Participants

A total of 224 participants (55 infants and 169 toddlers) were assessed using the K M-B CDI (Pae & Kwak, 2011). The K M-B CDI reports consists of two types: ‘ words and gestures’ and ‘ words and sentences’. Words and gestures are for infants from 8 to 17 months; words and sentences are for toddlers whose ages are from 18 months to 36 months.

Out of 224 reports, 196 reports were from the Wordbank database (http://wordbank.stanford.edu/). Thirty-eight participants were additionally recruited from daycare centers in Seoul, Seong-nam, and Pusan in Korea. Among 224 participants, 31 were assessed as late talkers (7 were infants, and 24 were toddlers).

Participants were selected who meet the following criteria: (1) must have been born in Korea, (2) must use Korean at home with Korean speaking parents, and (3) must exhibit normal hearing and neurological development. The above information was collected through the Alberta Language Development Parent questionnaire (ALDeQ; Paradis, Emmerzael, & Duncan, 2010). Table 1 shows the descriptive statistics of participants’ characteristics— number of participants, gender, age, and KM-BCDI score in expressive and comprehensive vocabularies.

Table 1.

Participants' characteristics

The mean age for TD toddlers was 27±6 months, for LT toddlers was 23±4 months, for TD infants was 13±2 months, for LT infants was 12±2 months.

A set of one-way ANOVAs were conducted to investigate if there were group differences. There was no significant group difference in age in months, for both infants (F_{(1, 53)} =.452, p>.001) and toddlers (F_{(1, 167)} = 7.260, p>.001). There was no significant group difference in infants’ scores of KM-BCDI as well, for both expression (F_{(1, 53)} = 6.306, p>.001) and comprehension (F_{(1, 53)} =.676, p>.001). However, there was significant group difference in toddlers’ scores of K M-B CDI, for both expression (F_{(1, 167)} = 64.798, p < .001) and comprehension (F_{(1, 167)} = 26.488, p < .001).

This comparison of average scores evidently shows the necessity of more sophisticated methods to assess infants and toddlers. In the period of infants and toddlers, they show cascading language development. The difference between words production may not seem serious in early age but it distinguishes LT from TD, and the difference remains.

Measures

Data Scoring

Parents were asked to check both comprehension and expression lists, and researchers double- checked those reports. If one expresses a word for example “ kkokkio”, 1 was put, and if one does not express the word, 0 was put. The scores of comprehension and expression were separately inputted for each word. The lexical semantic categories were divided as K M-B CDI. One example of data arrangement is shown in Table 2.

Table 2.

Data coding for LT toddlers sounds categoy

After putting all answers to each word, the similarity was measured by the Jaccard similarity measure. When variables are bina-ry, co-occurrences are measured by the Jaccard coefficient similarity (Borg & Groenen, 2003). Jaccard’ s Index is measured by the following formulation.

In this study, Jaccard’ s coefficient was measured through SPSS version 22.0 for Windows, both words within lexical-semantic category and words without categories were measured.

Data Analysis

The statistical analyses were conducted using SPSS version 22.0 for Windows as well. First, Jaccard’ s coefficient was measured with absolute values. Three sets of one-way ANOVAs were conducted to compare group scores for similarity.

Group Differences in Similarity within Lexical-Semantic Categories

One hundred sixty-eight matrices were abstracted. The K M-B CDI for toddlers (words and sentences) has 24 semantic categories, therefore, 24 matrices for TD toddlers’ expression and comprehension, respectively, and for LT toddlers as well. The K M-B CDI for infants (words and gestures) has 18 categories, therefore, 18 matrices for TD infants’ expression and comprehension, respec-tively, and for LT infants as well. Since LT infants do not express even a single word, the comparison on similarities of infants’ expression was not possible. Therefore the comparison of infants’ expression was not conducted.

One example of similarity matrix is shown in Table 3.

Table 3.

Similarity matrix of ‘sounds’ category for TD toddlers' expression

The descriptive statistics of the average similarity of infants’ comprehension are provided in Table 4.

Table 1.

Mean similarity of infants' comprehension

To compare the difference of similarity of infants’ comprehension between TD and LT on each lexical semantic category, a one-way ANOVA was conducted with ‘ group’ as the within-subjects factor.

Fifteen categories out of eighteen were significantly different (p < .001) between those two groups listed above. In 5 categories, which included ‘ Body parts’, ‘ Households’, ‘ Furniture and rooms’, ‘ Places’, and ‘ Quantifiers’, the differences were not significant.

The group differences of similarity of infants’ comprehension between TD and LT group in each lexical category are shown in Figure 1.

Figure 1.

Mean similarity of infants' comprehension.

TD=typically developing children; LT=late talker.

The descriptive statistics of the mean similarity of toddlers’ expression are provided in Table 5.

Table 5.

Mean similarity of toddlers' expression

To compare the difference of similarity of the toddlers’ expression between the TD and LT group on each lexical semantic category, a one-way ANOVA was conducted with ‘ group’ as the with-in-subjects factor. Twenty-three categories out of 24 were significantly different (p < .001) between those two groups listed above. The mean difference of similarity in ‘ Helping verbs’ category does not exist, for the expression scores were all 0 for LT group. The group differences of similarity of toddlers’ expression between TD and LT group in each lexical category are shown in Figure 2.

Figure 2.

Mean similarity of toddlers' expression.

TD=typically developing children; LT=late talker

The descriptive statistics of the average similarity of toddlers’ comprehension are provided in Table 7.

Table 7.

Mean similarity of expression and comprehension without lexical-semantic categories

To compare the difference of similarity of toddlers’ comprehension between the TD and LT group on each lexical semantic category, a one-way ANOVA was conducted with ‘ group’ as the with-in-subjects factor. Twenty-four categories out of 24 were significantly different (p < .001) between those two groups listed above. The group differences of similarity of toddlers’ comprehension between TD and LT group in each lexical category are shown in Figure 3 and Table 6.

Figure 3.

Mean similarity of toddlers' comprehension.

TD=typically developing children; LT=late talker.

Table 6.

Mean similarity of toddlers' comprehension

Group Differences in Similarity without Lexical- Semantic Categories in K M-B CDI

The similarities between words without lexical categories were measured. Six hundred and forty-one words for toddlers’ expression and comprehension, and 284 words for infants’ expression. As conducted within category, three pairs of ANOVAs, infants’ comprehension, toddlers’ expression, and toddlers’ comprehension, were conducted. Since the average score of LT infants’ expression is 0, the comparison of the similarity of infants’ expression was not possible.

The descriptive statistics of the average similarity of infants’ comprehension without lexical semantic categories are provided in Table 7.

To compare the difference of similarity of infants’ comprehension between TD group and LT group without lexical semantic category, a one-way ANOVA was conducted with ‘ group’ as the within-subjects factors.

Without lexical semantic categories, two groups were significantly different (p < .001) as listed above. The group differences of similarity of infants’ comprehension between TD and LT group without the lexical semantic category are shown in Figure 4.

Figure 4.

Mean similarity of infants' comprehension without categories.

TD=typically developing children; LT=late talker

The descriptive statistics of the average similarity of toddlers’ expression without lexical semantic categories are provided in Table 7. Without lexical semantic categories, the two groups were significantly different (p < .001) as listed above. The group differences of similarity of toddlers’ expression between the TD and LT group without lexical semantic categories are shown in Figure 5.

Figure 5.

Mean similarity of toddlers' expression without categories.

TD=typically developing children; LT=late talker.

The descriptive statistics of the average similarity of toddlers’ comprehension without lexical semantic categories are provided in Table 7.

To compare the difference of similarity of toddlers’ comprehension between the TD and LT group without lexical semantic category, a one-way ANOVA was conducted with ‘ group’ as the with-in-subjects factor. Without lexical semantic categories, the two groups were significantly different (p < .001) as listed above. The group differences of similarity of toddlers’ comprehension between TD and LT group without lexical semantic categories are shown in Figure 6.

Figure 6.

Mean similarity of toddlers' comprehension without categories.

TD=typically developing children; LT=late talker.

Differences of Semantic Relatedness within Categories

K M-B CDI (Pae & Kwak, 2011) has its lexical semantic categories. Words and gestures for infants have 18 lexical semantic categories, and words and sentences for toddlers has 24.

First, the mean similarity of infants’ comprehension was COMPAREd. The mean similarity of words in fifteen categories was significantly different. TD’ s mean similarity was higher than LT’ s. This indicates that relatedness of LT’ s is significantly weaker than TD’ s. The 5 categories which did not show significance were ‘ Body parts’, ‘ Households’, ‘ Furniture and rooms,’ ‘ Places’, and ‘ Quantifiers’.

Early vocabularies tend to consist of nouns while verbs and closed-class forms are typically acquired later (Bates et al., 1994). According to Hills, Maouene, Maouene, Sheya, & Smith (2009), the fact that “ early noun learning is slow at first and becomes fast” suggests that “ already learned nouns might help in learning new nouns” (p. 729). This means that both infants and toddlers should show their increased numbers of acquired words, once they learned how to speak a few words. However, the difference of semantic relatedness has become more significant as infants get older. Even though the comparison of a few categories did not show significant differences in infants’ comprehension, that all the categories of toddlers showed significant differences could give further evidence of residual deficits in late talkers.

Differences of Semantic Relatedness without Categories

All three pairs resulted in having significant differences in comparison without categories. The most outstanding difference was evident in toddlers’ expressive vocabulary, and the most obscure difference was observed in infants’ comprehension.

However, there were variables which were laid at the bottom in the infants’ comprehension graph. These variables were not dis-tinctively evident in the within categories comparison. These variables, laid at the bottom, could be the evidence of residual deficits which lead late talkers to linguistic delay. Those lowest similarity words (p < .004) in LT infants are arranged in Appendix 1.

Clinical Implications and Limitations

Approximately half of the late talkers catch up with their peers by age three, and these children are designated ‘ late bloomers’ (Paul, 1991; Rescorla & Schwartz, 1990; Rescorla et al., 1997). Even after vocabulary reaches the normal range, a number of late talkers continue to demonstrate production delays in other areas, including phonology, morphology, syntax and narrative abilities (Paul, 1991; Paul et al., 1996; Paul & Smith, 1993; Rescorla & Schwartz, 1990; Robertz, Rescorla, Giroux, & Stevens, 1998). Furthermore, follow-up study on late talkers has revealed that even though they reach the normal range of linguistic ability in the elementary school-aged period, they receive significant lower scores in language assessment than their peers in the higher grades in school. This suggests that there could be linguistic residual deficits in late talkers. The present study hypothesized that there are linguistic residual deficits in late talkers and tried to verify the evidence of residual deficits by comparing the similarity of TD and LT.

The idea of comparing the similarity of two groups is based on the Spreading Activation Model. The Spreading Activation Model explains semantic processing with the network model. According to the Spreading Activation Model, concepts which are related to each other activate stronger and faster. Collins & Loftus (1975) created this model, and they suggested that concepts are featured by nodes in a semantic network. Concepts which have a similar meaning are connected with each other within the network.

Early links between language and cognition provide the foundation for processing information (Ferguson & Waxman, 2017). A study of Smith & Yu (2008) provided that the link between language and categories is established early in infancy. Categorization is fundamental to word learning, which makes it easier to learn novel words (Chomsky, 2011).

The hypothesis of this study was that the residual deficits of LT came from the weak ability of semantic relatedness. In the semantic networks, links between words imply relatedness. Categorizing ability, which connects concepts towards languages, could be the reason for the linguistic residual deficits of LT. Both TD and LT showed lower similarity index in the comparison of without categories than within categories. The words laid in the bottom in the similarity graph (Figure 4) can also be the evidence of the weak categorizing ability of infants.

This study did not offer the learning mechanism of words via semantic networks. The small number of LT infants was a weak-ness in this study. Due to the small number of LT infant subjects, the results in some categories were distorted; such as ‘ Function words’, which showed higher similarity in LT than TD. Another limitation of this study is that the words inventory should not have been limited to the K M-B CDI. There could have been more words that LT could speak, but those words were not considered.

However, the current study proposed that semantic relatedness is a key factor in word learning in infancy and that weak semantic relatedness in early age could lead to remaining linguistic residual deficits. Therefore, this study can be the academic evidence for urging early intervention of ‘ at risk’ children and suggests the direction of intervention on late talkers.