Learning to Read Connections—

Sensitivity to Collocation Frequency

Links Vocabulary Size and Reading
Comprehension in Middle Childhood
Alexandra M. A. Schmitterer ABSTR ACT
DIPF, Leibniz Institute for Research and The collocation frequency of words in the language environment contributes
Information in Education, Frankfurt am to early vocabulary development. Vocabulary size, in turn, predicts children’s
Main, Germany reading comprehension skills later in development. Both collocation frequen-
cy and reading comprehension have been connected to inferential reason-
ing at different time points in development. Here, it was hypothesized that
IDeA Center for Individual Development 8-­year-­old children’s (N = 147; 76 female) sensitivity to collocation frequency
and Adaptive Education, Frankfurt am would be related to vocabulary size and reading comprehension skills of vary-
Main, Germany ing complexity. Participants completed an auditory thematic judgment task
to assess their sensitivity to collocation frequency (response accuracy or
speed). In the task, children were presented with a short sentence containing
Caterina Gawrilow a reference word (e.g., “John sees the cloud.”) and asked to judge which of
IDeA Center for Individual Development two subsequent words best fit the sentence (e.g., “rain” or “lip”). Semantic
and Adaptive Education, Frankfurt am relatedness between reference words and test words was operationalized
Main, Germany in three levels (strong, weak, and distant) based on a corpus-­based analysis
of collocation frequency. Multilevel and mediation analyses confirmed that
thematic judgment responses were related to corpus-­based measures of col-
University of Tübingen, Tübingen,
location frequency and were associated with vocabulary size and reading
Germany comprehension skills at the sentence and text level. Furthermore, thematic
judgment predicted vocabulary size and reading comprehension when the
Claudia Friedrich relation of decoding and reading comprehension was taken into account. The
study highlights sensitivity to collocation frequency as a link between early
University of Tübingen, Tübingen, language comprehension development (i.e., lexical retrieval and inferential
Germany reasoning) and reading comprehension in middle childhood. It also integrates
theoretical approaches from computational network or distributional seman-
tics studies and behavioral experimental studies.

R
eading comprehension plays an important role in everyday life.
The ability to read and comprehend text fluently involves many
complex cognitive processes that develop in early and middle
childhood. Longitudinal studies in multiple alphabetic languages broadly
distinguish two underlying cognitive components: decoding and lan-
guage comprehension (Caravolas et al., 2019; Florit & Cain, 2011; Hjet-
land et al., 2020; Hoover & Tunmer, 2018; Nation, 2019). Together, they
have been shown to explain more than 85% of the variability in reading
comprehension abilities in middle childhood (Chiu, 2018; Lervåg
Reading Research Quarterly, 59(3) et al., 2018; Lonigan et al., 2018).
pp. 448–467 | doi:10.1002/rrq.548
© 2024 The Author(s). Reading Research Quarterly
Decoding skills (i.e., word recognition) are explained by underlying
published by Wiley Periodicals LLC on behalf of phonological and phoneme-­grapheme convergence processes (e.g., pho-
International Literacy Association. This is an open access
article under the terms of the Creative Commons
nological awareness, letter-­sound knowledge, rapid naming; Caravolas
Attribution-NonCommercial License, which permits use, et al., 2019; Moll et al., 2014; Schmitterer & Schroeder, 2019a; Zugarra-
distribution and reproduction in any medium, provided
the original work is properly cited and is not used for
murdi et al., 2022). However, disentangling the cognitive processes
commercial purposes. underlying the relation between early language comprehension

448
 19362722, 2024, 3, Downloaded from https://ila.onlinelibrary.wiley.com/doi/10.1002/rrq.548 by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
development and reading comprehension skills in middle Only few computational studies have studied early
childhood is still considered a complex problem (Castles vocabulary development in other languages than English.
et al., 2018; Duke & Cartwright, 2021). Furthermore, while Here, we introduce studies using network growth model
decoding skills explain more variability in reading com- approaches that have been replicated across multiple lan-
prehension in early reading development, the contribution guages. These studies have investigated learning princi-
of language comprehension increases throughout reading ples that aim to explain early vocabulary development
development (Hjetland et al., 2019, 2020; Hulme (Fourtassi et al., 2020; Hills et al., 2009). In these studies,
et al., 2015). Thus, understanding how early language semantic networks were fitted for 130 high-­frequency
comprehension development relates to reading compre- nouns typically acquired before 30 months of age, and
hension is relevant for reading acquisition in the long vocabulary growth was modeled based on Age-­ of-­
term. In this study, we aim to investigate the role that the Acquisition norms (AoA; MCDI; Dale & Fenson, 1996).
early acquired cognitive architecture of semantic knowl- AoA norms provide information from parent question-
edge based on co-­occurrences might play in the relation naires that qualify the age at which each noun was
between vocabulary size and reading comprehension abil- acquired for at least 50% of children (Braginsky et al.,
ities in middle childhood. 2019). The structure of semantic networks was fitted
using feature-­based semantic relations (McRae et al.,
2005) and association norms (Nelson et al., 1998). In
Lexical Structure, Vocabulary addition, it was investigated whether the strength of
Development and Collocation relatedness between nodes and the number of nodes con-
Frequency nected to each other (i.e., connectivity) were consistent
Computational models of semantic representations with co-­ occurrence statistics from corpora of child-­
assume that lexical knowledge (i.e., vocabulary) is stored directed speech (CHILDES; MacWhinney, 2014; Word-
in networks or vector spaces (Collins & Loftus, 1975; Firth, bank; Frank et al., 2017). The networks followed these
1957; Harris, 1954; Landauer & Dumais, 1997; Mikolov principles: A new word is more likely to be learned if it is
et al., 2013; Pennington et al., 2014; McClelland & Rogers, semantically related to other words. For example, the
2003; Stella et al., 2017; Steyvers & Tenenbaum, 2005; Tur- word “rain” is learned faster if the word “cloud” is already
ney & Pantel, 2010). In such structures, words are repre- known or appears together with “rain” in the child’s envi-
sented by nodes, which are connected by links (or vectors) ronment. Furthermore, a new word is more likely to be
that represent specific relations between the words. An learned if it is semantically related to words that are con-
underlying assumption is that the processing and meaning nected to many other words (i.e., high connectivity and
representation of any given word is specified by the words similar contexts). For example, the word “rain” will be
it relates to. learned more quickly if it is semantically related to a hub
Links between words in these models can be semantic including “cloud,” “thunder,” “mountain,” “lightning,” and
relations based on features (e.g., cucumber—is edible; “water”.
cucumber—is a vegetable, McRae et al., 2005), free asso- Seminal network growth studies by Hills et al. (2009)
ciation norms (e.g., dog—bone; Nelson et al., 1998), or and Fourtassi et al. (2020) found that young children’s lexi-
co-­occurrence statistics (e.g., cloud—rain; Fourtassi cal structure in 10 languages represented associative rather
et al., 2020; Fourtassi, 2020). Because words often co-­ than feature-­based relations. Since association norms have
occur within specific thematic contexts (e.g., clouds and been explained by co-­ occurrence statistics (Four-
rain during a walk), semantic relations based on co-­ tassi, 2020; Lund et al., 1996; Spence & Owens, 1990;
occurrence are also referred to as thematic relations (Mir- Unger et al., 2023), this points to the importance of collo-
man et al., 2017). Particularly in distributional semantic cation frequency in early vocabulary development. In
models (mainly vector-­based), co-­occurrence measures addition, the growth of the networks has been explained
are used as the basic statistical information to specify the by the number of co-­occurring words with the target
models (Flores et al., 2020; Fourtassi, 2020; Pennington nouns (i.e., connectivity) in the language environment
et al., 2014; Unger et al., 2023). The underlying assump- (i.e., preferential acquisition). Thus, in these models words
tion is that words with similar meanings or roles tend to that were linked to many co-­occurring words in child-­
appear in similar linguistic contexts. In this study, we use directed speech were acquired earlier than words that were
the term collocation frequency to highlight that sensitiv- linked to few co-­occurring words. Studies with adults
ity to co-­occurrence measures is connected to the fre- show similar growth effects that highlight the importance
quency with which two words co-­occur (e.g., within 5 of semantic relatedness and connectivity for lexical growth
words, within a sentence; Evert, 2008; Unger et al., 2023). (Steyvers & Tenenbaum, 2005). Based on this evidence for
In this study, we hypothesize that sensitivity to colloca- lexical growth in early childhood and adulthood, we
tion frequency in the input language links vocabulary size assume that for middle childhood, computational models
and reading comprehension. would show that collocation frequency in the language

Learning to Read Connections | 449

 19362722, 2024, 3, Downloaded from https://ila.onlinelibrary.wiley.com/doi/10.1002/rrq.548 by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
environment and the structure of children’s individual lex- relatedness can also be inhibitory (see, e.g., Friedrich
ica contribute to vocabulary growth. et al., 2013). For example, in a triad paradigm, when par-
The assumption that semantic connectivity based on ticipants must decide which of two semantic relations is
collocation frequency plays an important role in early stronger (i.e., a semantic matching process), the decision
vocabulary development is corroborated by computa- will be more difficult when two relations are similar in
tional studies focusing on interindividual differences in strength than when they are dissimilar in strength.
early vocabulary development. Beckage et al. (2011) con- In line with the findings of computational studies,
ducted semantic network analyses based on AoA data behavioral studies suggest that adults, toddlers, younger
from typically developing children and children with sig- children, and older children are sensitive to semantic
nificant delays in vocabulary development (late talkers). relations between words (Lany & Saffran, 2013; Savic
Individual networks were constructed by linking words in et al., 2023; Unger et al., 2023). Semantic relations in this
each child’s vocabulary to co-­occurrence statistics from context include relatedness based on association norms
child-­directed speech (CHILDES; MacWhinney, 2014). or co-­ occurrence measures (Arias-­ Trejo & Plun-
For each individual network, various measures of con- kett, 2013; Buchanan et al., 2001; Landrigan & Mir-
nectivity were computed, and mean connectivity was man, 2018; Mirman et al., 2017; Savic et al., 2022, 2023;
compared between typically developing children and late Unger et al., 2020; Wojcik & Kandadhai, 2020). For
talkers. Results indicated that, while controlling for example, in a triad paradigm, Vales and Fisher (2019)
vocabulary size, late talkers’ semantic networks had lower asked 3-­to 9-­year-­old children to judge whether a panel
connectivity and greater distances between nodes than of three pictures contained a reference to a spoken target
typically developing children’s semantic networks. The (e.g., “bone”). In addition to the target picture, panels
increased network connectivity of typically developing included either two unrelated distractors (e.g., a flower
children compared to late talkers may indicate either a and a broom) or one unrelated and one related distractor
higher sensitivity to the connectivity of co-­ occurring (e.g., a flower and a puppy). The target-­distractor associ-
words in their language environment or higher connec- ation was based on adult association ratings. Children
tivity in their own lexicon. Taken together, the results of took longer to decide whether the target was part of the
these studies indicate that children’s word representations triad when a related word was included. The inhibitory
are structured according to collocation frequency and effect varied as a function of the strength of the related-
children may use their sensitivity to thematic connectiv- ness between the target and the related distractor: The
ity to increase vocabulary size. more related the target and distractor were, the longer it
took children to make their judgment. Thus, because
children had to inhibit the distractor to make their judg-
Sensitivity to Collocation Frequency ment, an inhibitory effect was observed. In the context of
across Development semantic networks, this is explained by the path length
Sensitivity to thematic relations has been investigated in between nodes: The shorter the path length, the stronger
behavioral studies in childhood and adulthood (Lany & the relatedness, the more effortful the decision, and the
Saffran, 2013; Savic et al., 2022, 2023; Unger et al., 2023; stronger the inhibitory effect.
Wojcik & Kandhadai, 2020). At the behavioral level, cogni- Sensitivity to thematic relatedness in children was also
tive structures are often investigated using priming para- found in a longitudinal study (Schmitterer & Schro-
digms or tasks that tap into semantic matching processes eder, 2019b) with a similar design. Notably, this study used
(e.g., triad paradigms). In priming paradigms, participants co-­occurrence measures from a corpus of child-­directed
are often asked to decide whether a word exists or not (i.e., literature to operationalize semantic relatedness (Schro-
lexical decision). In triad paradigms, participants are pre- eder et al., 2015). In their thematic judgment task, 4-­to
sented with three words and asked to judge the difference 7-­year-­olds were presented with a sentence containing a
between word relations or to identify a target word among target word (e.g., “Jan is looking at the cloud.”). Children
all three words. Stimuli in both paradigms include unre- were then presented with two words that were either
lated control items and primes or distractors that are strongly (strong: “rain”), weakly (weak: “mountain”), or
semantically related to the target. Judgment latencies in distantly related (control: “lip”) to the target word (“cloud”).
semantic priming paradigms are typically faster on primed The children were asked to judge which word would better
trials than on control trials. This is explained by the paral- fit the sentence. The compared relations were either close
lel activation of related items during lexical retrieval to each other in strength of relatedness (strong-­ weak:
(McClelland & Rogers, 2003; Rogers & McClelland, 2004) “cloud-­rain” vs. “cloud-­mountain”) or distant to each other
In priming paradigms, a related prime is presented just in strength of relatedness (strong-­distant: “cloud-­rain” vs.
before the target. Thus, the prime co-­activates the target “cloud-­ lip”). Across development, children were more
and facilitates subsequent retrieval. However, in task likely to choose the most strongly related word (“rain”) in
designs where distractors must be inhibited, the effects of the condition where the relatedness distance between the

450 | Reading Research Quarterly, 59(3)

 19362722, 2024, 3, Downloaded from https://ila.onlinelibrary.wiley.com/doi/10.1002/rrq.548 by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
word pairs was greater than in the condition where the In consequence, this could indicate less connectivity or
relatedness strength was closer. Furthermore, because the fewer entries in the lexica of children with reading com-
material in the study was based on a corpus of child-­ prehension difficulties.
directed literature rather than child-­directed speech, the Nation and Snowling’s (1999) study shows an associa-
results suggest a relation between sensitivity to collocation tion between reading comprehension difficulties at the
frequency in verbal input and exposure to written sentence or text level and reduced sensitivity to the con-
language. nectivity of thematic relations at the word level. However,
even when considering these findings it remains unclear
Sensitivity to Collocation Frequency whether thematic judgment would predict reading com-
prehension at the sentence or text level in addition to other
and Reading Comprehension semantic measures. For example, because measures of
The suggestion that thematic judgments are related to vocabulary size and thematic judgments should be related,
reading comprehension was further supported by an addi- and both measure lexical retrieval at the word level, the-
tional analysis of the longitudinal study by Schmitterer matic judgments may not add to the well-­documented
and Schroeder (2019b): thematic judgments in 4-­and relation between vocabulary size and reading comprehen-
5-­year-­olds predicted early reading comprehension skills sion (Duke & Cartwright, 2021; Hjetland et al., 2020;
(i.e., word-­ picture matching) in the same children Lervåg et al., 2018; Verhoeven et al., 2011). However, there
14 months later at the end of first grade. Responses on the is another cognitive process that may link complex reading
thematic judgment task predicted children’s reading abili- comprehension skills and thematic judgment: inferential
ties beyond children’s results in phonological working reasoning.
memory, letter-­sound knowledge, and syntactic compre-
hension tasks.
Sensitivity to Collocation Frequency,
To date, a relation between thematic judgment and
reading has only been found for beginning word-­level Reading Comprehension, and Inferential
reading skills (i.e., word-­ picture matching). However, Reasoning
reading theories suggest that there may also be a link to Inferential reasoning has been found to be relevant to
more complex reading comprehension abilities. For exam- reading comprehension abilities (Duke & Cartwright,
ple, the reading system framework (Perfetti & Sta- 2021; Elleman, 2017) and, in early development, has been
fura, 2014), which posits cascading processing, argues that linked to collocation frequency (Fisher et al., 2011). In
lexical processing at the word level is a pressure point for reading, an example of inferential reasoning is the ability
more complex comprehension processes at the sentence or to perceive and use indicators of local coherence within or
text level. Thus, if lexical processes at the word level, such between sentences to infer syntactic or semantic relations
as the retrieval of semantic information from the lexicon, (Ahmed et al., 2016; Kintsch, 1998). One example would
are impaired, higher-­level processes at the sentence and be identifying referents (e.g., animal in boat) to a main
text level will also be impaired. character (e.g., tiger) throughout a text or sentence (e.g.,
This link between sensitivity to collocation frequency Life of Pi). Co-­occurrences and a sensitivity thereof might
at the word-­level and reading comprehension skills is aid the ability to identify referents that are associated. If
illustrated by a study focusing on interindividual differ- poor comprehenders have less entries or less connectivity
ences in reading comprehension in 10-­year-­old children in their individual lexica, they might also be less likely to
(Nation & Snowling, 1999). Typically developing readers use statistical information from co-­occurrences to infer
and children with a specific reading comprehension defi- semantic relatedness in reading.
cit (i.e., poor comprehenders) were presented with In young children, sensitivity to collocation frequency
recordings of strongly associated prime-­target pairs (e.g., has been linked to inferential reasoning (Fisher et al., 2011;
“dog”—“cat”) or weakly associated pairs (e.g., “airplane”— Godwin et al., 2013; Matlen et al., 2015). For example,
“train”). Children were asked to decide whether the tar- Fisher et al. (2011) asked 4-­year-­olds to judge whether
get was a real word or not (i.e., lexical decision task). word pairs share features. Half of the word pairs regularly
Associations were based on co-­occurrence measures of occurred together in child-­directed speech (e.g., “bunny-­
child-­directed literature (Carroll et al., 1971). Children’s rabbit”), and half of them did not (e.g., “alligator-­
reaction times were faster in conditions with strong asso- crocodile”). Children were more likely to recognize that
ciations than in conditions with weak associations. For the word pairs shared common features when the words
poor comprehenders, the facilitatory effect of association also co-­occurred regularly. This suggests that sensitivity to
strength was weaker than for children without reading collocation frequency aids inferential reasoning. However,
difficulties. This suggests that children with reading whether inferential reasoning as indicated by thematic
comprehension difficulties may co-­ activate fewer judgment responses is related to higher-­level reading skills
strongly related items than typically developing children. in middle childhood remains to be investigated.

Learning to Read Connections | 451

 19362722, 2024, 3, Downloaded from https://ila.onlinelibrary.wiley.com/doi/10.1002/rrq.548 by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
This Study thematic judgment and vocabulary size and may weaken
In this study, we focused on the relation between thematic either contribution to reading comprehension. Moreover,
judgment, vocabulary size, and various reading comprehen- recent reading models treat vocabulary as a bridging pro-
sion skills in a group of older children. We collected data cess between decoding and language comprehension skills
from 147 German speaking third-­grade students. Our sam- (i.e., inferential reasoning; Duke & Cartwright, 2021).
ple was diverse regarding their reading comprehension abili- Therefore, if the results of the first model variant would
ties. We obtained teacher ratings for about half of the sample. not hold, we sought to explore a model in which vocabu-
These ratings indicated that teachers would definitely or lary mediates the relation between thematic judgment and
potentially recommend reading intervention for approxi- reading comprehension. This would indicate that a com-
mately 68% of the participants. We anticipated a high degree ponent representing inferential reasoning within thematic
of variability in decoding and comprehension skills that judgment would directly contribute to reading compre-
would allow us to identify underlying cognitive processes in hension and would predict vocabulary size.
typically developing and struggling readers. Regarding the first research question, we expected that
We used an adapted version of Schmitterer and Schro- children would be sensitive to differences in the strength
eder’s (2019b) thematic judgment task. The item selection of thematic relations between word pairs and this sensitiv-
process is shown in Figure 1. First, we analyzed whether ity to differences would be connected to collocation fre-
responses to this task measured children’s ability to dis- quency. Our specific hypotheses were as follows:
criminate between different strengths of thematic related- 1a Children will be more likely to identify a word pair as a
ness in our item set. Second, we examined the relation “better fit” than a distractor word pair (mean accuracy) when
between interindividual differences in thematic judgment the difference in collocation frequency between the contrasting
responses and interindividual differences in vocabulary word pairs in the input language is large versus small.
size, and reading comprehension at the word, sentence, 1b Children are more likely to show fast accurate responses
and text level. Third, we analyzed whether thematic judg- (mean latency of accurate decisions) when the maximum col-
ment mediated the relation between vocabulary size and location frequency between the contrasting word pairs is high
complex reading comprehension abilities. Thus, thematic (i.e., short path) versus low (i.e., long path).
judgment would represent a link between word-­level lexi- Second, we expected to find a relation between overall
cal retrieval and inferential reasoning required for higher-­ thematic judgment accuracy and vocabulary size. In addi-
level comprehension skills.In addition, we examined tion, we examined whether there would be a relation
whether these results hold, when we add decoding. With between accuracy or speed of thematic judgment
this approach, we aimed to investigate whether thematic responses and various reading skills. Our specific hypoth-
judgment would emerge as a relevant mediator in a com- eses were as follows:
mon modeling approach in reading research (i.e., Hjetland
2a Children with large vocabulary sizes will be more likely to
et al., 2020). Furthermore, with our model we explored the
respond more accurately in the thematic judgment task overall
relation of decoding to vocabulary size and thematic judg- than children with small vocabulary sizes.
ment: Decoding relies strongly on phonological processes
which also play an important role in the built up of seman- 2b We examine whether children with good decoding or read-
tic networks in early vocabulary development (Fourtassi ing comprehension skills at the word, sentence, or text level
et al., 2020). Thus, decoding may be related to both respond overall faster in accurate responses or more accurately

FIGURE 1
Schematic Description of Stimuli Selection in Thematic Judgment Task. Note. Co-­occurrence measure = t-­score
(Evert, 2008); The higher the t-­score, the shorter the path length.

452 | Reading Research Quarterly, 59(3)

 19362722, 2024, 3, Downloaded from https://ila.onlinelibrary.wiley.com/doi/10.1002/rrq.548 by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
in the thematic judgment task than children with poor decod- of up to eight students for individual testing. In addition,
ing or reading comprehension skills. we asked them to rate the need for reading intervention for
each student in three levels: “not necessary”, “potentially
Finally, we conducted a mediation analysis to deter-
necessary”, or “definitely necessary”. Teachers reported
mine whether thematic judgment responses would medi-
that in their ratings they orientated to the largest extent on
ate the relation between vocabulary size and reading
their students’ basic reading skills (i.e., decoding) and
comprehension. In addition, we examined whether these
advanced reading skills (i.e., reading comprehension on a
results hold, when we add decoding. If the results would
text level). They reported to orientate to a lesser extent on
not hold, we aimed to explore a model where vocabulary
other linguistic prerequisites (i.e., phonological awareness
mediates the relation between thematic judgment and
or grammatical knowledge) or vocabulary knowledge (see
reading comprehension. Our hypotheses were as follows:
Schmitterer & Brod, 2021, table 5). In addition, empirical
3a Children’s vocabulary size (X) predicts children’s thematic data suggest that teachers’ ratings were most strongly pre-
judgment responses (M), which in turn predicts reading com-
prehension skills (Y).
dicted by students’ spelling skills followed by reading com-
prehension on a text-­level (Schmitterer & Brod, 2021).
3b The effects in (3a) will remain stable when we add decoding For the 162 students, we obtained teacher judgments
ability as a covariate predicting reading comprehension (W). for 52% of the students (n = 83). For these students, the
teachers’ judgments were that 68% definitely (n = 26) or
3c Depending on the results, we may explore a variant where potentially (n = 30) needed a reading intervention. From
thematic judgment (X) predicts vocabulary size (M), which in the 162 participants, 15 participants were excluded because
turn predicts reading comprehension skills (Y). they scored below the chance level on thematic judgment
responses (n = 6), scored below clinical criteria for phono-
logical working memory and nonverbal intelligence
(n = 4), had a clinical condition that affected their responses
Method (i.e., hearing impairment, ADHD, n = 2), or exhibited
avoidant behavior during group or individual testing,
Participants resulting in data loss (n = 3).
The data analyzed in this study were part of the iLearn pre-­
and post-­test study (see Schmitterer & Brod, 2021). For
this study, a cross-­sectional subsample of children who Thematic Judgment Task Assessing
participated at the beginning of the school year (in 2019) Sensitivity to Collocation Frequency
was used. Testing took place in group and individual ses-
sions at the children’s schools and was conducted by Design and Procedure
trained student experimenters. The children and a legal The design, stimulus selection, and procedure of the
guardian consented to participate in the study. The study thematic judgment experiment are described in more
was approved by an institutional review board and the eth- detail in the supplementary material and in the original
ics committee of the authorized ministry of education study (Schmitterer & Schroeder, 2019b). The thematic
prior to data collection. judgment task used in this study was an adapted task. For
We initially included data from 162 children from 17 this study, we excluded homonymous reference words (i.e.,
primary schools and 29 classes from the region of Hesse in words with multiple meanings) and added a third condi-
Germany. The mean age of the children was 8 years and tion: strong-­distant [originally: unrelated], strong-­weak
10 months (SD = 6.5 months), 50.6% were female and [originally: association], and weak-­distant. The task con-
49.4% were male (no other gender was reported). The sisted of 15 context sentences that followed a subject-­verb-­
children came from different socio-­ economic back- object structure. The object of the sentence was one of 15
grounds (MHISEI = 54.79, SDHISEI = 16.75, rangeHISEI = 17–89; preselected high-­ frequency reference words. Based on
Ganzeboom, 2010). Fifty-­seven percent of the participants these 15 reference words, we computed a co-­occurrence
spoke German as their first language, 17% spoke German frequency measure in a corpus of child-­directed literature
as one of their first languages, and 26% spoke German as to determine the relationship between the reference word
their second language. Parents of about 37% of the chil- and the test words in the language input at three levels of
dren reported that their children had an impairment (e.g., relationship strength: strong, weak, or distant (Schroeder
visual impairment, mental disorders). However, parents of et al., 2015; t-­score; Evert, 2008). Sentence conditions and
only 10% of these children (n = 6) reported that their chil- verbs were manipulated based on t-­score (see descriptions
dren did not receive any form of intervention. in supplementary material; Table S1; Figure 1; Figure 2,
The thematic judgment response was administered in panels b and d) and controlled for lemma frequency, num-
individual sessions. As part of the project, teachers were ber of phonemes, phonological neighborhood density, and
asked to select students with reading difficulties and a total taxonomic relatedness (Tables S2 & S3). Familiarity was

Learning to Read Connections | 453

 19362722, 2024, 3, Downloaded from https://ila.onlinelibrary.wiley.com/doi/10.1002/rrq.548 by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
FIGURE 2
Between Condition Effects of Participant Responses and Stimuli Statistics. Note. Panels (a) and (c) show responses
of participants (N = 147), panels (b) and (d) show between condition effects of thematic relatedness measures per
reference word (N = 15); arrows represent standard errors.

controlled for by parent AoA ratings in the original design. Stimuli presentation within blocks was random.
study.The experiment followed a mixed stimulus-­within-­ Each block was followed by a short break. Prior to analy-
condition design. Each child responded to all stimuli in ses, we conducted power analyses (Westfall et al., 2014) on
each condition (45 stimuli; fully crossed). Each target sen- stimuli-­within-­condition and fully crossed designs with an
tence was presented three times (once for each condition) estimated 100 participants and 45 stimuli (d = .05) and
to each child (see Figure S1). Items were nested within tar- found power estimates greater than 0.8.
get sentences and conditions (stimuli-­within-­condition).
Participants were instructed as follows: “Soon you will Standardized Assessments
hear a sentence followed by two words. Please judge which
of the two words fits the sentence better. If you think the Vocabulary Size
first word fits the sentence better, press 1. If you think the Participants were asked to produce 40 nouns, verbs, or
second word fits the sentence better, press 2.” The children adjectives based on pictures on a computer screen (Worts-
indicated their judgments by pressing the appropriate key chatz-­und Wortfindungstest [WWT]; Glück & Glück,
on a laptop. In total, there were 90 pairs (45 counterbal- 2011). The raw score was the sum of all correct responses.
anced pairs) organized into six lists of 15 trials, each con- The test provided a list of alternative correct responses for
taining each target sentence once, of which three lists (= each item, derived from the normative sample, to increase
blocks) were presented to each child in a Latin-­square the objectivity of assessment. Criteria for item selection

454 | Reading Research Quarterly, 59(3)

 19362722, 2024, 3, Downloaded from https://ila.onlinelibrary.wiley.com/doi/10.1002/rrq.548 by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
were word length, word frequency, parts of speech, and Reading Comprehension
semantic category (categories derived from Dorn- Reading comprehension was assessed on the word, sen-
seiff, 2020). Semantic categories included nature, body tence, and text level (ELFE-­II, Lenhard et al., 2017). All
parts (including health and exercise), school, leisure, three subtests were speeded (2× 6 min, 1× 7 min). For the
kitchen (including food and cooking), and social interac- word level, children had to choose the matching word to a
tion. In total, items from 15 semantic categories were picture from a list of five words. For the sentence level,
included to index vocabulary breadth. Internal consistency children had to fill a gap in a sentence by choosing the cor-
was good (Cronbach’s α = .87). Convergent validity was rect word from a list of five words. For the text level, chil-
reported with a receptive vocabulary subtest in the same dren were presented with short text passages describing
normative sample (r = .72). In our sample, expressive familiar situations (e.g., play date) or natural phenomena
vocabulary also had a strong correlation with a receptive (e.g., diamonds made of coal). Children answered multiple
test of syntactic comprehension (r = .59; TSVK; Siegmüller choice questions about the content of each text passage.
et al., 2011). The children’s standardized scores on expres- The raw score for all three tests was the sum of correct
sive vocabulary were low (see Table 1). We assume that this responses. Split-­ half reliability for each subtest was
is explained by differences between the normative sample reported to be acceptable to high (word level: r = .98; sen-
(monolingual only, different region) and our sample. The tence level r = .97; text level: r = .87).
distribution of raw scores was acceptable (see Figure S3).

Decoding Covariates
Participants were asked to correctly read pseudowords Phonological Working Memory
(1 min) from a list of 156 items (Salzburger Lese-­und Participants completed a digit span task (Arbe-
Rechtschreibtest [SLRT-­II]; Moll & Landerl, 2010). The itsgedächtnistestbatterie [AGTB]; von Hasselhorn
raw score was the number of correctly read pseudowords. et al., 2012) consisting of eight adaptive test trials: they
Parallel test reliability was reported as r = .90 for third would advance to the next level after correctly answering
grade. Standardized scores were reported for the middle of two consecutive items with the same digit span. The raw
third grade. score was calculated as the mean of the last two correctly

TABLE 1
Descriptive Statistics of Experimental and Standardized Assessments
Variable M SD Range Max
Thematic judgment accuracy (%) 81.32 9.06 56–98 43a

Thematic judgment latency (ms) 2465 1151.74 451–6000 —

Vocabulary size 16.75 8.05 1–33 40

32.57 19.5 0–62

Decoding of pseudowords 40.55 14.97 8–88 156b

41.48 8.85 27–66

Reading comprehension word level 34.21 9.27 10–58 75b

45.92 8.18 25–66

Reading comprehension sentence level 12.11 4.98 0–24 36b

43.11 8.14 25–62

Reading comprehension text level 6.69 3.76 0–17 26b

41.84 8.89 25–63

Phonological working memory 3.91 0.62 2.4–5.5 —

46.16 8.6 29–67

Note. N = 147; for standardized assessments the first row depicts raw scores, the second row depicts standardized scores (T).
a
Two items were excluded from the analysis.
b
Speeded tests.

Learning to Read Connections | 455

 19362722, 2024, 3, Downloaded from https://ila.onlinelibrary.wiley.com/doi/10.1002/rrq.548 by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
answered span sizes. The internal consistency was good standardized, and conditions were dummy-­coded with the
(Cronbach’s α = .68–.78, depending on age). strong-­distant condition as the reference level.
To validate our procedure, we then ran models with
experimental lists, presentation blocks, or the thematic
Analyses relation between the verb and the reference word in the
All analyses were carried out in RStudio (2022.12.0) using target sentence, as a fixed effect (see code r1 on OSF). The
R 4.1.0 (R Core Team, 2021). Codebooks, code, data, and results showed a list effect for response accuracy models
supplementary material with additional descriptions of but not for response latency models. However, because
the material, tables, and figures are available on OSF (link these list effects were not systematic, we concluded that
provided in Data Availability Statement). this effect was random. There was also a presentation
block effect, with children responding significantly more
Thematic Judgment Experiment accurately and faster during the first presentation block
than during the last presentation block. Since the thematic
Preprocessing judgment experiment was the last task of the individual
In the experimental analysis, affirmative responses to the session and we assume that the children knew this, this
stronger related words are referred to as “accurate” could have influenced their concentration. Therefore, we
responses. Participants’ accurate and inaccurate responses controlled for block effects in the final models. Finally, we
were distributed across all items. We identified six partici- did not find any effects of the verb-­reference word rela-
pants and two items that were outliers regarding response tion, suggesting that our target sentences provided plausi-
accuracy (below chance). The reported results are without ble but not idiomatic contexts for the reference word.
outliers. The final models for hypotheses (1a) and (1b),
Latencies of inaccurate responses were significantly included condition and phonological working memory as
longer than latencies of accurate responses (see Figure S2). fixed effects. The final model structure was as follows: (g)
Inaccurate responses provided too few data points to draw lmer (av ~ condition + ph. working memory + block + (1 |
significant conclusions from (e.g., Fazio, 1990). Thus, we participant) + (1 | reference word) + (1 | item) + (1|class),
only included latencies of correct responses (80.7%). For data). To check for multicollinearity, we calculated vari-
accurate responses, we excluded latencies below 300 and ance inflation factors for each model using the {vif} func-
above 8000 ms. In addition, we excluded latencies that tion from the car package (Fox & Weisberg, 2019). All
deviated more than 2.5 SD from the log-­transformed par- variance inflation factors in these and all subsequent anal-
ticipant or item mean. In total, appr. 1.3% of data points yses were less than 2. Intra-­class correlation coefficients
for correct responses were discarded. (ICC), the explained variability for each random effect and
conditional and marginal R2 for each model were calcu-
Analyses lated using the {tab_model} function from the sjPlot pack-
For hypotheses (1a) and (1b), we analyzed between-­ age (Lüdecke et al., 2023). Post hoc analyses were
condition effects (i.e., strong-­distant vs. strong-­weak vs. conducted by applying single-­degree-­of-­freedom contrasts
weak-­distant pairs, see Figure 1) in response accuracy and based on the cell mean estimates in separate models with
latency with (generalized) linear mixed-­ effects models the intercept set to zero and using the same parameters as
using the {(g)lmer} function from the lme4 package (Bates in the base models.
et al., 2015). The response accuracy model was based on a To validate the relation between our results and the
binomial error distribution and used a logit link transfor- collocation frequency measures, we ran models in which
mation (Cohen et al., 2013). Response latencies were log-­ we replaced the condition factor with the t-­score differ-
transformed prior to analysis. Intercepts for participants ence and the t-­score maximum value. We entered both
and items were included as crossed random effects. t-­score measures into both models to see whether the
Because item structure was crossed (target sentences measure we hypothesized (hypotheses 1a and 1b) would
crossed with participants) and nested (items nested within predict the respective thematic judgment response over
target sentences), we included target sentences as an addi- the other measure for the respective response type. We
tional random effect. To account for contextual informa- assumed that the t-­score difference measure would be a
tion and because participants were nested in classes, we predictor of response accuracy because, in the response
also included class as an additional random effect. We accuracy model, all responses (correct or incorrect) were
conducted an a priori analysis with null models that entered into the model. Thus, the responses represented
included the condition as a singular fixed effect using dif- children’s sensitivity to the relation between the reference
ferent random-­effect variations and compared model fit word and the test words. In the latency analysis, only cor-
with the {anova} function from the car package (Fox & rect responses were included, and responses represented
Weisberg, 2019). We chose one of the most parsimonious children’s speed in choosing the more strongly related
random effect structures. All continuous fixed effects were word. Therefore, we assumed that the t-­score maximum

456 | Reading Research Quarterly, 59(3)

 19362722, 2024, 3, Downloaded from https://ila.onlinelibrary.wiley.com/doi/10.1002/rrq.548 by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
would be the stronger predictor here. The results are sum- For hypotheses (2a) and (2b) we added vocabulary
marized in Table 2. Figure 2 highlights the relation between size, decoding, and several reading comprehension tasks
collocation measures and behavioral responses per of varying complexity as fixed effect in the model. The
condition. final model structure was: (g)lmer (av ~ condition +
TABLE 2
Model Summaries for Effects of Condition and Collocation Statistics
Accuracy Latency
Fixed effects Odds ratio SE p Estimate SE p
Intercept 7.92 1.51 <.001 2132.15 63.34 <.001

Condition effects

Strong—Weak 0.58 0.14 .020 1.01 0.02 .684

Weak—Distant 0.55 0.13 .010 1.18 0.03 <.001

Main effects covariates

Ph. working memory 1.18 0.06 .002 1.01 0.02 .532

Presentation block 0.88 0.03 <.001 0.94 0.01 <.001

Random effects

σ2 3.290 0.146

τ00 participants 0.229 0.046

τ00 items 0.329 0.003

τ00 class 0.007 0.002

τ00 reference words 0.111 0.003

ICC 0.171 0.271

Marginal R /Conditional R
2 2
0.029/0.195 0.045/0.304

Fixed effects Odds ratio SE p Estimate SE p

Intercept 5.47 0.73 <.001 2255.83 63.93 <.001

Effects of collocation statistics

t-­score difference 1.52 0.29 .029 1.01 0.02 .795

t-­score maximum 0.80 0.15 .232 0.92 0.02 <.001

Main effects covariates

Ph. working memory 1.18 0.06 .002 1.01 0.02 .532

Presentation block 0.88 0.03 <.001 0.94 0.01 <.001

Random effects

σ2 3.290 0.146

τ00 participants 0.229 0.046

τ00 items 0.355 0.004

τ00 class 0.007 0.002

τ00 reference words 0.098 0.004

ICC 0.173 0.278

Marginal R2/conditional R2 0.027/ 0.195 0.044/0.310

Note. Accuracy models: 6321 observations; Latency models: 4673 observations; all models: 147 participants, 29 classes, 43 items (trials) and 15
reference words. Bold values represent significant effects.

Learning to Read Connections | 457

 19362722, 2024, 3, Downloaded from https://ila.onlinelibrary.wiley.com/doi/10.1002/rrq.548 by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
vocabulary size/decoding/reading comprehension + ph. et al., 2011) and conducted the analyses for different sam-
working memory + (1 | participant) + (1 | reference ple sizes (100, 120, 150, 162). The results showed power
word) + (1 | item) + (1| class), data). We again checked for estimates of above .8 for all effects of interest when the
unobtrusive multicollinearity. Furthermore, we used the sample size was <= 120. We did not calculate the media-
{anova} function from the car package (Fox & Weis- tion analysis as multilevel models. The explained variabil-
berg, 2019) to compare model fit between the base models ity at the class level for thematic judgment was very small
(Table 2) and the model with the added fixed effects to (less than 2%) which contributed to convergence issues in
determine whether a significant amount of additional multi-­level models.
variance was explained by the added variable. Results are
reported in Tables S5–S7 and shown in Figure 3.
Results
Mediation Analysis
Mediation analyses were conducted using the lavaan pack- Thematic Judgment Task in Middle
age (Rosseel et al., 2017). Prior to our analyses, we con- Childhood
ducted a power analysis for two models on the pwrSEM For hypothesis (1a), we hypothesized that children would
application (Wang & Rhemtulla, 2021). In the first model, be more likely to identify a word pair as “better matched”
we tested whether thematic judgment (M) mediated the (mean accuracy) than a distractor word pair, if the differ-
relation between vocabulary size (X) and reading compre- ence in collocation frequency between the contrasting
hension (Y). In the second model, we added a direct effect word pairs in the input language is large (i.e., low competi-
from decoding (W) on reading comprehension. In an tion with the distractor) versus small (i.e., high competi-
additional model, we explored a variant, where vocabulary tion with the distractor). Therefore, the difference in
size mediated the relation between thematic judgment and thematic relations (i.e., the t-­score difference) between
reading comprehension. We selected the smallest effect conditions is the strongest indicator of our hypothesis. The
sizes of interest (a = .35; b = .35, c = .35, and c’ = .3; factor results were consistent with this hypothesis. Children
loadings = .8; Lakens et al., 2018) based on effects reported responded more accurately in the strong-­distant condition
in previous literature (e.g., Hjetland et al., 2020; Verhoeven (M = 86%, SE = 1%) compared to the weak-­ distant

FIGURE 3
Effects of Vocabulary Size and Reading Abilities on Thematic Judgment Responses. Note. Plot based on raw data;
arrows represent standard errors.

458 | Reading Research Quarterly, 59(3)

 19362722, 2024, 3, Downloaded from https://ila.onlinelibrary.wiley.com/doi/10.1002/rrq.548 by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
condition (M = 79%, SE = 1%, Δ = 7%, β = .60, SE = .23, task. The same was true for children’s reading comprehen-
t = 2.59, p < .01) and in the strong-­distant condition com- sion on sentence, and text level, as well as in tendency on
pared to the strong-­weak condition (M = 79%, SE = 1% the word level. Among reading comprehension abilities,
Δ = 7%, β = .54, SE = .23, t = 2.32, p < .05). However, the size of the effect increased with task complexity. Thus,
response accuracy was not significantly different between we were able to affirm hypotheses (2a) and refine assump-
the strong-­ weak and the weak-­ distant conditions tions regarding hypothesis (2b).
(Δ = 0.1%, β = .06, SE = .23, t = .26, p > .05). This was a simi-
lar pattern to the between-­condition effects shown by the
t-­score difference values, indicating that the operational- Thematic Judgment as Link between
ization of the task was consistent with children’s response Vocabulary Size and Reading
patterns (Figure 2, panels a and b). Furthermore, we ran an Comprehension
additional model with a standardized value of the t-­score Based on our findings, we selected two reading measures
difference as a continuous fixed effect. while controlling for mediation analysis. First, a reading comprehension
for the t-­score maximum and found a main effect for composite score based on the scores of the reading com-
t-­score difference (see Table 2). Thus, the larger the differ- prehension subtests (i.e., word, sentence, and text level).
ence, the more likely children were to choose the more The mean accuracy score of thematic judgment correlated
strongly related test word. significantly with the outcome of all subtests (see
In hypothesis (1b), we hypothesized that children Figure S4). Second, we chose reading comprehension on
would be more likely to show a fast accurate response the text-­level as a manifest variable because it had the
(mean latency of accurate decisions) when the maximum strongest correlation with thematic judgment and corre-
collocation frequency between the contrasting word pairs lated significantly with vocabulary size. We focused on
was high (short path) versus low (long path). This hypoth- response accuracy because response latency or a response/
esis was also supported. Children were faster in the strong-­ accuracy measure (LISAS; Vandierendonck, 2017) did not
distant (M = 2327 ms, SE = 26 ms) and the strong-­weak correlate significantly with vocabulary size or reading com-
(M = 2391 ms, SE = 29 ms) condition compared to the prehension (see code r2 on OSF). We ran two models for
weak-­ distant condition (M = 2712 ms, SE = 33 ms; each dependent reading variable. One for the basic media-
(Δstrong-­distant = 385 ms, β = .17, SE = .03, t = 6.9, p < .001; tion analysis to see if thematic judgment (M) mediated the
Δstrong-­weak = 321 ms, β = .16, SE = .03, t = 6.33, p < .001)). relation of vocabulary size (X) and reading comprehension
Response latencies of accurate responses did not differ (Y; Figure 4a) and the second one with decoding to see if
between the strong-­distant and strong-­weak conditions the mediation effect would remain stable (Figure 4b).
(Δ = 64 ms, β = .01, SE = .02, t = 0.41, p > .05). This pattern Because effects changed between the first and the second
also matched the between-­condition pattern of the maxi- model, we added another exploratory analysis (Figure 4c)
mum t-­score (Figure 2, panels c and d). Finally, in line with to clarify the relations between the variables. All variables
our assumptions, the analysis with continuous collocation were entered into the models as z-­standardized raw scores.
frequency measures found that children gave faster accu-
rate responses on trials with higher t-­score maximums
Accuracy of Thematic Judgment as
than on trials with lower t-­score maximums (Table 2).
Finally, we found a significant main effect for phono- Mediator for Reading Comprehension
logical working memory in the accuracy models (Table 2). Thematic judgment responses were not standardized.
To rule out the possibility that this finding affected Therefore, we performed internal consistency analyses
between-­condition effects, we conducted additional analy- and a cluster analysis using the {alpha} and {iclust} func-
ses with data that excluded children with phonological tions of the psych package (Revelle, 2022). We removed 15
working memory problems (T < 35). These analyses items that were either negatively correlated with the scale
yielded the same between-­ condition pattern of results or showed item-­scale-­correlations of below 0.2. The final
without a main effect of phonological working memory item set (29 items; strong-­ distant: 11; strong-­ weak: 8;
(see code r1 on OSF). weak-­distant: 10) had an acceptable internal consistency
(Cronbach’s α = .73). The cluster analysis indicated that
internal consistency was highest if one cluster was assumed
Thematic Judgment and Vocabulary (Table S4). We entered thematic judgment as latent con-
Size, Decoding and Reading struct into the models. The latent variables represented the
Comprehension conditions (see Figure 4).
All results for this section are reported in Tables S5–S7 and
are summarized in Figure 3. In line with our hypotheses, Composite Reading Comprehension Score
we found that children with larger vocabulary sizes also For the composite reading score, the word and sentence
gave more accurate responses in the thematic judgment level tasks had a good fit (>.8) while the text level task

Learning to Read Connections | 459

 19362722, 2024, 3, Downloaded from https://ila.onlinelibrary.wiley.com/doi/10.1002/rrq.548 by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
FIGURE 4
Mediation Models for Reading Comprehension on the Text Level and Reading Fluency. Note. p < .05*; p < .01 **;
p < .001.

had a lower but acceptable fit (>.6). Both models for this Reading Comprehension on Text Level
score are shown in Figure S5. The mediation model of The mediation model for text-­level reading comprehen-
the composite reading score had an acceptable fit sion had a good fit (CFI = .99; RMSEA = .06 [.00, .15];
(CFI = .96; RMSEA = .09 [.04, .14]; SRMR = .07; SRMR = .04; 𝝌2 = 6.43 (4), p =. 17; see Figure 4a). Vocabu-
𝝌2 = 25.72 (12), p =. 01). Vocabulary size did not signifi- lary size did not significantly predict reading comprehen-
cantly predict reading comprehension, either before or sion before, but after thematic judgment responses were
after accounting for thematic judgment responses (c: considered (c: β = .13, SE = .10, z = 1.28, p > .05; c’: β = .30,
β = −.02, SE = .09, z = −.21, p > .05; c’: β = 12, SE = .07, SE = .08, z = 3.76, p < .001). Vocabulary size predicted the-
z = 1.731, p = .08). However, vocabulary size predicted matic judgment (a: β = .48, SE = .07, z = 6.79, p < .001) and
thematic judgment (a: β = .47, SE = .07, z = 6.78, p < .001), thematic judgment in turn predicted reading comprehen-
and thematic judgment in turn predicted reading com- sion (b: β = .35, SE = .14, z = 2.41, p < .05). The mediation
prehension (b: β = .30, SE = .13, z = 2.27, p < .05), leading effect was significant, indicating that thematic judgment
to a significant mediation effect (a*b: β = .14, SE = .06, contributed to the significant relation between vocabulary
z = 2.20, p < .05). size and reading comprehension (a*b: β = .16, SE = .07,
The second model included decoding. The model also z = 2.32, p < .05).
had an acceptable fit (CFI = .97; RMSEA = .08 [.04, .12]; The model including decoding had an acceptable fit
SRMR = .10; 𝝌2 = 34.11 (17), p =. 01). However, including (CFI = .97; RMSEA = .09 [.02, .15]; SRMR = .07; 𝝌2 = 14.52
decoding changed the pattern of results. Vocabulary size (7), p = .04). However, the effect of thematic judgment on
now significantly predicted reading comprehension both reading comprehension and the mediation effect were not
before and after considering thematic judgments (c: β = 18, significant (b: β = .13, SE = .12, z = 1.11, p > .05; a*b: β = .06,
SE = .05, z = 3.29, p < .01; c’: β = 19, SE = .04, z = 4.5, p < .001) SE = .05, z = 1.10, p > .05; see Figure 4b). Vocabulary size
and still predicted thematic judgment (a: β = .47, SE = .07, predicted thematic judgment (a: β = .48, SE = .07, z = 6.77,
z = 6.76, p < .001). The mediation effect was not significant p < .001) and reading comprehension (c: β = 28, SE = .08,
(a*b: β = .01, SE = .03, z = .26, p > .05) as thematic judgment z = 3.34, p < .01; c’: β = .34, SE = .06, z = 5.31, p < .001). In
no longer predicted reading comprehension (a: β = .02, addition, decoding strongly predicted reading compre-
SE = .07, z = .26, p > .05). Expectedly, decoding strongly hension (β = .55, SE = .06, z = 8.56, p < .001).
predicted reading comprehension (β = .73, SE = .05, Despite a moderate correlation with text-­level reading
z = 13.51, p < .001). comprehension (r = .31), thematic judgment no longer

460 | Reading Research Quarterly, 59(3)

 19362722, 2024, 3, Downloaded from https://ila.onlinelibrary.wiley.com/doi/10.1002/rrq.548 by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
contributed to the explanation of reading comprehension Fisher, 2019; Wojcik & Kandhadai, 2020). Our study adds
or the relation between vocabulary size and reading com- to previous studies by directly linking measures of co-­
prehension, when decoding was added. Thus, we went on occurrence to children’s responses between conditions
to explore a variant where vocabulary size would mediate (t-­score; Evert, 2008; Unger et al., 2023). Thus, we confirm
the relation between thematic judgment and reading and expand previous findings that find co-­occurrence sta-
comprehension. tistics to be directly related to behavioral outcomes in word
We computed an additional model in which vocabu- knowledge (Unger et al., 2023). Of note for future studies,
lary size was entered as a mediator of thematic judgment we used an item subset of a study with 4-­to 6-­year-­olds
and reading comprehension and included decoding (see (Schmitterer & Schroeder, 2019b). In the present study
Figure 4c). This model had an acceptable fit (CFI = .97; with 8-­year-­olds, we found similar results. We used the-
RMSEA = .08 [.01, .14]; SRMR = .07; 𝝌2 = 15.51 (8), p = .05). matic relatedness of high frequent words with an early age
Thematic judgment did not initially predict text-­ level of acquisition onset (e.g., Braginsky et al., 2019). This
reading comprehension (c: β = 13, SE = .11, z = 1.11, p > .05). group of words has been found to be similarly represented
However, thematic judgment strongly predicted vocabu- in semantic networks of toddlers and adults (Hills
lary size (a: β = 71, SE = .11, z = 6.1, p < .001) and after et al., 2008) and could serve as an indicator for sensitivity
accounting for vocabulary size, significantly predicted to collocation frequency across development.
reading comprehension (c’: β = .33, SE = .10, z = 3.38,
p < .01). The mediation effect of vocabulary size on the
relation between thematic judgment and reading compre-
The Relation of Thematic Judgment
hension was significant (β = .20, SE = .06, z = 3.00, p < .01). Responses and Vocabulary Size
Vocabulary size (b: β = .28, SE = .08, z = 3.34, p < .01) and Our results also suggest that vocabulary size and thematic
decoding (β = .55, SE = .06, z = 8.59, p < .001) both also sig- judgments are linked. This finding is consistent with pre-
nificantly predicted reading comprehension. vious work linking the structure and growth of semantic
networks based on co-­ occurrence measures to early
vocabulary development (Flores et al., 2020; Four-
tassi, 2020; Fourtassi et al., 2020; Hills et al., 2009). Chil-
Discussion dren were more likely to respond in the direction of the
This study focused on the relation between sensitivity to strongest related word in the thematic judgment task when
collocation frequency, vocabulary size, and reading com- they had a larger compared to a smaller vocabulary. This is
prehension skills in third-­grade students with low reading consistent with our assumption that children with higher
comprehension skills. Our results support the hypothesis compared to lower sensitivity to collocation frequency in
that sensitivity to collocation frequency is related to both the input language are also more likely to form highly
vocabulary size and reading comprehension. Thus, the interconnected semantic networks and thus learn more
ability to use collocation frequency as a source of semantic new words in the same amount of time. At the same time,
information is positively related to the acquisition of com- high connectivity may facilitate lexical retrieval of word-­
prehension skills throughout development. We also con- level conceptual information or semantic relatedness
clude that collocation sensitivity connects comprehension information. The average path length of links in the net-
abilities on the word level (i.e., requiring lexical retrieval) work of children with highly interconnected vocabularies
and more complex comprehension tasks (i.e., requiring should be shorter than in children with sparsely intercon-
inferential reasoning) on the sentence or text level. Over- nected semantic networks (see e.g., Beckage et al., 2011).
all, the results clearly indicate that thematic judgment (i.e., This should make retrieval less effortful.
a measure of sensitivity to collocation frequency), in addi-
tion to vocabulary size contributed to explaining variance Thematic Judgment, Inferential
in reading comprehension ability, even when other rele-
vant cognitive processes (i.e., decoding) are controlled.
Reasoning, and Reading
Comprehension
Our results indicate that thematic judgments in middle
Thematic Judgment as Indicator for childhood are related to reading comprehension skills of
Semantic Processing Throughout varying complexity (see Figure 3). These findings are con-
Childhood sistent with a cascading understanding of comprehension
Our results confirm the results of previous studies show- processes in reading. For example, the reading system
ing young children, older children, and adults are sensitive framework (Perfetti & Stafura, 2014) suggests that word-­
to thematic relations (Lany & Saffran, 2013; Savic level comprehension processes are a pressure point for
et al., 2023; Schmitterer & Schroeder, 2019b; Unger higher-­level reading processes. In the case of thematic
et al., 2023, 2020; Unger & Fisher, 2021; Vales & judgments, information about word relations is retrieved.

Learning to Read Connections | 461

 19362722, 2024, 3, Downloaded from https://ila.onlinelibrary.wiley.com/doi/10.1002/rrq.548 by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Thus, if retrieval of information about word relations is primarily semantic processes with reading comprehen-
impaired, reading comprehension at the word, sentence, sion. This led to an attenuated mediation effect of the the-
and text level will also be low. matic judgment on the relation of vocabulary size and
In addition, our results show that word-­level thematic reading comprehension because the relation between
judgment responses in 8-­year-­olds tended to be better vocabulary and reading comprehension then explained
explained by individual differences in complex reading semantic lexical retrieval processes to a larger extend.
comprehension tasks on the text-­level than by a word-­level Because we hypothesized that the thematic judgment
reading comprehension task. We suggest that this is task measures not only lexical retrieval but also inferen-
explained by the fact that thematic judgments measure at tial reasoning, and thus should explain an additional
least two cognitive processes. First, thematic judgments amount of variance in reading comprehension, we tried
foster lexical retrieval of semantic information. This ability an alternative approach. We examined whether the rela-
is relevant in vocabulary and reading comprehension tasks tion between thematic judgment and reading compre-
of any complexity. Second, thematic judgments require hension was mediated by vocabulary size. In this scenario,
inferential reasoning. Inferential reasoning is relevant for vocabulary size would bridge semantic and phonological
complex reading comprehension tasks (Ahmed et al., 2016; processes in lexical retrieval at the word level. In addi-
Duke & Cartwright, 2021) and has been linked to colloca- tion, thematic judgment would predict vocabulary size
tion frequency in young children (Fisher et al., 2011). In because sensitivity to collocation frequency may precede
line with our assumptions, our findings suggest that chil- (productive) vocabulary growth in development (Four-
dren who are more sensitive to collocation frequency, as tassi et al., 2020; Hills et al., 2009). Furthermore, thematic
indicated by thematic judgment responses, are also better judgment would be related to reading comprehension
in reading tasks that require inferential reasoning than based on shared variance explained by inferential reason-
children who are less sensitive to collocation frequency. ing that does not overlap with variance explained by lexi-
cal retrieval of concepts (i.e., vocabulary size) or
Thematic Judgment linking phonological processing (i.e., decoding). This explor-
atory model (Figure 4c) then showed that all three vari-
Vocabulary Size and Reading ables contributed significantly to explaining variance in
Comprehension text-­level reading comprehension, and that vocabulary
Our mediation analysis initially indicated that response size mediated the relation between thematic judgment
accuracy in thematic judgments mediated the relation and reading comprehension.
between vocabulary size and reading comprehension We concluded that this exploratory approach fits well
(Figure 4). However, when we added decoding as an addi- with research on early language comprehension develop-
tional predictor of reading comprehension skills, the effect ment (Fisher et al., 2011; Fourtassi et al., 2020; Savic
of thematic judgment on reading comprehension became et al., 2023; Unger & Fisher, 2021). In addition, this
insignificant while the effect of vocabulary size on reading approach supports middle childhood reading research in
comprehension increased. In line with reading research, several ways. First, it shows that both language compre-
children’s decoding skills explained a large amount of vari- hension and decoding contribute to reading comprehen-
ance in reading comprehension (Hjetland et al., 2020; sion (Hjetland et al., 2020; Hoover & Tunmer, 2018).
Hoover & Tunmer, 2018). Because decoding relies on Second, it shows that word-­level lexical processes predict
phoneme-­ grapheme conversion (see e.g., Castles higher-­level reading comprehension skills (Perfetti & Sta-
et al., 2018; Hjetland et al., 2020), we conclude that a large fura, 2014). Third, it confirms that cognitive processes
amount of variance in reading comprehension that is behind language comprehension are complex (Castles
explained by decoding, represents phonological processes. et al., 2018), and finally, that vocabulary may bridge cogni-
Our productive vocabulary size measure included phono- tive processes in reading comprehension (Duke &
logical processes, but we controlled phonological pro- Cartwright, 2021).
cesses in the thematic judgment task. Thus, we assume
that the inclusion of decoding shifted the shared variance
that was explained by vocabulary size in reading compre- Limitations
hension. Before including decoding, vocabulary size Our study had several limitations. First, performance on
explained some phonological and some semantic variance our measure of vocabulary size was lower than expected
in reading comprehension. In addition, thematic judg- (although with an acceptable distribution; see Figure S3),
ment responses then mediated semantic lexical retrieval and vocabulary size did not correlate with some typically
processes between vocabulary size and reading compre- related measures (i.e., reading comprehension on word or
hension. However, after the addition of decoding skills sentence level; see Figure S4). We suspect that this is due to
accounted for a large portion of the phonological variance differences between the normative sample and our sample
in reading comprehension, vocabulary size shared (i.e., region and language background). Furthermore, our

462 | Reading Research Quarterly, 59(3)

 19362722, 2024, 3, Downloaded from https://ila.onlinelibrary.wiley.com/doi/10.1002/rrq.548 by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
vocabulary measure was a measure of productive vocabu- between this ability and vocabulary size and various read-
lary, whereas the reading comprehension measures were ing comprehension skills changes across development and
receptive. However, most of our results that included what external influences (i.e., access to books, quality of
vocabulary size were in line with previous literature. We explanations) are associated with these relations.
would recommend replicating our findings with other Second, we found that children who are likely to ben-
vocabulary measures. efit from reading intervention have better reading compre-
Second, the item set for the thematic judgment task hension skills and larger vocabularies if they are also more
was not standardized, and after reducing the item set, the accurate in their thematic judgments. In addition,
internal consistency and the loadings for the latent vari- laboratory-­ based training studies that rely on co-­
able did not go beyond the interpretation of an acceptable occurrence statistics as a baseline for their design already
fit. This is not uncommon for meaning-­based measures yielded positive results (Savic et al., 2022, 2023). Moreover,
(see, e.g., the measure for morphological awareness in some training approaches for special needs populations
Caravolas et al., 2019). However, even though the item set (e.g., autistic children) use network visualization to make
was highly controlled for frequency, phonological and the shared origin of semantic relatedness palpable (i.e.,
orthographic features, and taxonomic relatedness, some Thinking Maps; Mashal & Kasirer, 2011). These
additional semantic features like idiomatic or composi- approaches could be combined with existing reading com-
tional multiword expressions may have also influenced prehension intervention programs (see, e.g., Duke
lexical processing in some items (Jiang et al., 2020; Kessler et al., 2021).
et al., 2021; Ulicheva et al., 2021). Thus, more, and differ- Third, experts agree that the comprehension processes
ent item sets for thematic judgments should be developed in reading models are more complex than decoding pro-
and tested. cesses (Castles et al., 2018). Some studies of reading devel-
Third, our experimental analysis showed that the opment in recent years have expanded the language
largest amount of explained variability was between par- component in reading models by including separate
ticipants or between items, while second-­level effects of meaning-­ related lexical components at the word-­ level,
variability between reference words or between classes or such as morphological awareness (e.g., Caravolas
schools were very small. These results suggest that our et al., 2019; Lervåg et al., 2018). Thematic judgment could
assessment of thematic judgment captured a sensitivity to be an additional semantic component that provides an
linguistic contexts that was relatively independent of informative contribution to language comprehension.
social contexts. This may be partly explained by our selec- Finally, the influence of collocation frequency on chil-
tion of high-­frequency words. We are not aware of any dren’s language processing has been studied more in oral
other study that has reported class 1 and class 2 level language than in written language. However, some studies
effects for both items and participants. This makes com- focusing on multiword expressions already indicate that
parisons difficult. However, it would be interesting to see collocation frequency affects children’s reading also in
more studies reporting similar random effects structures written language processing (Jiang et al., 2020). Thematic
in the future. Regarding the small amount of variability judgment could be used as an operationalization to exam-
explained by class or school level effects, we suspect that ine the effects of collocation frequency in written material
the fact that children were pre-­selected by teachers for as well.
individual sessions based on their reading ability may
have contributed to this finding. In addition, only up to
eight children per class participated in individual ses-
sions, which meant that we had a small number of chil- Summary
dren per class in the analysis. Future studies could address In this study, we examined the relation between a mea-
contextual variability by administering similar tasks to sure of sensitivity to collocation frequency in spoken lan-
whole classes. guage (i.e., thematic judgments) with vocabulary size and
reading comprehension measures of varying complexity
in 8-­year-­old children (N = 147). Our results indicate that
Future Directions responses in thematic judgments are related to co-­
The results of our study, together with previous literature occurrence statistics from child-­directed literature and
suggest that thematic judgment could be an indicator link- are associated with individual differences in vocabulary
ing comprehension processes in early language develop- size and reading comprehension skills of varying com-
ment and early reading development (e.g., Schmitterer & plexity. Moreover, interindividual differences in thematic
Schroeder, 2019b; Unger et al., 2020; Unger & Fisher, 2021; judgments predicted vocabulary size and reading com-
Vales & Fisher, 2019). Longitudinal studies that focus on prehension even when decoding was controlled. The
the role of sensitivity to collocation frequency across results link studies of lexical growth and inferential rea-
development could further explore whether the relation soning in early development with comprehension

Learning to Read Connections | 463

 19362722, 2024, 3, Downloaded from https://ila.onlinelibrary.wiley.com/doi/10.1002/rrq.548 by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
processes during reading acquisition in middle child- Data Availability Statement
hood. We conclude that thematic judgments are based on
a sensitivity to collocation frequency and tap into seman- The analyses presented here were preregistered: https://​
tic processes of lexical retrieval and inferential reasoning. osf.​io/​eajfc/​​. Data, codebook, code, and supplementary
Finally, sensitivities to collocation frequency have yet to material are available here: https://​osf.​io/​m9fbw/​​.
be addressed in combined language and reading compre-
hension interventions. REFERENCES
Ahmed, Y., Francis, D. J., York, M., Fletcher, J. M., Barnes, M., & Kulesz,
P. (2016). Validation of the direct and inferential mediation (DIME)
model of reading comprehension in grades 7 through 12. Contempo-
Author Contributions rary Educational Psychology, 44, 68–82. https://​doi.​org/​10.​1016/j.​
cedps​ych.​2016.​02.​002
Alexandra Schmitterer was responsible for conceptualiza- Arias-­Trejo, N., & Plunkett, K. (2013). What’s in a link: Associative and
tion, methodology, formal analysis, resources, data curation, taxonomic priming effects in the infant lexicon. Cognition, 128(2),
project administration, writing the original draft, and visual- 214–227. https://​doi.​org/​10.​1016/j.​cogni​tion.​2013.​03.​008
ization. Caterina Gawrilow and Claudia Friedrich contrib- Bates, D., Maechler, M., Bolker, B., & Walker, S. (2015). Fitting linear
mixed-­effects models using lme4. Journal of Statistical Software,
uted through writing, reviewing, editing, and supervision. 67(1), 1–48. https://​doi.​org/​10.​18637/​​jss.​v067.​i01
Beckage, N., Smith, L., & Hills, T. (2011). Small worlds and semantic
network growth in typical and late talkers. PLoS One, 6(5), e19348.

Acknowledgments https://​doi.​org/​10.​1371/​journ​al.​pone.​0019348
Braginsky, M., Yurovsky, D., Marchman, V. A., & Frank, M. C. (2019).
We thank our student research assistants for their support Consistency and variability in children’s word learning across lan-
guages. Open Mind, 3, 52–67. https://​doi.​org/​10.​1162/​opmi_a_​00026​
in data collection. Furthermore, we thank Marcus Hassel- Buchanan, L., Westbury, C., & Burgess, C. (2001). Characterizing
horn, Garvin Brod, Leonard Tetzlaff, and all collaborators semantic space: Neighborhood effects in word recognition. Psycho-
for their role in supporting the iLearn project. Open nomic Bulletin & Review, 8, 531–544. https://​doi.​org/​10.​3758/​BF031​
Access funding enabled and organized by Projekt DEAL. 96189​
Caravolas, M., Lervåg, A., Mikulajová, M., Defior, S., Seidlová-­Málková,
G., & Hulme, C. (2019). A cross-­linguistic, longitudinal study of the
foundations of decoding and reading comprehension ability. Scien-
Funding Information tific Studies of Reading, 23(5), 386–402. https://​doi.​org/​10.​1080/​
10888​438.​2019.​1580284
This research was funded by the German Federal Ministry Carroll, J. B., Davies, P., & Richman, B. (1971). Word frequency book.
of Education and Research, as part of the LONDI research American Heritage.
program and its subproject iLearn (grant number: Castles, A., Rastle, K., & Nation, K. (2018). Ending the Reading wars:
Reading acquisition from novice to expert. Psychological Science in
01GJ1710B). the Public Interest, 19(1), 5–51. https://​doi.​org/​10.​1177/​15291​00618​
772271
Chiu, Y. D. (2018). The simple view of Reading across development:

Ethics Approval and Conflict of Prediction of grade 3 Reading comprehension from prekindergarten
skills. Remedial and Special Education, 39(5), 289–303. https://​doi.​
Interest Disclosure org/​10.​1177/​07419​32518​762055
Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2013). Applied multiple
This study was approved by the Institutional Review Board regression/correlation analysis for the behavioral sciences. Routledge.
(Ethics Committee) of the DIPF | Leibniz Institute for Collins, A. M., & Loftus, E. F. (1975). A spreading-­activation theory of
Research and Information in Education (reference num- semantic processing. Psychological Review, 82(6), 407–428. https://​
doi.​org/​10.​1037/​0033-­​295X.​82.6.​407
ber: iLearn) and the Ethics Committee of the Hessian Dale, P. S., & Fenson, L. (1996). Lexical development norms for young
Ministry for Culture, Education and Opportunities (refer- children. Behavior Research Methods, Instruments, & Computers, 28,
ence number: GWU 695). Parents or legal guardians gave 125–127. https://​doi.​org/​10.​3758/​BF032​03646​
written consent for their children to participate prior to Dornseiff, F. (2020). Der deutsche Wortschatz nach Sachgruppen. De
data collection. Children could also agree to participate in Gruyter. https://​doi.​org/​10.​1515/​97831​10457742
Duke, N. K., & Cartwright, K. B. (2021). The science of reading
the study with a simplified version of the consent form and progresses: Communicating advances beyond the simple view of
were informed before data collection about their right to reading. Reading Research Quarterly, 56, S25–S44. https://​doi.​org/​10.​
refuse or stop participation at any time without negative 1002/​rrq.​411
consequences. In addition, teachers gave written consent Duke, N. K., Ward, A. E., & Pearson, P. D. (2021). The science of
for their own participation in the study prior to data col- reading comprehension instruction. The Reading Teacher, 74(6),
663–672. https://​doi.​org/​10.​1002/​trtr.​1993
lection. The consent forms included permission to use the Elleman, A. M. (2017). Examining the impact of inference instruction
collected data in scientific research projects, including on the literal and inferential comprehension of skilled and less
publications, in an anonymized form. The authors have no skilled readers: A meta-­analytic review. Journal of Educational Psy-
conflicts of interest to declare. chology, 109(6), 761–781. https://​doi.​org/​10.​1037/​edu00​00180​

464 | Reading Research Quarterly, 59(3)

 19362722, 2024, 3, Downloaded from https://ila.onlinelibrary.wiley.com/doi/10.1002/rrq.548 by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Evert, S. (2008). Corpora and collocations. In A. Lüdeling & M. Kytö Hjetland, H., Brinchmann, E., Scherer, R., Hulme, C., & Melby-­Lervåg,
(Eds.), Corpus linguistics. An international handbook, 1212–1248. M. (2020). Preschool pathways to reading comprehension: A sys-
Mouton de Gruyter. https://​cites​eerx.​ist.​psu.​edu/​viewd​oc/​downl​ tematic meta-­ analytic review. Educational Research Review, 30,
oad?​doi=​10.1.​1.​159.​6220&​rep=​rep1&​type=​pdf retrieved from Janu- 100323. https://​doi.​org/​10.​1016/j.​edurev.​2020.​100323
ary 01, 2021 Hjetland, H., Lervåg, A., Lyster, S. A. H., Hagtvet, B. E., Hulme, C., &
Fazio, R. H. (1990). A practical guide to the use of response latency in Melby-­Lervåg, M. (2019). Pathways to reading comprehension: A
social psychological research. In C. Hendrick & M. S. Clark (Eds.), longitudinal study from 4 to 9 years of age. Journal of Educational
Research methods in personality and social psychology (pp. 74–97). Psychology, 111(5), 751–763. https://​doi.​org/​10.​1037/​edu00​00321​
Sage Publications, Inc. Hoover, W. A., & Tunmer, W. E. (2018). The simple view of reading:
Firth, J. R. (1957). A synopsis of linguistic theory 1930–55. In Studies in Three assessments of its adequacy. Remedial and Special Education,
linguistic analysis (pp. 1–32). The Philological Society. Reprinted in 39(5), 304–312. https://​doi.​org/​10.​1177/​07419​32518​773154
Palmer (1968), pp. 168–205. Hulme, C., Nash, H., Gooch, D., Lervåg, A., & Snowling, M. J. (2015).
Fisher, A. V., Matlen, B. J., & Godwin, K. E. (2011). Semantic similarity of The foundations of literacy development in children at familial risk
labels and inductive generalization: Taking a second look. Cognition, of dyslexia. Journals.Sagepub.Com, 26(12), 1877–1886. https://​doi.​
118(3), 432–438. https://​doi.​org/​10.​1016/j.​cogni​tion.​2010.​12.​008 org/​10.​1177/​09567​97615​603702
Flores, A. Z., Montag, J., & Willits, J. (2020). Using known words to Jiang, S., Jiang, X., & Siyanova-­Chanturia, A. (2020). The processing of
learn more words: A distributional analysis of child vocabulary multiword expressions in children and adults: An eye-­tracking study
development. arXiv preprint arXiv:2009.06810. of Chinese. Applied PsychoLinguistics, 41(4), 901–931. https://​doi.​
Florit, E., & Cain, K. (2011). The simple view of reading: Is it valid for org/​10.​1017/​S0142​71642​0000296
different types of alphabetic orthographies? Educational Psychology Kessler, R., Weber, A., & Friedrich, C. K. (2021). Activation of literal
Review, 23(4), 553–576. https://​doi.​org/​10.​1007/​S1064​8-­​011-­​9175-­​6 word meanings in idioms: Evidence from eye-­tracking and ERP
Fourtassi, A. (2020). Word co-­occurrence in child-­directed speech predicts experiments. Language and Speech, 64(3), 594–624. https://​doi.​org/​
children’s free word associations. In Proceedings of the workshop on 10.​1177/​00238​30920​943625
cognitive modeling and computational linguistics (pp. 49–53). Associa- Kintsch, W. (1998). Comprehension: A paradigm for cognition.
tion for Computational Linguistics. https://​doi.​org/​10.​18653/​​v1/​P17 Cambridge University Press.
Fourtassi, A., Bian, Y., & Frank, M. C. (2020). The growth of children’s Lakens, D., Scheel, A. M., & Isager, P. M. (2018). Equivalence testing for
semantic and phonological networks: Insight from 10 languages. psychological research: A tutorial. Advances in Methods and Prac-
Cognitive Science, 44(7), e12847. https://​doi.​org/​10.​1111/​cogs.​12847​ tices in Psychological Science, 1(2), 259–269. https://​doi.​org/​10.​1177/​
Fox, J., & Weisberg, S. (2019). An {R} companion to applied regression 25152​45918​770963
(Third ed.). Sage. https://​socia​lscie​nces.​mcmas​ter.​ca/​jfox/​Books/​​ Landauer, T. K., & Dumais, S. T. (1997). A solution to Plato’s problem:
Compa​nion/​ The latent semantic analysis theory of acquisition, induction, and
Frank, M. C., Braginsky, M., Yurovsky, D., & Marchman, V. A. (2017). representation of knowledge. Psychological Review, 104(2), 211–240.
Wordbank: An open repository for developmental vocabulary data. https://​doi.​org/​10.​1037/​0033-​295X.​104.2.​211
Journal of Child Language, 44(3), 677–694. https://​doi.​org/​10.​1017/​ Landrigan, J. F., & Mirman, D. (2018). The cost of switching between
S0305​00091​6000209 taxonomic and thematic semantics. Memory and Cognition, 46,
Friedrich, C. K., Felder, V., Lahiri, A., & Eulitz, C. (2013). Activation of 191–203. https://​doi.​org/​10.​3758/​s1342​1-­​017-­​0757-­​5
words with phonological overlap. Frontiers in Psychology, 4, 556. Lany, J., & Saffran, J. (2013). Statistical learning mechanisms in infancy.
https://​doi.​org/​10.​3389/​fpsyg.​2013.​00556​ Comprehensive Developmental Neuroscience: Neural Circuit Develop-
Ganzeboom, H. B. G. (2010). A new international socio-­economic ment and Function in the Brain, 3, 231–248.
index (ISEI) of occupational status for the international standard Lenhard, W., Lenhard, A., & Schneider, W. (2017). ELFE II—Ein
classification of occupation 2008 (ISCO-­08) constructed with data Leseverständnistest für Erst-­bis Siebtklässler. Hogrefe.
from the ISSP 2002-­2007. In Paper presented at the annual confer- Lervåg, A., Hulme, C., & Melby-­Lervåg, M. (2018). Unpicking the
ence of international social survey Programme, Lisbon, Portugal. developmental relationship between oral language skills and reading
http://​www.​harry​ganze​boom.​nl/​Pdf/​2010%​20-%​20Gan​zeboom%​ comprehension: It’s simple, but complex. Child Development, 89(5),
20-%​20A%​20New%​20Int​ernat​ional%​20Soc​io-​Econo​mic%​20Ind​ 1821–1838. https://​doi.​org/​10.​1111/​cdev.​12861​
ex%​20ISEI%​20of%​20occ​upati​onal%​20sta​tus%​20for%​20the%​20Int​ Lonigan, C. J., Burgess, S. R., & Schatschneider, C. (2018). Examining
ernat​ional%​20Sta​ndard%​20Cla​ssifi​cation%​20of%​20Occ​upati​on.​pdf the simple view of Reading with elementary school children: Still
Glück, C. W., & Glück, C. W. (2011). Wortschatz-­und Wortfindungstest simple after all these years. Remedial and Special Education, 39(5),
für 6-­bis 10-­Jährige (WWT 6–10). Urban & Fischer. 260–273. https://​doi.​org/​10.​1177/​07419​32518​764833
Godwin, K. E., Matlen, B. J., & Fisher, A. V. (2013). Development of Lüdecke, D., Bartel, A., Schwemmer, C., Powell, C., Djalovski, A., &
category-­based reasoning in 4-­to 7-­year-­old children: The influence Titz, J. (2023). _sjPlot: Data Visualization for Statistics in Social Sci-
of label co-­occurrence and kinship knowledge. Journal of Experi- ence. R package version 2.8.14. https://​CRAN.​R-­​proje​ct.​org/​packa​
mental Child Psychology, 115(1), 74–90. https://​doi.​org/​10.​1016/j.​ ge=​sjPlot
jecp.​2012.​11.​008 Lund, K., Burgess, C., & Audet, C. (1996). Dissociating semantic and
Harris, Z. S. (1954). Distributional structure. Word, 10(3), 146–162. associative word relationships using high-­ dimensional semantic
https://​doi.​org/​10.​1080/​00437​956.​1954.​11659520 space. In G. W. Cottrell (Ed.), Proceedings of the 18th annual confer-
Hills, T., Maouene, M., Maouene, J., Sheya, A., & Smith, L. (2009). ence of the cognitive science society (pp. 603–608). Erlbaum.
Longitudinal analysis of early semantic networks: Preferential MacWhinney, B. (2014). The CHILDES project: Tools for analyzing talk,
attachment or preferential acquisition? Psychological Science, 20(6), Volume II. Psychology Press.
729–739. https://​doi.​org/​10.​1111/j.​1467-­​9280.​2009.​02365.​x Mashal, N., & Kasirer, A. (2011). Thinking maps enhance metaphoric
Hills, T., Maouene, M., Maouene, J., Sheya, A., & Smith, L. B. (2008). Is competence in children with autism and learning disabilities.
there preferential attachment in the growth of early semantic noun Research in Developmental Disabilities, 32(6), 2045–2054. https://​
networks? Proceedings of the 30th Annual Conference of the 57 Cog- doi.​org/​10.​1016/j.​ridd.​2011.​08.​012
nitive Science Society, October, 1–6. papers://c941067e-36da-4589- Matlen, B. J., Fisher, A. V., & Godwin, K. E. (2015). The influence of
acf1-2f5738fdb5a1/Paper/p406. label co-­occurrence and semantic similarity on children’s inductive

Learning to Read Connections | 465

 19362722, 2024, 3, Downloaded from https://ila.onlinelibrary.wiley.com/doi/10.1002/rrq.548 by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
generalization. Frontiers in Psychology, 6, 1146. https://​doi.​org/​10.​ Schmitterer, A. M. A., & Schroeder, S. (2019a). Effects of reading and
3389/​fpsyg.​2015.​01146​ spelling predictors before and after school entry: Evidence from a
McClelland, J., & Rogers, T. (2003). The parallel distributed processing German longitudinal study. Learning and Instruction, 59, 46–53.
approach to semantic cognition. Nature Reviews. Neuroscience, 4, https://​doi.​org/​10.​1016/j.​learn​instr​uc.​2018.​09.​005
310–322. https://​doi.​org/​10.​1038/​nrn1076 Schmitterer, A. M. A., & Schroeder, S. (2019b). Young Children’s
McRae, K., Cree, G. S., Seidenberg, M. S., & McNorgan, C. (2005). semantic judgment of thematic relations predicts word Reading:
Semantic feature production norms for a large set of living and non- Evidence from a longitudinal study. Cognitive Development, 50, 22–
living things. Behavior Research Methods, 37(4), 547–559. 35. https://​doi.​org/​10.​1016/j.​cogdev.​2019.​01.​002
Mikolov, T., Chen, K., Corrado, G., & Dean, J. (2013). Efficient Schroeder, S., Würzner, K. M., Heister, J., et al. (2015). childLex: A
estimation of word representations in vector space. arXiv preprint lexical database of German read by children. Behav Res, 47, 1085–
arXiv:1301.3781. 1094. https://​doi.​org/​10.​3758/​s1342​8-­​014-­​0528-­​1
Mirman, D., Landrigan, J.-­F., & Britt, A. E. (2017). Taxonomic and Siegmüller, J., Kauschke, C., von Minnen, S., & Bittner, D. (2011).
thematic semantic systems. Psychological Bulletin, 143(5), 499–520. TSVK. Test zum Satzverstehen von Kindern. In Eine profilorientierte
https://​doi.​org/​10.​1037/​bul00​00092​ Diagnostik der Syntax. Elsevier.
Moll, K., & Landerl, K. (2010). SLRT-­II: Lese-­und Rechtschreibtest. Spence, D. P., & Owens, K. C. (1990). Lexical co-­occurrence and
Weiterentwicklung des Salzburger Lese-und Rechtschreibtests association strength. Journal of Psycholinguistic Research, 19, 317–
(SLRT). Huber. 330. https://​doi.​org/​10.​1007/​BF010​74363​
Moll, K., Ramus, F., Bartling, J., Bruder, J., Kunze, S., Neuhoff, N., Stella, M., Beckage, N. M., & Brede, M. (2017). Multiplex lexical
Streiftau, S., Lyytinen, H., Leppänen, P. H. T., Lohvansuu, K., Tóth, networks reveal patterns in early word acquisition in children.
D., Honbolygó, F., Csépe, V., Bogliotti, C., Iannuzzi, S., Démonet, J. Scientific Reports, 7(1), 46730. https://​doi.​org/​10.​1038/​srep4​6730
F., Longeras, E., Valdois, S., George, F., … Landerl, K. (2014). Steyvers, M., & Tenenbaum, J. B. (2005). The large-­scale structure of
Cognitive mechanisms underlying reading and spelling development semantic networks: Statistical analyses and a model of semantic
in five European orthographies. Learning and Instruction, 29, 65–77. growth. Cognitive Science, 29(1), 41–78. https://​doi.​org/​10.​1207/​
https://​doi.​org/​10.​1016/J.​LEARN​INSTR​UC.​2013.​09.​003 s1551​6709c​og2901_​3
Nation, K. (2019). Children’s reading difficulties, language, and reflections Turney, P. D., & Pantel, P. (2010). From frequency to meaning: Vector
on the simple view of reading. Australian Journal of Learning Difficulties, space models of semantics. Journal of Artificial Intelligence Research,
24(1), 47–73. https://​doi.​org/​10.​1080/​19404​158.​2019.​1609272 37, 141–188. https://​doi.​org/​10.​1613/​jair.​2934
Nation, K., & Snowling, M. J. (1999). Developmental differences in Ulicheva, A., Marelli, M., & Rastle, K. (2021). Sensitivity to meaningful
sensitivity to semantic relations among good and poor regularities acquired through experience. Morphology, 31(3), 275–
comprehenders: Evidence from semantic priming. Cognition, 70(1), 296. https://​doi.​org/​10.​1007/​S1152​5-­​020-­​09363​-­​5
4–9. https://​doi.​org/​10.​1016/​S0010​-­​0277(99)​00004​-­​9 Unger, L., & Fisher, A. V. (2021). The emergence of richly organized
Nelson, D. L., McEvoy, C. L., & Schreiber, T. A. (1998). The University of semantic knowledge from simple statistics: A synthetic review.
South Florida word association, rhyme, and word fragment norms. Developmental Review, 60, 100949. https://​doi.​org/​10.​1016/j.​dr.​2021.​
Retrieved from http://​w3.​usf.​edu/​FreeA​ssoci​ation/​​ 100949
Pennington, J., Socher, R., & Manning, C. D. (2014). Glove: Global Unger, L., Vales, C., & Fisher, A. V. (2020). The role of Co-­occurrence
vectors for word representation. In Proceedings of the 2014 conference statistics in developing semantic knowledge. Cognitive Science,
on empirical methods in natural language processing (EMNLP) (pp. 44(9), e12894. https://​doi.​org/​10.​1111/​cogs.​12894​
1532–1543). Association for Computational Linguistics. https://​ Unger, L., Yim, H., Savic, O., Dennis, S., & Sloutsky, V. M. (2023). No
aclan​tholo​gy.​org/​D14-​1162.​pdf frills: Simple regularities in language can go a long way in the devel-
Perfetti, C., & Stafura, J. (2014). Word knowledge in a theory of reading opment of word knowledge. Developmental Science, 26, e13373.
comprehension. Scientific Studies of Reading, 18(1), 22–37. https://​ https://​doi.​org/​10.​1111/​desc.​13373​
doi.​org/​10.​1080/​10888​438.​2013.​827687 Vales, C., & Fisher, A. V. (2019). When stronger knowledge slows you
R Core Team. (2021). R: A language and environment for statistical down: Semantic relatedness predicts children’s co-­ activation of
computing. R Foundation for Statistical Computing. https://​www.​R-­​ related items in a visual search paradigm. Cognitive Science, 43(6),
proje​ct.​org/​ e12746. https://​doi.​org/​10.​1111/​cogs.​12746​
Revelle, W. (2022). Package “psych.” Cran.Rstudio.Org. https://​cran.​ Vandierendonck, A. (2017). A comparison of methods to combine
rstud​io.​org/​web/​packa​ges/​psych/​​psych.​pdf speed and accuracy measures of performance: A rejoinder on the
Rogers, T. T., & McClelland, J. L. (2004). Semantic cognition: A parallel binning procedure. Behavior Research Methods, 49(2), 653–673.
distributed processing approach. MIT press. https://​doi.​org/​10.​3758/​s1342​8-­​016-­​0721-­​5
Rosseel, Y., Yorgensen, D. T., De Wilde, L., Oberski, D., Byrnes, J., Verhoeven, L., Van Leeuwe, J., & Vermeer, A. (2011). Vocabulary
Vanbrabant, L., Savalei, V., Merkle, E., Hallquist, M., Rhemtulla, M., growth and reading development across the elementary school
Katsikatsou, M., Barendse, M., Rockwood, N., Scharf, F., Du, H., & years. Scientific Studies of Reading, 15(1), 8–25. https://​doi.​org/​10.​
Jamil, H. (2017). Package ‘lavaan’. 1080/​10888​438.​2011.​536125
Savic, O., Unger, L., & Sloutsky, V. M. (2022). Exposure to co-­occurrence von Hasselhorn, M., Schumann-­Hengsteler, R., Gronauer, J., Grube, D.,
regularities in language drives semantic integration of new words. Mähler, C., Schmid, I., Seitz-Stein, K., & Zoelch, C. (2012). Arbe-
Journal of Experimental Psychology: Learning, Memory, and Cogni- itsgedächtnistestbatterie für Kinder von fünf bis zwölf Jahren:(AGTB
tion, 48(7), 1064–1081. https://​doi.​org/​10.​1037/​xlm00​01122​ 5–12). Hogrefe.
Savic, O., Unger, L., & Sloutsky, V. M. (2023). Experience and Wang, Y. A., & Rhemtulla, M. (2021). Power analysis for parameter
maturation: The contribution of co-­ occurrence regularities in estimation in structural equation modeling: A discussion and tutorial.
language to the development of semantic organization. Child Advances in Methods and Practices in Psychological Science, 4(1),
Development, 94(1), 142–158. https://​doi.​org/​10.​1111/​cdev.​13844​ 2515245920918253. https://​doi.​org/​10.​1177/​25152​45920​918253
Schmitterer, A. M. A., & Brod, G. (2021). Which data do elementary Westfall, J., Kenny, D. A., & Judd, C. M. (2014). Statistical power and
school teachers use to determine reading difficulties in their stu- optimal design in experiments in which samples of participants
dents? Journal of Learning Disabilities, 54, 349–364. https://​doi.​org/​ respond to samples of stimuli. Journal of Experimental Psychology:
10.​1177/​00222​19420​981990 General, 143(5), 2020–2045. https://​doi.​org/​10.​1037/​xge00​00014​

466 | Reading Research Quarterly, 59(3)

 19362722, 2024, 3, Downloaded from https://ila.onlinelibrary.wiley.com/doi/10.1002/rrq.548 by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Wojcik, E. H., & Kandhadai, P. (2020). Paradigmatic associations and Caterina Gawrilow is Professor of School Psychology at the
individual variability in early lexical–semantic networks: Evidence University of Tübingen. Her main areas of interest are ADHD
from a free association task. Developmental Psychology, 56(1), 53–69. in children, adolescents, and adults, as well as experimental and
https://​doi.​org/​10.​1037/​dev00​00844​ applied research on self-­regulation and executive functions.
Zugarramurdi, C., Fernández, L., Lallier, M., Valle-­Lisboa, J. C., &
Email: [email protected]
Carreiras, M. (2022). Mind the orthography: Revisiting the
contribution of Prereading phonological awareness to Reading
Claudia Friedrich is Professor of Developmental Psychology
acquisition. Developmental Psychology, 58(6), 1003–1016. https://​
doi.​org/​10.​1037/​dev00​01341​
at the University of Tübingen. Her main area of interest is
language acquisition with a focus on the developing lexicon and
its cognitive foundations. Email: claudia.friedrich@psycho.
Submitted April 25, 2023 uni-tuebingen.de
Final revision received April 26, 2024
Accepted May 9, 2024

Alexandra Schmitterer (corresponding author) is an Associated Supporting Information

Postdoc Researcher at the DIPF|Leibniz Institute for Research and
Information in Education. She studies how language is processed in Additional supporting information may be found in the
the mind and how language processes are connected to literacy and online version of this article on the publisher’s web-
learning development. Email: [email protected] site:10.1002/rrq.548/suppinfo

Learning to Read Connections | 467