Automatic Morphological Processing in Middle School Students with and without Word Reading Difficulties

Abstract

Morphological processing is the use of morphological structure during word reading. This study investigated whether middle school students applied morphological structure automatically when reading words. In addition, this study asked whether students with word reading difficulties (WRD) applied morphological structure in a way that differed from proficient word readers. Participants were seventh- and eighth-grade students (n = 80). Students were divided into two reading ability groups: proficient word readers (n = 55) and students with word reading difficulties (n = 25). Four computer-administered experimental tasks measured automaticity in reading morphologically complex words and morphologically simple words. A backward masking measure assessed whether students were applying morphological structure automatically to support task accuracy. Students were significantly more accurate in masked performance with morphologically complex words than with morphologically simple words on an oral word reading task. Students with WRD benefitted more from morphological structure on this task than proficient readers did. Findings suggest that proficient word readers and students with WRD automatically apply morphological structure when reading words aloud. In addition, middle school students with WRD may rely more on morphological structure than their proficient peers. However, there may be differences in morphological processing based on the nature of word reading tasks.

1. Introduction

In middle school, academic success depends on the ability to learn content by reading. Throughout middle school, the texts assigned to students increase in complexity [1] and, relatedly, contain increasing numbers of complex words, such as multisyllabic words or those that have multiple morphemes [2]. Yet, many middle school students in the United States struggle to read proficiently. Only 31% of eighth-grade students achieved proficiency on the most recent National Assessment of Educational Progress (NAEP) reading assessment [3]. A lack of reading proficiency continues to impact students long after middle school; those who do not read proficiently will continue to lag behind their peers as their schooling progresses [4].

Sometimes it is assumed that older students only struggle with high-level reading skills, like comprehension, perhaps because instruction in secondary school focuses on reading text to learn content. It is true that the majority of middle school students with reading difficulties have comprehension difficulties [5,6]. However, middle school students with reading difficulties also struggle with foundational, word-level skills [5,6,7]. It is estimated that almost half of students with reading comprehension difficulties struggle with word-level skills (e.g., decoding, word reading fluency) that hinder their ability to derive meaning from text [5,6]. Given the complexity of words and texts required for all students to succeed in middle school and the prevalence of reading difficulties in this population, research on factors that contribute to word reading skills is needed to best support students with word reading difficulties (WRD). The present study concerns one potential factor: morphological processing.

1.1. Morphological Processing

Morphemes are the smallest units of meaning within a word. They comprise stems (e.g., red, frost), affixes (i.e., prefixes and suffixes), roots (e.g., duc, ject), and combining forms (e.g., chrom, geo). Free morphemes can stand alone as words (e.g., eager), whereas bound morphemes must be attached to another morpheme (e.g., -ness, which cannot stand alone). The majority of words in the English language are morphologically complex (i.e., consist of two or more bound or free morphemes [8,9]). Morphologically complex words can be classified into three types: compound words, inflected words, and derived words [8]. Compound words are generally defined as those that consist of at least two free morphemes that are combined to create one word (e.g., basket + ball = basketball). However, two or more combining forms can also create a compound word (e.g., bio + logy = biology). Inflected words are created when at least one free morpheme is combined with a suffix that changes the number, tense, person, gender, mood, voice, or case of the stem (e.g., walk + -ed = walked; cousin + -s = cousins).

The third type, derived words, result from the combination of at least one free or bound stem and one or more affixes that change the meaning of the word (rather than the number, tense, person, gender, mood, voice, or case of the stem) and, unlike inflections, can change the part of speech (e.g., drive + -en = driven). Derived words occur more frequently in English than inflected words and compound words [8] and are used more frequently in academic texts [2,10,11]. Moreover, recent findings have indicated that a small number of derivational suffixes occur frequently in the academic language that is critical to success in multiple curricular areas (English language arts, science, social studies, and math [10]). These words become increasingly important in the secondary grades, as the texts students encounter contain increasing numbers of derived words [11]. Consequently, derived words have been identified as critical to academic success across the secondary school curriculum [2,12,13]. Derived words may provide readers with orthographic structures that support automatic word reading, thus supporting fluent reading and comprehension of the increasingly complex texts encountered in middle school and beyond.

This study concerns morphological processing, a type of morphological ability in which students use morphological structure in a relatively implicit way to read words. In particular, this study examines morphological processing of derived words.

Morphemes have been identified as rich sources of information employed during reading. This includes the obvious high-level information contained in the morpheme itself (e.g., semantics, syntactic function) but also low-level orthographic information. Because morphemes occur frequently in multiple words, they can be learned as orthographic chunks that can be accessed more quickly [14].

First, because morphemes are the smallest units of meaning in a word, they are theorized to provide semantic and syntactic information that supports vocabulary learning, as well as comprehension of sentences and passages [15,16,17,18]. As such, the majority of prior research on the role of morphological ability in the reading abilities of middle school students has typically focused on the contribution of morphological awareness to reading comprehension outcomes. Findings have consistently indicated that morphological awareness explains a unique portion of variance in middle school students’ reading comprehension over and above other reading and language abilities [16,17,18,19,20,21]. This is often interpreted as support for students’ use of morpho-semantic and morpho-syntactic information to make meaning from text.

Second, morphemes also provide orthographic regularities in complex words—morphemes frequently occur in English and may serve as chunks that help students efficiently process words encountered in text. In most theories of reading, there are two pathways in which visual word processing occurs (e.g., [22,23]). In the first pathway, orthography is directly mapped onto semantics, or the meaning of the word (orthography → semantics). In the second, less direct pathway, orthography is mapped onto phonology before the meaning is retrieved (orthography → phonology → semantics). Morphological processing may support automatic processing in both pathways. Students may rapidly activate learned morpho-orthographic chunks (e.g., the suffix -ation, the prefix post-) and then map them to their corresponding meanings (e.g., -ation turns a verb into a noun and means “an action or process”), thus supporting automatic word reading in the direct orthography → semantics pathway [24]. However, morphemes may support automatic word reading even if they are not linked to any higher-level knowledge [25,26,27]. Students may also rapidly map stored morpho-orthographic chunks onto phonology within the orthography → phonology → semantics pathway. For example, they could learn that -tion is pronounced “shun”, despite the fact that this is a fairly irregular spelling of that chunk. But even morphemes with regular spelling–sound mappings (e.g., post-) may still be able to read as a whole, rather than letter-by-letter, because of the frequent co-occurrence of these letters. This would lead to more efficient processing than relying on smaller, non-morpho-orthographic units (e.g., individual letter-sound mappings [28]).

Consequently, it has been suggested by some researchers that the presence of multiple morphemes in some multisyllabic words (e.g., the stem eager and the suffix -ness in eagerness) may help students read them more automatically than words with similar characteristics (e.g., length, number of syllables, surface frequency) containing one morpheme (e.g., embarrass [24,26,29]). Importantly, there is evidence that middle school is the crux at which morphological processing begins to contribute to word reading efficiency [30,31,32,33,34]. This suggests that some middle school students may be applying knowledge of morphological structure automatically, thus improving word-level reading skills.

1.2. Morphological Ability and Students with Word Reading Difficulties

In general, students with WRD exhibit lower levels of morphological ability compared to controls [35,36,37]. In particular, students with WRD have difficulty recognizing and spelling derived words [36,38]. They also show reduced knowledge of syntactic properties of suffixes [39] and of relational (semantic) knowledge of stems and derivations [35]. These students also may especially struggle with written tasks of morphological ability and perform at higher levels in tasks of oral morphological ability compared to their chronological-age-matched peers [35,40]. In addition, there is some evidence that younger adolescents with WRD rely more on explicit (and more effortful) morphological skills to comprehend text than their typically achieving peers [41].

An important question is why students with WRD show these deficits. Some have posited that lower morphological ability in this population reflects phonological deficits [35,42]. Yet, findings consistently indicate that morphological awareness contributes to reading outcomes over and above phonological awareness of students with and without reading difficulties [19,34,43]. This suggests that morphological ability cannot be explained solely by phonological skills. Instead, students with WRD may not have had sufficient practice with morphologically complex words and/or more difficult morphemes (e.g., opaque derivations) in written formats to exhibit levels of performance similar to that of their peers [26,44]. This is consistent with the idea that some morphological learning is grounded in a form of chunk learning.

Students with WRD also demonstrate relative strengths in their morphological ability, particularly in morphological processing. Morphological processing in reading involves using information afforded by morphemes (orthography, phonology, and semantics) to read morphologically complex words [45]. Unlike more explicit forms of morphological ability (e.g., morphological awareness) morphological processing is thought to occur in a relatively implicit manner. For example, younger elementary school students with dyslexia exhibited sensitivity to morphological structure at similar levels to those of their reading-age-matched peers in a probe detection task [46]. In addition, older elementary school students with WRD outperformed peers matched by comprehension ability on producing words from the same morphological family [35]. In a masked priming study with a sample of adolescents (mean age = 13), students with dyslexia exhibited effects for the priming of derived words that were similar to reading-age-matched peers [47].

Thus, there is evidence that students with WRD are sensitive to morphological structure, beginning at a young age and continuing into adolescence. Some have suggested that students with WRD and word reading disabilities (e.g., dyslexia) may harness their relative strengths in morphological processing by relying on morphological structure to compensate for phonological deficits [48]. For example, these students may particularly benefit from processing words in the orthography → semantics pathway (thus bypassing phonology [35]) or using the large-grained orthographic units offered by morphemes to process words more efficiently in the orthography → phonology → semantics pathway (thus avoiding effortful decoding via individual letter-sound mappings [40,48].

This hypothesis is supported by findings from an early study of morphological processing, reading rate, and sentence comprehension [49]. This study used a self-paced reading paradigm in which text was chunked in several ways: by morphemes, by syllables, and by words. Older Dutch adolescents with dyslexia improved their reading rate and sentence comprehension when a text was chunked into morphemes, rather than into syllables, and displayed similar levels of performance for morpheme-chunked text and text presented word-by-word. These effects were not found for reading-age-matched peers, who performed better when text was presented word-by-word. Oral reading rate and sentence comprehension were superior for students without dyslexia, regardless of text presentation. Consequently, students with WRD may rely on morpho-orthographic chunks more than proficient readers—however, it is unclear whether this actually helps their overall reading ability [41,48].

Taken together, these findings indicate that students with WRD may be appropriately sensitive to morphological structure but perhaps have less frequent and less varied experiences with morphologically complex words (as well as morphologically simple words [48]). Lower levels of reading comprehension and vocabulary (and related decreases in reading experience) may lead to incomplete or unstable representations of morpho-orthographic, morpho-phonological, and morpho-semantic regularities afforded by these words, even if relatively explicit skills like morphological awareness are intact [41,50]. Thus, similar to younger students, older students with WRD (e.g., middle school students) may have acquired morphological regularities but cannot yet apply them with automaticity [24,26,29]. For example, as with younger students, it has been suggested that students with WRD may only chunk stems when reading morphologically complex words, rather than fully chunking words into their stem(s) and affix(es) [46]. This may lead to the incomplete acquisition and application of morpho-phonological and morpho-semantic regularities—students may only learn and apply the pronunciations and meanings of stems, not affixes—further contributing to the instability of morpho-orthographic representations [26,51]. Accordingly, older students with WRD (e.g., middle school students) may apply knowledge of morphological structure in a relatively strategic (and effortful) way, but may not be able to automatically apply this knowledge while reading [32].

1.3. Automaticity

Capturing the automaticity of morphological processing is difficult, and typical measures introduce variables that may confound the deployment of morphological structure during word reading. For example, word reading fluency and latency-based measures are subject to the rate/accuracy tradeoff: students can increase their reading rate at the expense of accuracy or decrease their rate to improve accuracy [52]. In addition, these measures may also be impacted by a student’s articulation rate or decision-making speed [53,54].

Two recent studies by Roembke and colleagues have employed a novel backward masking measure to isolate automaticity from knowledge during word reading [55,56]. In this measure, words are briefly presented and then covered with a visual mask (e.g., a series of hashtags). The participant is then asked to provide a response related to the word (e.g., select the matching picture). Consequently, phonological and semantic information must be automatically activated and stored while the visual input is present in order for the participant to generate a correct response [55,57]. In contrast, when participants complete unmasked versions of the same tasks, the visual input remains, and they can use their knowledge of related phonological and semantic information to produce a response, rather than relying on automatic activation of this information [58]. Thus, masked versions of a task require knowledge and automaticity. Overall accuracy showed a masking decrement that was consistent across experimental tasks (i.e., performance was lower in masked versions than in unmasked versions).

1.4. The Present Study

The present study used this novel backward masking measure to not only examine the extent to which middle school students with WRD used morphological structure in word reading but also whether they could apply this structure automatically (as opposed to relying on knowledge of morphemes). Our investigation was guided by the following research questions:

• Does the performance decrement for masking differ for morphologically complex words and morphologically simple words? If students are using morphological structure to read words, they should show a smaller decrement for morphologically complex words than morphologically simple words matched on other factors.
• Is the masking decrement moderated by group (i.e., students with WRD vs. proficient word readers)? If students with word reading difficulties rely more on morphological structure to automatically read words, there should be an interaction between morphological status, masking, and reading ability group.

2. Materials and Methods

2.1. Participants

Data were collected from November 2021 to February 2022. Eighty participants were recruited through school districts, a university email list, word of mouth, social media, and community organizations. Participants were screened for inclusion in the study with caregiver-provided information. They were excluded if they had an intellectual disability, were multilingual, or had an uncorrected vision or hearing impairment.

Participants for the present study were part of a larger study on the role of automatic morphological processing in the reading abilities of middle school students. For the present study, participants were divided into two reading ability groups: students with WRD difficulties and proficient word readers. Word reading ability was classified based on decoding outcomes. Participants were classified as having WRD if they scored at or below the 35th percentile on the WRMT-III Basic Skills cluster (a combination of the WRMT-III Word Attack and Word Identification subtests [59]). Proficient readers were classified as those above the 35th percentile on the WRMT-III Basic Skills cluster. The 35th percentile was chosen as a cutoff because it is often used to classify middle school students as having reading difficulties (e.g., [60,61]) and encompasses students with low-average to below-average reading abilities.

All participants were enrolled in Grades 7 (66.25%) or 8 (33.75%) and were between the ages of 12 and 14 (M = 12.76). The sample comprised mostly Caucasian students (92.5%), but also included students who were Hispanic (2.5%), Asian (2.5%), and those who identified as multiple ethnicities (2.5%). There were slightly more female (52.5%) than male participants (48.75%). Almost one-third of participants (31.25%) had WRD, and approximately 69% were classified as proficient word readers.

2.2. Assessments

Standardized assessments were administered remotely over Zoom and audio recorded by the first author. Participants were required to use a computer or tablet for the assessments to ensure the screen size was sufficient to view the stimuli.

Two subtests of the WRMT-III (Form A) were administered to assess decoding: Word Identification and Word Attack. These untimed subtests comprise the Basic Skills cluster of the WRMT-III. In Word Identification, students read aloud 46 high-frequency words of increasing difficulty. Split-half reliability is 0.89 for Grades 7 and 8. In Word Attack, students read aloud 26 nonsense words of increasing difficulty, from simple consonant-vowel combinations to multisyllabic nonsense words. For both assessments, students’ accuracy in reading the stimuli was measured. Standard scores were calculated for the Basic Skills cluster. Split-half reliability for the Basic Skills cluster is 0.92 and 0.91 for Grade 7 and 8, respectively.

2.3. Experimental Measures

All experimental tasks were completed by students via an online research platform that was developed and delivered by Foundations in Learning, LCC. The research platform was derived from WordFlight, Version 2.111, a commercially available literacy solution from Foundations in Learning, and it incorporates the same engaging, student-friendly interface. A similar platform from FIL was used in the original studies on backward masking by Roembke et al. [55,56].

Participants completed four tasks, each in a masked and unmasked version: Find the Picture, Verify the Word, Count the Number of Syllables, and Say the Word. Find the Picture and Verify the Word were similar to tasks used by Roembke et al. [55,56]. Count the Number of Syllables and Say the Word were newly developed for this study. To facilitate readability, these tasks are referred to as the following: Find the Picture = Picture; Count the Number of Syllables = Syllables; Say the Word = Say; and Verify the Word = Verify. These four tasks were chosen to diversify the task response mode (oral vs. silent) and types of information (orthographic, phonological, and semantic) students may harness from morphological structure during reading. The experimental tasks are summarized in

Table 1. Experimental tasks.

Task	Response	Pathway(s) Targeted
Count the Number of Syllables	Written target word presented. Participant clicks on the correct number of syllables in the word.	O → P → S
Find the Picture	Written target word presented. Participant clicks on picture that best represents target.	O → S
Say the Word	Written target word presented. Participant says the word aloud and clicks on a button when finished.	O → P → S O → S
Verify the Word	Spoken and written target word presented. Participant clicks green or red button to indicate whether the stimuli matched.	O → P → S O → S

Note. O → P → S = orthography → phonology → semantic pathway; O → S = orthography → semantic pathway.

. Screenshots of the tasks are displayed in Figure 1.

In all tasks, the participant was presented with a written target word and provided a response. In unmasked versions of tasks, the target word remained on the screen, and participants had unlimited time to respond. In the masked versions of all four tasks, the target word was presented for 90 ms and then covered with a series of 11 hashtags, ensuring that the number of hashtags in the mask exceeded the number of letters in the longest items used in the study [55].

Items used in all tasks varied in morphological status such that some words were morphologically complex, and others were morphologically simple. Consequently, in tasks with masked and unmasked versions, there were two within-subject factors of interest: morphological status (complex vs. simple words) and masking (masked vs. unmasked tasks). Morphological status was determined using the WebCELEX database [62]. Simple words were defined as those that were monomorphemic and could not be morphologically decomposed (e.g., kitchen). Complex words were defined as bimorphemic words composed of one stem and one derivational suffix (e.g., king + -dom → kingdom). All items were real English words likely to be known by middle school students, with an age of acquisition of 11.0 years of age or less. Words contained 2 or 3 syllables and had a length of 5 to 10 letters. Items were balanced on four factors that impact visual word recognition: length, number of syllables, surface frequency, and orthographic neighborhood size.

Within each task, items in the complex and simple conditions, and in the masked and unmasked task versions, were equated on these factors. Items across Syllables, Picture, Say, and Verify were equated on the four factors, as well.

Table 2. Item Properties.

Property	Syllables		Picture		Say		Verify
	C	S	C	S	C	S	C	S
Length	7.40	7.38	7.25	7.25	7.35	7.35	7.3	7.28
Syllables	2.50	2.50	2.50	2.50	2.50	2.50	2.50	2.50
Log Frequency	7.59	7.79	7.59	7.88	7.92	7.83	7.66	7.42
Orthographic Neighborhood	0.40	0.38	0.45	0.43	0.43	0.43	0.43	0.43

Note. C = morphologically complex words; S = morphologically simple words.

contains the means for each of these factors within the two morphological status conditions and across the four tasks.

Participants completed 320 experimental trials during the session, divided evenly across the four tasks (i.e., completed 80 trials of each task) and the two masking conditions. For each task, 40 of the items were complex words, and 40 were simple words. Similarly, 40 of the trials were masked, and 40 were unmasked. A separate word list for each task was used (i.e., each list contained 80 unique words—see

Table A1. Words used in experimental tasks.

Syllables		Picture		Say		Verify
C	S	C	S	C	S	C	S
applause	altar	accordion	appetite	addition	agenda	actor	bargain
arrival	cantaloupe	angelic	calendar	agency	artifact	allergic	broccoli
botanist	captain	armor	canyon	brownie	babble	anxiety	burrow
burglarize	chimpanzee	artist	caravan	buzzer	biscuit	authorize	cadet
contagious	clever	athletic	carnival	careful	campaign	boredom	cannon
continent	college	baggage	casserole	childhood	cardigan	bravery	carousel
cottage	confetti	bracelet	collar	clueless	catapult	cheerful	cellar
disastrous	embarrass	cashier	corridor	dangerous	challenge	chilly	chipmunk
eagerness	formula	cavalry	crocodile	dentistry	cinnamon	classify	clarinet
eyeful	funeral	cavity	curtain	devilish	citizen	clerical	custom
greedy	gallon	circular	daisy	edible	concrete	closet	eclipse
harden	hurdle	cooler	dinosaur	entrance	connect	colorful	emerald
leakage	legend	cutlery	flamingo	erosion	elephant	contestant	faucet
lonely	lettuce	dinner	garden	favorite	garlic	emotion	gargoyle
merciful	minister	dynamite	guitar	heroic	habitat	eruption	horizon
nursery	molasses	explorer	helmet	infancy	indigo	explosion	insect
objection	moustache	ferocious	honey	metallic	magazine	famous	kangaroo
opposite	mussel	flatten	hospital	mighty	magnet	gorgeous	leprechaun
outer	obnoxious	freezer	hurricane	numeral	mannequin	justice	magenta
pacify	occupy	golden	iguana	parental	monkey	listener	mistletoe
package	orchestra	kingdom	kitchen	passable	nourish	moisture	ocean
parchment	pamphlet	lemonade	leopard	porous	papyrus	musical	ordinance
piglet	papaya	lioness	mandolin	pressure	peculiar	novelist	ornament
piracy	parliament	magician	mountain	reporter	pheasant	outage	pedestal
reflection	pigeon	motorist	neighbor	retirement	porcelain	passage	potato
rivalry	prohibit	mouthful	nozzle	rotation	progress	persuasion	privilege
runner	revenue	pavement	octopus	rotten	rally	poetic	protocol
servant	rhythm	physician	paradise	roughly	restaurant	precedent	proton
seventh	salon	pianist	pelican	scary	retina	raider	python
silence	scribble	prisoner	pilgrim	sparkle	ribbon	robber	rabies
snowy	serpent	referee	porridge	spiral	somersault	seasonal	reckon
stroller	signature	sculptor	pyramid	splatter	sparrow	shortage	sarcasm
tablet	silhouette	skater	science	suspicion	specimen	smelly	sergeant
terminal	stomach	student	spatula	tolerant	statistic	therapist	sinister
threaten	throttle	sweater	sprinkle	tourist	sugar	venomous	skeleton
traumatic	torment	tambourine	surgeon	urgent	tribute	vicious	squirrel
tropical	turtle	thirsty	terrain	usage	umbrella	weaken	thesaurus
typical	vehicle	victory	thermostat	wilderness	walrus	widen	ventricle
verbal	vinegar	windy	vitamin	wishful	whistle	wooden	village
visitor	volcano	wizard	weasel	wringer	wrangle	worthy	villain

Note. C = morphologically complex; S = morphologically simple.

). The order in which the words were presented was randomized within each 20-trial block. Each complex word was matched with a simple word on length, syllables, log HAL frequency (a measure of word frequency), and orthographic neighborhood size. The matched pairs were assigned to tasks together, allowing these factors to be roughly counterbalanced within tasks and across masked and unmasked versions of each task.

In Syllables (Figure 1, Panel A), a target word was presented in a written format. The participant was presented with four choices for the number of syllables that comprised the target word (i.e., 1, 2, 3, 4) and clicked on the correct number of syllables.

In the Picture task (Figure 1, Panel B), participants saw a written target word along with four pictures to represent that word. The participant clicked on the picture that best matched the target. Foils included words that were orthographically or phonologically similar to the target (e.g., cardinal as a foil for carnival). Semantic competitors (e.g., festival as a foil for carnival) were not used, in order to avoid confusion. The order in which the pictures were presented to participants was randomized. Each picture was used in just one trial; pictures were not reused across trials.

In Verify (Figure 1, Panel C), one spoken and one written target were presented simultaneously. The participant clicked on a green or red button to indicate whether the stimuli matched or did not match, respectively. In 50% of the trials, the auditory and visual stimuli matched, but in the other 50% of trials, the two types of stimuli did not match. Foils (i.e., the spoken targets presented when the stimuli did not match) consisted of words that were close phonological or orthographic competitors. Some had similar onsets (e.g., freezer and freedom), and some had similar offsets (e.g., freezer and blazer). As in Picture, words used as foils were only used once and were not presented as targets in later trials. Importantly, matched lists were used such that the match and mismatch conditions were counterbalanced across subjects: a word that appeared in the match condition for some participants was used in the mismatch condition for others.

In Say (Figure 1, Panel D), a target word was presented in a written format. The participant read the word aloud and clicked a button below the target word to indicate the word had been read. The participant’s oral readings of the target words were recorded and saved as audio files, which were later scored for accuracy.

For the unmasked version of Say, the first trial in the first block for each participant was removed from the study due to a technical error in the FIL platform (i.e., 80 unmasked trials were not included in the analyses). In addition, for 157 of the trials, there were problems with the recording due to user error, microphone issues, or Internet connectivity. In such cases, the first author used the audio recording of the session to find the participant’s reading of the target word. This rectified most of the issues (99 trials). However, there were 58 trials remaining for which a response could not be found or clearly heard—these trials were not included in the analyses. Consequently, 138 trials of the task (approximately 2.2% of trials) were removed in total.

2.4. Inter-Rater Reliability and Procedural Fidelity

Prior to the experimental task sessions, experimental task supervisors completed a 2-hr training session conducted by the first author. In this session, procedures were modeled, and supervisors practiced and received feedback on their implementation of the procedures. The first author attended the first experimental task session for each supervisor and evaluated their adherence to the experimental task session procedures with a procedural fidelity checklist. All supervisors achieved 100% fidelity to all procedures in the first session and thus supervised the remainder of the sessions independently.

Administration of the assessments and experimental tasks were audio recorded to allow for reliability scoring and procedural fidelity monitoring. The first author periodically listened to experimental task sessions for each of the administrators and provided them with feedback on their implementation of the procedures.

Inter-rater reliability was assessed for all measures requiring a rater judgment (WRMT-III Word Identification, WRMT-III Word Attack, and Say). Inter-rater reliability for these measures was assessed for 20% of participants (n = 16), randomly selected on a biweekly basis over the course of data collection. For all measures, each word or pseudoword was counted as a response, and point-by-point agreement was measured. Reliability was calculated by dividing the total number of agreements by the sum of agreements and disagreements. Overall, inter-rater reliability for the measures was high (range = 95% to 99.6%).

Procedural fidelity for the administration of the assessments and experimental tasks was measured for 20% of the participants (n = 16). Two participants were randomly selected from each biweekly period of data collection to ensure that procedural fidelity was measured over the course of the entire study. Procedural fidelity was measured with two types of checklists: one checklist for administration of the experimental tasks and one checklist for the administration of each assessment. Procedural fidelity for each assessment or experimental task session was calculated by dividing the total number of completed items by the number of items possible in the checklist. Overall, procedural fidelity was high for all assessments and experimental task sessions (range = 99% to 100%).

2.5. Analyses

We answered our research questions using a series of sequential data analytic steps. All data were analyzed using jamovi [63] and RStudio [64] including the afex and emmeans packages [65,66].

A series of mixed analyses of variance (ANOVAs) were conducted to examine the effects of two within-subject factors (masking [masked vs. unmasked] and morphological status [complex vs. simple]) and one between-subjects factor (reading ability group [students with WRD vs. proficient word readers]). RQ1 and RQ2 were both addressed by different sets of factors within the same model(s).

All analyses were conducted separately on the experimental tasks (Syllables, Picture, Say, and Verify). This was done for two reasons. First, this study was not concerned with directly comparing accuracy across tasks. Second, each task inherently required different numbers of response options and thus had different chance levels. Picture had 4 response options; Verify had 2 response options; Syllables had 3 response options; and Say had infinite response options (i.e., is a production task). Thus, each ANOVA should be seen as a replication/extension of results from the others.

After examining the results of each ANOVA, we performed Tukey’s follow-up significance test [67] only for significant interaction effects aligned to the study’s purpose and research questions. For instances in which both main and interaction effects were significant, we performed the follow-up test only for the interaction effect since main effects cannot be reliably interpreted in the presence of a significant interaction.

We also calculated partial eta-squared (η²_p) values [68] for all significant differences. Partial eta-squared quantifies the extent to which an independent variable (i.e., masked status, morphological status) affects the magnitude of the outcome variable (i.e., word reading accuracy). Partial eta-squared values range from 0 to 1, with larger values indicating greater magnitude. Interpretations of eta-squared may differ, but the following interpretation is common [68]: values equaling between 0.01 and 0.05 constitute a small effect, values between 0.06 and 0.13 constitute a medium effect, and values larger than 0.14 constitute a large effect.

Minimum detectable effect (MDE) analyses were conducted using G*Power (Version 3.1.9.6, [69]) to determine the minimum effect sizes that could be detected given the sample size that was obtained in the study (n = 80). For the ANOVAs, results of the MDE analyses indicated that a sample size of n = 80 would allow the detection of an effect of η²p = 0.091 (assuming power of 0.8 and α = 0.05), which would constitute a medium effect [70].

3. Results

Means and standard deviations of performance on the experimental tasks are displayed in

Table 3. Means and standard deviations of accuracy for the full sample, proficient word readers, and students with word reading difficulties.

	Masked		Unmasked
	Complex	Simple	Complex	Simple
		Full Sample
Say	0.87 (0.13)	0.78 (0.15)	0.93 (0.07)	0.93 (0.08)
Syllables	0.86 (0.12)	0.86 (0.12)	0.94 (0.09)	0.94 (0.09)
Picture	0.94 (0.06)	0.94 (0.06)	0.98 (0.03)	0.99 (0.02)
Verify	0.93 (0.07)	0.93 (0.08)	0.98 (0.06)	0.99 (0.07)
Proficient Word Readers
Say	0.91 (0.1)	0.84 (0.10)	0.95 (0.05)	0.96 (0.07)
Syllables	0.89 (0.09)	0.90 (0.11)	0.96 (0.08)	0.96 (0.09)
Picture	0.96 (0.04)	0.96 (0.04)	0.98 (0.03)	0.99 (0.03)
Verify	0.94 (0.07)	0.95 (0.08)	0.98 (0.06)	0.99 (0.09)
Students with WRD
Say	0.80 (0.16)	0.64 (0.13)	0.87 (0.08)	0.88 (0.09)
Syllables	0.79 (0.14)	0.91 (0.12)	0.92 (0.11)	0.90 (0.10)
Picture	0.89 (0.08)	0.89 (0.08)	0.97 (0.04)	0.99 (0.03)
Verify	0.92 (0.08)	0.91 (0.09)	0.99 (0.04)	0.99 (0.02)

Note. Standard deviations are the values presented in parentheses.

. They are provided for the full sample, proficient readers, and students with WRD.

The full results for the first stage of analysis can be found in

Table A2. Results of ANOVAs.

Task	SS	df	MS	F	p	η2p
SAY
Masking	1.007	1	1.007	119.345	<0.001	0.605
Morphological Status	0.179	1	0.179	57.676	<0.001	0.425
Difficulty	0.960	1	0.960	49.106	<0.001	0.386
Masking $\times$ Morphological Status	0.264	1	0.264	53.772	<0.001	0.408
Morphological Status $\times$ Difficulty	0.025	1	0.025	8.005	0.006	0.093
Masking $\times$ Morphological Status $\times$ Difficulty	0.048	1	0.048	9.826	0.002	0.112
SYLLABLES
Masking	0.572	1	0.572	66.167	<0.001	0.459
Morphological Status	0	1	0	0.004	0.953	0
Difficulty	0.386	1	0.386	14.5	<0.001	0.157
Masking $\times$ Morphological Status	0.003	1	0.003	1.165	0.284	0.015
Morphological Status $\times$ Difficulty	0	1	0	0.246	0.621	0.003
Masking $\times$ Morphological Status $\times$ Difficulty	0.002	1	0.002	0.64	0.426	0.008
PICTURE
Masking	0.222	1	0.222	148.727	<0.001	0.656
Morphological Status	0.006	1	0.006	5.541	0.021	0.066
Difficulty	0.127	1	0.127	34.4	<0.001	0.306
Masking $\times$ Morphological Status	0.005	1	0.005	4.015	0.049	0.049
Morphological Status $\times$ Difficulty	0	1	0	0.641	0.426	0.008
Masking $\times$ MorphologicalStatus $\times$ Difficulty	0	1	0	0.006	0.939	0
VERIFY
Masking	0.218	1	0.218	48.906	<0.001	0.385
Morphological Status	0	1	0	0.058	0.811	0.001
Difficulty	0.013	1	0.0123	0.994	0.322	0.013
Masking $\times$ Morphological Status	0	1	0	0.246	0.622	0.003
Morphological Status $\times$ Difficulty	0.003	1	0.003	1.612	0.208	0.020
Masking $\times$ Morphological Status $\times$ Difficulty	0.002	1	0.002	1.017	0.316	0.013

Note. SS = type III sums of squares; df = degrees of freedom; MS = mean squares; F = f-statistic value; bold font = p < 0.05.

(ANOVAs). The results of the ANOVAs for the four task types are described below.

Overall, Syllables showed only a main effect of masking and a main effect of difficulty. Similarly, Verify only showed a main effect of masking. These effects were expected and replicate prior work [55,56]. The lack of other effects suggests this task was not differentially sensitive to morphological status or to differences in children with WRD. Thus, our discussion focuses on the Say and Picture tasks where interactions were observed.

3.1. RQ1: The Effect of Morphological Status on Masked and Unmasked Performance

With respect to RQ1, the main effect of masking was significant across all four tasks. This indicated that participants exhibited lower performance on masked tasks than unmasked tasks, and replicated the effect for masking from Roembke et al. [55,56]. The main effect of morphological status was significant on two tasks: Picture: F(1, 78) = 5.541, p = 0.021, η²p = 0.066 and Say: F(1, 78) = 57.676, p < 0.001, η²p = 0.043. This was because accuracy was significantly higher for complex words than simple words in Say, while the opposite was true for Picture.

There was a significant interaction between masking and morphological status on two tasks: Say: F(1, 78 = 53.772, p < 0.001, η²p = 0.408) and Picture: F(1, 78 = 4.015, p = 0.049, η²p = 0.049). Thus, this interaction accounted for approximately 41% and 5% of the variance in accuracy scores on Say and Picture, respectively. These interactions are displayed in Figure 2.

As depicted in Figure 2, Panel A, the masking decrement was substantially lessened for complex words than for simple words in Say, resulting in a crossover interaction. Pairwise comparisons revealed an effect of morphological status that was significant for masked performance (t(78) = 9.336, p < 0.001), but not for unmasked performance (t(78) = −1.175, p = 0.664). This means that on masked trials, performance for complex words was significantly greater than for simple words on Say (mean difference = 0.113; SE = 0.012).

In contrast, for Picture, the effect of morphological status was driven by unmasked performance. As shown in Figure 2, Panel B, there was no difference in the masking decrement for complex and simple words (t(78) = −0.09, p = 1). Instead, unmasked performance for complex words was significantly lower than the performance for simple words (t(78) = −4.67, p < 0.001; mean difference = −0.017; SE = 0.004).

3.2. RQ2: Differences in Students With and Without WRD

With respect to RQ2, the main effect of having a word reading difficulty was significant for three tasks: Syllables, Picture, and Say. As expected, the performance of students with WRD was significantly lower than the performance of proficient readers on those three tasks. On Verify, the difference in performance between students with WRD and proficient word readers was not significant.

There was a significant interaction between having WRD and morphological status on Say: F(1, 78) = 8.005, p = 0.006, η²p = 0.093. This interaction is depicted in Figure 3. Across all trials of Say, both groups demonstrated better performance with complex compared to simple words. However, morphological complexity lessened the masking decrement more for students with WRD than proficient word readers, as illustrated by the difference in slopes in Figure 3. The estimated marginal mean difference was larger for students with WRD (0.070; t(78) = 6.286, p < 0.001) than for proficient readers (0.032; t(78) = 4.262, p < 0.001).

In addition, on Say, there was a three-way interaction between morphological status, masking, and reading ability group: F(1, 78) = 9.826, p = 0.002, η²p = 0.112, indicating that this interaction accounted for 11% of the variance in word reading accuracy. This is depicted in Figure 4. Pairwise comparisons revealed that this interaction was driven by differences in masked performance with complex and simple words. Students in both reading ability groups demonstrated higher masked performance with complex words compared to simple words (proficient readers: t(78) = 4.988, p < 0.001; students with WRD: t(78) = 7.897, p < 0.001). Importantly, the estimated marginal mean difference for complex words and simple words was greater in magnitude for students with WRD (0.158, SE = 0.02) than for proficient readers (.067, SE = 0.014). As Figure 4 suggests, as a whole this pattern is consistent with the idea that morphologically complex words tend to be less disrupted by masking than simple words, and that this effect is highly enhanced in students with WRD, who showed substantial difficulties with masked morphologically simple words.

In sum, across tasks there was a consistent decrement for masking but an inconsistent effect of morphological status. There was a significant interaction between masking and morphological status on two tasks, indicating that students were significantly more accurate in masked performance with complex words than masked performance with simple words on Say, and in unmasked performance with simple words than in unmasked performance with complex words on Picture. However, there was no significant effect of morphological status on Syllables or Verify. Students with WRD benefitted more from morphological structure on Say than proficient readers did, particularly on masked trials, but did not differ from proficient readers in the impact of morphological structure on accuracy in the other three tasks.

4. Discussion

Before discussing the results, it is important to note several limitations of the study. First, the sample was relatively homogeneous, comprising mostly Caucasian students, all of whom were monolingual. Because morphological ability and reading development may differ for students who are culturally and/or linguistically diverse (e.g., [71,72,73]), caution should be used in generalizing the results to the broader population of middle school students. Second, the sample size acquired for the study was insufficient to detect small effects. Third, due to technical errors, there was a small loss of data for one of the tasks (Say) that may have reduced reliability and impacted results for that task. Finally, the WRMT-III Word Identification and Word Attack assessments were administered remotely, which may have impacted student performance on this assessment. However, because this study concerned relative scores, rather than absolute scores, it is unlikely that deviations in performance by the sample as a whole would have impacted results.

Previous studies have demonstrated a facilitatory effect of morphological structure on word reading accuracy and speed [31,34,44]. One aim of the present study was to understand whether middle school students were more accurate at reading morphologically complex words than morphologically simple words and, if so, whether morphological structure fostered more automatic processing of multisyllabic words. Our results suggest it may matter what the child is trying to do with the words; that is, it is affected by tasks. Results from the Say task—the task most aligned with tasks used in prior work—supported this claim. Morphologically complex words were read aloud more accurately, particularly in the masked condition, where automaticity was important (see Table A2). However, results from the Picture task—which may capture the processes of mapping words onto meaning that are most engaged in silent reading—showed the opposite pattern. Here (in the unmasked condition), morphologically complex words impeded performance. Moreover, the remaining tasks showed no effect of morphological status.

These results make sense when we consider the varied ways that morphological structure may contribute to word reading. On Say, all students were more accurate at pronouncing complex words than simple words, particularly when the target words were masked. This may be because additional morphological structure allowed students to quickly activate the word in the orthography → phonology pathway via morpho-orthographic chunking. That is, students may have decomposed the word into large morpho-orthographic chunks and activated the relevant pronunciations for the chunks, thus allowing for more efficient processing afforded by orthographic chunks of morphologically simple words (i.e., letters, syllables). Alternatively, it is possible that mapping morpho-orthographic units onto morpho-semantic units led to stronger representations of the words, thus supporting the use of the more direct orthography → semantics pathway [14]. However, it may also indicate that morphological structure served a compensatory role for improving accuracy on the only production task in the study (i.e., one of the most difficult tasks [48,74]). Under this view, students in the present study may not have been automatic enough to apply their word reading knowledge to process the entire word before the mask. Consequently, they may have used familiar orthographic chunks to compensate for incomplete processing of the word and produce the best possible response.

This explanation is supported by the types of errors students made when reading words aloud. We did not perform a systematic analysis of students’ errors. However, informally, students’ errors often reflected words with similar orthographic chunks (and logography) as the target word. For example, students provided the word magnetize for the morphologically simple target word magazine, and destiny for the morphological word dentistry. These errors support the role of chunk learning in word reading—students used frequently co-occurring chunks of letters to process the word as completely as possible before the mask. For complex words, though, this also involved using morpho-orthographic structure to improve performance on the task. Students often committed “stem errors” [75]: they provided incorrect responses that contained the correct stem but an incorrect suffix (e.g., suspicious for the target word suspicion) or just the stem (e.g., rough for the target word roughly).

These errors are aligned with findings that students with lower word reading ability (and less reading experience) may not fully chunk morphologically complex words and over rely on stems [46]. After identifying known orthographic or morpho-orthographic chunks, students could have activated the corresponding phonological chunks, resulting in higher overall performance for masked performance with complex words than masked performance with simple words. This is because the complex words provided larger orthographic chunks of co-occurring letters and have lower transitional probabilities between morpho-orthographic units than those between other non-morpho-orthographic chunks [14,25,26], thus increasing the likelihood that students would process the word quickly and accurately enough to beat the mask. Although this strategy may have benefitted student performance in favor of morphemes, it still resulted in relatively low performance overall, particularly for students with WRD (see Table 3). This indicates that students with WRD may not have developed sufficient orthographic, semantic, and/or phonological representations to read these words automatically, and thus defaulted to relatively strategic applications of morphological information [14,26]. Future research should systematically investigate the types of error middle school students make in morphological processing tasks (both masked and unmasked), and whether these errors reflect the use of morpho-orthographic chunks.

In contrast, in the Picture task, students showed poorer performance with morphologically complex words than they did with simple words. This may be because the emphasis on meaning led children to decompose the words into their constituent morphemes. However, in many cases, these constituents were not needed and may even interfere. For example, in a word like dangerous, the suffix -ous indicates that the adjective is to be treated as a noun. However, in the picture display, there are likely to be three other nouns. The semantic features of danger alone would have been enough to select the correct picture. Notably, under conditions of masking—where there may not have been enough time to do this decomposition—the cost of morphological complexity disappears.

It is an open question whether this cascade of processes is unique to the task or maybe engaged in silent reading where the mapping between print and phonology is less important. Even absent the specific picture matching, when reading sentences, the context may provide information redundant to the morphemes (e.g., if a word appears in a subject or object position, a suffix like -ation may not be needed to indicate that it is a noun). Particularly, if sentence processing is “Good Enough” [76], full morphological decomposition may not be necessary (much like in our Picture task). Thus, the kind of redundancy afforded by the morpheme may appear in typical reading. However, the typical speed of reading may mean that the masked condition (where effects were not seen) could be more representative. Eye-movement studies of carefully controlled sentences with and without morphologically complex words could help disentangle this.

In the other two tasks (Syllables and Verify), there was not a clear benefit of morphological structure. The lack of effects for these tasks may have occurred for several reasons. First, overall accuracy was almost at ceiling, particularly for Verify. High accuracy was expected, given that the words used in the tasks were chosen to be easily recognized by seventh and eighth graders, and thus appropriate for divorcing knowledge from automaticity. Moreover, these tasks were judgment tasks (i.e., students had to select a response from an array of choices) which had much higher chance levels than Say, and even Picture where there were four choices. Thus, there may have been a greater influence of guessing on accuracy. High levels of accuracy on these tasks may have reduced the variability that was possible to attribute to other factors (e.g., morphological status). Future investigations with the novel backward masking paradigm should examine the impact of increasing task complexity (and thus decreasing accuracy) in multiple ways: changing the response type (i.e., using production rather than judgment tasks) and increasing item complexity (e.g., increasing the age of acquisition, decreasing surface frequency).

Second, the demands of the tasks may not have required students to fully take advantage of additional morphological structure to improve performance. In Verify, it may have been sufficient for students to compare just a few sounds in the written and auditory stimuli to produce a correct response, rather than activating the codes for both words in their entirety. Similarly, on Syllables, it is possible that morphological complexity might not matter for average accuracy in each morphological status condition because students still had to decompose even multisyllabic morphemes (e.g., dan-ger in dangerous) to produce the correct response. Moreover, students could have engaged in rough chunking of orthographic (and morpho-orthographic) units, rather than activating the related phonological and/or semantic information for the whole word. Thus, we may not expect to see large benefits of additional morphological structure in these three tasks due to ceiling effects and processing strategies that did not require full decomposition of the morphologically complex words.

More broadly, however, this complex breakdown of results suggests that it would be wise for reading research to not over-rely on any one task (e.g., reading words aloud) when trying to understand broader constructs like morphological processing. Skilled readers must engage a broader array of processes with written words: mapping them onto both sound and meaning, fitting them into sentence contexts, and identifying whether they are known or unknown words. As we observed here, constructs like morphological processing may play out differently as a function of the differences in natural (and experimental) task demands and goals.

The second, related, aim of the present study was to determine whether morphological processing similarly impacted students with WRD and proficient word readers. We hypothesized that students with WRD would benefit more from morphological structure than their proficient peers, as indicated by previous findings [49]. Again, this effect varied by task. In the Say task, the masking decrement was smaller for morphologically complex words than for morphologically simple words, indicating that students were able to automatically process morphological structure to support word reading. In addition, students with WRD took advantage of morphological structure more than the proficient word readers, particularly on masked trials. That is, morphological structure appeared to buffer the severe masking decrement students with WRD showed for unmasked words more than it did for their proficient peers.

Even in Picture, we saw that the masking decrement was reduced for morphologically complex words (Figure 2B) (even though they were more challenging on the whole). This supports the idea that morphological structure can help students achieve automaticity. Here, the fact that there was no interaction with WRD suggests that this route was equally available to children with WRD (though it was not differentially helpful, as we observed in Say).

Yet, there was no significant benefit of morphological structure for students with or without WRD on the other two tasks. The differences in results across the tasks may be related to the types of responses and related information emphasized by each.

5. Conclusions

Previous work has suggested that automatic morphological processing is a critical facet of skilled reading that develops throughout secondary school [30,31,32,33,34] and that students with WRD harness morphological structure during word reading to compensate for phonological deficits [49]. Using the novel backward masking measure implemented in previous studies on automaticity in middle school students [55,56], this study extended these findings to show that middle school students automatically process morphological structure when reading words aloud. In addition, middle school students with WRD relied more on morphological structure than proficient word readers did to automatically process words when reading aloud. However, this reliance on morphological structure actually hindered students when completing picture-based tasks on which full morphological decomposition was not necessary and did not impact accuracy on two tasks that emphasized orthography and phonology. Thus, there appear to be differences in the role of automatic morphological processing based on the presence of WRD and the nature of the word reading task.