INFORME
Consistency and Variability in Children’s Word
Learning Across Languages
Mika Braginsky1, Daniel Yurovsky2, Virginia A. Marchman3, and Michael C. Frank3
1Department of Brain and Cognitive Sciences, Instituto de Tecnología de Massachusetts
2Department of Psychology, University of Chicago
3Department of Psychology, Universidad Stanford
un acceso abierto
diario
Palabras clave: word learning, language acquisition, corpus analysis
yo
D
oh
w
norte
oh
a
d
mi
d
F
r
oh
metro
h
t
t
pag
:
/
/
d
i
r
mi
C
t
.
metro
i
t
.
/
mi
d
tu
oh
pag
metro
i
/
yo
a
r
t
i
C
mi
–
pag
d
F
/
d
oh
i
/
i
/
.
/
1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
2
6
1
8
6
8
3
6
8
oh
pag
metro
_
a
_
0
0
0
2
6
pag
d
.
/
i
ABSTRACTO
Why do children learn some words earlier than others? The order in which words are
acquired can provide clues about the mechanisms of word learning. In a large-scale corpus
análisis, we use parent-report data from over 32,000 children to estimate the acquisition
trajectories of around 400 words in each of 10 idiomas, predicting them on the basis of
independently derived properties of the words’ linguistic environment (from corpora) y
significado (from adult judgments). We examine the consistency and variability of these
predictors across languages, by lexical category, and over development. The patterning of
predictors across languages is quite similar, suggesting similar processes in operation. En
contrast, the patterning of predictors across different lexical categories is distinct, in line with
theories that posit different factors at play in the acquisition of content words and function
palabras. By leveraging data at a significantly larger scale than previous work, our analyses
identify candidate generalizations about the processes underlying word learning across
idiomas.
F
b
y
gramo
tu
mi
s
t
t
oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3
INTRODUCCIÓN
Despite tremendous individual variation in children’s rate of development (Fenson et al., 2007),
the first words that they utter are strikingly consistent (Schneider, Yurovsky, & Franco, 2015;
Tardif et al., 2008): they tend to talk about important people in their life (“mom,” “dad”),
social routines (“hi,” “uh-oh”), animals (“dog,” “duck”), and foods (“milk,” “banana”). Incluso
as children learn from their own experiences and according to their own interests (Mayor &
Plunkett, 2014; nelson, 1973), their vocabulary grows rapidly, typically adding more nouns, pero
also verbs (“go”) and other predicates (“hot”) to their repertoires. Over just their first 3 años,
children learn hundreds, even thousands of words (Fenson et al., 1994; Mayor & Plunkett,
2011).
One classic approach to word learning focuses on the specific mechanisms that children
bring to bear on the learning problem. Por ejemplo, across many laboratory experiments, a
variety of mechanisms have been identified as plausible drivers of early word learning, incluir-
ing co-occurrence based and cross-situational word learning (Schwartz & Terrell, 1983; Yu &
Ballard, 2007), social cue use (Baldwin, 1993), and syntactic bootstrapping (Gleitman, 1990;
Mintz, 2003). The ability to identify which of these mechanisms is most explanatory has been
Citación: Braginsky, METRO., Yurovsky, D.,
Marchman, V. A., & Franco, METRO. C. (2019).
Consistency and Variability in
Children’s Word Learning Across
Idiomas. Mente abierta: Discoveries in
Ciencia cognitiva, 3, 52–67.
https://doi.org/10.1162/opmi_a_00026
DOI:
https://doi.org/10.1162/opmi_a_00026
Materiales suplementarios:
https://doi.org/10.1162/opmi_a_00026;
https://github.com/mikabr/
aoa-prediction
Recibió: 21 Septiembre 2017
Aceptado: 8 Abril 2019
Conflicto de intereses: Los autores tienen
declaró que no hay intereses en competencia
exists.
Autor correspondiente:
Mika Braginsky
mikabr@mit.edu
Derechos de autor: © 2019
Instituto de Tecnología de Massachusetts
Publicado bajo Creative Commons
Atribución 4.0 Internacional
(CC POR 4.0) licencia
La prensa del MIT
Word Learning Consistency and Variability Braginsky et al.
challenging. En efecto, many theories of early word learning take multiplicity of cue types and
mechanisms as a central feature (p.ej., Bloom, 2000; Hollich et al., 2000). As important as this
work is, aunque, these studies are typically aimed at understanding how one or a small handful
of words are learned in the laboratory under precisely defined learning conditions. They do not
directly address questions regarding the developmental composition and ordering of growth
in the lexicon across many different children in their natural environments, nor whether these
patterns are consistent across different languages.
An alternate approach to word learning asks why some words are learned so early and
some much later. This question about the order of the acquisition of first words can provide a
different window into the nature of children’s language learning. Posed as a statistical problem,
the challenge is to find what set of variables best predicts the age at which different words
are acquired. Previous work using this approach has revealed that, en Inglés, within a lexical
categoría (p.ej., nouns, verbos), words that are more frequent in speech to children are likely to be
learned earlier (Buen hombre, Valle, & li, 2008). Further studies have found evidence that a variety
of other semantic and linguistic factors are related to word acquisition, such as salience and
iconicity (Hills, Maouene, Maouene, Sheya, & Herrero, 2009; Perry, Perlman, & Lupyan, 2015;
roy, Franco, DeCamp, Molinero, & roy, 2015; alimenta, 2010; Swingley & Humphrey, 2018).
But these exciting findings are limited in their generality because each study used a
different dataset and focused on different predictors. Además, nearly all studies to date
have exclusively analyzed data from English-learning children, providing no opportunity for
cross-linguistic comparison of the relative importance of the many relevant factors under con-
sideration. Cross-linguistic comparisons are critical to identifying the universal mechanisms
that are in play for all children and differentiating them from patterns of acquisition that emerge
due to the particulars of a given language or culture (mi. Bates & MacWhinney, 1987; Slobin,
1985). Our goal here is to extend these classic approaches by assessing the degree to which
the predictors of word learning are consistent across different languages, as well as whether
there are similar patterns across different lexical categories.
The primary tool for characterizing the breadth of children’s early vocabularies in these
previous studies has been structured parent report. Naturalistic language samples and experi-
mental methods are both valuable methods for assessing aspects of child language (Bornstein
& Haynes, 1998; Fernald, Perfors, & Marchman, 2006). But outside of a few ultra-dense tran-
scripts (p.ej., Roy et al., 2015), neither method typically provides the kind of holistic and com-
prehensive view that comes from parent report. We focus in particular on the MacArthur-Bates
Communicative Development Inventory (CDI; Fenson et al., 2007), a family of parent-report
vocabulary checklists in which parents are asked whether their child “understands” or “under-
stands and says” a large set of individual words.
The CDIs are an inexpensive and widely used method for gathering reliable and valid
data about the nature and extent of young children’s productive and receptive vocabularies
(see Fenson et al., 1994, for review; cf. Feldman et al., 2000; Fenson et al., 2000). A pesar de
CDIs cannot exhaustively capture all words in a child’s vocabulary (Mayor & Plunkett, 2011),
they do give an estimate of a child’s knowledge about several hundred words, far more than
the handful that are typically tested in a lab experiment. CDI estimates of vocabulary size are
highly correlated with children’s vocabulary knowledge as assessed with naturalistic obser-
vation or using standardized tests (Fenson et al., 2007). Por supuesto, any parent report mea-
sure is subject to reporting biases. The CDIs were designed to minimize these by asking
parents to report only on observable behaviors that are currently (rather than retrospectively)
MENTE ABIERTA: Descubrimientos en ciencia cognitiva
53
yo
D
oh
w
norte
oh
a
d
mi
d
F
r
oh
metro
h
t
t
pag
:
/
/
d
i
r
mi
C
t
.
metro
i
t
.
/
mi
d
tu
oh
pag
metro
i
/
yo
a
r
t
i
C
mi
–
pag
d
F
/
d
oh
i
/
i
/
/
.
1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
2
6
1
8
6
8
3
6
8
oh
pag
metro
_
a
_
0
0
0
2
6
pag
d
/
.
i
F
b
y
gramo
tu
mi
s
t
t
oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3
Word Learning Consistency and Variability Braginsky et al.
demonstrated and to identify words from a preselected list (rather than having them recall
them on their own).
Because of the low cost of administering CDI instruments, it is relatively easy to gather
samples containing data about hundreds or thousands of children. Such large samples in turn
make it possible to recover stable estimates of the average difficulty of individual words, incluso
if individual children’s data may be noisy. De este modo, CDI data are typically the dataset of choice
for the studies of vocabulary composition described above.
Finalmente, CDI instruments have been adapted in dozens of different languages, providing
an opportunity for cross-linguistic comparison. The American English CDI is not simply trans-
lated to other languages verbatim; en cambio, expert groups of researchers adapt the form for their
particular linguistic and cultural situation. This process leads to a wide range of forms that share
a common structure, but contain sets of words that are customized to a particular language and
cultura. De este modo, cross-linguistic comparisons do not reflect children’s acquisition of a single set
of words, but instead capture relevant information regarding patterns of children’s vocabulary
development using instruments designed specifically for each language.1
In our study, we conduct cross-linguistic comparisons of the acquisition trajectories
of children’s early-learned words using Wordbank (wordbank.stanford.edu; Franco, Braginsky,
Yurovsky, & Marchman, 2016), an open repository that aggregates administrations of the CDI
across languages. We integrate these acquisition trajectory data with independently derived
characterizations of the word-learning environment from other datasets. The use of secondary
datasets is warranted because no currently available resource provides data on both children’s
language environments and their learning outcomes for more than a small handful of children.
En particular, we derive our estimates of the language environment from transcripts of speech
to children in the CHILDES database (MacWhinney, 2000) and measures of meaning-related
word properties from available psycholinguistic databases. This data-integration methodology
was originated by Goodman et al. (2008); it relies on large samples to average out the (substan-
tial) differences among children and care environments. While introducing additional sources
of variability, this approach allows for analyses that cannot be performed on smaller datasets
that measure only children or environments but not both.
yo
D
oh
w
norte
oh
a
d
mi
d
F
r
oh
metro
h
t
t
pag
:
/
/
d
i
r
mi
C
t
.
metro
i
t
.
/
mi
d
tu
oh
pag
metro
i
/
yo
a
r
t
i
C
mi
–
pag
d
F
/
d
oh
i
/
i
/
/
.
1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
2
6
1
8
6
8
3
6
8
oh
pag
metro
_
a
_
0
0
0
2
6
pag
d
.
/
i
To measure environmental input, we used existing adult speech data from the CHILDES
database to estimate each word’s frequency (a) in speech to children, (b) as a sole utterance
constituent, (C) in utterance-final position, y el (d) mean length in words of utterances
(MLU-w) containing that word. While crude, these measures are both easy to compute and
relatively comparable across languages. To derive proxies for the meaning-based properties of
each word, we accessed available psycholinguistic norms using adult ratings of each word’s
(a) concreteness, (b) valence, (C) arousal, y (d) association with babies. Integrating these
estimados, we predict each word’s acquisition trajectory, assessing the relative contributions
of each predictor, how predictors change over development, and how predictors differ by
lexical category. Since vocabulary composition differs in comprehension and production (p.ej.,
Benedict, 1979), we conduct our analyses independently on each.
F
b
y
gramo
tu
mi
s
t
t
oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3
These analyses address two questions. Primero, we ask about the degree of consistency
across languages in the relative importance of each predictor. para hacerlo, we compare the
1 Por supuesto, observational data of this type are still open to other sources of bias, a point we return to in the
Discusión.
MENTE ABIERTA: Descubrimientos en ciencia cognitiva
54
Word Learning Consistency and Variability Braginsky et al.
estimates for the effect of each predictor for each language and conduct analyses that deter-
mine the likelihood that the consistency of the estimates did not occur by chance. Consistency
in the patterning of predictors would suggest that similar information sources are important for
learners, regardless of language, and that linguistic dissimilarities (p.ej., greater morphological
complexity in Russian, greater phonological complexity in Danish) do not dramatically alter
the course of acquisition. En cambio, evidence for variability across languages would show
the degree to which learners face different challenges in learning different languages, posing
a challenge for more universalist accounts. Más, systematicity in the variability between
languages would reveal which languages are more similar than others in the structure of these
different challenges.
Segundo, we ask which lexical categories are most influenced by linguistic environment
factores, like frequency and utterance length, compared with meaning-based factors like con-
creteness and valence. Division of dominance theory suggests that nouns might be more
sensitive to meaning factors, while predicates and closed-class words might be more sensitive
to linguistic environment factors (Gentner & Boroditsky, 2001). And on syntactic bootstrapping
teorías (Gleitman, 1990), nouns are argued to be learned via frequent co-occurrence (opera-
tionalized by frequency), while verbs might be more sensitive to syntactic factors (operational-
ized here by utterance length; Snedeker, Geren, & Shafto, 2007). De este modo, examining the relative
contribution of different predictors across lexical categories can help test the predictions of
influential theories of acquisition.
MÉTODOS
The code and data for these analyses are available on GitHub (Braginsky, Yurovsky, Marchman,
& Franco, 2019a).
Acquisition Trajectories
To estimate the trajectory of words’ acquisition, we used vocabulary data collected using
CDI instruments adapted in many different languages, including both Words & Gestures
(WG) and Words & Sentences (WS) formas. When filling out a CDI form, parents are either
asked to indicate whether their child “understands” (comprensión) or “understands and
says” (producción) each of around 400–700 words. Both comprehension and production are
queried for younger children and only production is queried for older children. We included
data from the items on the WG form for comprehension, and data from the items in com-
mon between the WG and WS forms for production. Placeholder items, such as “child’s
own name,” were excluded. Mesa 1 gives an overview of our acquisition data (Acarlar et al.,
2008; Bleses et al., 2008; Boudreault, Cabirol, Poulin-Dubois, suton, & Trudeau, 2007;
Caselli et al., 1995; Caselli, Rinaldi, Stefanini, & Volterra, 2012; Eliseeva & Vershinina, 2009;
Eriksson & Berglund, 2002;
Jackson-Maldonado et al., 2003; Kovacevic, Babic, & Brozovic,
1996; Simonsen, Kristoffersen, Bleses, Wehberg, & Jørgensen, 2014; Trudeau & suton, 2011;
Vershinina, Eliseeva, Lavrova, Ryskina, & Zeitlin, 2011; also see Supplemental Information
Figure SI.1, Braginsky, Yurovsky, Marchman, & Franco, 2019b, for the age distributions). Cada uno de
the datasets was collected in the language of the community, p.ej., the Mexican Spanish CDI
data were collected in several areas of Mexico; longitudinal administrations were excluded.
For each word, the CDI data yield a trajectory reflecting the number of children that
are reported to understand or produce the word at each age covered by the instrument (ver
Cifra 1 for some examples).
MENTE ABIERTA: Descubrimientos en ciencia cognitiva
55
yo
D
oh
w
norte
oh
a
d
mi
d
F
r
oh
metro
h
t
t
pag
:
/
/
d
i
r
mi
C
t
.
metro
i
t
.
/
mi
d
tu
oh
pag
metro
i
/
yo
a
r
t
i
C
mi
–
pag
d
F
/
d
oh
i
/
i
/
/
.
1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
2
6
1
8
6
8
3
6
8
oh
pag
metro
_
a
_
0
0
0
2
6
pag
d
/
.
i
F
b
y
gramo
tu
mi
s
t
t
oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3
Word Learning Consistency and Variability Braginsky et al.
Mesa 1. Statistics for data from Wordbank and CHILDES. N indicates number of children.
Production
Comprehension
CHILDES
Idioma
Croatian
Danish
Inglés (Americano)
Francés (Quebec)
italiano
Norwegian
Russian
Español (Mexican)
Swedish
Turkish
CDI items
388
381
393
396
392
380
410
399
371
395
norte
627
6,112
7,312
1,364
1,400
7,466
1,805
1,891
1,367
3,537
Ages
8−30
8−36
8−30
8−30
7−36
8−36
8−36
8−30
8−28
8−36
norte
250
2,398
1,792
537
648
2,374
768
788
467
1,115
Nota. CDI = MacArthur-Bates Communicative Development Inventory.
Ages
8−16
8−20
8−18
8−16
7−24
8−20
8−18
8−18
8−16
8−16
Types
12,064
4,956
45,597
28,819
7,544
10,670
5,191
33,529
8,815
6,503
Tokens
218,775
195,658
7,679,042
2,551,113
188,879
231,763
32,398
1,609,614
359,155
44,347
yo
D
oh
w
norte
oh
a
d
mi
d
F
r
oh
metro
h
t
t
pag
:
/
/
d
i
r
mi
C
t
.
metro
i
t
.
/
mi
d
tu
oh
pag
metro
i
/
yo
a
r
t
i
C
mi
–
pag
d
F
/
d
oh
i
/
i
/
/
.
1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
2
6
1
8
6
8
3
6
8
oh
pag
metro
_
a
_
0
0
0
2
6
pag
d
/
.
i
F
b
y
gramo
tu
mi
s
t
t
oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3
Cifra 1. Example production trajectories for the words “dog” and “jump” across languages. Points show the proportion of children
producing each word for each one-month age group. Lines show the best-fitting logistic curve. Labels show the forms of the words in each
idioma.
Word Properties
For each word in each of our 10 idiomas, we used corpora of child-directed
Overview.
speech in that language from CHILDES to obtain an estimate of its frequency, the mean length
of utterances in which it appears, its frequency as the sole constituent of utterance, and its
frequency in utterance final position. We also computed each word’s length in phonemes.
MENTE ABIERTA: Descubrimientos en ciencia cognitiva
56
Word Learning Consistency and Variability Braginsky et al.
Mesa 2.
Items with the highest and lowest values for each predictor in English.
Predictor
Arousal
Babiness
Concreteness
Final frequency
Frecuencia
MLU-w
Number phonemes
Solo frequency
Valence
Highest
Lowest
naughty, dinero, scared
baby, bib, bottle
apple, baby, ball
libro, él, allá
tú, él, eso
daddy, cuando, día
refrigerator, cockadoodledoo, babysitter
No, Sí, thank you
happy, hug, love
hoy, asleep, shh
jeans, penny, donkey
eso, now, cómo
put, cuando, give
babysitter, rocking chair, grrr
ouch, thank you, peekaboo
i, ojo, ear
feed, bathroom, tooth
ouch, hurt, sick
Nota. MLU-w = mean length of utterances.
Además, each word’s concreteness, valence, arousal, and relatedness to babies2 were
compiled from ratings based on previous studies using adult raters. Since existing ratings are
primarily available for English, we mapped all words onto translation equivalents across CDI
formas, verified by native speaker judgment, allowing us to use the English ratings across lan-
calibres. Of the resulting translation equivalent meanings, 35% occur only in one language,
51% occur in more than one but not all languages, y 14% occur in all languages. Mientras
necessarily imperfect, this method allows us to examine languages for which limited resources
existir. Example words for these predictors in English are shown in Table 2 (also see Figures SI.2
and SI.3, Braginsky et al., 2019b, for the distributions of values of each predictor).
Each numeric predictor was centered and scaled (within language) so that all predictors
would have comparable units.
For each language, we derived unigram counts based on adult speech in all cor-
Frecuencia.
pora in CHILDES for that language. Frequencies varied widely both within and across lexical
categories (see Figure SI.4, Braginsky et al., 2019b). Each word’s count was summed across
inflected forms (p.ej., “dogs” counts as “dog”) and synonyms (p.ej., “father” counts as “daddy”).
For polysemous words (p.ej., “orange” as in color or fruit), occurrences were split uniformly be-
tween the senses on the CDI (there were only between 1 y 10 such word pairs in the various
idiomas; in the absence of cross-linguistic corpus resources for sense disambiguation, este
is a necessary simplification). Counts were normalized to the length of each corpus, Laplace
smoothed (es decir., counts of 0 were replaced with counts of 1), and log transformed.
Solo and Final Frequencies. Using the same dataset as for frequency, we estimated the fre-
quency with which each word occurred as the sole word in an utterance, and the final word
of an utterance (not counting single-word utterances). Solo and final counts were normalized
to the length of each corpus, Laplace smoothed, and log transformed. Since both of these
2 Previous studies have shown robust consistency in the types of words that children learn very early (Tardif
et al., 2008). These words seem to describe concepts that are important or exciting in the lives of infants in a
way that standard psycholinguistic features like concreteness do not. Capturing this intuition quantitatively is
difficult, but Perry et al. (2015) provide a proxy measure as a first step. This measure is simply the degree to
which a particular word was “associated with babies.” Intuitively, we expect this measure to capture the degree
to which words like “ball” or “bottle” feature heavily in the environment (and presumably, mental life) of many
babies.
MENTE ABIERTA: Descubrimientos en ciencia cognitiva
57
yo
D
oh
w
norte
oh
a
d
mi
d
F
r
oh
metro
h
t
t
pag
:
/
/
d
i
r
mi
C
t
.
metro
i
t
.
/
mi
d
tu
oh
pag
metro
i
/
yo
a
r
t
i
C
mi
–
pag
d
F
/
d
oh
i
/
i
.
/
/
1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
2
6
1
8
6
8
3
6
8
oh
pag
metro
_
a
_
0
0
0
2
6
pag
d
/
.
i
F
b
y
gramo
tu
mi
s
t
t
oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3
Word Learning Consistency and Variability Braginsky et al.
estimates are by necessity highly correlated with frequency, we then residualized unigram fre-
quency out of both, so that values reflect an estimate of the effects of solo frequency and final
frequency over and above frequency.
MLU-w. MLU-w is only a rough proxy for syntactic complexity, but is relatively straight-
forward to compute across languages (in contrast to other metrics). For each language, nosotros
estimated each word’s MLU-w by calculating the mean length in words of the utterances in
which that word appeared, for all corpora for that language. For words that occurred fewer
than 10 veces, MLU-w estimates were treated as missing.
In the absence of consistent resources for cross-linguistic pronuncia-
Number of Phonemes.
ción, we computed the number of phonemes in each word in each language based on phone-
mic transcriptions of each word obtained using the eSpeak tool (Duddington, 2012). Nosotros entonces
spot-checked these transcriptions for accuracy.
Concreteness. We used previously collected norms for concreteness (Brysbaert, Warriner, &
Kuperman, 2014), which were gathered by asking adult participants to rate how concrete the
meaning of each word is on a 5-point scale from abstract to concrete.
Valence and Arousal. We also used previously collected norms for valence and arousal
(Warriner, Kuperman, & Brysbaert, 2013), for which adult participants were asked to rate words
on a 1–9 happy-unhappy scale (valence) and 1–9 excited-calm scale (arousal).
Babiness. We used previously collected norms of “babiness,” a measure of association with
infancy (Perry et al., 2015) for which adult participants were asked to judge a word’s association
with babies on a 1–10 scale.
Category was determined on the basis of the conceptual categories pre-
Lexical Category.
sented on the CDI form (p.ej., “Animals,” “Action Words”), such that the Nouns category
contains common nouns, Predicates contains verbs, adjectives, and adverbs, Function Words
contains closed-class words (following E. Bates et al., 1994), and the remaining items are
binned as Other.
The resulting set of predictor value for each language had varying numbers
Imputation.
of missing values, depending on resource availability (number phonemes 0%, concreteness
0%–1%, arousal and valence 8%–13%, [solo/final] frequency 2%–14%, babiness 10%–33%,
MLU-w 2%–53%). We used iterative regression imputation within each language to fill in these
missing values by first replacing missing values with samples drawn randomly with replace-
ment from the observed values, and then iteratively imputing values for each predictor based
on a linear regression fitting that predictor from all others.
A potential concern for comparing coefficient estimates is predictor collinear-
Collinearity.
idad. Fortunately, in every language, the only relatively high correlations were between MLU-w
and solo frequency (mean over languages r = −0.44), as expected given the similarity of
these factors, along with modest correlations between frequency and concreteness (significar
over languages r = −0.36) and between frequency and number of phonemes (mean over
languages r = −0.33), a reflection of Zipf’s Law (Zipf, 1935). More importantly, the variance
MENTE ABIERTA: Descubrimientos en ciencia cognitiva
58
yo
D
oh
w
norte
oh
a
d
mi
d
F
r
oh
metro
h
t
t
pag
:
/
/
d
i
r
mi
C
t
.
metro
i
t
.
/
mi
d
tu
oh
pag
metro
i
/
yo
a
r
t
i
C
mi
–
pag
d
F
/
d
oh
i
/
i
/
/
.
1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
2
6
1
8
6
8
3
6
8
oh
pag
metro
_
a
_
0
0
0
2
6
pag
d
/
.
i
F
b
y
gramo
tu
mi
s
t
t
oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3
Word Learning Consistency and Variability Braginsky et al.
inflation factor for each predictor in each language was no greater than 2.27, Indicando que
multicollinearity among the predictors is low (see Figure SI.5 for the full set of pairwise corre-
lations and Figure SI.6 for the variance inflation factors, Braginsky et al., 2019b).
Análisis
We used mixed-effects logistic regression models (fit with the MixedModels package in Julia;
D. Bates et al., 2018) to predict whether each child understands/produces each word from
the child’s age, properties of the word, interactions between each property and age, and inter-
actions between each property and lexical category (which was contrast coded). Each model was
fit to all data from a particular language and included a random intercept for each word and a
random slope of age for each word. Computational and technical limitations prevented us from
including random effects for child or including data from all languages in one joint model.
The magnitude of the standardized coefficient on each property gives an estimate of its
independent contribution to words being understood/produced by more children. Interactions
between properties and age give estimates of how this effect is modulated for earlier-learned
and later-learned words. Por ejemplo, a positive effect of babiness means that words as-
sociated with babies are learned earlier; a negative interaction with age means that high
babiness primarily leads to higher rates of production and comprehension for younger children.
Similarmente, interactions between properties and lexical category give estimates of how the effect
differs among nouns, predicates, and function words.
yo
D
oh
w
norte
oh
a
d
mi
d
F
r
oh
metro
h
t
t
pag
:
/
/
d
i
r
mi
C
t
.
metro
i
t
.
/
mi
d
tu
oh
pag
metro
i
/
yo
a
r
t
i
C
mi
–
pag
d
F
/
d
oh
i
/
i
/
.
/
1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
2
6
1
8
6
8
3
6
8
oh
pag
metro
_
a
_
0
0
0
2
6
pag
d
.
/
i
F
b
y
gramo
tu
mi
s
t
t
oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3
Cifra 2. Estimates of coefficients in predicting words’ developmental trajectories for English
comprehension and production data. Larger coefficient values indicate a greater effect of the pre-
dictor on acquisition: positive main effects indicate that words with higher values of the predictor
tend to be understood/produced by more children, while negative main effects indicate that words
with lower values of the predictor tend to be understood/produced by more children; positive age
interactions indicate that the predictor’s effect increases with age, while negative age interactions
indicate the predictor’s effect decreases with age. Line ranges indicate 95% confidence intervals;
filled in points indicate coefficients for which p < .05.
OPEN MIND: Discoveries in Cognitive Science
59
Word Learning Consistency and Variability Braginsky et al.
RESULTS
English Predictor Effects
To illustrate the structure of our analysis, we first describe the results for English data, shown in
Figure 2 as the main effect and age interaction coefficient estimates and 95% confidence inter-
vals, for comprehension and production. For main effects, words are more likely to be known
by more children if they are higher in frequency or concreteness, as well as in babiness for
comprehension and in sentence-final frequency or sole-constituent frequency for production.
In contrast, words that appear in shorter sentences (MLU-w) are more likely to be reported
as understood or produced. For age interactions, while most predictors have consistent effects
over age, words that are higher in frequency or concreteness are more likely to be known more
by older children, while words that are higher in valence have a greater effect on acquisition
in younger children, with an additional negative interaction with babiness in comprehension
and positive interaction with MLU-w in production.
Cross-linguistic Predictor Effects
Figure 3 shows the coefficient estimate for each predictor in each language and measure (for
additional visualizations of the coefficients, see Figures SI.7, SI.8, and SI.9, Braginsky et al.,
2019b). We find that frequency is the strongest predictor of acquisition (mean across languages
and measures ¯β = 0.23). Other relatively strong overall predictors include concreteness ( ¯β =
0.18), solo frequency ( ¯β = 0.17), MLU-w ( ¯β = −0.14), and final frequency ( ¯β = 0.13). Number
of phonemes is comparatively large for production ( ¯β = −0.31) but not comprehension ( ¯β =
−0.07); conversely, babiness is comparatively large for comprehension ( ¯β = 0.19) but not
production ( ¯β = 0.08). Finally, valence ( ¯β = 0.06) and arousal ( ¯β = 0.003) have much smaller
effects.
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
o
p
m
i
/
l
a
r
t
i
c
e
-
p
d
f
/
d
o
i
/
i
/
.
/
1
0
1
1
6
2
o
p
m
_
a
_
0
0
0
2
6
1
8
6
8
3
6
8
o
p
m
_
a
_
0
0
0
2
6
p
d
/
.
i
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Figure 3. Estimates of coefficients in predicting words’ developmental trajectories for all lan-
guages and measures. Each point represents a predictor’s coefficient in one language, with the bar
showing the mean across languages. Filled in points indicate coefficients for which p < .05.
OPEN MIND: Discoveries in Cognitive Science
60
Word Learning Consistency and Variability Braginsky et al.
Given the emphasis on frequency effects in the literature (Ambridge, Kidd, Rowland, &
Theakston, 2015), one might have expected frequency to dominate, but several other predictors
are also quite strong. In addition, some factors previously argued to be important for word
learning, namely valence and arousal (Moors et al., 2013), appear to have limited relevance
when compared to other factors. These results provide a strong argument for our approach of
including multiple predictors and languages in our analysis.
Consistency
Apart from valence and arousal, all other predictors have the same the direction of effect in
all or almost all languages and measures (at least 17 of the 20). Thus, across languages, words
are likely to be understood and produced by more children if they are more frequent, shorter,
more concrete, more frequently the only word in an utterance, more associated with babies,
more frequently the final word in an utterance, and appear in shorter utterances.
Additionally, there is considerable consistency in the magnitudes of predictors across
languages. A priori it could have been the case that different languages have wildly different
effects of various factors (due to linguistic or cultural differences), but this pattern is not what
we observe. Instead, there is more consistency in the correlations between coefficients across
languages than would be expected by chance. As shown in Figure 4, each language’s mean
pairwise correlation with other languages’ coefficients (i.e., the correlation of coefficients for
English with coefficients for Russian, for Spanish, and so on) is outside of bootstrapped esti-
mates in a randomized baseline created by shuffling predictor coefficients within language. The
pairwise correlations are more consistent for production (mean 0.72) than for comprehension
(mean 0.56), in which French and Russian effects are more idiosyncratic.
Variability
While some particular coefficients differ substantially from the trend across languages (e.g.,
the effect of frequency for comprehension in Spanish is near 0), these individual datapoints are
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
o
p
m
i
/
l
a
r
t
i
c
e
-
p
d
f
/
d
o
i
/
i
.
/
/
1
0
1
1
6
2
o
p
m
_
a
_
0
0
0
2
6
1
8
6
8
3
6
8
o
p
m
_
a
_
0
0
0
2
6
p
d
/
.
i
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Figure 4. Correlations of coefficient estimates between languages. Each point represents the
mean of one language’s coefficients’ correlation with each other language’s coefficients, with the
vertical line indicating the overall mean across languages. The shaded region and line show a boot-
strapped 95% confidence interval for a randomized baseline where predictor coefficients are shuf-
fled within language.
OPEN MIND: Discoveries in Cognitive Science
61
Word Learning Consistency and Variability Braginsky et al.
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
o
p
m
i
/
l
a
r
t
i
c
e
-
p
d
f
/
d
o
i
/
i
/
.
/
1
0
1
1
6
2
o
p
m
_
a
_
0
0
0
2
6
1
8
6
8
3
6
8
o
p
m
_
a
_
0
0
0
2
6
p
d
/
.
i
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Figure 5. Dendrograms of the similarity structure among languages’ coefficients.
difficult to interpret. Many unmeasurable factors could potentially account for these dif-
ferences: Spanish frequency estimates could be less accurate due to corpus sparsity or idio-
syncrasy, the samples of children in the Spanish CDI or CHILDES data could differ more
demographically, or Spanish-learning children could in fact rely less on frequency. Rather than
attempting to interpret individual coefficients, we instead ask how the patterns of difference
among languages reflect systematic substructure in the variability of the effects.
To examine the substructure of predictor variability, we used hierarchical clustering
analysis to find the similarity structure among the pairwise correlations between languages’
predictors. The resulting dendrograms are shown in Figure 5; these broadly reflect language
typology, especially for production data. This result suggests that some language-to-language
similarity is captured by the profile of coefficient magnitudes our analysis returns.
Comprehension vs. Production
As mentioned above, word length is the one predictor of acquisition that varied substantially
between measures: it is far more predictive for production than comprehension. Thus, as mea-
sured here, length seems to reflect effects of production constraints (i.e., how difficult a word
is to say) rather than comprehension constraints (i.e., how difficult it is to store or access).
This result may explain why the hierarchical clustering analysis above appears more similar
to linguistic typology in production than comprehension, that is, the role of production diffi-
culty may be more similar for more typologically related languages. Another possibility is that
since the measures are confounded with age (comprehension is only measured for younger
children), word length may play a larger role later in acquisition. Similarly, the stronger effect
of babiness in comprehension over production could be due to its larger prominence earlier
in development.
OPEN MIND: Discoveries in Cognitive Science
62
Word Learning Consistency and Variability Braginsky et al.
Developmental Change
For both comprehension and production, positive age interactions can be seen in at least 9 out
of 10 languages for concreteness and frequency. Conversely, there are negative age interactions
for babiness and valence for comprehension in at least 9 out of 10 languages. This sug-
gests that concreteness and frequency facilitate learning more so later in development, while
babiness and valence facilitate learning earlier in development. This result is consistent with
the speculation above that the babiness predictor captures meanings that have special salience
to very young infants.
Lexical Categories
Previous work suggests that predictors’ relationship with age of acquisition differs among lex-
ical categories (Goodman et al., 2008). We investigate these differences by including lexical
category interaction terms in our model. Figure 6 shows the resulting effects for each lexical
category, combining the main effect of a given predictor with the main effect of the lexical cate-
gory and the interaction between that predictor and that lexical category (see also Figures SI.10
and SI.11, Braginsky et al., 2019b).
Across languages, the strongest predictors of acquisition for both nouns and predicates
are concreteness (nouns ¯β = 0.44; predicates ¯β = 0.28) and frequency (nouns ¯β = 0.36; pred-
icates ¯β = 0.41). Thus content words are most likely to be known by more children if they are
more frequent or more concrete. Conversely, function words are most influenced by number
of phonemes ( ¯β = −0.74), babiness ( ¯β = −0.61), and MLU-w ( ¯β = −0.61), meaning that func-
tion words are most likely to be known by more children if they are shorter, less associated
with babies, or appear in shorter sentences. These patterns are supportive of the hypothesis that
different word classes are learned in different ways, or at least that the bottleneck on learning
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
o
p
m
i
/
l
a
r
t
i
c
e
-
p
d
f
/
d
o
i
/
i
/
/
.
1
0
1
1
6
2
o
p
m
_
a
_
0
0
0
2
6
1
8
6
8
3
6
8
o
p
m
_
a
_
0
0
0
2
6
p
d
.
/
i
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Figure 6. Estimates of effects in predicting words’ developmental trajectories for each language,
measure, and lexical category (main effect of predictor + main effect of lexical category + inter-
action between predictor and lexical category). Each point represents a predictor’s effect in one
language, with the bar showing the mean across languages.
OPEN MIND: Discoveries in Cognitive Science
63
Word Learning Consistency and Variability Braginsky et al.
tends to be different, leading to different information sources being more or less important
across categories.
Additionally, the mean pairwise correlation of coefficients between languages is much
larger for nouns (0.68) and predicates (0.54) than for function words (0.29). The higher
between-language variability for function words suggests the learning processes differ substan-
tially more across languages for function words than they do for content words (see Figure SI.12,
Braginsky et al., 2019b).
DISCUSSION
What makes words easier or harder for young children to learn? Previous experimental work
has largely addressed this question using small-scale laboratory studies. While such experi-
ments can identify sources of variation, they typically do not allow for different sources to
be compared directly. In contrast, observational studies allow the effects of individual factors
to be measured across ages and lexical categories (e.g., Goodman et al., 2008; Hills et al.,
2009; Swingley & Humphrey, 2018), but are limited in the size and scope of the datasets and
languages that can be directly compared. The current analyses take advantage of recent inno-
vative approaches via Wordbank, a large, cross-linguistic dataset of parent report instruments.
By compiling data regarding early lexical development across 10 languages and examining
patterns of acquisition in relation to 9 predictors, our work expands the scope of these studies
dramatically, leading to several new findings.
First, we found consistency in the patterning of predictors across languages at a level
substantially greater than the predictions of a chance model. This consistency supports the
idea that differences in culture or language structure do not lead to fundamentally different
acquisition strategies, at least at the level of detail we were able to examine. Instead, they are
likely produced by processes that are similar across populations and languages. Such processes
could include learning mechanisms or biases internal to children, or interactional dynamics
between children or caregivers. We believe these consistencies should be an important topic
for future investigation.
Second, predictors varied substantially in their weights across lexical categories. Fre-
quent, concrete nouns were learned earlier, consistent with theories that emphasize the im-
portance of early referential speech (e.g., Baldwin, 1995). For predicates, concreteness was
somewhat less important and frequency was somewhat more important. And for function
words, length and MLU-w were more predictive, perhaps because it is easiest to decode the
meanings of function words that are used in short sentences (or because such words have
meanings that are easiest to decode). Overall, these findings are consistent with some predic-
tions of both division of dominance theory, which highlights the role of conceptual structure
in noun acquisition (Gentner & Boroditsky, 2001), and syntactic bootstrapping theory, which
emphasizes linguistic structure over conceptual complexity in the acquisition of lexical cate-
gories other than nouns (Snedeker et al., 2007). More generally, our methods here provide a
way forward for testing the predictions of these theories across languages and at the level of
the entire lexicon rather than individual words.
In addition to these new insights, several findings emerged that confirm and expand
previous reports. Environmental frequency was an important predictor of learning, with more
frequently heard words learned earlier (Goodman et al., 2008; Swingley & Humphrey, 2018).
Predictors also changed in relative importance across development. For example, certain words
whose meanings were more strongly associated with babies appeared to be learned early for
OPEN MIND: Discoveries in Cognitive Science
64
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
o
p
m
i
/
l
a
r
t
i
c
e
-
p
d
f
/
d
o
i
/
i
/
.
/
1
0
1
1
6
2
o
p
m
_
a
_
0
0
0
2
6
1
8
6
8
3
6
8
o
p
m
_
a
_
0
0
0
2
6
p
d
.
/
i
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Word Learning Consistency and Variability Braginsky et al.
children across the languages in our sample (as in Tardif et al., 2008). Finally, word length
showed a dissociation between comprehension and production, suggesting that challenges in
production do not carry over to comprehension (at least in parent-report data).
Despite its larger scope, our work shares a number of important limitations with pre-
vious studies. First and foremost, our approach is to predict acquisition data for one set of
individuals from data about the experiences of a completely different set of individuals and
from conceptual ratings gathered from yet others. In contrast to dense-data analyses (Roy et al.,
2015), this approach fundamentally limits the amount of variability we will be able to capture.
Second, the granularity of the predictors that can be extracted from corpus data and applied to
every word is necessarily quite coarse. Ideally, predictors could be targeted more specifically
at particular theoretical constructs of interest (e.g., the patterns of use for specific predicates).
Third, our analyses are conducted within language, so to the extent that the predictors can
have differing ranges in different languages, cross-linguistic patterns in predictor effects could
be obscured.
Finally, our data are observations gleaned from parent report. CDI instruments are both
reliable and valid, and the cross-linguistic adaptations we used contain the original researchers’
best attempts to create culturally appropriate word lists. Nevertheless, this observational de-
sign introduces many sources of uncertainty and bias. First, the open data format of Wordbank
reflects the sampling and administration methods of many groups around the world; these in-
troduce many unknown biases that we cannot control (though they would likely not contribute
to observed consistencies). Second, language and culture covary completely in our sample and
so variability that we observe cannot be attributed to one or the other. Finally, some observed
consistencies could arise from consistency in parental reporting biases. For example, across
languages, parents might be generally biased to underreport comprehension of function words.
Despite the quantity of data analyzed here, our conclusions will require further testing through
converging evidence from both laboratory experiments and direct observation.
In sum, by examining predictors of early word learning across languages, we identified
substantial cross-linguistic consistency in the factors contributing to the ease or difficulty of
learning individual words. This suggests that common learning mechanisms and/or environ-
mental supports for learning are shared across all of these languages. These findings also testify
to the importance of building open, shared resources in the study of child language learning—
without the efforts of many research groups across many language communities, studies like
ours would be impossible. Additionally, we hope that our work here provides a baseline for
the building of future predictive models that allow theories of language learning to be tested
at scale.
ACKNOWLEDGMENTS
Thank you to the labs and individuals who contributed data to Wordbank.
FUNDING INFORMATION
National Science Foundation, Award ID 1528526 (to MCF); Zhou Fund for Language and
Cognition Research, Stanford University (to MCF); Jacobs Foundation Research Fellowship
(to MCF); National Institutes of Health National Research Service Award (to DY); James S.
McDonnell Scholar Award (to DY).
OPEN MIND: Discoveries in Cognitive Science
65
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
o
p
m
i
/
l
a
r
t
i
c
e
-
p
d
f
/
d
o
i
/
i
.
/
/
1
0
1
1
6
2
o
p
m
_
a
_
0
0
0
2
6
1
8
6
8
3
6
8
o
p
m
_
a
_
0
0
0
2
6
p
d
.
/
i
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Word Learning Consistency and Variability Braginsky et al.
AUTHOR CONTRIBUTIONS
MB: Conceptualization: Equal; Data curation: Lead; Formal analysis: Lead; Funding acqui-
sition: Supporting; Methodology: Equal; Visualization: Lead; Writing - Original Draft: Lead;
Writing - Review & Editing: Equal. DY: Conceptualization: Equal; Data curation: Equal; For-
mal analysis: Equal; Funding acquisition: Supporting; Methodology: Equal; Visualization: Sup-
porting; Writing - Original Draft: Supporting; Writing - Review & Editing: Supporting. VAM:
Conceptualization: Equal; Data curation: Supporting; Formal analysis: Equal; Funding acqui-
sition: Equal; Methodology: Equal; Visualization: Supporting; Writing - Original Draft: Equal;
Writing - Review & Editing: Equal. MCF: Conceptualization: Lead; Data curation: Support-
ing; Formal analysis: Equal; Funding acquisition: Lead; Methodology: Equal; Visualization:
Supporting; Writing - Original Draft: Equal; Writing - Review & Editing: Lead.
REFERENCES
Acarlar, F., Aksu-Koç, A., Küntay, A. C., Mavi¸s, ˙I., Sofu, H., Topba¸s,
S., & Turan, F. (2008). Adapting MB-CDI to Turkish: The first
phase. In S. Ay, Ö. Aydn, I. Ergenç, S. Gökmen, S. ˙I¸ssever, &
D. Peçenek (Eds.), Essays of Turkish linguistics: Proceedings of
the 14th International Conference on Turkish Linguistics, Ankara,
Turkey, August 6–8, 2008 (pp. 313–320). Wiesbaden, Germany:
Harrassowitz Verlag.
Ambridge, B., Kidd, E., Rowland, C. F., & Theakston, A. L. (2015).
The ubiquity of frequency effects in first language acquisition.
Journal of Child Language, 42, 239–273.
Baldwin, D. A. (1993). Infants’ ability to consult the speaker for clues
to word reference. Journal of Child Language, 20, 395–418.
Baldwin, D. A.
(1995). Understanding the link between joint
attention and language. In C. Moore & P. J. Dunham (Eds.), Joint
attention: Its origins and role in development (pp. 131–158).
Hillsdale, NJ: Erlbaum.
Bates, D., Kelman, T., Kleinschmidt, D., AB, S., Mogensen, P. K.,
Bouchet-Valat, M., . . . Noack, A. (2018). dmbates/MixedModels.jl:
Add adaptive Gauss-Hermite quadrature (Version v0.81.1).
Zenodo. doi:10.5281/zenodo.1303403
Bates, E., & MacWhinney, B. (1987). Competition, variation, and
language learning. In B. MacWhinney (Ed.), Mechanisms of lan-
guage acquisition: The 20th Annual Carnegie Mellon Symposium
on Cognition (pp. 157–193). Hillsdale, NJ: Erlbaum.
Bates, E., Marchman, V., Thal, D., Fenson, L., Dale, P., Reznick, J. S.,
. . . Hartung, J. (1994). Developmental and stylistic variation in
the composition of early vocabulary. Journal of Child Language,
21, 85–123.
Benedict, H. (1979). Early lexical development: Comprehension and
production. Journal of Child Language, 6, 183–200.
Bleses, D., Vach, W., Slott, M., Wehberg, S., Thomsen, P., Madsen,
T. O., . . . Basbøll, H. (2008). The Danish Communicative
Developmental Inventories: Validity and main developmental
trends. Journal of Child Language, 35, 651–669.
Bloom, P. (2000). How children learn the meanings of words.
Cambridge, MA: MIT Press.
Bornstein, M. H., & Haynes, O. M. (1998). Vocabulary competence
in early childhood: Measurement, latent construct, and predic-
tive validity. Child Development, 69, 654–671.
Boudreault, M., Cabirol, E., Poulin-Dubois, D., Sutton, A., &
Trudeau, N. (2007). MacArthur Communicative Development
Inventories: Validity and preliminary normative data. Revue
canadienne d’orthophonie et d’audiologie, 31, 27–37.
Braginsky, M., Yurovsky, D., Marchman, V. A., & Frank, M. C. (2019a,
April 4). Estimating and predicting words’ age of acquisition. Re-
trieved from https://github.com/mikabr/aoa-prediction
Braginsky, M., Yurovsky, D., Marchman, V. A., & Frank, M. C.
(2019b). Supplemental material for “Consistency and variability
in children’s word learning across languages.” Open Mind: Dis-
coveries in Cognitive Science, 3. doi:10.1162/opmi_a_00026
Brysbaert, M., Warriner, A. B., & Kuperman, V. (2014). Concreteness
ratings for 40 thousand generally known English word lemmas.
Behavioral Research Methods, 46, 904–911.
Caselli, M. C., Bates, E., Casadio, P., Fenson, J., Fenson, L., Sanderl,
L., & Weir, J. (1995). A cross-linguistic study of early lexical de-
velopment. Cognitive Development, 10, 159–199.
Caselli, M. C., Rinaldi, P., Stefanini, S., & Volterra, V. (2012). Early
action and gesture “vocabulary” and its relation with word com-
prehension and production. Child Development, 83, 526–542.
Duddington, J. (2012). eSpeak text to speech. espeak.sourceforge.net
Eliseeva, M. B., & Vershinina, E. A.
(2009).
(cid:0)(cid:2)(cid:3)(cid:4)(cid:5)(cid:4)(cid:6)(cid:7)(cid:2)
(cid:8)(cid:4)(cid:6)(cid:9)(cid:10)(cid:5)(cid:11)(cid:12)(cid:7) (cid:6)(cid:2)(cid:13)(cid:2)(cid:14)(cid:4)(cid:15)(cid:4) (cid:6)(cid:10)(cid:12)(cid:14)(cid:11)(cid:5)(cid:11)(cid:16) (cid:17)(cid:2)(cid:5)(cid:18) (cid:4)(cid:5)
(cid:17)(cid:4)
(cid:9)(cid:2)(cid:19)
18
36
(cid:16)(cid:20)(cid:2)(cid:14) (cid:21)(cid:22)(cid:4) (cid:9)(cid:10)(cid:5)(cid:2)(cid:6)(cid:11)(cid:10)(cid:23)(cid:10)(cid:9) (cid:24)(cid:10)(cid:3)(cid:25)(cid:6)(cid:5)(cid:26)(cid:6)(cid:4)(cid:14)(cid:19)(cid:3)(cid:4)(cid:5)(cid:4) (cid:4)(cid:22)(cid:6)(cid:4)(cid:19)
(cid:8)(cid:11)(cid:3)(cid:10)(cid:27)
[Some norms of
language development of children
the
from 18 to 36 months (based on the materials of
MacArthur questionnaire)]. In T. A. Kruglakova, M. B. Elesseva,
M. A. Elivanova, & I. N. Levina (Eds.),
(cid:28)(cid:6)(cid:4)(cid:29)(cid:23)(cid:2)(cid:9)(cid:7)
(cid:4)(cid:8)(cid:5)(cid:4)(cid:23)(cid:11)(cid:8)(cid:15)(cid:14)(cid:11)(cid:19)(cid:5)(cid:11)(cid:3)(cid:11) (cid:30) (cid:31) !" (cid:9)(cid:10)(cid:5)(cid:2)(cid:6)(cid:11)(cid:10)(cid:23)(cid:7) (cid:9)(cid:2)#(cid:17)(cid:26)(cid:8)(cid:10)(cid:6)(cid:4)(cid:17)(cid:8)(cid:4)(cid:18)
–
(cid:3)(cid:4)$(cid:2)(cid:6)(cid:2)(cid:8)(cid:20)(cid:11)(cid:11) (cid:21)%&
%! (cid:11)’(cid:8)(cid:16) (cid:31) ! (cid:15)( )(cid:10)(cid:8)(cid:3)(cid:5)(cid:30)(cid:28)(cid:2)(cid:5)(cid:2)(cid:6)(cid:29)(cid:26)(cid:6)(cid:15)(cid:27)(
[Problems in Developmental Linguistics—2009: Materials of the
International Conference (June 17–19, 2009, Saint Petersburg,
Russia)] (pp. 22–29). Saint Petersburg, Russia: Zlatoust.
Eriksson, M., & Berglund, E. (2002). Instruments, scoring manual
and percentile levels of the Swedish Early Communicative Devel-
opment Inventory, SECDI (Volume 17 of FoU-rapport). Gävle,
Sweden: Högskolan i Gävle.
Feldman, H. M., Dollaghan, C. A., Campbell, T. F., Kurs-Lasky, M.,
Janosky, J. E., & Paradise, J. L. (2000). Measurement properties of
the MacArthur Communicative Development Inventories at ages
one and two years. Child Development, 71, 310–322.
Fenson, L., Bates, E., Dale, P., Goodman, J. C., Reznick, J. S., & Thal,
D. (2000). Reply: Measuring variability in early child language:
Don’t shoot the messenger. Child Development, 71, 323–328.
OPEN MIND: Discoveries in Cognitive Science
66
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
o
p
m
i
/
l
a
r
t
i
c
e
-
p
d
f
/
d
o
i
/
i
.
/
/
1
0
1
1
6
2
o
p
m
_
a
_
0
0
0
2
6
1
8
6
8
3
6
8
o
p
m
_
a
_
0
0
0
2
6
p
d
/
.
i
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Word Learning Consistency and Variability Braginsky et al.
Fenson, L., Dale, P., Reznick, J., Bates, E., Thal, D., Pethick, S., . . .
Stiles, J. (1994). Variability in early communicative development.
Monographs of the Society for Research in Child Development,
59(5, Series No. 242).
Fenson, L., Marchman, V. A., Thal, D., Dale, P. S., Reznick, J. S.,
& Bates, E. (2007). MacArthur-Bates Communicative Develop-
ment Inventories: User’s guide and technical manual (2nd ed.).
Baltimore, MD: Brookes Publishing.
Fernald, A., Perfors, A., & Marchman, V. A. (2006). Picking up speed in
understanding: Speech processing efficiency andvocabulary growth
across the 2nd year. Developmental Psychology, 42, 98–116.
Frank, M. C., Braginsky, M., Yurovsky, D., & Marchman, V. A. (2016).
Wordbank: An open repository for developmental vocabulary
data. Journal of Child Language, 44, 677–694.
Gentner, D., & Boroditsky, L. (2001). Individuation, relativity, and early
word learning. In M. Bowerman & S. Levinson (Eds.), Language
acquisition and conceptual development. (Vol. 3, pp. 215–256).
Cambridge, England: Cambridge University Press.
Gleitman, L.
(1990). The structural sources of verb meanings.
Language Acquisition, 1, 3–55.
Goodman, J. C., Dale, P. S., & Li, P. (2008). Does frequency count?
Journal of
Parental input and the acquisuisition of vocabulary.
Child Language, 35, 515–531.
Hills, T. T., Maouene, M., Maouene, J., Sheya, A., & Smith, L. (2009).
Longitudinal analysis of early semantic networks: Preferential
attachment or preferential acquisition? Psychological Science,
20, 729–739.
Hollich, G. J., Hirsh-Pasek, K., Golinkoff, R. M., Brand, R. J., Brown,
E., Chung, H. L., . . . Bloom, L.
(2000). Breaking the lan-
guage barrier: An emergentist coalition model for the origins of
word learning. Monographs of the Society for Research in Child
Development, 65(3, Serial No. 262).
Jackson-Maldonado, D., Thal, D. J., Fenson, L., Marchman, V. A.,
Newton, T., & Conboy, B. T. (2003). Macarthur Inventarios del
Desarrollo de Habilidades Comunicativas: User’s guide and
technical manual. Baltimore, MD: Brookes Publishing.
Kovacevic, M., Babic, Z., & Brozovic, B. (1996). A Croatian language
parent report study: Lexical and grammatical development. Paper
presented at the Seventh International Congress for the Study of
Child Language, July 14–19, 1996, Istanbul, Turkey.
MacWhinney, B. (2000). The CHILDES project: Tools for analyzing
talk (3rd ed.). Mahwah, NJ: Erlbaum.
Mayor, J., & Plunkett, K. (2011). A statistical estimate of infant and
toddler vocabulary size from CDI analysis. Developmental Sci-
ence, 14, 769–785.
Mayor, J., & Plunkett, K. (2014). Shared understanding and idio-
syncratic expression in early vocabularies. Developmental Science,
17, 412–423.
Mintz, T. H. (2003). Frequent frames as a cue for grammatical cat-
egories in child directed speech. Cognition, 90, 91–117.
Moors, A., De Houwer, J., Hermans, D., Wanmaker, S., Van Schie,
K., Van Harmelen, A.-L., . . . Brysbaert, M. (2013). Norms of
valence, arousal, dominance, and age of acquisition for 4,300
Dutch words. Behavioral Research Methods, 45, 169–177.
Nelson, K. (1973). Structure and strategy in learning to talk. Mono-
the Society for Research in Child Development,
graphs of
38(1–2, Serial No. 149), 1–135.
Perry, L. K., Perlman, M., & Lupyan, G. (2015). Iconicity in English
and Spanish and its relation to lexical category and age of
acquisition. PloS ONE, 10(9), e0137147.
Roy, B. C., Frank, M. C., DeCamp, P., Miller, M., & Roy, D. (2015).
Predicting the birth of a spoken word. Proceedings of the Na-
tional Academy of Sciences, 112, 12663–12668.
Schneider, R., Yurovsky, D., & Frank, M. C. (2015). Large-scale in-
vestigations of variability in children’s first words. In D. C. Noelle
et al. (Ed.), Proceedings of the 37th Annual Meeting of the Cog-
nitive Science Society, Pasadena, California, July 22–25, 2015
(pp. 2110–2115). Austin, TX: Cognitive Science Society.
Schwartz, R. G., & Terrell, B. Y. (1983). The role of input frequency
in lexical acquisition. Journal of Child Language, 10, 57–64.
Simonsen, H. G., Kristoffersen, K. E., Bleses, D., Wehberg, S., &
Jørgensen, R. N. (2014). The Norwegian Communicative Devel-
opment Inventories: Reliability, main developmental trends and
gender differences. First Language, 34, 3–23.
Slobin, D. I. (1985). The crosslinguistic study of language acquisition
(Vol. 2, Theoretical issues). Hillsdale, NJ: Erlbaum.
Snedeker, J., Geren, J., & Shafto, C. L. (2007). Starting over: Inter-
national adoption as a natural experiment in language develop-
ment. Psychological Science, 18, 79–87.
Stokes, S. F. (2010). Neighborhood density and word frequency pre-
dict vocabulary size in toddlers. Journal of Speech, Language,
and Hearing Research, 53, 670–683.
Swingley, D., & Humphrey, C. (2018). Quantitative linguistic pre-
learning of specific English words. Child
dictors of
infants’
Development, 89, 1247–1267.
Tardif, T., Fletcher, P., Liang, W., Zhang, Z., Kaciroti, N., &
Marchman, V. A. (2008). Baby’s first 10 words. Developmental
Psychology, 44, 929–938.
Trudeau, N., & Sutton, A. (2011). Expressive vocabulary and early
grammar of 16- to 30-month-old children acquiring Quebec
French. First Language, 31, 480–507.
Vershinina, E. A., Eliseeva, M. B., Lavrova, T. S., Ryskina, V. L., &
Zeitlin, C. N. (2011).
(cid:0)(cid:2)(cid:3)(cid:4)(cid:5)(cid:4)(cid:6)(cid:7)(cid:2) (cid:8)(cid:4)(cid:6)(cid:9)(cid:10)(cid:5)(cid:11)(cid:14)(cid:7) (cid:6)(cid:2)(cid:13)(cid:2)(cid:14)(cid:4)(cid:15)(cid:4)
[Some norms of
(cid:6)(cid:10)(cid:12)(cid:14)(cid:11)(cid:5)(cid:11)(cid:16) (cid:17)(cid:2)(cid:5)(cid:2)(cid:18) (cid:4)(cid:5) * (cid:17)(cid:4) %* (cid:9)(cid:2)(cid:19)(cid:16)(cid:20)(cid:2)(cid:14)
language development of children from 8 to 18 months]. In
)(cid:22)(cid:2)(cid:30)
(cid:20)(cid:11)(cid:10)(cid:23)(cid:7)(cid:8)(cid:4)(cid:2) (cid:4)(cid:29)(cid:6)(cid:10)(cid:12)(cid:4)(cid:14)(cid:10)(cid:8)(cid:11)(cid:2)" (cid:5)(cid:6)(cid:10)(cid:17)(cid:11)(cid:20)(cid:11)(cid:11) (cid:11) (cid:11)(cid:8)(cid:8)(cid:4)(cid:14)(cid:10)(cid:20)(cid:11)(cid:11)
[Spe-
cial education: Traditions and innovations]. Saint Petersburg,
Russia: Publication House of Herzen State Pedagogical Univer-
sity of Russia.
Warriner, A. B., Kuperman, V., & Brysbaert, M. (2013). Norms of
valence, arousal, and dominance for 13,915 English lemmas.
Behavioral Research Methods, 45, 1191–1207.
Yu, C., & Ballard, D. H. (2007). A unified model of early word learn-
ing: Integrating statistical and social cues. Neurocomputing, 70,
2149–2165.
Zipf, G. K.
(1935). The psycho-biology of
language. Oxford,
England: Houghton, Mifflin.
OPEN MIND: Discoveries in Cognitive Science
67
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
o
p
m
i
/
l
a
r
t
i
c
e
-
p
d
f
/
d
o
i
/
i
/
/
.
1
0
1
1
6
2
o
p
m
_
a
_
0
0
0
2
6
1
8
6
8
3
6
8
o
p
m
_
a
_
0
0
0
2
6
p
d
/
.
i
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
