zit op voetbal
Well I play football
2a En dat vind ik leuk
And that I like
2b omdat ik iedere dag oefen voor voetbal
because I practice every day for football
3 En dan word ik steeds beter. [cause]
and then I am getting better and better
4 Dan spelen we echte wedstrijden [list]
Then we play real games
 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026 2013

5 En dan zit het hele stadion vol [list]

And then the stadium is crowded
6 En dan verliezen we soms. [list]
And then we loose sometimes
7 Maar het meeste winnen we.
But mostly we win
[3–1; Jordy; 6;7]

In this fragment Jordy uses a strategy commonly found among young children: many
utterances are connected by the temporal marker en dan (‘and then’), suggesting a temporal
sequence of events. However, sometimes a differently relation seems intended. For example, in
line 3 en dan is used to signal a causal relationship (he probably intends to express a complex
relationship that another reason why he likes football is that because of practicing every day he
gets better and better at it). The example has been analyzed as a causal relation, signaled by a
temporal connective (a so-called implicit use of connectives, see Spooren, 1997). This is an
indication that Jordy has not acquired the intricate possibilities of using en dan (literally a
temporal connective) to express causal relations. Still, he does know how to use the intended
causal relationship.
In total, the children produced 1355 utterances (112.9 utterances per child), 756 in
the description task (63.0 per child) and 599 in the conversation task (49.9 per child).
1097 coherence relations were analyzed (526 in the description task, 571 in the conversation
task). The children in grade 1 produced 566 coherence relations (mean of 94.3 relations
per child), the children in grade 6 produced 531 relations (mean of 88.5 relations per
child).
Per child, the percentage of categories mentioned in the hypotheses and research questions
was calculated (e.g., the percentage of content relations in basic order). These percentages were
analyzed using:

a. the Mann–Whitney rank order sum test for differences between two age groups (for instance,
the difference in percentage of content relations between grade 1 and grade 6);
b. the binomial test for evaluating one-sample distributions (for instance, the observed
percentage content relations compared to the expected percentage if content and non-content
relations were to occur equally frequently).

In addition, for the analysis of frequencies x2-tests were used.

2.2. Results

2.2.1. Reliability of the analyses

The analyses were carried out by two analysts independent from each other, after careful
instruction and on the basis of a coding scheme (see Appendix C for an example of the coding
scheme). Cohen’s kappa coefficient of agreement between two analysts was calculated separately
for each child’s data. Agreement was lowest for Order of the segments (with K varying from .40
to .83 in grade 1, and from .78 to 1.00 in grade 6). For Basic Operation K varied from .66 to .91 for
grade 1 and from .88 to 1.00 for grade 6. For Source of Coherence K varied from .66 to .88 for
grade 1, and from .90 to .98 for grade 6. Finally, for Polarity K varied form .54 to 1.00 for grade 1
and from .87 to 1.00 for grade 6.
2014 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026

Table 1
Coherence relations in our corpus classified as content, epistemic or speech act (both by grade and by task) (row
percentage in parentheses) (Study 1)
Type of relation Content Epistemic Speech Act
Grade
1 375 (67.4) 21 (3.8) 160 (28.8)
6 279 (53.9) 37 (7.1) 202 (39.0)
Task
Description 487 (94.2) 26 (5.0) 4 (.8)
Conversation 167 (30.0) 32 (5.7) 358 (64.3)
Note. In 23 cases the epistemic/speech act character of the relation was unclear.

2.2.2. Tests of the hypotheses

COMPLEX RELATIONS. It was expected that causal, negative and negative causal relations, which
– in terms of the classification – are more complex than their additive and/or positive counterparts
– are more frequently used by older children. This turned out not to be the case: Causal relations:
grade 1: 56%, grade 6: 50% (z = 1.28, p = .28); negative relations: grade 1: 6%, grade 6: 10%
(z = 1.44, p = .16); negative causal relations: grade 1: 2%, grade 6: 2%; z = .16, p = .88).
SOURCE OF COHERENCE AND COMPLEXITY. The data are summarized in Table 1. Summed over both
tasks, the children produced more content relations than non-content relations (content relations:
59%; binomial test: p < .05). Moreover, younger children produced more content relations than
older children (grade 1: 66% content relations; grade 6: 52% content relations; z = 2.56, p < .01).
If we compare epistemic or speech act relations separately, the two age groups show no
difference: grade 1 produced 181 non-content relations, 21 of which (11.6%) were epistemic.
Grade 6 produced 239 non-content relations, 37 of which (15.5%) were epistemic (z = .48,
p = .63). Therefore, there is no indication that younger children have more difficulty with
epistemic relations than older children.
ROLE OF CONTEXT. As expected, it was found that the task determines the type of relations
produced by the children: in the descriptive task more content relations (92.6%), in the
conversational task more non-content (epistemic and speech act) relations (72.3%) (in a binomial
test, both percentages differ from chance, p < .001).
As can be seen in Table 1, the predominance of non-content relations in the conversational
task is mainly due to the number of speech act relations: in the description task there are 30 non-
content relations, of which 26 (86.7%) are epistemic and 4 (13.3%) are speech act. In the
conversation task, there are 319 non-content relations, of which 32 (8.2%) are epistemic, and 358
(91.8%) are speech act (x2(1) = 483.84, p < .001).
RELATION BETWEEN SOURCE OF COHERENCE AND ORDER OF THE SEGMENTS. As expected, content
relations showed a preference for basic order (87%; binomial test: p < .05) and non-content
relations showed a preference for non-basic order (78%; binomial test: p < .05).

2.3. Discussion

We did not find evidence for the hypothesis that causal, negative and negative causal relations
were more difficult for younger children. Does this imply that complex relations are not acquired
later, contrary to what was expected on the basis of the classification? We believe there is another
explanation, at least for the use of the causal and negative additive relations: It is likely that even
the youngest age group has already acquired causal and negative relations. Evers-Vermeul (2005)
 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026 2015

and Evers-Vermeul and Sanders (submitted for publication) have looked at the first occurrence of
connectives in naturalistic data from children of age 1;6-up. Although they studied connectives and
not coherence relations, their data suggest that causal and contrastive connectives occur very early
indeed.5 Before they are three, most children have acquired causal (want, omdat) and negative
(maar) connectives. Hence, these results would explain why we do not find a difference between the
age groups in this study. At the same time, however, these data of younger children show direct
evidence for the idea of cumulative cognitive complexity: negative connectives emerge later than
positive connectives and causal connectives are preceded by additive connectives.
In the case of negative causal relations, we seem to be dealing with a bottom effect: only 2% of
the relations produced was negative causal. Consequently, it is not possible to say whether the
relation as such is too complex for even the oldest children or whether the tasks employed did not
elicit the use of negative causal relations. This point will be taken up in the next section.
For Source of Coherence, we did find that content relations were indeed simpler than non-
content relations. Our data accord in this respect with the suggestions by Bates (1976) and Bloom
and Capatides (1987) that content relations are simpler than non-content ones. Yet, they contrast
with the findings of Kyratzis et al. (1990), suggesting that their findings (speech act relations are
acquired before content relations) were indeed biased by the communicative context of the task they
used. This interpretation is strongly corroborated by another new finding in this study: the type of
relations children use depends on the communicative task. In a descriptive task they mainly use
content relations, while non-content relations predominate in conversational tasks. Finally, our data
provide convincing evidence for the status of Order of the Segments as an independent primitive in
our classification: non-content relations tend to be realized in a non-basic order, a bias which cannot
be accounted for in terms of a general Iconicity principle: a temporal ordering of the events
described by the segments is irrelevant in the case of non-content relations.
In conclusion, our data provide only moderate support for the strong prediction. There is no
indication that relations that are classified as complex by Sanders et al. (1992) are acquired later.
This holds both for causal relations as opposed to additive relations, for negative relations as
opposed to positive relations (although here the difference is marginally significant), and for
negative causal relations. It seems that even the youngest age group has acquired causal relations,
so that differences between age groups do not occur. About half of the relations produced by both
age groups are causal. This suggests that in future research younger children should be tested.

3. Study 2: children’s use of contrastive cause–consequence relations

The data of Study 1 were inconclusive with respect to the question of complex relations, more
specifically, with respect to the children’s proficiency in dealing with negative causal relations.
Example (13) is a classic example of such a relation: in the terminology of Sanders et al. (1992),
this example is an instance of the Contrastive Cause–Consequence (CCC) relation (also known
as the Denial of Expectation relation).

(13) Although John is a Republican, he is honest.

According to most analyses of such examples (e.g., Lakoff, 1971; Van Dijk, 1977; Abraham,
1979; Blakemore, 1989; Spooren, 1989; Oversteegen, 1997), a Contrastive Cause–Consequence

5
As we have stated repeatedly, this does not necessarily mean that causal coherence relations also occur very early, but
that inference is very likely.
2016 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026

relation presents two states of affairs (John being a Republican and John being honest), in which
the second state-of-affairs runs counter to what could be expected on the basis of the first. In other
words, a Contrastive Cause–Consequence relation like (13) can only be used if there is some sort
of presupposition or expectation that ‘‘generally Republicans are dishonest’’.
Because of the implicational character of the presupposition underlying (13) this relation is
classified as causal, and because the implication can only be related to the actual utterance
through a negation, the relation is classified as negative. Hence, the Contrastive Cause-
Consequence relation is an example of a negative causal relation.
Ever since Piaget, it is known that CCC relations, or, as he calls them, relations of discordance,
are notoriously difficult to learn. Yet, the precise cause of the complexity is difficult to assess,
because the factors that may determine the complexity have not been isolated carefully. For
instance, (13) may be complex for syntactic reasons (it is a subordinate construction), for lexical
reasons (because of the conjunction although), or for semantics reasons (because of the presence
of the negation), or for reasons of discourse coherence (the complexity of the relation-itself).
Furthermore, different researchers give different assessments of the exact age at which children
have fully acquired the relation. Some put it as early as age 7 (Bloom et al., 1980; Peterson, 1986),
others put it as late as age 13 (Piaget, 1924/1969; Katz and Brent, 1968; Scott, 1984). In spontaneous
observational data, the CCC relation is hardly ever found. This is partly due to the fact that most
researchers focus on connectives rather than coherence relations, and the connective although is
very infrequent in child language. But also in the research discussed above, explicitly aimed at
analyzing relations rather than connectives, there were few occurrences of a CCC relation. This
makes it uncertain whether children do not have the proficiency to use the relation or whether the
communicative context in which the corpus was elicited did not give rise to the use CCC relations.
Because of the infrequency of CCC relations in spontaneous speech, the question of the child’s
proficiency in dealing with these relations needed to be answered via an experiment.
The experiment was designed to investigate the children’s proficiency with respect to the CCC
relation.6 What we wanted to know was whether children display developmental patterns, and if
so, whether these patterns reflect the analysis of coherence relations in terms of the
categorization. In the experiment, children had to complete forced-choice items. In those items,
they had to choose between two possible completions of a discourse fragment. The items were
designed in such a manner that the choice reflects understanding of the coherence relation.
The question under study is whether CCC relations are more complex than other relations.
This leads to two hypotheses:
Hypothesis 1. Children perform worse on items involving CCC relations than on items
involving non-causal and/or non-negative relations.
Hypothesis 2. Younger children perform worse on items involving CCC relations than older
children.

3.1. Method

3.1.1. Material
Two sorts of item sets were developed. In the design of the sets, the analysis of CCC relations
implying (presupposing) a general rule was made use of. The first set consisted of items like the
following.

6
The experiment was designed in close collaboration with and carried out by Hanny den Ouden (cf. Den Ouden, 1995).
 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026 2017

(14) Contrastive Consequence–Cause (negative causal: sentence completion)

People who are lok wear sweaters.
Els is lok but she is . . .
a. not wearing a sweater.
b. wearing a sweater.

That is, the subjects were given a general rule, and on the basis of this rule they had to choose
between two solutions to complete a sequence containing but. This task was chosen because it
more or less resembles the production process: On the basis of an available rule, the speaker
makes a formulation choice. The items contained nonsense words (e.g., the word lok) to prevent
the children to base their choice on real world knowledge in stead of on knowledge of the
linguistic connective but and the coherence relation it encodes.
The second set of items was of the following type:

(15) Contrastive Consequence–Cause (negative causal: inferring a general rule)

Manel has a lot of money, but she is not bap.
a. People who have much money are not bap.
b. People who have much money are bap.

In this case the subjects were given a but-sequence. On the basis of this sequence, they had to
choose which of two alternatives they thought to be the case as a rule. In a way, this task
resembles the interpretation process: The underlying idea is that a correct interpretation of a CCC
relation implies inferring the general rule.
Hypothesis 1 predicts that CCC relations are more complex than positive causal relations
and negative additive relations. To test this prediction, other items were used that contain
Result relations with the marker and therefore and Contrast relations with the marker but.
Examples are:

(16) Result relation (positive causal): sentence completion

Tall people are rast.
Broes is tall and therefore she is . . .
a. rast
b. not rast
(17) Result relation (positive causal): inferring a general rule
Jani has black hair and therefore he is iepen.
a. People with black hair are not iepen.
b. People with black hair are iepen.
(18) Contrast relation (negative additive): sentence completion
Jakki goes to ploef, but Boli goes to . . .
a. pluik
b. ploef

Because the interpretation of the Contrast relation does not imply inferring a general rule (see
Spooren, 1989), there were no Contrast items in the inference task.
The presence (present/absent) and position of a negation in the items (first clause, second
clause), and the position of the nonsense word (first clause/second clause), were systematically
varied. In total, there were 24 items in the completion task, and 16 items in the inference task.
2018 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026

Apart from these 40 nonsense items, there were 20 meaningful items, without a nonsense
word. These items were added to study whether the children were able to perform the task if they
could use real world knowledge. Twelve meaningful items were presented in the completion task,
and eight items in the inference task.

3.1.2. Participants
On the basis of a number of pretests, it was decided to use children from grade 3 as the younger
age group (younger children did not succeed in performing the task) and children from grade 6 as
the older age group. In the experiment, 32 children from grade 3 and 29 children from grade 6
cooperated. The children were recruited from three Dutch elementary schools. The reading level
of the children varied equally between age groups. The younger group varied in age from 8;0 to
9;9 years (mean: 8;7), the older group varied in age from 10;11 to 13;0 years (mean: 11;8). In both
age groups, the number of boys was somewhat larger than the number of girls (grade 3: 17 boys,
grade 6: 16 boys).

3.1.3. Design
The items were presented to the subjects in two blocks. In the first block the nonsense items
were presented, in the second block the regular items. This order was fixed to prevent the subjects
from developing a strategy of using real world knowledge (as induced by the meaningful items)
to respond to the nonsense items.
Each block consisted of two subblocks: a completion task block and an inference task block.
The order of these subblocks varied between lists. The order of the items within each subblock
was randomized, but did not vary between subjects. The factor Relation Type (for completion
task: CCC, Result, Contrast; for inference task: CCC, Result) was repeated with respect to both
subjects and items. The factor Age Group (grade 3, grade 6) was a between-subjects, within-
items factor.

3.1.4. Procedure and data analysis

The experiment was carried out individually. It consisted of 60 two-choice items. The
experiment was preceded by a lengthy instruction to the task. Then the children were presented with
test items until it was clear that they understood the task. After that, the 60 experimental items were
presented to the children. Presentation of the items differed for the two age groups. For the younger
group the test items and the experimental items were presented both orally and in writing, because a
pretest had shown that this age group could not cope with a written-only task. The older group were
given the test items orally and in writing, but the experimental items only in writing.
In the case of sentence completion items the children chose between two possible
continuations, as in example (14). In the case of general inference items the children chose
between two possible rules, see example (15).
A percentage correct responses was calculated per type of relation per child. This was done
separately for nonsense and meaningful items and for the completion task and the inference task.
These percentages were analyzed using Mann–Whitney tests to test differences between age
groups and Wilcoxon matched-pairs tests to test differences within age groups.

3.2. Results

The results are summarized in Table 2.

These data will be discussed separately for the nonsense items and for the meaningful items.
 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026 2019

Table 2
Percentages correct responses as a function of age, task, and relation type (Study 2)
Task Completion Inference
NegAdd PosCau NegCau PosCau NegCau
Items containing nonsense words
Age
8–9 79 82 41 75 25
11–12 93 92 62 93 53
Meaningful items
Age
8–9 88 95 44 95 59
11–12 97 97 78 100 78

3.2.1. Nonsense items

Hypothesis 1 states that CCC relations are more difficult than the other relations. This was found
in both the inference task and the completion task. For the inference task: CCC relations: 39%
correct, positive causal relations: 84% correct (z = 5.56, p < .001; for the completion task: CCC
relations: 51% correct, positive causal relations: 86% correct (difference with CCC-relations:
z = 5.10, p < .001), negative additive relations 86% correct (difference with CCC-relations:
z = 5.53, p < .001).
The second hypothesis states that younger children have more trouble with the CCC relations
than older children. This was also found for both the inference task (grade 3: 25% correct, grade
6: 53% correct, z = 2.66, p < .01) and the completion task (grade 3: 41% correct, grade 6: 62%
correct, z = 2.32, p < .05).

3.2.2. Meaningful items

As in the case of the nonsense items, CCC relations were more difficult than other relations,
both in the inference task and the completion task (inference task: CCC relations: 68% correct,
positive causal relations: 97% correct, z = 5.30, p < .001; completion task: CCC relations: 60%
correct, positive causal relations: 96% correct (difference with CCC relations: z = 4.95, p < .001,
negative additive items: 92% correct (difference with CCC relations: z = 4.71, p < .001).
It is also the case that younger children find the CCC relation more difficult than the older
children (inference task: grade 3: 59% correct, grade 6: 78% correct, z = 2.37, p < .05;
completion task: grade 3: 44% correct, grade 6: 78% correct, z = 3.45, p < .01).

3.3. Discussion
The results show that both age groups can deal with the concept of polarity and the concept of
causality per se. Even the younger of the two age groups performs substantially better than
chance at both negative additive items and positive causal items. The complexity is in the
combination of the two. Items that are both negative AND causal create difficulties. Note that this
effect cannot be attributed to the complexity of the marker of the CCC relation because in
Contrast items and in CCC items the same marker has been used (namely the marker but). Nor
can it be attributed to the syntactic or semantic complexity of the constructions, because in all
items the same syntactic construction was used, namely a coordinate conjunction, and in all items
the presence or absence of a negation was varied systematically. The conclusion must be that it is
the complexity of the coherence relation that creates the complexity. And this is precisely in line
with the predictions following from our classification of coherence relations.
2020 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026

The same conclusion, namely that negative causal relations are more complex than the other
relations in the experiment, can be drawn from the fact that, with age, children learn to interpret the
CCC relation better. This may not be so obvious from the mean scores, because these are all
relatively low, specifically for the nonsense items. But it is certainly obvious if we look at the
individual response patterns of the subjects. For instance there were 13 subjects who gave correct
answers to all or nearly all items. Of these, 12 were from the older group and 1 from the younger
group. By contrast, the 14 children that seemed to answer randomly nearly all came from the
younger age group.
In conclusion, it seems that these data lend support to the thesis that CCC relations are more
complex than the others. This is in line with our cognitive theory of coherence relations which
analyzes CCC relations as most complex.

4. General discussion

The child language literature provides ample support for both the weak and the strong
prediction that follows from the classification of coherence relations as proposed by Sanders et al.
(1992). Additive relations are acquired before causal relations and positive relations are acquired
before negative relations. New empirical data were gathered in the two studies reported in this
paper. Study 1 was an analysis of spoken data in two communicative settings (a descriptive task
and a conversational task). It has shown a strong interaction between Source of Coherence and
Order of the Segments, thus corroborating these basic concepts. The experiment in Study 2 has
shown that negative causal relations are acquired later than negative additive and positive causal
relations, irrespective of syntactic construction and semantic or lexical complexity, in line with
the strong prediction following from the classification. Therefore, these data support a cognitive
interpretation of the classification: Language users make use of their (tacit) knowledge of these
basic concepts when they process language.
Do the results of Study 1 really support the idea of cognitive complexity? After all, older children
do not use complex relations like causals more often than younger children. We believe that this
result is not in conflict with the complexity idea, because even the younger children in our study had
already acquired the concept of causality. The implication is that the utterances of younger children
should be studied to find the kind of differences under discussion. This is corroborated by Evers-
Vermeul’s (2005) analysis of causal connectives produced by very young Dutch children (ages
between 1;6 and 6;0), who predominantly showed the acquisition order additive–temporal–causal.
The ecological validity of the task in Study is admittedly relatively low. The children had to
make inferences on the basis of nonsense words. However, we believe that in combination with
the more naturalistic tasks in Study 1, this somewhat artificial nature of the task is compensated
for. Moreover, the design of Study 2 allows a more precise investigation of questions that
remained unanswered when using natural settings: It taps communicative behavior that does not
manifest itself in more or less spontaneous settings. The converging nature of these different sets
of data allow for the conclusions that we drew.
Nevertheless, sufficient questions for future research remain. Additive relations (or at least the
content additive relations) can be further subdivided into purely additive and additive-temporal
relations. Studying children’s narratives, Berman (1996) found that younger children (3-year
olds) use Hebrew ve (roughly equivalent to English and) as a clause-chaining device, but that
older children (5- and 9-year olds) increasingly use it to express temporal sequentiality. They also
use relatively more other lexical markers of sequentiality (then, afterwards, after that). This
tendency may be expected because temporal sequence is taken to be more complex than purely
 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026 2021

additive relations: interpretation involves the temporality feature. Obviously, there are many
degrees of temporality (overlap, (partial) precedence, (partial) ‘postcedence’), and there are many
(mostly formal semantically inspired) studies of how natural language is sensitive to expressing
these distinctions (see Oversteegen, 1989 for an analysis of Dutch temporal connectives). The order
in which such distinctions are acquired may shed light on the validity of the Sanders et al. (1992)
claim that the temporal distinction is secondary to the additive–causal distinction.
As a methodological point, we see the fact that the analysis of discourse coherence crucially
involves conceptual relations rather than linguistic connections as an urgent issue for future
research. In his paper on the methodology of cognitive linguistics, Langacker (1999) takes as a
starting point for linguistic research the analysis of explicit linguistic elements. This seems to put
the analysis of coherence relations outside of the realm of linguistics, by definition. We believe
that such a move would be precipitate and that linguistic regularities behind coherence relations
can and should be studied within linguistics. But we are also convinced that the analysis of
discourse coherence is in urgent need of reliable and valid analyses of coherence relations. As we
have argued elsewhere (Sanders and Spooren, 2007), the strategy of ‘converging evidence’
followed in this article may be beneficial to that goal.

Acknowledgements

The research reported here would have been impossible without the contributions of Rinda den
Besten, Hanny den Ouden, Kiek Tates, and Janneke Visser. Leo Noordman and Jacqueline Evers-
Vermeul commented on earlier versions of this article. Aleth Bolt corrected our English. We also
like to thank two anonymous reviewers of Journal of Pragmatics for their helpful comments.

Appendix A. Overview of the classification (Sanders et al., 1993)

Overview of the taxonomy and some prototypical RST-relations

Basic Source of Order Polarity Class Relation in Rhetorical
Operation Coherence Structure Theory (RST)
Causal Content Basic Positive 1.a (Non-)volitional cause
1.b Condition
Causal Content Basic Negative 2. Contrast
Causal Content Non-basic Positive 3.a (Non-)volitional result
3.b Condition
Causal Content Non-basic Negative 4. Contrast
Causal Epist/SA Basic Positive 5.a Evaluation
5.b Interpretation
Causal Epist/SA Basic Negative 6. Anti-thesis
Causal Epist/SA Non-basic Positive 7.a Evidence
7.b Motivation
Causal Epist/SA Non-basic Negative 8. Anti-thesis
Additive Content – Positive 9.a Sequence
9.b Joint
Additive Content – Negative 10. Contrast
Additive Epist/SA – Positive 11. Presentational sequence
Additive Epist/SA – Negative 12. Concession
Note. Epist/SA: Epistemic / Speech Act. In the original classifications we distinguished between Semantic and Pragmatic
relations, where the latter category included both Epistemic and Speech Act relations. For reasons of presentation
(consistency with other literature) we have avoided the original terminology in this article.
2022 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026

Below, each class of relations is illustrated with example of a typical RST-relation. Linguistic
markers are in italics.

(1) Non-volitional result

Because there was a heavy storm, the tiles fell from of the roof.
(2) Contrast
Although there was a storm, no tiles have fallen from the roof.
(3) Non-volitional cause
The tiles fell from the roof, because there was a heavy storm. ees.
(4) Contrast
No tiles have fallen from the roof, even though there was a heavy storm.
(5) Interpretation
A lot of tiles have fallen from the roof. There must have a been a heavy storm.
(6) Anti-thesis
Although there was a heavy storm, I am sure not a tile will have fallen from the roof.
(7) Evidence
I am absolute sure there was a heavy storm. All the tiles have fallen from the roof.
(8) Anti-thesis
I am sure not a tile will have fallen from the roof, even though there was a heavy storm.
(9) Sequence
There was a heavy storm last Tuesday. On wednesday we had snow.
(10) Contrast
There was a heavy storm last Tuesday. But the next day we had beautiful sunny weather.
(11) Presentational Sequence
(There was a heavy storm last Tuesday). The tiles have fallen from the roof. Furthermore,
I saw a tree lying on the street.
(12) Concession
Some tiles have fallen from the roof. But I still think the storm was not that bad.

Appendix B. Sample from the corpus (Study 1)

Utterance Classification RST-relation

Grade 1
(a) Die meneer schiet een pijl. En dan klapt de band. ca con ba po CA
‘That man shoots an arrow. And then the tire bursts.’
(b) Die mensen huilen allemaal. Omdat die vioolspeler ca con nba po RE
zo mooi speelt.
‘Those people are all crying. Because that violinist is
playing so beautifully.’
(c) Ze zijn ramen aan het wassen. En dan laat hij zijn ad con na po CI
emmertje vallen.
‘They are cleaning the windows. And then he drops his
little bucket.’
(d) We verliezen soms [met voetbal]. Maar meestal ad con na neg OP
winnen we.
‘Sometimes we lose [at soccer]. But usually we win.’
(e) Ik heb liever een juf. Want een juf is veel liever ca sa nba pos EV
als een meester.
‘I prefer a female teacher. Because a female teacher is
much sweeter than a male teacher.’
(f) Bassie en Adriaan is leuk. Skippy vind ik niet leuk. ad sa na neg OP
‘Bassie and Adrian [a children’s TV program] is nice.
Skippy isn’t nice.’
 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026 2023

Appendix B (Continued )
Utterance Classification RST-relation

Grade 6
(g) Een klein jongetje duwt de vazen van de tafel af. ca con ba pos CA
En daardoor vallen de vazen kapot.
‘A little boy pushes the vases of the table. And because
of that the vases fall to pieces.’
(h) Hier denken de mensen dat er een haai komt. Want ca con nba pos RE
er zijn twee jochies die houden een stok met een vin omhoog.
‘Here the people think that a shark is coming. Because
there are two kids who hold up a stick with a fin.’
(i) Hier heb je een skelet van een dinosaurus. En daar ad con na pos CI
staan allemaal holbewoners bij.
‘Here you have the skeleton of a dinosaure. And a lot of cavemen
are standing close to it.’
(j) Die ene visser heeft een grote boot met allemaal visnetten. ca con ba neg CO
En die vangt helemaal geen vis.
‘That one fisherman has a large boat with all sorts of fishing nets.
And that one isn’t catching any fish.’
(k) Mij maakt het nooit uit [of ik merkkleding draag]. ca sa nba pos CD
Als je maar leuke dingen hebt.
‘To me it never matters [whether I’m wearing brand clothing.
As long as you [i.e. ‘‘I’’] have nice things.’
(l) Ik vind het ook wel spannend [om van school te gaan]. ca ep nba pos EV
Want daar is alles nieuw.
‘I think it is also quite exciting [to leave school]. Because
everything is new there.’
Note. Classification categories: ad: additive; ca: causal; ba: basic order; nba: non-basic order; na: not applicable; con:
content; sa: speech act; ep: epistemic; pos: positive; neg: negative.
Relations: CA: cause; CI: circumstance; OP: opposition; RE: result; CO: contrast; CD: condition; EV: evidence.

Appendix C. Analysis of content, epistemic and speech act relations (Study 1)

The paraphrase criteria sketched below were introduced in Sanders (1997). A relation was
considered a content relation if it passed the following paraphrase test:

(C1) Paraphrase criteria for content relations

a. S1. This fact causes/caused the fact S2.
b. S1. This fact is/was caused by the fact that S2.

The categorization as epistemic or speech act relations was done with a satisfactory amount of
interrater agreement (Cohen’s k = .69, n = 430). Non-content relations were considered epis-
temic if they could be paraphrased with any of the following phrases:

(C2) Paraphrase criteria for epistemic relations

a. S1. And on the basis of this I conclude here and now that S2.
b. I conclude here and now that S1. I do this on the basis of S2.
c. It must be the case that S1. My evidence for that is S2.
d. It probably is the case that S1. My evidence for that is S2.
2024 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026

Another distinguishing feature of epistemic relations is that they involve the generation/inference
of new knowledge (as opposed to the conveying of existing knowledge that occurs frequently in
speech act relations). A typical epistemic relation is:

(C3) [Chantal points to a mummy walking around; in the corner of the picture there is an open
coffin]
a. Daar komt een mummie
‘There comes a mummy’
b. Misschien hebben ze de kist opengedaan
‘Maybe they opened the coffin’
(Chantal 12:6)

The relationship between (C2)a and (C2)b is epistemic because the speaker concludes here and
now that a possible explanation for the mummy walking around is that someone opened the
coffin.
Relationships were categorized as speech act relations if they could be paraphrased as in (C3).

(C3) Paraphrase criteria for speech act relations

a. I say to you, order you, ask you . . . (etc.) S1. The basis for my speech act is
that S2.
b. Given that S1 is the case, I say to you, order you, ask you . . . (etc.) S2.

A typical example of a speech act relation is the following:

(C4) a. En Sesamstraat vind ik niet meer zo leuk

‘And I don’t like Sesame Street very much anymore’
b. Want daar ben ik nu te groot voor.
‘Because I’m too old for that’
(Rose 12:3)

The relation between (C4)a. and b. is a speech act relation because the speaker is telling the
interviewer that she does not like Sesame Street very much anymore, and the basis for this
statement is that she has outgrown the program.

References

Abraham, Werner, 1979. But. Studia Linguistica 33, 89–119.

Asher, Nicholas, Lascarides, Alex, 1998. Bridging. Journal of Semantics 15 (1), 83–113.
Bates, Elisabeth A., 1976. Language in Context: The Acquisition of Pragmatics. Academic Press, New York.
Bebout, Linda J., Segalowitz, Sidney J., White, G.J., 1980. Children’s comprehension of causal constructions with
BECAUSE and SO. Child Development 51, 565–568.
Berman, Ruth A., 1996. Form and function in developing narrative abilities: the case of AND. In: Slobin, D., Gerhardt,
D.J., Kyratzis, A., Guo, J. (Eds.), Social Interaction, Context and Language. Erlbaum, Mawah, NJ, pp. 243–268.
Blakemore, Diana, 1989. Denial and contrast: a Relevance Theoretic analysis of BUT. Linguistics and Philosophy 12, 15–
37.
Bloom, Lois, Capatides, Joanne B., 1987. Sources of meaning in the acquisition of complex syntax: the sample of
causality. Journal of Experimental Child Psychology 43, 112–128.
 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026 2025

Bloom, Lois, Lahey, Margaret, Hood, Lois, Lifter, Karin, Fiess, Kathleen, 1980. Complex sentences: acquisition of
syntactic connectives and the semantic relations they encode. Journal of Child Language 7, 235–261.
Clark, Herb H., Clark, Eve V., 1977. Psychology and Language: an Introduction to Psycholinguistics. Harcourt Brace
Jovanovich, New York.
Corrigan, Roberta L., 1975. A scalogram analysis of the development of the use and comprehension of BECAUSE in
children. Child Development 46, 195–201.
Dancygier, Barbara, Sweetser, Eve E., 2000. Constructions with IF, SINCE and BECAUSE: Causality, epistemic stance,
and clause order. In: Couper-Kuhlen, E., Kortmann, B. (Eds.), CAUSE–CONDITION–CONCESSION–CONTRAST.
Cognitive and Discourse Perspectives. Mouton de Gruyter, Berlin, pp. 111–142.
Den Ouden, Hanny, 1995. De verwerving van de negatief causale relatie [The acquisition of the negative causal relation].
Toegepaste Taalwetenschap in Artikelen 53, 83–94.
Eisenberg, Ann R., 1980. A syntactic, semantic and pragmatic analysis of conjunction. Stanford Papers and Reports on
Child Language Development 19, 70–78.
Evers-Vermeul, Jacqueline, 2005. Connections Between Form and Function of Dutch Connectives. Change and
Acquisition as Windows on Form-Function Relations. PhD Thesis, UiL OTS, Utrecht University.
Evers-Vermeul, Jacqueline, Sanders, Ted. The emergence of Dutch connectives; how cumulative cognitive complexity
explains the order of acquisition, submitted for publication.
French, Lucia A., Nelson, Katherine, 1985. Young Children’s Knowledge of Relational Terms: Some Ifs, Ors and Buts.
Springer Verlag, New York.
Garnham, Alan, Oakhill, Jane (Eds.), 1992. Discourse Representation and Text Processing. A Special Issue of Language
and Cognitive Processes. Lawrence Erlbaum Associates, Hove, UK.
Gernsbacher, Morton A., Givón, Talmy (Eds.), 1995. Coherence in Spontaneous Text. Benjamins, Amsterdam.
Graesser, Arthur C., McNamara, Danielle S., Louwerse, Max M., 2003. What do readers need to learn in order to process
coherence relations in narrative and expository text? In: Sweet, A.P., Snow, C.E. (Eds.), Rethinking Reading
Comprehension. Guilford, New York, pp. 82–98.
Graesser, Arthur C., Millis, Keith K., Zwaan, Rolf, 1997. Discourse comprehension. Annual Review of Psychology 48,
163–189.
Halliday, Michal A.K., Hasan, Ruqaiya, 1976. Cohesion in English. Longman, London.
Handford, Martin, 1991. Waar is Wally? [Where’s Waldo?] M&P Uitgeverij bv., Weert.
Hickmann, Maya, 2003. Children’s Discourse: Person, Space and Time Across Languages. Cambridge University Press,
Cambridge, UK.
Hobbs, Jerry R., 1979. Coherence and coreference. Cognitive Science 3, 67–90.
Hood, Lois, Bloom, Lois, 1979. What, When, and How about Why: A Longitudinal Study of Early Expressions of
Causality. Monographs of the Society for Research in Child Development 44 (Serial No. 181).
Katz, Evelyn W., Brent, Sandor D., 1968. Understanding connectives. Journal of Verbal Learning and Verbal Behavior 7,
501–509.
Kyratzis, Amy, Guo, Jiansheng, Ervin-Tripp, Susan, 1990. Pragmatic conventions influencing children’s use of
causal constructions in natural discourse. In: Hall, K., Koenig, J.P., Meacham, M., Reinman, S., Sutton, L.A.
(Eds.), Proceedings of the 16th Annual Meeting of the Berkeley Linguistics Society, BLS, Berkeley, CA, pp. 205–
214.
Lakoff, Robin, 1971. If’s, and’s, and but’s about conjunction. In: Fillmore, C.J., Langendoen, D.T. (Eds.), Studies in
Linguistic Semantics. Holt, Rhinehart & Winston, New York, pp. 114–149.
Langacker, Ronald, 1999. Assessing the cognitive linguistic enterprise. In: Janssen Th., A.J.M., Redeker, G. (Eds.),
Cognitive Linguistics: Foundations, Scope and Methodology. Mouton de Gruyter, Berlin, Cognitive Linguistics
Research 15., pp. 13–59.
Mann, William C., Thompson, Sandra A., 1988. Rhetorical Structure Theory: toward a functional theory of text
organization. Text 8, 243–281.
McCabe, Allyssa, Peterson, Carole, 1985. A naturalistic study of the production of causal connectives by children. Journal
of Child Language 12, 145–159.
Oversteegen, Eleonore, 1989. Tracking Time, a Proposal for the Representation of Temporal Expressions in Dutch. PhD
Thesis, University of Utrecht.
Oversteegen, Eleonore, 1997. On the pragmatic nature of causal and contrastive concessives. Discourse Processes 24, 51–
86.
Peterson, Carole, 1986. Semantic and pragmatic uses of BUT. Journal of Child Language 13, 583–590.
Peterson, Carole, McCabe, Allyssa, 1987. The connective AND: do older children use it less as they learn connectives?
Journal of Child Language 14, 375–381.
2026 W. Spooren, T. Sanders / Journal of Pragmatics 40 (2008) 2003–2026

Piaget, Jean, 1969. Judgement and Reasoning in the Child. Routledge & Kegan Paul, London. [Translation of: Jean Piaget,
1924. Le jugement et le raisonnement chez l’enfant. Delachaux et Niestlé, Neuchatel, etc.].
Piaget, Jean, 1971. The Language and Thought of the Child. Routledge & Kegan Paul, London. [Translation of: Jean
Piaget, 1923. Le langage et le pensée chez l’enfant. Delachaux et Niestlé, Neuchatel, etc.].
Sanders, Ted, 1997. Semantic and pragmatic sources of coherence. On the categorization of coherence relations in
context. Discourse Processes 24, 119–147.
Sanders, Ted, Pander Maat, Henk, 2006. Cohesion and coherence: Linguistic approaches. In: Brown, K., et al. (Eds.),
Encyclopedia of Language and Linguistics. 2nd ed. Elsevier, London.
Sanders, Ted, Spooren, Wilbert, 2001. Text representation as an interface between language and its users. In: Sanders,
T., Schilperoord, J., Spooren, J. (Eds.), Text Representation: Linguistic and Psycholinguistic Aspects. Benjamins,
Amsterdam, pp. 1–25.
Sanders, Ted, Spooren, Wilbert, 2007. Discourse and text structure. In: Geeraerts, D., Cuykens, H. (Eds.), Handbook of
Cognitive Linguistics. Oxford University Press, Oxford, pp. 916–943.
Sanders, Ted, Spooren, Wilbert, Noordman, Leo, 1992. Toward a taxonomy of coherence relations. Discourse Processes
15, 1–35.
Sanders, Ted, Spooren, Wilbert, Noordman, Leo, 1993. Coherence relations in a cognitive theory of discourse
representation. Cognitive Linguistics 4, 93–134.
Scott, Cheryl M., 1984. Adverbial connectivity in conversations of children 6 to 12. Journal of Child Language 11, 423–
452.
Schiffrin, Deborah, 1986. Functions of and in discourse. Journal of Pragmatics 10, 41–66.
Spooren, Wilbert, 1989. Some Aspects of the Form and Interpretation of Global Contrastive Coherence Relations.
Unpublished Ph.D. Dissertation, K.U. Nijmegen.
Sweetser, Eve E., 1990. From Etymology to Pragmatics. Cambridge University Press, Cambridge.
Taboada, Maite, Mann, William, 2006. Applications of Rhetorical Structure Theory. Discourse Studies 8 (4), 567–588.
Tabor, Whitney, Traugott, Elizabeth C., 1998. Structural scope expansion and grammaticalization. In: Ramat, A.G.,
Hopper, P.J. (Eds.), The Limits of Grammaticalization. Benjamins, Amsterdam, pp. 227–270.
Traugott, Elizabeth C., 1997. Review of Christian Lehmann, thoughts on grammaticalization. Journal of Linguistics 33,
171–211.
Van Dijk, Teun A., 1977. Connectives in text grammar and text logic. In: Van Dijk, T.A., Petöfi, J. (Eds.), Grammars and
Descriptions: Studies in Text Theory and Text Analysis. De Gruyter, Berlin, pp. 11–63.
Verhagen, Arie, 2000. Achter het Nederlands. Over interacties tussen taal en achtergrondcognitie [Beyond Dutch: on
interactions between language and background cognition]. Inaugural Speech, Leiden University.
Wing, Clara S., Scholnick, Ellin K., 1981. Children’s comprehension of pragmatic concepts expressed in BECAUSE,
ALTHOUGH, IF and UNLESS. Journal of Child Language 8, 347–365.

Wilbert Spooren (born 1956, Ph.D. Nijmegen University, 1989) is professor of Language and Communication at
the Faculty of Arts of Vrije Universiteit Amsterdam, The Netherlands. His group teaches language and communication
at several BA- and MA-levels. His research focuses on the relation between genre, discourse structure and text quality.
He has published on these issues nationally and internationally in journals and edited volumes, see www.let.vu.nl/staf/
w.spooren.

Ted Sanders (born 1963, Ph.D. Tilburg University, 1992) is professor of Dutch language use and discourse studies at
Utrecht University, The Netherlands. His group teaches language and communication at several BA- and MA-levels. His
research focuses on discourse structure and coherence, from a linguistic and a psycholinguistic perspective. He is the
(co-)author of many journal publications and the (co-)editor of several edited books and special issues, see
http://www.let.uu.nl/ted.sanders/personal/index.php.