Later language development, typically occurring between the ages of 9 and adulthood, plays a critical role in preparing individuals for active participation as communicators in¹ the adult speech community, ultimately leading to mature native language proficiency (Berman, 2004, 2017; Berman and Ravid, 2009; Nippold, 2016). This period is also characterized by significant brain reorganization, with changes in brain structures and functions that are particularly pronounced during adolescence (Fuhrmann et al., 2015; Kadosh et al., 2013; Paus, 2005; Simmonds et al., 2014). Multi-modal brain areas undergo both structural and functional transformations, which enhance brain connectivity and activation. These changes support the integration of areas responsible for higher cognitive functions (Kilford et al., 2016; Larsen and Luna, 2018). As a result, these neural developments have significant socio-cognitive effects which improve attention, cognitive flexibility, executive functions, goal management, memory, and processing abilities in older children, adolescents and young adults (Gooch et al., 2016; Lecce et al., 2017; Osterhaus et al., 2017; Tolchinsky and Berman, 2023). Due to these structural changes, adolescent brains are especially sensitive to environmental influences, such as peer pressure.

One key cognitive outcome of adolescent brain development is a dramatic improvement in processing abilities. While younger children rely on limited cognitive resources (Elman, 1993), focusing on frequent, salient, and predictable language categories (Dattner et al., 2023; De Ruiter et al., 2018), older learners have enhanced cognitive capacities—particularly in attention, memory, processing, and literacy skills—that allow them to process more linguistic complexity (Kilford et al., 2016; Larsen and Luna, 2018). As a result, older children and adolescents are better able to focus on less frequent or prominent linguistic features (Ramscar et al., 2018) and adapt to the changing structure of language input, particularly in the shift from spoken to written language (Berman, 2008; Onnis and Thiessen, 2013). These cognitive developments allow older learners to extract information from larger and more diverse data sets, enabling them to learn less prominent language categories and items.

Many of the changes in adolescent brains are associated with areas involved in social cognition (Best and Ban, 2021; Blakemore, 2012; Padilla-Walker and Van der Graaff, 2023), making socio-affective contexts—particularly peer interactions—crucial in shaping development. Notably, these changes contribute to the maturation of Theory of Mind (ToM), including a growing understanding of one's own and others' emotions and behaviors, which facilitates cooperation and interaction with peers. Adolescents also undergo significant psychological changes, including developments in self-awareness and identity, which influence how they engage with others. Compared to younger children, adolescents are more sociable and form more complex, hierarchical peer relationships (Giletta et al., 2021; Meeus, 2016).

Given these developmental shifts, the current study focused on peer talk as a key arena for examining language as reflecting social-cognitive development during later childhood and adolescence.

The neural and socio-cognitive changes that occur during adolescence culminate in the emergence of mature, complex language abilities during the phase known as Later Language Development (Berman, 2017; Clark, 2004; Kuhn, 2011; Proverbio and Zani, 2005). Language skills continue to evolve through the school years into young adulthood and beyond (Berman, 2007; Berninger et al., 2017; Gooch et al., 2016; Nippold, 2016). These developments support the acquisition of automatic, flexible, and abstract language use capabilities, alongside enhanced metalinguistic abilities and the capacity for higher-order, non-literal language (Berman and Ravid, 2010; Ravid, 2006a; Karmiloff-Smith, 1994; Nippold and Duthie, 2003). In parallel, adolescents gain access to a broader range of new lexical items and linguistic constructions (Ravid and Tolchinsky, 2002; Tolchinsky, 2022).

A key aspect of language proficiency during this period is lexical development, particularly within the content and referential domains (Strömqvist et al., 2002), with a focus on nouns, adjectives, and verbs (Beitchman et al., 2008). As cognitive, social, and linguistic skills mature, individuals develop greater abstract reasoning and analytical abilities (Hollarek and Lee, 2022). These advancements are reflected in their use of more complex vocabulary, which is often characteristic of academic language (Anglin et al., 1993; Nagy and Townsend, 2012). Recent research by Atanasova et al. (2020) further highlights significant differences in lexical knowledge between children and adults, both in terms of accuracy and processing speed. Their findings suggest that, regarding the Age of Acquisition (AoA) of words in production, adolescents (ages 14–16) exhibit an intermediate pattern between younger children and adults, indicating ongoing brain maturation.

Studies of Hebrew, where the lexicon is organized by and around morphological systems, further illustrate these developments. In Hebrew, command of derivational morphology and morpho-lexical abilities increases exponentially throughout adolescence. This includes the acquisition of morphologically structured mental words, abstract and derived nouns (Ravid, 2006a; Ravid and Avidor, 1998), denominal adjectives (Cutillas and Tolchinsky, 2017; Ravid et al., 2016), and specialized vocabulary across all lexical classes (Ben-Zvi and Levie, 2016; Berman and Nir-Sagiv, 2010).

Most prior studies investigating language development during the school years have focused on the written mode (Anglin et al., 1993; Berman et al., 2011; Dabrowska, 2018; Larsen and Nippold, 2007; Ravid and Levie, 2010), regarding written texts as “optimal hunting grounds” for unveiling later linguistic abilities (Haim and Ravid, 2022), especially in the context of genre-specific text types (Jisa and Tolchinsky, 2009). The idea is that the written mode, occupying a privileged cognitive position for literate individuals, imposes cognitive demands on memory, executive functions, and top-down processing, promoting the retrieval of high-register lexical items and the creation of complex syntactic units (Chafe, 1992; Kärkkäinen et al., 2007; White, 2014). Indeed, there is ample evidence regarding the lexical and syntactic density and complexity of the written mode in the later development of text production abilities (Nippold and Sun, 2008; Ravid and Berman, 2010; Tolchinsky et al., 2010). In contrast, little work has been done on the lexical characteristics of spoken interaction in school-aged children and adolescents. It makes sense to assume that Later Language conversation might be constrained by the characteristics of online discourse (Ravid and Chen-Djemal, 2015), with lesser density and diversity of lexical items than the written mode; however there can be no doubt that participants' age and schooling experience must drive lexical acquisition in ways that can be discerned even in the contexts of spoken interaction (Levie et al., 2020).

Against this background, the current study focuses on conversation as the core of inter-personal linguistic communication, the most natural and immediate implementation of language abilities (Casillas and Frank, 2017; Levinson, 2019)—a shared, timed communicative activity requiring attention, memory and ToM resources (Hari and Kujala, 2009). It is an established fact that early language learning directly derives from the wiring of brain social interaction pathways (Baldwin, 1995; Brooks and Meltzoff, 2005; Conboy et al., 2008; Kuhl, 2007). Specifically, turntaking and communication repairs provide rich socio-cognitive experience which drives linguistic acquisition (Dingemanse et al., 2015; Kendrick et al., 2023). As these social interaction pathways continue to develop with age and schooling, constituting an important socio-cognitive achievement at teenage, they provide another venue for investigating the development later lexical abilities.

The current study investigates lexical distributions in children and adolescents in peer talk—i.e., spontaneous spoken interactions with their close friends (see Methods Section below). Peer interactions (Ellis and Zarbatany, 2017) are characterized by several properties that can be taken to reflect children's foundational abilities, unassisted by teachers or parents, and may also benefit and enhance lexical expression. First, peer talk among children reflects self-management with no external intervention or support by adults (Forrester and Cherrington, 2009; Veneziano, 2018). Second, this communicative context provides the most motivation for participants to talk to each other and carry out transactional interactions (Veneziano, 2010). Third, peer talk enables spontaneous play, which is critical in constructing social interactions (Blum-Kulka et al., 2010). Fourth, this is a non-scripted, speedy exchange of speech and activities, requiring constant monitoring and mutual attention to interlocutors' needs, urge management and control – necessitating the usage of ToM vocabulary (Clark and Krych, 2004; Rendle-Short et al., 2014). And finally, this context enables meaningful learning by children from children, which may involve more advanced linguistic resources in expository-like explanations, persuasion and discussion (Zadunaisky Ehrlich, 2011; Zadunaisky Ehrlich and Blum-Kulka, 2010).

This study investigates the characteristics and development of lexical production in children and adolescents aged 6–14 during peer interactions. Lexical acquisition is a key component of mastering mother-tongue proficiency (Berman and Ravid, 2009; Dockrell and Messer, 2004; Nippold, 2002; Ravid and Ginat-Heimann, 2014), encompassing aspects such as lemma frequency, token use, and lexical categories (Ambridge et al., 2015), as well as lexical depth and associations (Caro and Mendinueta, 2017; Schmitt, 2014; Wolter, 2001). In this context, we focused on token analyses of two primary lexical components: the content lexicon and the discourse lexicon.

We categorized content words into two main groups: core content words and extended content words, which we define and explain below, along with their implications for our lexical analyses.

The discourse lexicon plays a crucial role in lexical development, complementing the content lexicon and highlighting the flexible, dynamic cognitive abilities humans use to convey and interpret meaningful intentions (Heintz and Scott-Phillips, 2023). From a linguistic perspective, the discourse lexicon refers to words that are not directly referential or lexical but serve essential communicative, discursive, and social-pragmatic functions (Dickey, 1997). This includes discourse markers, address forms, and various social words expressing affect and stance (Da Costa and Bond, 2016). These discourse elements are critical in interaction throughout all stages of language development (Choi, 2007; Lieven et al., 1992) and continue to play a key role in adults' social and linguistic understanding during communication (Oyetade, 1995). In this framework, we categorized discourse words into three groups: discourse markers, social words, and address forms.

For the current study, a new corpus of children and adolescents engaged in peer conversations was collected, recorded, and transcribed. This age range was chosen for the study based on prior research about age cut-off points that mark significant changes in Later Language Development (Berman, 2005, 2008; Berman and Verhoeven, 2002; Ravid and Hershkovitz, 2017; Ravid, 2006a; Ravid and Levie, 2010). The total research population consisted of 72 participants, 34 boys and 38 girls, from middle- high socio-economic background in central Israel, all native Hebrew speakers with no diagnosed developmental disorders in language or communication, and no history of hearing, vision, attention, learning, or emotional difficulties.

In dividing the participants into age/schooling groups, we relied on the general psycholinguistic and developmental literature, as well as literature on the acquisition of Hebrew, reviewed in the Introduction section (Berman, 2005; Nippold, 2016; Ravid, 1995). This review indicated that language development during elementary and middle school years progresses more slowly compared to early childhood acquisition. Therefore, we combined every two consecutive schooling levels into a single age/grade level group, such that the research participants were divided into four age/grade-level groups based on adjacent two-year age ranges as follows:

From this point onward, our reference framework for the participants is the educational systems that host the four research groups: early elementary school (1st−2nd grades), middle elementary school (3rd−4th grades), late elementary school (5th−6th grades), and middle school (7th−8th grades). Triadic conversation was selected for this study as requiring complex turn-taking, the usage of pronominal reference, and the comprehension of the role of a third person in conversation (Dattner et al., 2019; Salazar Orvig et al., 2010). Each age/grade level group thus contained 6 triads (groups of three), i.e., 18 children, and the total number of triads in the study was 24. The triads were formed with a variety of gender combinations, based on the participants' expressed preferences, to ensure a smooth and natural flow of conversation. Each triad conversed for 35 min (a total of 3 h for each age/grade level group), and the total amount of recording time was 12 h.

The recruitment of participants was done through direct contact with their parents. In each triad, the children recruited their close friends, who were likely to be at a similar developmental level and accustomed to close, shared interaction. Each triad was recorded on both video and audio for a total of 35 min, during which they interacted naturally with their peers in their homes. They were instructed to talk about anything that came to their minds, to play with anything they chose, to engage in any activity, or to have a conversation. The only restriction was not to use any screen devices such as computers, phones, iPads and the like. No intervention was made by the researcher, who sat outside the children's field of vision. Before the recordings began, participants were told about the importance of participating in the study and the potential benefits that could result from this research in the field of language acquisition, conversation, and language instruction. The explanation to the children emphasized how their participation could contribute to understanding the development of conversation and communication skills. This explanation helped motivate children to engage in conversation in the same location for the entire recording and to actively participate.

The video and audio files were transcribed into Hebrew by three communication disorders students, under the full supervision and guidance of both authors. The reliability of the transcription was ensured by having the first author transcribe 20% of the recordings herself and comparing these transcriptions with those made by the students, with emphasis on accurately conveying the phonetic information in context and clarifying overlapping speech. Since the transcription was in unvowelized Hebrew, disambiguation of homographs was carried out when needed using vowelization or phonetic writing (e.g., ראית could be ra'íta “you, Masc. saw” or ra'it “you, Fm. saw”). The transcription was done while watching the video, with notes and clarifications added about what was happening in the room, participants' body and facial gestures, and physical activity observed, such as wrestling between the children. As part of the transcription process, the text was segmented into utterances, with particular attention given to prosodic intonation patterns that highlight text content (Hill et al., 2021; Wynn et al., 2023). An utterance (in peer talk) is defined as a segment of speech by one speaker. Prosodically it has a distinct melodic pattern, including pauses, rhythm, and loudness changes. Content-wisean utterance is defined as conveying a relatively complete thought or serving a specific communicative function within the interaction, often constituting a turn or part of a turn. This pragmatic unit is highly relevant to samples of spoken language; whereas clauses, which are a syntactic unit, are more relevant to studies of written language (Berman, 2008; Berman and Slobin, 1994).

Lexical frequencies are typically considered from two perspectives: lexical tokens and lexical types. Lexical tokens refer to the inflected forms of words as they appear in the transcription (e.g., hevánti “I've understood”, or macxikot “funny, Fm.Pl”). Token frequency counts the total number of occurrences of these lexical tokens in a given category within the corpus, reflecting their actual usage by participants. Type frequency, on the other hand, is a more abstract measure. It involves extracting the lemma from each token (e.g., the root b-y-n and the verb pattern hif'il, yielding the abstract verb lemma “understand”), and counting each lemma only once. This measure reflects both the size of the category and its diversity within the participants' mental lexicon.

While both measures are important in the developmental psycholinguistic literature (Ambridge et al., 2015), the current study focuses on token frequencies, which are directly influenced by and reflect the syntactic and pragmatic structure of discourse. Accordingly, all content word tokens—nouns, verbs, adjectives, and adverbs—along with all discourse-related tokens, including discourse markers, social words, and address forms, were identified based on their syntactic and discursive contexts. The first author performed the identification and coding, while the second author reviewed 50% of the texts to ensure consistency in lexical classification. For challenging cases, such as determining whether a beynoni present-tense form was a verb, noun, or adjective (Ravid, 1995, 2019b; Ravid and Levie, 2010; Ravid et al., 2016), decisions were made using the syntactic context as well as through discussion and reference to the literature.

Table 1 presents the general distributions of the study categories in our database, as against two counts: the number of word tokens and the number of utterances in the corpus.

Table 1 shows the general information regarding lexical frequencies in the study database, consisting of 12 h of triadic conversation transcribed into 24 transcripts, as described in the Participants and Materials section above. The meanings and implications of these lexical characteristics of peer talk are explored in the Discussion section. Figures 1, 2 illustrate these findings.

Figure 1 shows the lexical density of the text in terms of the number of nouns, verbs, adjectives, and adverbials out of all word tokens.

Figure 2 shows the distributions of three lexicons in the database: the extended content lexicon (nouns, verbs, adjectives, and adverbials); the discourse lexicon (discourse markers, social words, and address forms); and the rest of the word tokens, which constitute the function word lexicon—out of all word tokens.

Given the study's focus on lexical development between ages 6 and 14 years, we conducted developmental statistical analyses on the data in Table 1. Tables 2–4 present the results of the parametric and nonparametric tests regarding the distributions of words of different classes, utterances, and MLU by the four age/grade level study groups. Please note that each triad was recorded for the same amount of time−35 min.

According to Table 2, the number of words and utterances rises significantly with age and schooling, with the youngest group of 1st−2nd graders producing the least number of words and the smallest number of utterances. MLU rises too significantly with age and schooling, with the oldest group producing the longest utterances.

As shown in Table 3 and illustrated in Figure 3, the number of both core and extended content words increased significantly with age and schooling, with 7th−8th graders producing more content words than 1st−2nd graders. In terms of proportions, content words maintained consistent percentages across all age groups—~32% for core content words and 37% for extended content words. The number of nouns and verbs also increased significantly with age and schooling. For nouns, there was also a notable increase in their proportion, with a clear distinction between the youngest group and the older groups. For verbs, however, the proportional pattern was less clear and did not indicate any significant developmental changes. The numbers and proportions of adjectives and adverbials generally remained stable across age groups, although non-parametric tests revealed a significant increase in the number of adjectives, particularly from the youngest group to the older groups.

Table 4 shows that the discourse words class increases significantly in number with age and schooling, with the youngest group producing fewer discourse words than the rest of the groups. The numbers of discourse markers and of social words rise significantly with age and schooling, with the oldest group producing the highest number of these words; whereas the number and proportions of address forms stay constant across the age/grade level groups. These results are illustrated in Figure 4.

Figure 5 illustrates the distributions of the content word classes and the discourse word class across the different age/grade level groups, as against the distributions of all word tokens.

Turning to the core of our study, the following sections outline the evolving distributions of content and discourse words and discuss their implications for Later Lexical Development. To date, we have not identified any directly comparable studies that examine word classes from a developmental psycholinguistic perspective in Hebrew or any other language. However, for several of the phenomena analyzed, we did find relevant comparisons that help to illuminate and enrich our findings.

This section presents entirely new analyses, focusing on the results of an exhaustive examination of three types of discourse words in the database: discourse markers, social words, and address forms. Discourse words enhance the pragmatic dimension of language, revealing speakers‘ attitudes, intentions, and their understanding of the social dynamics at play. In the current study, these three categories accounted for ~12% of the word tokens in the text, or about 0.4 discourse words per utterance. Among these, social words made up the largest proportion, close to 8% of the word tokens in the corpus, followed by discourse markers at around 3% and address forms at over 1%. These words, which convey discourse stance, affect, interactional relationships, and textual structure, are part of the very early emerging lexicon (Gopnik, 1988). Our analysis of schoolgoing children and adolescents revealed that the total discourse lexicon expands with age and schooling, with the oldest age group using the most discourse words and the youngest group using the fewest. Specifically, discourse markers increased with age, with the two older groups using more than the two younger groups. In the same way, the number of social words was highest in the 7th−8th grade group (ages 13–14), compared to the 1st−2nd grade group (ages 6–7). These findings clearly indicate the growing social motivation and conversational abilities of our participants, reflecting their increasing capacity to attend to, initiate, and sustain conversational topics, as well as to respond to and engage with the mental and positional states expressed in their interlocutors' linguistic productions.

These numbers reflect communicatively accurate use of socially oriented discourse items that facilitate and advance triadic interactions. For instance, 1st−2nd graders frequently used the discourse marker bo “literally, come”, which is meant to encourage interlocutors to take action, as well as enhancing discourse markers like récax (“literally, murder”), as in xazaka récax “extremely strong”, and the complimentary máshehu máshehu (“literally, something something”), as in ata máshehu máshehu “you're something else”. Examples of discourse items used by 3rd−4th graders include the discourse markers ke'ilu “like” and háyush “[informal and intimate} hi”, along with interjections like íma'le “[literally, mommy] oh my goodness!” and yummy.

5th−6th graders employed the discourse marker be'emet “[literally, in-truth] = really” in three distinct ways: ma'aréxet xisunit be'emet al ha-panim “an immune system really [literally, on-the-face] in horrible condition”; the derogatory nu be'emet “come on, seriously?”; and the pretend-horror interjection lo be'emet “not really”. These nuanced uses are difficult to translate with full communicative accuracy, but they hopefully convey the pragmatics of the interaction.

Finally, the oldest group, 7th−8th graders, also used the discourse marker bo “literally, come”, however, unlike its more concrete function in the younger group, the oldest participants used bo to invite their interlocutors to engage in shared contemplation – a clearly mental activity. Moreover, this group demonstrated expert use of hedges and fillers in the utterance involving the academic term questionnaires: hu asa ka'éle pá'am stam she'elonim ve-ze “he did like-these once just like that [random] questionnaires and-stuff”; and in the following description: sabába, aval yesh lo kéta kaze she-hu mashmin ve-marze, mashmin ve-marze “alright, but he has this thing where he keeps gaining weight and losing weight, gaining weight and losing weight”. This last utterance uses discourse markers to express satisfaction (sabába), followed by a roundabout way of getting to the issue of the fluctuating weight of their friend. Hopefully, this translation succeeds in capturing the nuances of the original discourse while maintaining the intended pragmatic meaning.

The increasing use of discourse words closely reflects the demands of spoken interaction, where they function as vital tools for creating coherence and adding subtle layers of meaning and perspectives. Therefore, the rise in discourse word frequency within our data may reflect our participants' developing skill in navigating a three-way conversation. This involves strategies such as soliciting agreement, expressing alignment or opposition, sustaining conversational threads or introducing new ones, and highlighting their individual viewpoints on occurrences, personal stance, and how they believe others perceive them.

The current study offers a novel and comprehensive analysis of two key components of the lexicon as reflected in the spontaneous speech interactions of older children and adolescents: four categories of content words and three categories of discourse words. To the best of our knowledge, this is the first psycholinguistic study of its kind to examine the evolution of these lexicons in Modern Hebrew.

However, there are (at least) two limitations to this study. First, we focused on tokens rather than lemmas in our analysis. This means we included all instances of the lexical categories under investigation that appeared in the text, rather than examining abstracted types (e.g., different nouns, adverbials, etc.). This token-based approach to lexical development may provide a different perspective on the distribution of content and discourse words compared to analyses based on lexical density and diversity in terms of lemmas.

A second limitation is the absence of in-depth semantic and morphological analyses of relevant age- and grade-level lexical sub-classes in development, such as abstract nominals, mental verbs, roots and patterns, denominal adjectives, and others—all beyond the scope of the current article. These analyses, which we intend to publish in the future, will hopefully complement, expand, refine, and deepen our understanding of later stages of lexical development during the school years.