How Morphology, Context, Vocabulary and Reading Shape Lexical Inference in Typical and Dyslexic Readers

Abstract

Children’s ability to infer meanings of unfamiliar words during reading is thought to rely on the interplay between decoding, morphological awareness, contextual support, and vocabulary knowledge, but it remains unclear how these sources operate in typically developing (TD) readers compared to those with developmental dyslexia (DD). This study examined whether morphological cues (suffixes) or/and contextual information facilitate meaning inference and which variables predict performance. Sixty children (30 TD, 30 DD; aged 9–12) completed a battery of tasks assessing pseudoword decoding, expressive vocabulary (breadth) synonyms, antonyms (depth), morphological awareness (deriving and decomposing words), and reading comprehension. The main inference task consisted of 20 short stories in which pseudowords replaced target words; in half the stories, pseudowords included derivational suffixes, while in the other half no such clues were available. Results showed that TD children performed significantly better than DD peers across all tasks. Regression analyses revealed that vocabulary depth and morphological awareness predicted inferencing in both groups, but decoding was uniquely predictive for DD children and reading comprehension only for TD children. These findings suggest that while lexical inference in both groups appears to draw on vocabulary and morphology, TD children may additionally integrate higher-order comprehension, whereas DD children seem to remain more influenced by decoding and partial lexical cues.

1. Introduction

Children’s ability to infer the meanings of unfamiliar words during reading relies on the dynamic interplay of decoding, vocabulary knowledge, morphological awareness, and contextual analysis. Morphological awareness—the recognition that words are composed of morphemes such as roots, prefixes, and suffixes—has long been identified as a predictor of vocabulary growth and reading comprehension (Carlisle, 1995; Melloni & Vender, 2022; Dosi, 2025). More recent research emphasizes that lexical inferencing is a multi-component process in which phonological decoding, vocabulary breadth and depth, and morphological analysis converge with contextual cues to support accurate meaning construction (Perfetti & Stafura, 2014; Raudszus et al., 2021; Ardanouy & Deacon, 2024).

1.1. Broader Theoretical Frameworks

Lexical inferencing can be understood through several complementary theoretical perspectives that, taken together, highlight the interplay of decoding, vocabulary, morphology, and context in reading development and show how these components jointly support the ability to infer word meanings during reading.

In line with this view, the Lexical Quality Hypothesis (LQH) (Perfetti & Stafura, 2014) argues that efficient reading depends on the precision of lexical representations, which integrate orthographic, phonological, semantic, and morphological information. When such representations are underspecified—as is often the case in children with developmental dyslexia (DD)—access to meaning becomes fragile, and readers struggle to combine word-internal cues with contextual information (Nation & Snowling, 1998; Rothou & Padeliadu, 2019). This framework thus predicts that decoding weaknesses hamper the construction of high-quality lexical entries, limiting inferencing. Whereas the LQH emphasizes the quality of individual lexical representations, the Simple View of Reading (SVR) (Florit & Cain, 2011) extends this logic to the sentence and text level, proposing that reading comprehension reflects the interaction between decoding and linguistic comprehension. In transparent orthographies such as Greek, decoding skills are relatively efficient but continue to constrain comprehension until around age ten (Protopapas et al., 2007). According to this model, children with DD show persistent decoding difficulties while maintaining relatively intact comprehension (Sleeman et al., 2022), which may alter the balance of resources they bring to inferencing tasks. Extending the SVR, Scarborough’s (2001) Reading Rope illustrates how strands of word recognition (e.g., phonological awareness, decoding, orthographic processing) intertwine with strands of language comprehension (e.g., vocabulary, morphology, syntax, background knowledge) to support fluent reading. This perspective underscores that morphological awareness and vocabulary depth are not isolated skills but integrated components of comprehension. Finally, the Construction–Integration model (Kintsch, 1998) emphasizes the integration of word-level cues with sentence- and discourse-level context to build coherent mental representations of text. From this viewpoint, readers with weaker lexical representations—such as children with DD—may fail to coordinate morphology and context effectively, resulting in partial or inaccurate inferences.

Review and meta-analytic work reinforces these theoretical claims. Nation (2017) stresses the central role of reading experience and lexical quality in inference-making, Oakhill and Cain (2011) highlight comprehension monitoring and inference generation as key drivers of reading development, and Marks et al. (2022) provide neurocognitive evidence that morphological processing is a critical differentiator between typical and impaired readers. Taken together, these perspectives converge on three predictions: first, decoding acts as a gateway to word meaning and remains a bottleneck for children with DD; second, vocabulary breadth and depth provide the semantic resources needed for inferencing, with depth often exerting the stronger influence on comprehension; and third, morphology and context allow readers to extend or refine interpretations, though integration of these cues is more successful when lexical representations are well specified. These predictions directly inform the hypotheses of the present study outlined below.

1.2. Decoding and Inferencing

Decoding provides the entry point to word meaning. Efficient decoding facilitates recognition of morphemes within words, while poor decoding reduces resources available for higher-level integration. In transparent orthographies, decoding shapes reading until the age of 10 (Protopapas et al., 2007; Florit & Cain, 2011). Decoding ability also influences the extent to which children can benefit from morphological cues: those with stronger pseudoword decoding are more likely to exploit morphological structure (Raudszus et al., 2021).

For children with DD, persistent phonological decoding deficits constrain both direct word recognition and the ability to exploit morphology and context (Rothou & Padeliadu, 2019; Giazitzidou & Padeliadu, 2022). These weaknesses also reduce access to vocabulary and contextual cues, limiting inferencing efficiency. Evidence from Croatian, another transparent Slavic orthography, aligns with this pattern. Kałdonek-Crnjaković (2019) showed that DD learners of English as a foreign language experience persistent spelling difficulties despite multisensory instruction, particularly with complex or less frequent words. Kuvač Kraljević et al. (2024) further demonstrated that poor Croatian readers rely disproportionately on decoding alongside comprehension, with many showing either decoding-specific or mixed decoding–comprehension deficits. Thus, decoding remains a key point for inference even in later elementary school.

1.3. Vocabulary Breadth and Depth

Vocabulary knowledge provides the semantic foundation for inferencing. Vocabulary breadth (the number of words known) increases the likelihood of recognizing contextual cues, while vocabulary depth (the richness of semantic, syntactic, and morphological representations) supports fine-grained inferencing through synonyms, antonyms, and derivational families (Nation, 2017).

Studies consistently show that both breadth and depth contribute uniquely to reading comprehension and inferencing. For example, Ouellette (2006) found that breadth was more strongly associated with word recognition, whereas depth uniquely predicted comprehension. Similarly, Guerra and Kronmüller (2024) demonstrated that children with richer semantic networks are more adept at bridging context gaps and generating accurate inferences.

Cross-linguistic studies confirm these roles. In transparent orthographies such as Greek and Hebrew, vocabulary depth—together with morphological knowledge—predicts later fluency and comprehension beyond phonological awareness (Manolitsis et al., 2017; Cohen-Mimran et al., 2023). In opaque orthographies such as English and French, breadth is critical in early stages, but depth becomes the stronger predictor of higher-order comprehension as texts grow more complex (Kuo & Anderson, 2006; Nagy et al., 2006).

Children with DD and language disorders typically show weaknesses in both breadth and depth, limiting their ability to leverage context and exacerbating decoding difficulties (Dosi & Gavriilidou, 2020; McGregor et al., 2013). These deficits restrict their inferencing options, making them more reliant on surface cues or frequency-based strategies.

1.4. Morphological Awareness

Morphological awareness equips readers with strategies to interpret unfamiliar words by decomposing complex forms into familiar morphemes or generating new derivations. Awareness of derivational morphology supports vocabulary growth and inferencing (Carlisle, 2000; Kirby et al., 2012). Longitudinal studies demonstrate that morphological abilities in preschool predict later fluency and comprehension (Cohen-Mimran et al., 2023). Cross-linguistic research shows that sensitivity to derivational morphology provides a powerful mechanism for extending lexical knowledge (Kuo & Anderson, 2006). In Greek, both inflectional and derivational awareness predict reading comprehension independently of phonological awareness (Manolitsis et al., 2017).

For children with DD, morphological processing is often impaired (Burani et al., 2008; Tsesmeli & Seymour, 2006; Diamanti et al., 2018). Yet morphology may also serve as a compensatory scaffold; thus, readers with DD can sometimes use morpheme boundaries to support decoding (Casalis et al., 2004; Melloni & Vender, 2022). Whether children with DD benefit from morphological cues in the same way as their peers remains an open question, particularly in languages with rich morphology.

1.5. The Role of Context

Contextual inferencing enables readers to use syntactic and semantic information from text to refine interpretations. When encountering an unfamiliar word, children typically draw on both word-internal cues (morphology) and word-external cues (context) simultaneously (Perfetti & Stafura, 2014). Effective contextual inferencing requires not only vocabulary and syntax but also working memory to integrate information across sentences (Cain et al., 2004; Nation, 2017).

Research shows that children flexibly combine context and morphology, compensating for weaknesses in one by relying on the other. For example, Cain et al. (2004) found that good comprehenders could use sentence context to infer novel word meanings, whereas poor comprehenders often failed when clues were distant or subtle. Schneider et al. (2022) reported that stronger readers shift flexibly between cues, integrating morphological and contextual information simultaneously.

Cross-linguistic evidence supports the universality of this strategy. German- and Dutch-speaking children with stronger vocabularies and reading proficiency were more successful at using context (Gough Kenyon et al., 2018; Raudszus et al., 2021). In L2 Chinese, learners exploited supportive contexts more effectively when they had sufficient morphological knowledge (Xu & Zhang, 2020).

For readers with DD, context often plays a compensatory role. Nation and Snowling (1998) and Rothou and Padeliadu (2019) showed that children with DD rely heavily on contextual cues, though not always effectively, because their reduced vocabulary and decoding difficulties constrain integration.

Despite consistent evidence for the role of decoding, vocabulary, morphology, and context, few studies have examined how these components jointly predict lexical inferencing, and fewer still have compared their predictive weight in typically developing (TD) and DD readers. This study extends prior work in three ways: it focuses on Greek, a transparent, morphologically rich orthography where direct TD–DD comparisons remain scarce, thus contributing cross-linguistic evidence beyond the predominantly English-based literature; it evaluates the combined contribution of decoding, vocabulary breadth and depth, morphological awareness, and reading comprehension within a single design, rather than in isolation; and it integrates quantitative analyses with qualitative error examination to capture the strategies children use when confronted with unfamiliar words. Together, these features illuminate both the shared foundations of inferencing and the distinct developmental pathways of TD and DD readers, offering a more nuanced account of the mechanisms shaping lexical growth in childhood.

2. The Present Study

Building on this rationale, the objectives of the present study were twofold: first, to determine whether TD and DD readers differ in vocabulary knowledge, decoding, morphological awareness, and lexical inferencing; and second, to identify the extent to which these linguistic and reading-related skills predict inferencing performance within each group. To this end, group comparisons and hierarchical regression analyses were complemented by qualitative error analysis, capturing both the strength of predictors and the strategies children employed when constructing or attempting to construct meaning.

Through this design, it was sought to provide a more comprehensive account of how multiple linguistic and reading processes converge in lexical inferencing, and to clarify whether the pathways into word meaning are similar or distinct for children with and without DD.

Research Questions and Hypotheses

Two interrelated questions along with their hypotheses were formed:

RQ1. Do children with DD differ from TD peers in their performance on baseline language and reading measures (vocabulary breadth and depth, morphological awareness, decoding and reading comprehension) and in their ability to infer word meanings from text?

RH1. Consistent with prior evidence (Giazitzidou & Padeliadu, 2022; Melloni & Vender, 2022; Rothou & Padeliadu, 2019; Schneider et al., 2022), children with DD will perform significantly lower than TD peers on vocabulary breadth, vocabulary depth, morphological awareness, decoding measures and reading comprehension, as well as on the lexical inferencing task. These differences are expected to reflect the constraints of underspecified lexical representations, as proposed by the LQH, and the documented decoding difficulties in DD described by the SVR.

RQ2. To what extent do decoding, morphological awareness, vocabulary knowledge, and reading comprehension skills predict children’s success in lexical inferencing, and does this predictive pattern differ between children with DD and TD peers?

RH2a. Across groups, morphological awareness and vocabulary knowledge will uniquely predict lexical inferencing, consistent with the LQH and supported by reviews showing their robust contributions to comprehension.

(Nation, 2017; Marks et al., 2022)

RH2b. For TD readers, once decoding is relatively automatized, morphological awareness and reading comprehension (reflecting the integration of context, as outlined by the Construction–Integration model and Reading Rope) will jointly support inferencing.

(Kintsch, 1998; Scarborough, 2001)

RH2c. For DD readers, persistent decoding weaknesses will continue to constrain lexical representations, making decoding, vocabulary breadth, and morphology stronger predictors than comprehension. This aligns with evidence that children with DD rely on partial lexical and semantic cues rather than flexible integration of context.

(Nation & Snowling, 1998; Rothou & Padeliadu, 2019)

3. Materials and Methods

3.1. Participants

The study included a total of 60 children (9–12 years old), divided into two groups: 30 TD children and 30 children with developmental DD. The TD group had a mean age of 9.97 years (SD = 1.07), while the DD group had a mean age of 10.25 years (SD = 1.05). Both groups were matched for age, minimizing potential confounding effects (t (58) = −1.027, p = 0.308), and for gender (TD: 16 males; DD: 18 males). Participants also did not differ in terms of their non-verbal intelligence (t (58) = 0.156, p = 0.876), tested by means of Colored Progressive Matrices (Raven et al., 2008).

All participants were native Greek monolingual speakers, with normal or corrected-to-normal vision and no reported neurological or sensory impairments. Participants were recruited from public schools in Athens, and written informed consent was obtained from parents or legal guardians.

Children in the DD group had an official diagnosis of DD from the Centre for Interdisciplinary Assessment, Counselling, and Support (KEDASY). In Greece, such diagnoses are issued exclusively by state diagnostic centers, based on standardized assessments of both accuracy and fluency in reading and spelling, together with phonological and language abilities. This multifaceted protocol aligns with international standards, where DD is defined as a persistent difficulty with word reading accuracy and fluency, ensuring the reliability of research involving Greek samples.

For the TD group, teachers and parents/guardians confirmed that children showed age-appropriate performance in reading and spelling and had no history of learning or language difficulties or special educational needs.

Inclusion criteria required enrollment in mainstream schools, native Greek proficiency, and parental consent. For the DD group, an official diagnosis from ΚΕDASΥ was mandatory, while TD children had to demonstrate typical reading development without a history of learning or language disorders. Exclusion criteria included neurological, sensory, or cognitive impairments, uncorrected vision or hearing deficits, non-native Greek speakers, or lack of parental consent.

The study was conducted in accordance with the Declaration of Helsinki and approved by the Ethical Committee of the University of Thrace (ΔΠΘ/ΕHΔΕ/38844/332/31-01-2025).

An a priori power analysis was conducted in GPower version 3.1 (Faul et al., 2009). We set α = 0.05 (two-tailed), the conventional threshold in behavioral and educational research. Effect-size assumptions followed Cohen’s (1988) guidelines and prior studies reporting comparable effects in reading and inferencing research (e.g., Cain et al., 2004; Nation, 2017; Melloni & Vender, 2022), where medium (d = 0.50) to large (d = 0.80) effects are typically observed. With N = 60 (30 per group), power is ≈0.86 to detect large between-group effects and ≈0.48 for medium effects (independent-samples t test).

3.2. Material

In addition to the large battery administered, all participants completed a non-verbal intelligence test (Raven’s Colored Progressive Matrices; Raven et al., 2008), as described in the Participants section. This was used as a screening tool to ensure no group differences in general reasoning abilities and to confirm comparable non-verbal intelligence performance between TD and DD children. Beyond this measure, participants were administered a set of tasks in Greek assessing vocabulary knowledge, morphological awareness, decoding, reading comprehension, and lexical inference. All tasks, except for reading comprehension and lexical inference, were administered orally.

3.2.1. Vocabulary Knowledge Tasks

To provide a comprehensive evaluation of vocabulary knowledge, both breadth and depth were measured. Breadth was assessed with the Greek version of the Expressive Vocabulary Test (Vogindroukas et al., 2009). The task comprises 50 colored images depicting common objects or actions. Children are asked to name each picture, and testing is discontinued once five consecutive errors occur. The maximum score is 50, corresponding to the number of correct responses.

Depth of vocabulary knowledge was examined with a multiple-choice task designed to measure knowledge of synonyms and antonyms. This task contained two sections: one with 20 synonyms and one with 20 antonyms. For each item, four possible answers were provided, and participants selected the single correct option. Each correct answer was scored as 1, yielding a total maximum score of 40 (20 per subtest).

3.2.2. Morphological Awareness Tasks

Morphological awareness was measured through two productive tasks designed to assess both morphological derivation and decomposition. Derivation refers to the ability to generate new words by applying affixes (e.g., happy → happiness), while decomposition involves analyzing complex words into their morphemic constituents (e.g., unhappiness → un- + happy + -ness). The first part of the task required participants to generate a derived form from a base word to appropriately complete a sentence (12 items). For example, given the prompt “Kostas was the … of the competition (win),” the correct response would be winner. The second part of the task presented participants with a derived word and asked them to retrieve and produce the corresponding base form, again in a sentence context (12 items). For example: “This cake has fresh … (buttery),” with the expected answer butter. Altogether, the morphological awareness measure contained 24 items, and the maximum score was 24 points.

3.2.3. Decoding Task

Word decoding ability was assessed with the DADA pseudoword reading test (Panteliadou et al., 2019). The full version of the test contains 40 pseudowords arranged in order of increasing complexity. Each pseudoword is phonotactically legal in Greek but does not correspond to an existing lexical item, thereby requiring children to rely on grapheme–phoneme correspondences rather than memory. The administration was discontinued if a child made five consecutive errors. Scoring was straightforward: one point was awarded for each correctly decoded pseudoword, with a maximum of 40 points.

3.2.4. Reading Comprehension Task

Reading comprehension was evaluated using two passages from the DADA test (Panteliadou et al., 2019). Each participant read both texts, either silently or aloud, depending on comfort and fluency. Following each passage, they answered seven multiple-choice questions targeting different levels of comprehension: literal understanding (e.g., recalling explicitly stated information), inferential comprehension (e.g., drawing conclusions beyond the given text), and evaluative comprehension (e.g., reasoning about intentions or implications). Each correct answer received 1 point, resulting in a maximum total score of 42.

3.2.5. Lexical Inference Task

The experimental task assessing meaning inference was specifically designed for this study and was based on Cain et al. (2004). Twenty short stories were created (see Appendix A,

Table A1. Examples of stories used in the lexical inferencing task. The table shows both the derived and simple forms, with suffixes in italics.

Simple Word	Derived Word
My family and I decided to go on a trip to the park with the wild βρα (pseudoword) on Sunday. As soon as we arrived, we sat on a bench to rest after the journey. We were very hungry and asked mom if we could eat something. We had a sandwich and drank some juice, and our parents had a coffee. Once we finished, we entered the main area. The park was huge. There were lakes, many trees, and pathways. There were many species from different countries around the world. Some lived on land, some flew, and others swam. We took lots of photos to remember this beautiful trip.	Last night, when it was very hot, the air conditioner in our house broke down and we couldn’t fix it. Luckily, though, in addition to the air conditioner, we also have a μαριστήρα (pseudoword). Our house is on the third floor of a new apartment building. It is very nice, bright, and has a big balcony. There, Mom has set up a cot that turns into a bed, a table with chairs, and pots with colorful flowers. My dad has plugged it in and placed it in a spot where it can cool the whole room. He asked if the air was reaching us, and my brother asked him to turn it a little more towards him. With that solution, the warm evening passed pleasantly for the whole family.
Target answer: ζώα (animals)	Target answer: ανεμιστήρα (fan)

), each embedding a pseudoword in place of a real target word. These pseudowords were carefully constructed (followed Kosmidis et al., 2025) to represent novel concepts and were not synonymous with existing Greek words. To examine the role of morphology, half of the pseudowords (10 items) contained derivational suffixes that could serve as morphological cues (e.g., τραπείο, μαριστήρα, σκευγευτής), while the remaining 10 were simple, morphologically opaque items with no such cues (e.g., παβά, βρα, γέζη). In both conditions, participants were expected to infer the possible meaning of the pseudoword by integrating either morphological or contextual information provided by the story. Participants read each story and were asked to infer the meaning of the pseudoword by selecting the real word that best matched it. Responses were scored as follows: 2 points for the correct target word, 1 point for a semantically related word (e.g., for the pseudoword τραπείο, which carried the derivational suffix -ειο, the target word was γραφείο “office,” while acceptable semantically related responses included θρανίο “desk” or τραπέζι “table”, which shares phonological similarity with the pseudoword), and 0 points for unrelated answers. The maximum possible score was 40. To ensure lexical appropriateness, all target words were selected from high- to mid-frequency levels based on the HelexKids database (Terzopoulos et al., 2017).

3.3. Data Collection Procedure

Each participant completed two individual assessment sessions of approximately one hour each, conducted in a quiet, designated room at their school during regular school hours. In the first session, participants completed measures of non-verbal intelligence, and a lexical inference task, decoding skills, and vocabulary depth, while the second session included the remaining ten stories of the lexical inference task, as well as assessments of morphological awareness, reading comprehension, and expressive vocabulary.

All responses of the oral tests were audio-recorded and subsequently transcribed to allow for accurate verification and analysis. To ensure scoring reliability, two independent researchers evaluated the participants’ responses. Any discrepancies in scoring were reviewed and resolved through discussion to reach consensus.

3.4. Data Analysis

Statistical analyses were conducted using IBM SPSS Statistics (version 29). For the first research question (RQ1), group differences between children with DD and TD peers on baseline language and reading measures—including breadth and depth of vocabulary, morphological awareness, decoding, reading comprehension, and lexical inference—were examined using independent samples t-tests. Effect sizes were calculated using Cohen’s d to evaluate the magnitude of group differences.

For the second research question (RQ2), the relationships between decoding, morphological awareness, vocabulary knowledge (breadth and depth), and lexical inferencing performance were first examined descriptively using Pearson correlation analyses within each group. To identify the unique contributions of each predictor to lexical inferencing ability, hierarchical regression analyses were then performed independently for the DD and TD groups. Predictors were entered in theoretically motivated blocks (decoding first, followed by vocabulary, morphology, and reading comprehension), allowing assessment of both shared and unique variance explained by the aforementioned variables. The significance level for all statistical tests was set at α = 0.05.

For the hierarchical regressions, the power analysis targeted the incremental R² of the morphological-awareness block, which comprised two tested predictors (derivational, decomposition) entered after decoding and vocabulary and before reading comprehension. In GPower (version 3.1) we specified two tested predictors and seven total predictors in the final model (decoding; vocabulary breadth & depth: synonyms, antonyms; morphological awareness: derivational, decomposition; reading comprehension). Under a large incremental effect (f² ≈ 0.35), N = 60 yields power ≈ 0.80.

4. Results

4.1. Group Differences in Baseline Tasks and the Lexical Inference Task (RQ1)

Independent samples t-tests were conducted to compare children with DD and TD peers on baseline tasks (decoding, reading comprehension, vocabulary and morphological awareness measures) and the lexical inferencing task (

Table 1. Mean accuracy scores (%) and standard deviations (SD) of TD and DD children across language and reading tasks.

	Tasks	Group	Mean (%)	SD
Decoding	Pseudowords reading	TD	90.3	4.3
		DD	58.5	4.9
Reading comprehension		TD	89.5	4.1
		DD	69.3	4.9
Vocabulary breadth	Expressive vocabulary	TD	89.8	3.9
		DD	78.6	4.8
Vocabulary depth	Synonyms	TD	92.5	10.9
		DD	77.5	4.1
	Antonyms	TD	95.5	6.2
		DD	77	3.4
Morphological awareness	Derivation	TD	87.5	7.8
		DD	79.2	6.8
	Decomposition	TD	96.7	5.6
		DD	75	5.4
Lexical inferencing	Stories	TD	89.3	5.3
		DD	76.3	6.5

).

Significant group differences emerged across all language and reading measures. Children with DD performed significantly lower than TD peers on decoding of pseudowords (t (58) = 6.579, p < 0.001), on vocabulary breadth (t (58) = 7.287, p < 0.001), and on vocabulary depth measures, including synonyms (t (58) = 7.962, p < 0.001), and antonyms (t (58) = 7.433, p < 0.001) on morphological derivation (t (58) = 6.917, p < 0.001), and morphological decomposition (t (58) = 5.517, p < 0.001), on reading comprehension (t (58) = 7.329, p < 0.001) and on the main lexical inference task (t (58) = 6.683, p < 0.001).

Effect sizes (Cohen’s d) indicated large differences for decoding (d = 1.60), reading comprehension (d = 1.89), vocabulary breadth (d = 2.05), synonyms (d = 1.82), antonyms (d = 1.70), morphological awareness (derivation: d = 1.55; decomposition: d = 1.22), and lexical inference (d = 1.52), confirming the practical significance of these group differences.

Error Patterns for Morphological Awareness and the Lexical Inferencing Tasks

As summarized in

Table 2. Summary of main error patterns in morphological awareness and lexical inferencing tasks with representative examples.

Task	Error Type	Description	Representative Examples	Group Pattern
Morphological awareness—production of derived words	Alternative derivational form; frequent-form preference	Produced a different but semantically related derivation from the base instead of the target form	Given word: φως (light) → target: φωτεινό (bright); children: φωτισμός (lighting)	Frequent in DD; rare in TD
	Alternative derivational/compound forms	Produced a diminutive or compound word from the base instead of the target low-frequency derivation	Given word: αμπέλι (vine) → target: αμπελώνας (vineyard); children: αμπελάκι (small vine), αμπελοχώραφο (vine field)	Frequent in DD; rare in TD
	Inflectional substitution/frequent-form preference	Supplied an inflected form of the base adjective/noun instead of producing the required derived form, often selecting the more frequent or similar form	Given word: αρχαίος (ancient, mask. sg.) → target: αρχαιότητα (antiquity); children: αρχαία (ancient, fem. sg.)	Frequent in DD; occasional in TD
Morphological awareness—decomposition	Derived-form substitution	Instead of retrieving the required base form, children produced another derived word from the same base	Given word: ψαρεύω (to fish) → target: ψάρια (fish); children: ψαρικά (fish products)	Frequent in DD; rare in TD
	Lateral derivation from stimulus	Instead of retrieving the required base form, children generated a new derivative based on the given word	Given word: χορευτής (dancer) → target: χορούς (dances); children: χορευτικό (dance performance)	Present in DD; rare in TD
Lexical inferencing	Semantic field activation	Related but imprecise lexical choice	Given pseudoword: ράδκεια → target: διάρκεια (duration) → children: χρόνος (time)	Frequent in DD; occasional in TD
	Phonological similarity	Phonologically similar but irrelevant word	Given pseudoword: ράδκεια → target: διάρκεια (duration) → children: ράδιο (radio)	Frequent in DD; absent in TD
	Frequent/concrete word preference	Favored familiar/simple words over abstract/academic ones	correct: ζώα (animals), βιβλίο (book); struggled: ομοιότητα (similarity)	Frequent in DD; rare in TD
	No response	Item left unanswered	–	Common in DD; rare in TD
	Semantically appropriate but morphologically inaccurate guesses	Meaning-related but not morphologically aligned with the target	Given pseudoword: εροιότητα → target: δραστηριότητα (activity) → children: ασχολία (hobby)	Occasional in TD

, children with DD produced systematic error patterns in both morphological awareness and lexical inferencing.

In the morphological awareness task, DD children often generated alternative derivations, diminutives, compounds, or inflectional forms instead of the required derived words, reflecting a tendency to rely on more frequent or semantically accessible forms rather than low-frequency or abstract targets. In decomposition, DD children frequently produced another derivative from the same root or extended derivation from the stimulus itself, rather than isolating the simple base form, suggesting difficulty in shifting from derivational to decompositional processing, which may indicate reduced cognitive flexibility, which was not tested in the present study. By contrast, TD children produced fewer, less systematic errors, and their deviations were usually morphologically plausible alternatives, indicating more precise morphological knowledge.

In the lexical inferencing task, DD children often produced semantically related but inaccurate responses, with phonological similarity, or frequent/concrete words, and they were more likely to omit answers on abstract or low-frequency words. TD children, in contrast, rarely made phonological errors, tended to attempt semantically plausible guesses rather than leave items unanswered, and were more successful in narrowing down to the intended lexical item. This pattern indicates not only more robust morphological representations and more efficient integration of context but also greater confidence and a more active inferencing approach. Overall, the two profiles suggest that while TD children demonstrate robust morphological representations and efficient integration of context, DD children’s strategies are constrained by partial morphological cues, semantic associations, and frequency-based preferences.

The observed differences in inferencing skills may also be influenced by broader cognitive factors not directly measured in this study, such as working memory, attentional control, or metacognitive strategies. These unassessed processes likely interact with linguistic skills during inferencing, representing a limitation of the current study.

4.2. Predictors of Lexical Inferencing

For TD children, lexical inferencing was positively correlated with morphological awareness, i.e., with decomposition (r (30) = 0.488, p = 0.006). It was also correlated with reading (r (30) = 0.373, p = 0.042), vocabulary breadth (expressive vocabulary; r (30) = 0.513, p = 0.004) and vocabulary depth (synonyms (r (30) = 0.762, p < 0.001), and antonyms (r (30) = 0.597, p < 0.001) (see Appendix B,

Table A2. Pearson correlations among language measures for the TD and DD groups (n = 30, respectively).

Measure	1	2	3	4	5	6	7	8
1. Pseudowords	—	0.27	0.28	0.40 *	0.19	0.61 ***	0.51 **	0.18
2. Reading	0.46 *	—	0.51 **	0.74 ***	0.66 ***	0.58 ***	0.41 *	0.37
3. Vocabulary breadth	0.56 ***	0.43 *	—	0.54 **	0.73 ***	0.46 *	0.68 ***	0.51 **
4. Synonyms	0.20	0.58 ***	0.50 **	—	0.78 ***	0.73 ***	0.63 ***	0.76 ***
5. Antonyms	0.35	0.27	0.62 ***	0.22	—	0.58 ***	0.49 **	0.60 ***
6. Derivation	0.39 *	0.22	0.50 **	0.23	0.37 *	—	0.49 **	0.23
7. Decomposition	0.08	0.24	0.27	0.20	0.48 **	0.48 **	—	0.49 **
8. Stories	0.37 *	0.63 ***	0.55 **	0.71 ***	0.30	0.22	0.00	—

Note. Values are Pearson’s r. The results for participants with DD were presented on the grey background. * p < 0.05, ** p < 0.01, *** p < 0.001.

). In the hierarchical regression analysis (

Table 3. Hierarchical regression predicting story comprehension in TD and DD groups.

Step	Predictors Entered	ΔR2	F Change	df1	df2	p	β (Final Model)
TD group
1	Decoding (pseudowords)	0.03	0.9	1	28	0.351	0.77
2	Vocabulary breadth & depth (synonyms, antonyms)	0.60 ***	13.66	3	25	<0.001	Synonyms = 0.90
3	Morphological awareness (derivational, decomposition)	0.36 ***	518.97	2	23	<0.001	Derivational = 1.94, Decomposition = 0.31
4	Reading comprehension	0.01 ***	32.81	1	22	<0.001	0.15
Total R2 = 0.997 *
DD group
1	Decoding (pseudowords)	0.14 *	4.48	1	28	0.043	0.37
2	Vocabulary breadth & depth (synonyms, antonyms)	0.48 ***	10.52	3	25	0.004	Synonyms = 0.58
3	Morphological awareness (derivational, decomposition)	0.36 ***	187.48	2	23	<0.001	Derivational = 0.60, Decomposition = 0.60
4	Reading comprehension	0.01 ***	11.3	1	22	0.003	0.14
Total R2 = 0.985 *

Note. Standardized β values are from the final step of each model. Synonyms consistently emerged as the strongest vocabulary predictor, and both derivational and decomposition tasks made significant unique contributions. * p < 0.05; *** p < 0.001.

), decoding entered at Step 1 did not significantly predict lexical inferencing, accounting for only 3% of the variance, R² = 0.03, F(1,28) = 0.90, p = 0.351. Adding vocabulary breadth and depth at Step 2 significantly improved the model, ΔR² = 0.60, F change(3,25) = 13.66, p < 0.001, with synonyms emerging as a strong predictor (β = 0.90, p < 0.001). At Step 3, morphological awareness (derivational and decomposition tasks) accounted for an additional 36% of the variance, ΔR² = 0.36, F change(2,23) = 519.0, p < 0.001, with both derivational (β = 1.94, p < 0.001) and decomposition (β = 0.31, p < 0.001) skills contributing. At Step 4, reading comprehension added only 0.5% variance, though still statistically significant, ΔR² = 0.005, F change(1,22) = 32.81, p < 0.001. The final model accounted for 99.7% of the variance in lexical inferencing (R² = 0.997).

For children with DD, lexical inferencing was significantly correlated with decoding (r (30) = 0.371, p = 0.043), reading (r (30) = 0.633, p < 0.001), vocabulary breadth (r (30) = 0.640, p < 0.001) and vocabulary depth (synonyms; r (30) = 0.708, p < 0.001) (see Appendix B, Table A2). In the hierarchical regression analysis (Table 3), decoding alone at Step 1 explained 14% of the variance, R² = 0.14, F(1,28) = 4.48, p = 0.043. Adding vocabulary at Step 2 significantly improved the model, ΔR² = 0.48, F change(3,25) = 10.52, p < 0.001, with synonyms again a strong predictor (β = 0.58, p = 0.004). Morphological awareness at Step 3 accounted for an additional 36% of the variance, ΔR² = 0.36, F change(2,23) = 187.48, p < 0.001, with both derivational (β = 0.60, p < 0.001) and decomposition (β = 0.60, p < 0.001) each emerging as significant predictors. Finally, reading comprehension at Step 4 explained an additional 0.7% of the variance, ΔR² = 0.007, F change(1,22) = 11.30, p = 0.003. The full model accounted for 98.5% of the variance in lexical inferencing (R² = 0.985).

To assess potential multicollinearity, we examined Variance Inflation Factor (VIF) and tolerance values for all predictors. Diagnostics indicated that while some predictors showed acceptable values (e.g., VIF ≈ 2.7–4.9), others exhibited elevated collinearity (VIF up to 13.9; tolerance as low as 0.07), particularly among vocabulary and morphological measures (e.g., synonyms, antonyms, derivational, decomposition). These results suggest redundancy among the linguistic predictors, which likely contributed to the very high R² values and inflated standardized coefficients observed in the models. Collinearity statistics are reported in

Table 4. Collinearity diagnostics (Variance Inflation Factor [VIF] and tolerance values) for predictors entered in the hierarchical regression models.

Predictor	VIF Range	Tolerance Range
Decoding (pseudowords)	1.0–5.7	0.18–1.0
Vocabulary breadth	2.9–13.8	0.07–0.34
Synonyms	6.3–12.8	0.08–0.16
Antonyms	4.9–11.0	0.09–0.20
Morphological derivation	7.0–11.8	0.08–0.14
Morphological decomposition	2.7–7.0	0.14–0.37
Reading comprehension	2.7–4.9	0.20–0.36

.

5. Discussion

The present study had two main goals. First, it aimed to determine whether TD and DD readers differ in vocabulary knowledge, decoding, morphological awareness, reading comprehension and lexical inferencing accuracy (RQ1). Second, it examined the extent to which these linguistic and reading-related skills predict inferencing performance within each group (RQ2). To address these questions, we combined quantitative analyses of group performance and regression models with qualitative analyses of error patterns, thereby capturing not only the strength of predictors but also the strategies children used successfully construct meaning.

The major results showed that children with DD performed significantly lower than their TD peers on all baseline measures (decoding, vocabulary breadth and depth, morphological awareness, and reading comprehension) as well as on the main task of lexical inferencing (RH1). These differences were generally large, suggesting potential practical significance. The qualitative error analyses pointed to distinct patterns: DD children often relied on frequent or semantically related forms, phonological similarity, or concrete word preferences, and more often omitted responses, whereas TD children produced fewer and more systematic errors, typically offering semantically plausible guesses that suggest somewhat more precise morphological and contextual integration.

Regression analyses suggested that in both groups, vocabulary depth (synonyms) and morphological awareness (derivation and decomposition) were associated with lexical inferencing (RH2a). However, the predictive pathways appeared to diverge: decoding contributed significantly only in the DD group (RH2c), while reading comprehension added variance only in the TD group (RH2b). Taken together, these patterns tentatively indicate that although vocabulary depth and morphology may support inferencing across groups, TD children may additionally draw on higher-order comprehension processes, whereas DD children seem to remain more influenced by decoding and rely more on partial morphological and semantic cues.

5.1. Groups’ Performance on Language and Reading Measures

Consistent with RH1, TD children significantly outperformed DD peers on all tasks, including decoding, reading comprehension, morphological awareness, vocabulary breadth and depth, and lexical inferencing. These differences echo evidence that DD extends beyond phonological decoding to multiple linguistic levels that jointly support comprehension and vocabulary growth (Giazitzidou & Padeliadu, 2022; Protopapas et al., 2007; Rothou & Padeliadu, 2019). In transparent orthographies such as Greek, decoding deficits constrain access to morphemic structure and reduce lexical quality, limiting children’s ability to exploit morphological and contextual cues (Florit & Cain, 2011; Raudszus et al., 2021). From the perspective of the LQH (Perfetti & Stafura, 2014), weaker decoding undermines precise lexical representations, with compounding effects on vocabulary and inferencing. In parallel, the SVR (Florit & Cain, 2011) and Scarborough’s (2001) Reading Rope explain how weaknesses in word recognition unravel broader comprehension strands, producing the large gaps observed.

Building on this, morphological awareness was a key differentiator, since TD children capitalized on morphological cues in the inferencing task, whereas DD children displayed less efficient strategies, consistent with prior evidence of difficulties in derivational analysis (Burani et al., 2008; Casalis et al., 2004; Melloni & Vender, 2022). This aligns with the Construction–Integration model (Kintsch, 1998), which emphasizes the need to integrate multiple lexical and contextual cues; TD children, with stronger morphological skills, achieved this more efficiently.

Error patterns further clarify these group differences. TD children’s errors were rare and typically plausible, reflecting relatively precise lexical representations. In contrast, DD children systematically relied on less efficient strategies, i.e., generating alternative derivations, diminutives, or inflected forms in production; producing related derivatives rather than the base in decomposition; and favoring phonological similarity, high-frequency words, or omissions in inferencing. This overreliance on surface cues and frequency reflects partial activation of representations, consistent with the LQH, SVR, and Construction–Integration model. By contrast, TD children’s greater reliance on plausible guesses underscores the role of lexical quality, morphology, and metacognitive monitoring in more effective inferencing.

Taken together, the findings show that TD children integrate morphology and context more successfully, while DD children remain restricted by weaker morphological awareness, confirming the group-level predictions set out in RH1. Beyond linguistic skills, however, broader cognitive mechanisms not assessed here—such as working memory, attentional control, and executive functions—likely constrain children’s ability to retain context, suppress irrelevant associations, and monitor interpretations (Cain et al., 2004; Oakhill & Cain, 2011). These processes may explain DD children’s tendency toward omissions, phonological confusions, and reliance on frequent forms. From the perspective of the Construction–Integration framework, successful inferencing requires not only lexical precision but also efficient coordination of cognitive resources. Future research should therefore examine how linguistic and domain-general skills jointly shape inferencing in TD and DD readers.

5.2. Predictors of Lexical Inferencing

Correlation and regression analyses highlighted both shared and distinct predictors of inferencing across groups. In line with RH2a, in both TD and DD children, lexical inferencing was strongly associated with vocabulary breadth and depth, confirming that a rich lexicon—both in quantity and semantic organization—is universally essential for constructing meaning from context (Kuo & Anderson, 2006; Nation, 2017; Nagy et al., 2006; Ouellette, 2006). The role of vocabulary depth, and particularly performance on synonyms, suggests that fine-grained semantic relations within the lexicon may support sensitivity to contextual cues (Cain et al., 2004; Dosi & Boni, 2023; Dosi, 2025; Guerra & Kronmüller, 2024; Nation, 2017; Nagy et al., 2006; Marks et al., 2022).

Group-specific predictors, however, revealed distinct developmental pathways. For TD children, inferencing was uniquely supported by morphological awareness (derivation and decomposition) and by reading comprehension, consistent with the Construction–Integration model and Scarborough’s Reading Rope (RH2b confirmed). This pattern suggests that once decoding is automatized, TD readers can allocate resources to integrating morphology and discourse-level information, consistent with the Construction–Integration model (Kintsch, 1998). For DD children, by contrast, decoding remained a significant predictor, suggesting that phonological access continues to constrain meaning construction into later ages, even though decoding no longer plays a decisive role for TD children in transparent orthographies (Protopapas et al., 2007; Florit & Cain, 2011), as predicted by the SVR and LQH (RH2c confirmed). This claim is further supported by the error analysis of the inferencing task: DD participants often relied on phonological similarity when producing responses, overlooking available morphological cues. Such a strategy reflects their continued dependence on surface-level phonological processing and reduced access to morphological structure, and it was not observed in the TD group. Consistent with this pattern, their inferencing also relied on expressive vocabulary breadth, pointing to compensatory strategies in which the sheer availability of lexical items helps offset weaker morphological parsing (Dosi & Gavriilidou, 2020; Melloni & Vender, 2022; Schneider et al., 2022).

At the same time, these predictive patterns may reflect the contribution of broader cognitive mechanisms not assessed in this study. Processes such as working memory, attentional control, and executive functions are critical for holding contextual information, suppressing irrelevant associations, and monitoring competing interpretations during inferencing (Cain et al., 2004; Oakhill & Cain, 2011). From the perspective of Scarborough’s (2001) Reading Rope and the SVR (Florit & Cain, 2011), decoding and linguistic comprehension interact with these domain-general skills, such that weaknesses in lower-level processes (e.g., phonological decoding in DD readers) place greater demands on cognitive control and limit higher-level integration. Similarly, the Construction–Integration framework emphasizes that successful inferencing requires coordination of multiple sources of information alongside efficient allocation of cognitive resources. Thus, while vocabulary depth appeared to be an important linguistic predictor in both groups (RH2a), the contrasting pathways in TD (RH2b) and DD (RH2c) children may reflect differences in how linguistic and domain-general cognitive systems interact to support meaning construction. These findings should be interpreted with caution, given the modest sample, large effect sizes, and evidence of multicollinearity, and regarded as preliminary until replicated in larger and cross-linguistic samples.

5.3. Educational Implications

The results indicate distinct instructional needs for TD and DD learners. For children with DD, decoding and vocabulary breadth predicted inferencing, while error patterns showed reliance on phonology, frequency, and omissions. Instruction for these students may therefore combine systematic decoding support with explicit vocabulary enrichment. Teachers can emphasize breadth by introducing new words across diverse semantic fields, while simultaneously strengthening depth through work on synonyms, antonyms, and multiple meanings. Morphological analysis can be practiced through guided word-building tasks (e.g., deriving new forms from base words, decomposing derived words), which help children recognize structural cues for meaning. Scaffolded inferencing exercises using short texts with unfamiliar words, accompanied by prompts to consider both morphology and context, may gradually shift children away from reliance on surface cues.

For TD children, inferencing was driven primarily by vocabulary depth, morphology, and comprehension, and their errors reflected semantically plausible guesses. Instruction for these learners may therefore focus on refining morphological decomposition and strategic integration of context. Activities that highlight low-frequency or specific morphemes of low-frequency and require students to infer meaning in varied discourse contexts can strengthen lexical precision. Teachers might also integrate think-alouds and metacognitive prompts (e.g., “What clues helped you guess this word?”), which encourage students to monitor and evaluate their inferencing strategies.

Across both groups, assessment may consider not only accuracy but also error patterns, as these provide insight into learners’ strategies and can guide tailored intervention. For example, issues in decoding suggest a need for phonological awareness reinforcement, while overreliance on frequent or concrete words highlights the importance of expanding students’ exposure to abstract and academic vocabulary.

By combining decoding instruction, explicit vocabulary and morphology teaching, and strategy-based inferencing activities, educators can foster both breadth and depth of lexical knowledge while supporting the development of flexible and efficient inferencing skills in diverse learners.

5.4. Limitations

Several limitations should be acknowledged. The sample was modest and cross-sectional, restricting causal claims and raising the possibility that the very large effect sizes observed (d > 1 across most measures) partly reflect sampling variability. This further limits generalizability, particularly given the highly specific linguistic context of Greek, a transparent and morphologically rich orthography. Because decoding predicted inferencing only in the DD group, future cross-linguistic studies are needed to determine whether this pathway is specific to transparent systems or applies more broadly.

Important cognitive variables—including working memory, processing speed, attention, and verbal IQ—were not assessed, though they may account for additional variance in inferencing. While groups did not differ in non-verbal intelligence (Raven’s CPM), the absence of a verbal IQ measure limits interpretive precision, as group differences in vocabulary and inferencing may partly reflect broader reasoning skills. In addition, reading comprehension was measured only through the written DADA test, which may confound comprehension with decoding fatigue, especially in the DD group. Future research should include both oral and written comprehension tasks.

Finally, the qualitative error analysis, while informative, was exploratory and based on relatively few responses, warranting replication with larger samples and tasks that systematically manipulate morphological, semantic, and phonological cues. A further statistical concern is the high collinearity among predictors (VIF values up to 13.9), which likely inflated R² (≈0.985–0.997) and produced unusually large standardized coefficients (β > 1). While consistent with prior evidence linking vocabulary and morphology to comprehension (e.g., Cain et al., 2004; Nation, 2017; Melloni & Vender, 2022), these estimates should be regarded with caution, and larger samples with dimensionality-reduction approaches are needed to obtain more conservative effects.

6. Conclusions

This study examined how decoding, vocabulary (breadth and depth), morphological awareness, and reading comprehension relate to lexical inferencing in TD and DD readers of Greek. DD readers performed lower across baseline measures and in inferencing, likely reflecting constraints from weaker decoding, vocabulary and morphological awareness. Across groups, vocabulary depth and morphological awareness emerged as potential predictors, with some group-specific patterns: TD readers additionally drew on reading comprehension, whereas DD readers remained influenced by decoding. While these findings are broadly in line with previous research, they should be regarded as preliminary given the relatively small sample and the specific linguistic context. Further studies with larger and more diverse samples are needed to verify these patterns and clarify the role of linguistic and domain-general skills in lexical inferencing.