## Word count: 1000

Introduction

Early vocabulary development is an indicator of later linguistic, cognitive and academic outcomes (Bleses et al., 2016; Weisleder & Fernald, 2013). Understanding how early vocabulary develops therefore provides insight into both a child’s language learning and the environments that shape it. Cross-linguistic research suggests that the overall trajectory of early vocabulary is broadly similar across languages, yet differences in input quantity and culture-influenced communicative norms may affect vocabulary size at 18 months (Caselli et al., 1999; Maital et al., 2000; Bleses et al., 2008, Hoff, 2006).

Existing literature often relies on comparisons between English with another language, limiting evaluation of broader cultural patterns. However, large cross-linguistic datasets provide an opportunity to address this limitation. The Wordbank database compiles communicative development inventories (CDI) data, providing a cross-linguistic database (Frank et al., 2017). Examining patterns within this database can allow understanding as to whether previous observed similarities in vocabulary development are sustained across cultural groups.

Therefore, the present study investigates whether vocabulary production between 18-24 months differs across macro-cultural regions defined by Krys et al (2025). 18-24 months is the focus as it represents a key period where produced vocabulary accelerates (Fernald et al., 1998)

Two hypotheses were tested. (H1) Production size at 18 months will differ across regions and (H2) Vocabulary growth rates for children aged 18-24 months will be similar across regions.

Methods

Data was obtained from Wordbank. (Frank et al., 2017). The original dataset contained 96,246 caregiver-reported records but was filtered to include children aged 18-24 months (M age = 21.68m, SD=2.2m). Duplicate child ID records and missing values were removed, ensuring each record was independent.

The dataset was subsetted to the following variables: language, age and production. Production was measured using the CDI production score, the number of words caregivers reported their child could produce. Age was recorded in months. Language referred to the CDI instrument language.

Languages were grouped together into macro-cultural regions proposed by Krys et al (2025). This produced a new categorical variable representing 6 cultural regions: WEIRD, Confucian, Latin America/Caribbean, Middle East/North Africa (MENA), Non-Confucian and Postcommunist states. Region groups with fewer than 100 records were removed to reduce imbalance across groups.

The final dataset included children aged 18-24 months, with 5 variables: age, age_c, cultural region, language, production score. It consisted of 31,308 records.

As the production scores represent count data and showed overdispersion relative to a Poisson distribution, negative binomial generalized linear models were used For H1, the dataset was filtered to include only children aged 18 months. Production was modelled as a function of cultural region. A likelihood ratio test evaluated whether cultural region improved model fit and Tukey-adjusted pairwi comparisons identified specific regional differences.

For H2, production was modelled as a function of centered age, cultural region and their interaction.

# Install the following packages in the console:
library(ggplot2)
library(knitr)
library(emmeans)
library(MASS)
library(kableExtra)
library(dplyr)

# Load dataset "wordbank.csv"
wb <- read.csv("wordbank.csv")
# Select to relevant columns: language, age, production
# Use na.omit to remove NAs from language, age, and production columns
# Subset age range to 18 to 24 months
wordb <- wb %>%
  dplyr::select(language, age, production) %>%
  na.omit() %>%
  filter(age >= 18 & age <=24)
# Check the number of distinct languages and list of languages present
length(unique(wordb$language))
unique(wordb$language)

# Define macro-cultural language lists with no repeated languages across groups 

## WEIRD (languages from Western, educated, industrialised, rich and democratic societies)
WEIRD <- c("American Sign Language", "British Sign Language", "Croatian", "Czech", "Danish", "Dutch", "English (American)", "English (Australian)", "English (British)", "Finnish", "French (French)", "French (Quebecois)", "Estonian","German", "Greek (Cypriot)", "Hebrew", "Hungarian", "Italian", "Latvian", "Norwegian", "Portuguese (European)", "Spanish (European)", "Swedish", "Slovak")

## Confucian-Inspired (East Asian Confucian)
Confucian <- c("Cantonese", "Mandarin (Beijing)", "Mandarin (Taiwanese)", "Korean", "Japanese")

## Latin America/Caribbean
Latin_America_Caribbean <- c(
  "Spanish (Mexican)", "Spanish (Argentinian)", "Spanish (Peruvian)", "Catalan")

## MENA (Middle East and North African)
MENA <- c("Persian", "Turkish")

## Postcommunist
Postcommunist <- c("Russian")

## Sub-Saharan Africa group
Sub_Saharan_Africa <- c(
  "Kigiriama", "Kiswahili")

# With no languages directly falling within the non-Confucian Asia/Oceania region, this language list was not created

# Run to make sure there are no overlaps -> overlaps should be empty in values
all_lists <- list(WEIRD, Confucian, Latin_America_Caribbean, MENA, Postcommunist, Sub_Saharan_Africa)
overlaps <- Reduce(intersect, all_lists)
# Create a region-level column and clean dataset
wordb <- wordb %>%
  mutate(
    region_group = case_when(
      language %in% WEIRD ~ "WEIRD",
      language %in% Confucian ~ "Confucian",
      language %in% Latin_America_Caribbean ~ "Latin America/Caribbean",
      language %in% MENA ~ "Middle East/North Africa", 
      language %in% Postcommunist ~ "Postcommunist States",
      language %in% Sub_Saharan_Africa ~ "Sub-Saharan Africa",
      TRUE ~ NA_character_)) %>%
  filter(!is.na(region_group))
# Keep region groups with sample size of 100 or above
region_counts <- table(wordb$region_group)
keep_regions <- names(region_counts[region_counts >= 100])
wordb <- subset(wordb, region_group %in% keep_regions)

Results

# Create descriptives table by cultural region
wordb %>%
  group_by(region_group) %>% # group by region_group
  summarise(N = n(), Percent = round(100 * n() / nrow(wordb), 1), Languages = n_distinct(language),
            
# Calculate sample size, % of total sample, no. of languages and mean/SD of vocabulary production (excluding missing values)
Mean= round(mean(production, na.rm = TRUE), 1),
SD = round(sd(production, na.rm = TRUE), 1),
.groups = "drop") %>%
arrange(desc(N)) %>% # organise by sample size
  rename(Region = region_group) %>% # Rename variable for clearer table presentation

# Format the table for the report
  kable(
    caption = "Table 1. Descriptive statistics for vocabulary production across cultural regions.",
    digits = 1
  ) %>%
  kable_styling(full_width = FALSE) %>%
  footnote(general = "Note. N = number of observations; Languages = number of languages represented in each region.",
           general_title = "")
Table 1. Descriptive statistics for vocabulary production across cultural regions.
Region N Percent Languages Mean SD
WEIRD 25501 81.5 24 112.0 124.5
Confucian 3006 9.6 5 208.0 191.0
Latin America/Caribbean 1521 4.9 4 164.8 156.7
Middle East/North Africa 878 2.8 2 173.4 169.9
Postcommunist States 402 1.3 1 169.7 155.7
Note. N = number of observations; Languages = number of languages represented in each region.

Descriptive statistics are displayed in Table 1.

To assess whether production differed at 18 months depending on the cultural region, a negative binomial generalized linear model was fitted. Production at 18 months significantly differed with cultural region, χ²(4) = 69.19, p < .001.

Pairwise comparisons with Tukey correction showed that production at 18 months were significantly higher in Confucian regions compared to Latin America/Caribbean (p = .043), Postcommunist States (p = .001), and WEIRD regions (p < .001). Production was also higher in Latin America/Caribbean compared to WEIRD regions (p = .033) and higher in Middle East/North Africa compared to WEIRD regions (p = .015). Other pairwise comparisons were not statistically significant.

To examine whether vocabulary growth differed across cultural regions between 18 and 24 months, a negative binomial generalized linear model including centered age, cultural region, and their interaction was fitted. Production increased significantly with age, β = 0.036, z = 12.77, p < .001. The interaction between age and cultural region was significant for all non-WEIRD groups, indicating that vocabulary growth rates differed across cultural regions. Relative to WEIRD children, growth rates were higher for Confucian (β = 0.153, z = 16.19, p < .001), Latin America/Caribbean (β = 0.131, z = 10.83, p < .001), Middle East/North Africa (β = 0.138, z = 8.19, p < .001), and Postcommunist regions (β = 0.222, z = 9.89, p < .001).

Growth trajectories by cultural region are illustrated in Figure 1. Children speaking languages identified with WEIRD societies have produce words a at a slower rate than children speaking languages from the other regions.

# geom_smooth(method = glm with a negative binomial family to match the distribution used in statistical models
# Theta parameter is extracted rom model_h2 to ensure consistency with the estimated dispersion in fitted model
ggplot(wordb, aes(x = age, y = production, colour = region_group)) +
  geom_smooth(method = "glm", method.args = list(family = MASS::negative.binomial(theta = model_h2$theta)),
    formula = y ~ x, se = FALSE, linewidth = 1) +
  labs(
    x = "Age (months)",
    y = "Vocabulary production (words)",
    colour = "Macro-cultural region",
    title = "Figure 1. Vocabulary production across age by cultural region"
  ) +
  theme_minimal(base_size = 12) +
  theme(
    legend.position = "right",
    plot.title = element_text(face = "bold")
  )

Discussion

This study investigated whether vocabulary production at 18 months and growth between 18 and 24 months differed across macro-cultural regions. Consistent with H1, regional differences in production score at 18 months were observed. Children speaking Confucian-inspired languages produced significantly more words than children speaking WEIRD-specified languages, whilst production with the Post Communist language produced fewer words. No significant differences were observed between WEIRD and Latin American/Carribean or MENA regions. One explanation is cross-linguistic variation in lexical structure. Languages with clearer word boundaries, e.g., Mandarin, may facilitate earlier word segmentation and higher reported vocabulary scores (Bleses et al., 2008).

Contrary to H2, production growth rates differed across regions, with steeper increases in Confucian, Latin American/Caribbean, MENA, and Post-Communist regions relative to WEIRD societies. This diverges from research suggesting that early vocabulary growth follows similar trajectories across languages (Frank et al., 2021). These differences may reflect variation in linguistic input and caregiver-child interaction patterns across cultures (Fernald et al., 2013; Hoff. 2006).

However, it must be noted that the flatter growth curve observed for WEIRD regions may partly reflect uneven sample distribution. As WEIRD languages accounted for 81.2% of the records, its estimates are likely more stable. Other regions therefore may produce steeper or more variable slope estimates.

Another limitation is the potential oversimplification of the regions as macro-cultural regions cover a variety of languages, socio-linguistic contexts, and caregiving practices which may be understated by macro-cultural classifications. Despite these limitations, this study demonstrates the value of large cross-linguistic datasets for researching early language development.

References

Bleses, D., Vach, W., Slott, M., Wehberg, S., Thomsen, P., Madsen, T. O., & Basbol, H. (2008). Early vocabulary development in Danish and other languages: A CDI-based comparison. Journal of Child Language, 35(3), 619–650. doi:10.1017/S0305000908008714

Bleses, D., Makransky, G., Dale, P. S., Hojen, A., & Ari, B. A. (2016). Early productive vocabulary predicts academic achievement 10 years later. Applied Psycholinguistics, 37(6), 1461–1476. doi:10.1017/S0142716416000060

Caselli, C., Casadio, P., & Bates, E. (1999). A comparison of the transition from first words to grammar in English and Italian. Journal of child language, 26(1), 69–111. https://doi.org/10.1017/s0305000998003687

Fernald, A., Pinto, J. P., Swingley, D., Weinbergy, A., & McRoberts, G. W. (1998). Rapid Gains in Speed of Verbal Processing by Infants in the 2nd Year. Psychological Science, 9(3), 228–231. https://doi.org/10.1111/1467-9280.00044 

Fernald, A., Marchman, V. A., & Weisleder, A. (2013). SES differences in language processing skill and vocabulary are evident at 18 months. Developmental science, 16(2), 234–248. https://doi.org/10.1111/desc.12019

Frank, M. C., Braginsky, M., Yurovsky, D., & Marchman, V. A. (2017). Wordbank: an open repository for developmental vocabulary data. Journal of child language, 44(3), 677–694. https://doi.org/10.1017/S0305000916000209

Frank, M. C., Braginsky, M., Yurovsky, D., & Marchman, V. A. (2021). Variability and consistency in early language learning: The Wordbank project. The MIT Press. https://doi.org/10.7551/mitpress/11577.001.0001

Henrich, J., Heine, S. J., & Norenzayan, A. (2010). The weirdest people in the world?. The Behavioral and brain sciences, 33(2-3), 61–135. https://doi.org/10.1017/S0140525X0999152X

Hoff, E. (2006). How social contexts support and shape language development. Developmental Review, 26(1), 55–88. https://doi.org/10.1016/j.dr.2005.11.002

Krys, K., de Almeida, I., Wasiel, A., & Vignoles, V. L. (2025). WEIRD–Confucian comparisons: Ongoing cultural biases in psychology’s evidence base and some recommendations for improving global representation. American Psychologist, 80(2), 247–263. https://doi.org/10.1037/amp0001298

Maital, S. L., Dromi, E., Sagi, A., & Bornstein, M. H. (2000). The Hebrew Communicative Development Inventory: language specific properties and cross-linguistic generalizations. Journal of child language, 27(1), 43–67. https://doi.org/10.1017/s0305000999004006

Weisleder, A., & Fernald, A. (2013). Talking to children matters: early language experience strengthens processing and builds vocabulary. Psychological science, 24(11), 2143–2152. https://doi.org/10.1177/0956797613488145