Capital Gains: Effects of Word Class and Sentence Position on Capitalization Use Across Age

ABSTRACT

Learning to capitalize in English requires identifying a word's type and sentence position. In two cloze studies (2021–2022), Australian students of all genders (95% White, monolingual) spelled words with one and two capitalization cues (proper nouns, sentence‐initial words) and no‐cue control words. High school (12–18 years, n = 59) and university students (18–63 years, n = 78) exhibited near‐perfect capitalization. Primary school students (8–12 years) writing single words (n = 99) used proper‐noun cues more than sentence‐initial cues (ds > 0.49), but when writing consecutive words (n = 101), capitalized more accurately with two cues than one (ds > 0.32). Early capitalization appears better with more cues, but task format influences children's use of grammatical context.

In complex writing systems such as English, many spelling rules are probabilistic: their use is determined by contextual factors that spellers acquire over time (Marinelli et al. 2021). For example, the sound /ɒ/ is usually spelled “o” (e.g., “font”), but is spelled “a” when preceded by “w” (e.g., “want”). Capitalization is one of the few deterministic spelling patterns in English, where rules bear (virtually) no exceptions: every proper noun should be capitalized, as should the first word of every sentence. Deterministic patterns are useful tools for assessing factors that influence spelling development, as well as theoretical claims about how we learn to write. Once a speller learns a deterministic pattern, they should make virtually no errors in applying it, because its rules are consistent. However, spellers need to integrate these patterns with other elements of spelling (Pacton et al. 2005). The use or avoidance of capital letters, for example, can represent knowledge about broader aspects of spelling acquisition, such as word class and syntax. The current study examined how capitalization use in English differs across age groups. Specifically, it assessed whether spellers used one cue to capitalization more than another or, as predicted by Integration of Multiple Patterns theory (Treiman and Kessler 2014), whether capitalization increased with the number of cues.

1. Capital Letters and Capitalization

More than one‐third of the world's population, across all (inhabited) continents, use a writing system that includes both uppercase and lowercase letter forms, such as the Latin, Cyrillic, or Armenian script (Daniels and Share 2018; WorldStandards 2024). In these writing systems, capitalization is the conventional use of an uppercase form for the first letter of a word. In German, for example, all nouns and noun phrases are capitalized. These rules require the simultaneous processing of syntactic (grammatical and sentence structure) and morphological information (word and meaning structure) (Daniels and Share 2018) and have thus attracted considerable research. However, there has been little research on capitalization in other orthographies. In English, as in many other writing systems, only proper nouns and the first word in a sentence are typically capitalized. In Australia, where this research was conducted, children are taught to capitalize proper nouns in Grade 2, having learned in Preparatory Grade (pre‐Grade 1) that personal names and sentence‐initial words should be capitalized (Australian Curriculum, Assessment and Reporting Authority (ACARA), n.d.).

To spell correctly in English, writers must coordinate cognitive processes based on prior learning, including syntactic processes and orthographic processes (encoding visual features of spelling). Because capitalization is a grammatical spelling feature, its correct use requires some processing of syntactic information: for the capitalization of sentence‐initial words, the speller must consider sentence structure, and for proper noun capitalization, the speller must identify word class. The task of determining that a word belongs to the class of specific naming words (i.e., proper nouns) may appear more complex than merely identifying the first word in a sentence. However, since most proper nouns take a capital regardless of sentence position (i.e., they occur in memory in a single orthographic form, e.g., “Meg”), no further cues may need to be considered. In contrast, most other word types have two orthographic forms (e.g., “Mug”, “mug”). In considering sentence position in capitalization, spellers must also choose between these two possible stored representations. Thus, there are two rules that spellers must take into account when deciding whether to capitalize a word: the word‐level cue of proper nouns and the sentence‐level cue of sentence‐initial words. It is an open question whether one cue identity provides stronger evidence for capitalization than the other or whether the two in combination work better than one.

In English, we know little about how well children internalize capitalization rules, how these skills progress with age, or how accurately adults maintain their skills. Although most capitalization research in English has taken place in the US, Australian children tend to apply letter case knowledge in a similar way to US children (Treiman and Kessler 2003). Specifically, young children show some awareness of the visual structure of letter case (Treiman and Kessler 2022), and by Grade 1 are more likely to capitalize the first letter of a word than letters in other positions (Treiman and Kessler 2004). Given that all capitalization teaching in Australia takes place between the Preparatory grade and Grade 2 (ACARA, n.d.), then in general, any age‐based improvement from Grade 3 onward can be attributed to developmental or exposure‐based differences, rather than to direct teaching.

To our knowledge, there are only two experimental studies on capitalization in English in the upper primary school years. Odom (1962) showed that US children in Grades 4–6 (n = 1818) capitalized about 70% of proper nouns and 60% of sentence‐initial words in an unspecified standardized task. In a dictated cloze task, Evans (2015) similarly found that Australian Grade 3 and 4 students (8–10 years, n = 55) correctly capitalized approximately 70% of proper nouns. However, they also incorrectly capitalized about 20% of common nouns, which suggests that these students were still acquiring the rules of capitalization. It is also possible that the nature of the task may have encouraged the overuse of word‐initial capitals: some children may have ignored the printed sentence context and treated the target words as single entities, which they then capitalized.

Adults appear to have internalized proper‐noun capitalization rules, but may still overgeneralize the use of capital letters. In the study by Evans (2015), undergraduates used capitals correctly more often overall than the Grade 3 and 4 students did. The adults capitalized almost all proper nouns (99%), but also incorrectly capitalized 11% of common nouns. It is difficult to know whether this represents a genuine misunderstanding of how capitals should be used, or whether some adults, like some children, did not always attend to sentence context and capitalized some common nouns written to dictation. Evans did not compare adults' and children's correct omission of capitals from common nouns separately from their correct application to proper nouns. However, if adults do use capitalization rules more deterministically than children do, as Evans's results would suggest, then we might expect the overgeneralization of capitals to common nouns to reduce with age. In English, we can find no previous study of adults' sentence‐initial capitalization use, nor any experimental studies of capitalization use in adolescence, and so the progression of these skills through later school years is unclear.

Two general spelling error analyses with adolescents found that capitalization errors were more common than most other error types (comprising 5%–8% of errors), occurring in about 2% of words (Lunsford and Lunsford 2008; Wilcox et al. 2013). In a study of children's epistolic letters, Geoghegan and Fitzgerald (1935) found that the omission of capitals from sentence‐initial words comprised a greater proportion of errors overall (about 5%) than did the omission of capitals from proper nouns (about 3%). However, this may be because there are more sentence‐initial words than proper nouns in general prose, and so for this reason, the findings from naturalistic studies must be considered with caution (Wilcox et al. 2013). There is currently no or limited experimental evidence comparing the influence of proper‐noun and sentence‐initial word cues on capitalization decision‐making, either within or between education levels. Therefore, it is unclear whether one cue identity provides clearer evidence for capitalization than the other.

2. Integration of Multiple Patterns

Integration of Multiple Patterns theory (IMP) suggests that spelling is motivated by learned or observed patterns, which provide evidence for the use of one spelling over another (Treiman and Kessler 2014). IMP emphasizes the role of statistical learning in spelling by considering the interactions between patterns: the likelihood of a spelling choice (e.g., choosing to capitalize) increases with the number of cues supporting its use. As noted earlier, cues to capitalization are that a word is (a) a proper noun, or (b) the first word in a sentence. IMP would thus predict greater capitalization for words combining these cues (e.g., “M” in the sentence, “Monday is tomorrow”) than for words with only one cue (e.g., “T” or “M” in “Tomorrow is Monday”) and less frequent capitalization again for words with neither cue (e.g., “i” in “Tomorrow is Monday”). IMP is yet to be broadly tested, although the theory has been retrospectively applied to explain why words are more often spelled correctly when morphological patterns align with phonological (Treiman et al. 1994) or graphotactic patterns (Kemp and Bryant 2003, in English; Pacton et al. 2005, in French) rather than with only a single pattern. In line with the predictions of IMP, spelling scores have also been shown to increase with the number of linguistic cues presented in learners of Dutch (de Bree et al. 2016; Van Der Ven and de Bree 2019; Van de Ven et al. 2022). Spellers produce more varied responses when cues conflict, whether these cues are deterministic (de Bree and van den Boer 2021) or probabilistic (de Bree et al. 2016). The claims of IMP can thus be applied to examine whether the number of cues affects capitalization use, and whether any such effect is more pronounced in childhood—when spelling skills are still developing—than in adulthood.

IMP focuses on the ordinal relation between numbers of cues; two cues provide more evidence than one cue, and one cue more than none (Treiman and Kessler 2014). By comparing the proportion of words capitalized in each cue condition (henceforth referred to as capitalization rates), we can test these claims of IMP for capitalization and compare the effects of each cue individually. This provides an opportunity to extend IMP, which makes no claim about whether one type of cue may be stronger than another. Further, we can investigate the mechanisms underlying capitalization use by examining whether there are differences between age groups in their capitalization of cued or no‐cue words.

With capitalization rules possessing virtually no exceptions, capitalization rates of considerably less than 1.00 for cued words (or considerably more than 0.00 for no‐cue words) would suggest a misunderstanding of the rules, or the parameters within which they apply. Thus, it is also important to test the theory's claims in a way that differentiates spelling performance from a tendency to respond with a certain bias. Consider a task in which half of a list of words require capitalization. Person A capitalizes all words, Person B capitalizes none, and Person C capitalizes half the words, but only half correctly. All would score 50%, illustrating how simple accuracy measures can fail to capture important differences in performance. One way to address this is Signal Detection Theory, which provides measures of two performance parameters: discriminability and response bias (Macmillan and Creelman 2004).

In the present context, discriminability measures how well a speller distinguishes where to use capitalization (words with one or two cues) from where to omit it (no‐cue words). It is not always clear what misunderstanding is underpinning an overgeneralization error. For example, a no‐cue word could be capitalized because the speller does not attend to sentence‐level information, or because they misunderstand what constitutes a proper noun. Thus, discriminability provides an indication of whether differences in capitalization rates between cued word types remain meaningful when accounting for this uncertainty. Response bias supplements discriminability by comparing age groups in the extent to which they under‐ or overgeneralize capitalization overall. Thus, separating signal detection from capitalization rates can evaluate the practical implications of capitalization decision‐making.

Poorer or younger spellers may have more difficulty than better or older spellers in distinguishing where to use, and where to omit, a spelling feature like capitalization, because their reduced writing fluency costs them cognitive resources (Sweller 1994). Therefore, compared to better spellers, poorer spellers may require stronger evidence for capitalization before they use it. It is possible, then, that absolute spelling ability could mediate any age‐based differences in capitalization use. IMP would predict that developing spellers would discriminate better for words with two cues, where the evidence for capitalization is stronger, than for those with one. However, the theory makes no definitive claim about whether the advantage of multiple cues is maintained in adulthood (Treiman and Kessler 2014), especially as adults' performance may be at ceiling for patterns such as capitalization.

3. The Current Studies

The current studies (preregistered on the Open Science Framework, https://osf.io/vdz9r) compared the use of capitalization for proper nouns and sentence‐initial words in children, adolescents, and adults. Sentence‐initial proper nouns provided two spelling cues to capitalization; sentence‐medial proper nouns and sentence‐initial common nouns provided one cue each, and sentence‐medial common nouns provided no cues. We presented the words in a cloze (fill‐in‐the‐blanks) dictation task to provide sentence context but without the extra burden of writing all the words in each sentence. Below, we examine the use of capitalization in all four word types by analyzing capitalization rates, and capitalization performance through the signal detection analyses, in which cued word types are compared against the no‐cue word type. We also examined whether spelling ability mediated the effect of education level or grade on capitalization discriminability, discussed in Supporting Information (https://osf.io/j2mdw/). The studies were relatively confirmatory, with specific and directional hypotheses, although we had limited evidence on which to base these predictions. Comparisons in the use of capitals for no‐cue words between education levels or grades were more exploratory.

Our aim was to examine whether students used one capitalization cue more than the other, or whether they capitalized words more often when supported by two cues than by one, and less often again with no cues (in line with the predictions of IMP). We also investigated how capitalization decision‐making differs between education levels. We had three hypotheses:

1. Primary and high school students would show significantly greater rates of capitalization and greater discriminability for two‐cue words than one‐cue words, but university students would show no significant differences in capitalization rate nor discriminability between word types. Across cued word types, university students would show significantly greater rates of capitalization and discriminability than high school students, and high school students greater than primary school students.

2. All education levels would capitalize one‐cue words significantly more often than no‐cue words. Primary school students would use capitals significantly more often for no‐cue words than high school students would, and high school students would use them significantly more often than university students.

3. The tendency to capitalize (response bias) would be significantly lower for primary school than for high school students, and for high school than for university students.

4. Study 1

4.1. Method

4.1.1. Participants

Initial participants were 243 students in South‐East Australia across three education levels: primary school (Grades 3–6, from two schools and one community group), high school (Grades 7–12, from another two schools and one community group), and university (from one local university). Prior to analysis, we excluded participants who read and wrote in an alphabetic first language (L1) whose capitalization rules differ from those in English (n = 4), who missed 30% or more of the words in the experimental task (n = 1), or who wrote in all capitals (n = 2). Characteristics of the remaining 236 participants are summarized in Table 1.

TABLE 1.

Means (and standard deviations) and counts for participant characteristics by education level, Study 1.

Characteristic	Primary school	High school	University
n	99	59	78
Age (years)	10.5 (1.1)	15.1 (1.7)	25.8 (8.7)
Range	8.8–12.8	12.6–18.4	20.0–63.0
Gender (n)
Female	48	37	59
Male	38	18	19
Other	—	2	—
No answer	13	2	—
White/Monolingual (n)	96	57	73
WRATrs (max 57)	32.7 (5.9)	42.8 (6.0)	48.0 (3.8)
WRATss	101.4 (16.6)	108.7 (14.9)	111.5 (10.6)

Note: WRATrs, raw spelling score; WRATss, standardized score.

These characteristics are representative of the areas in which we tested, and of the broader South‐East Australian population. We had no reason to expect that participants' capitalization use would differ systematically from other populations. The general spelling ability of each sample fell within the normal range on the spelling subtest of the Wide‐Range Achievement Test (WRAT‐5, Wilkinson and Robertson 2017, population mean standardized score 100, SD = 15). Primary and high school students received no remuneration for participating. University students participated as part of a class activity, but anonymously agreed for their results to be included in a later publication.

4.1.2. Materials

4.1.2.1. Cloze Task

We created a cloze dictation task with two sets of 20 sentences. Each sentence contained two target words: one proper noun and one common noun, with one word positioned as the first word in the sentence (sentence‐initial) and the other as the third word or later (sentence‐medial). Hence, each target word belonged to one of four word types: sentence‐initial proper nouns (two cues to capitalization), sentence‐medial proper nouns (one cue), sentence‐initial common nouns (one cue) and sentence‐medial common nouns (no cues). Target words were matched between sets for sentence structure and within sets for proper or common noun category (e.g., a day of the week), as shown in Table 2.

TABLE 2.

Example matched sentences, showing cue combinations, sentence sets and structures.

Noun type	Sentence position	Number of cues	Example sentence (target word in bold)	Set	Structure
Proper	Initial	2	April is when leaves fall	A	1
	Medial	1	Rain is coldest in July	A	2
Common	Initial	1	Leaves fall in April	B	2
	Medial	0	July is when rain is coldest	B	1

We also included four filler sentences whose target words were not analyzed. Therefore, each set contained 24 sentences (with 48 words to spell), as shown in the Supporting Information. The order of the sentences within the two sets was then randomized to create two sentence orders for each set to reduce any fatigue‐ or order‐based effects. Thus, there were four versions of the task in all. Where possible, we avoided using target words starting with letters whose upper‐ and lowercase forms are the same shape (e.g., “V” and “v”). We also avoided capitonyms (capitalizable homographs, e.g., “March” and “march”) and only used brand names whose logos are written with an initial capital letter.

We chose words of relatively high frequency (3.5 or higher on a logarithmic Zipf scale for SUBTLEX‐UK adult and child corpora, van Heuven et al. 2014, or at least 1 occurrence per million words in the SUBTLEX‐US corpus, Brysbaert and New 2009). All but five target words met at least one of the word frequency thresholds. We retained these five because they referred to the names of local food outlets and of Australian states that were familiar to Australian children, even if they did not occur in the available UK or US corpora.

4.1.3. Procedure

Data for this and the second study were collected in 2021 and 2022. After we had obtained ethics approval, we contacted local school principals, and those who were interested gave their consent to participate. Participating schools were based in lower‐ to middle‐class areas. Sessions took place in a quiet room, in groups of four to 28 participants, with a supervising teacher present with primary student groups, in addition to the researcher. Participants began with the cloze task and were randomly allocated to one of the four sentence set‐order combinations. Here, participants were instructed to fill in the dictated missing words exactly as they thought they should be written. The experimenter dictated each sentence twice, emphasizing the target words. Finally, participants completed the WRAT‐5 spelling subtest. Sessions took from 20 to 60 min, depending on group size and age.

4.1.3.1. Coding

Target words were scored 1 when the first letter was capitalized, or 0 when written in lowercase. Scoring was based solely on this word‐initial letter; we did not take whole‐word spelling into account. Two members of the research team independently coded words where the case of an initial letter was ambiguous (about 4% of words for children, and none for adolescents and adults), resolving disagreements through discussion. Final scores were calculated as proportions, as some participants occasionally missed a word. Raw WRAT spelling ability scores were calculated out of 57, and were converted into standardized scores.

4.2. Results

Analyses were completed using lme4, car, and multcomp packages in R (R Core Team 2025), and Bayesian paired‐samples t‐tests in jamovi (The jamovi project 2024). At the time of preregistration, we planned to conduct the capitalization rate analyses binomially. However, we later found that binomial analysis would not easily adapt to missed trials. To address this issue, we compared capitalization rates in the same way as the signal detection analyses, using the hit and false‐alarm rates rather than binomial scores. Linear mixed models were run with word type (sentence‐initial and ‐medial proper nouns, sentence‐initial common nouns; sentence‐medial common nouns were included for capitalization rate analyses only) and education level (primary school, high school, university) as fixed effects and with random intercepts for participants. Because the spread of grades was not uniform across schools, we were unable to include school as a blocking variable without confounding with age. Planned comparisons were run for capitalization rate analyses of word type (both overall and at each education level) to examine the claims of Integration of Multiple Patterns (IMP, Treiman and Kessler 2014). For signal detection analyses, post hoc tests were estimated through Tukey's HSD method for comparing a family of nine (discriminability) or three estimates (response bias). Cohen's d calculations, alongside transformed and non‐transformed model results for both studies, are provided in Supporting Information.

4.2.1. Capitalization Rates

Table 3 shows the proportion of trials for which participants used capitalization correctly (hit rates, for the cued conditions) and incorrectly (false‐alarm rates, for the no‐cue condition), by education level and overall. Where capitals were required, capitalization rates were high for one‐ and two‐cue proper nouns, but somewhat lower for one‐cue sentence‐initial words. High school and university students capitalized cued words at near‐ceiling‐level accuracy, considerably more often than primary school students, who used capitals approximately two‐thirds of the time for proper nouns and for less than half of the one‐cue sentence‐initial words. Students at lower levels of education appeared to incorrectly capitalize no‐cue words more often than those at higher levels.

TABLE 3.

Capitalization rates and mean discriminability scores (and standard deviations) by word type and education level, Study 1.

Education level	n	Proper nouns		Common nouns
		Sentence‐initial	Sentence‐medial	Sentence‐initial	Sentence‐medial
Capitalization rate a
Primary school	99	0.67 (0.30)	0.63 (0.27)	0.45 (0.37)	0.12 (0.14)
High school	59	0.96 (0.13)	0.94 (0.12)	0.88 (0.26)	0.08 (0.17)
University	78	0.99 (0.05)	0.97 (0.08)	0.93 (0.19)	0.06 (0.17)
Overall	236	0.85 (0.26)	0.82 (0.25)	0.72 (0.37)	0.09 (0.16)
Discriminability score b
Primary school	99	1.76 (1.09)	1.63 (0.94)	1.08 (1.24)
High school	59	2.89 (0.70)	2.79 (0.73)	2.63 (1.02)
University	78	3.04 (0.61)	2.97 (0.65)	2.84 (0.89)

Note: Where participants missed a word (0.3% of trials), their hit rate or false‐alarm rate was calculated as a proportion of completed trials.

^a The proportions represent hit rates (correct use of capitals) for the first three columns but false‐alarm rates for the fourth column.

^b Discriminability scores compare the relevant hit rate to the false‐alarm rate. Higher discriminability scores (d′) indicate a better ability to discriminate between words that need capitalization and those that do not.

We analyzed the use of capitalization between all four word types with a linear mixed model. Main effects of word type (χ²(3) = 639.42, p < 0.001, ${\eta }_p^2$ = 0.82) and education level (χ²(2) = 113.67 p < 0.001, ${\eta }_p^2$ = 0.40) were subsumed by a significant interaction, χ²(6) = 283.84, p < 0.001, ${\eta }_p^2$ = 0.29. The close‐to‐normal distribution of the residuals suggested that there would be no benefit to transforming the scores.

Overall, there was no difference between one‐ and two‐cue proper nouns (p = 0.109, d = 0.12). However, both proper‐noun word types were capitalized significantly more often than one‐cue sentence‐initial words (ps < 0.001, ds > 0.31), and both one‐cue word types were capitalized more often overall than no‐cue words (ps < 0.001, ds > 2.20). Indeed, there was no difference between one‐ and two‐cue proper nouns at any education level (ps > 0.190, ds < 0.31). Primary school students capitalized both proper‐noun word types significantly more often than they did one‐cue sentence‐initial words (ps < 0.001, ds > 0.55). However, although high school and university students capitalized two‐cue words more often than they did one‐cue sentence‐initial words (ps < 0.006, ds > 0.38), they showed no significant difference between one‐cue word types (ps > 0.051, ds < 0.31). Given that IMP does not theorize about the relative strength of individual cues, we conducted Bayesian paired‐samples t‐tests to examine these null results between one‐cue word types more closely. These yielded anecdotal evidence of no difference between one‐cue word types for university students (BF₀₁ = 1.29) but not for high school students (BF₀₁ = 0.72). Hence, the data do not support clear‐cut conclusions about the equivalence of capitalization rates for single cues among these groups.

All education levels capitalized both one‐cue word types significantly more often than the no‐cue words (ps < 0.001, ds > 1.17). High school students capitalized each cued word type more often than primary school students did (ps < 0.001, ds > 1.24), although there were no differences between high school and university for these word types (ps > 0.322, ds < 0.31). The use of capitals for no‐cue words did not differ significantly between primary school and high school, nor between high school and university (ps > 0.103, ds < 0.27).

4.2.2. Calculating Discriminability and Response Bias

Discriminability is indexed by d' scores, with higher positive values indicating better ability to distinguish words requiring capitalization from words that do not, and scores close to zero indicating negligible ability. Response bias scores indicate the position of a “choose‐to‐capitalize” decision‐making threshold: the higher the score, the more conservative a participant is in using capitalization. Therefore, larger positive scores indicate a stronger tendency to underuse capitals, larger negative scores indicate a stronger tendency to overuse capitals, and scores close to 0 indicate unbiased responding.

Discriminability (d') scores were calculated from hit and false‐alarm rates (see Table 3): d' was obtained for each cued word type by subtracting the z‐score of the false‐alarm rate from the z‐score of the relevant hit rate. Similarly, response bias scores were calculated as the negative sum of these two z‐scores, halved (Stanislaw and Todorov 1999). We used a standard adjustment to avoid ceiling and floor values: at a participant level, hit rates or false‐alarm rates of 1.00 were adjusted to 1–1/(2 N), and those of 0.00 to 1/(2 N), where N represents the number of trials the participant completed for that word type, as these integers produce z‐scores of positive or negative ∞ (Macmillan and Creelman 2004).

4.2.3. Discriminability Analyses

As shown in Table 3, d' scores were greater for high school and university students than for primary school students. High school and university students performed close to ceiling for discriminability (the maximum possible score was 3.29), resulting in non‐normality of residual variance. This was addressed via a square‐root transformation on these data for all education levels. A linear mixed model of the transformed data showed significant main effects of word type (χ²(2) = 161.70, p < 0.001, ${\eta }_p^2$ = 0.15) and of education level (χ²(2) = 65.34, p < 0.001, ${\eta }_p^2$ = 0.35) on discriminability, subsumed by a significant two‐way interaction (χ²(4) = 59.80, p < 0.001, ${\eta }_p^2$ = 0.11). Post hoc tests showed that primary school students discriminated proper nouns in either sentence position significantly more often than they did one‐cue sentence‐initial words (ps < 0.001, ds > 0.62), with no difference between proper‐noun conditions (p = 1.000, d = 0.04). There were no significant differences between word types for high school, nor for university (all ps > 0.233, ds < 0.36). High school students were significantly better discriminators of each cued word type than primary school students were (all ps < 0.001, ds > 0.64), but did not differ significantly from university students for any cued word type (ps > 0.979, ds < 0.21).

4.2.4. Response Bias Analyses

Primary school students' positive response bias score (M = 0.46, SD = 0.58) indicated that they underused capitalization overall. In comparison, high school (M = 0.00, SD = 0.40) and university students (M = −0.03, SD = 0.33) scored close to zero, showing virtually no response bias. As with discriminability, a linear mixed model was run with square‐root transformations to c scores to reduce positive skew of residual variance. This model exhibited a significant main effect of education level on response bias, χ²(2) = 70.02, p < 0.001, ${\eta }_p^2$ = 0.23. Primary school students were significantly more conservative in using capitals than high school students were (p < 0.001, d = 1.00). The difference between high school and university students was not significant (p = 1.00, d = 0.00).

A degree of caution must be used in interpreting d' and c when performance is close to ceiling: correcting for hit rates of 1.00 and false‐alarm rates of 0.00 can produce an underestimation of discriminability, especially when responding is biased (Brown and White 2005). However, we suggest that the conclusions drawn from our analyses of d' and c values are appropriate, given that (a) where differences occurred between conditions or groups, these were clear‐cut, and (b) the patterns of capitalization rates (see Table 3) align with conclusions drawn from d' and c analyses (e.g., high accuracy rates for the older groups).

The analyses showing that spelling ability mediated the effect of education level on discriminability are provided in Supporting Information.

4.3. Discussion

This study aimed to determine whether students (a) use one capitalization cue more than the other, or (b) accumulate cues to make their spelling decisions, in line with the Integration of Multiple Patterns theory (Treiman and Kessler 2014). We also examined whether capitalization decision‐making differed between levels of education. As expected, primary school students were substantially more conservative overall, and thus also poorer discriminators across word types than high school students were. However, high school students capitalized cued words and discriminated cued from no‐cue words just as well as university students did, showing a lack of response bias. This sat in contrast to our hypotheses, as did the lack of differences across age in the use of capitals for no‐cue words. Thus, capitalization performance did increase as age increased but reached its asymptote at an earlier age than we had predicted: like adults, high school students mostly made correct capitalization decisions.

IMP (Treiman and Kessler 2014) predicts that people make spelling decisions more easily with two cues than with one, and with one than none. We expected primary school and high school students to show this pattern of results in their rate of capitalization use, and in their capitalization discriminability. By university, we predicted that participants would capitalize one‐cue and two‐cue words equally often. Although all education levels capitalized one‐cue words more often than no‐cue words, none wholly supported the claim that capitalization performance would be better for two‐cue words than for one‐cue words.

Unexpectedly, primary school students capitalized both (one‐cue and two‐cue) proper‐noun word types more consistently than one‐cue sentence‐initial words. This result makes two things clear: first, that there was an advantage to sentence‐initial proper nouns (two cues) over sentence‐initial common nouns (sentence‐position cue only); second, that there was a lack of advantage to sentence‐initial proper nouns over sentence‐medial proper nouns (proper‐noun cue only). In line with Odom (1962), it seems that proper‐noun status provided a much stronger cue for capitalization than did sentence position. This suggests that primary school students used word‐level information better than sentence‐level information. They did not completely ignore sentence‐level information, given that they capitalized common nouns considerably more often in sentence‐initial position than in sentence‐medial position. However, children capitalized these one‐cue sentence‐initial words only about half the time. Spellers are likely to see common nouns written in both all‐lowercase and capitalized form, but proper nouns are almost always capitalized. Thus, a statistical learning perspective would suggest that the proper‐noun cue is clearer than the sentence‐initial word cue because proper nouns have only one orthographic form in memory (Peressotti et al. 2003).

High school and university students' use of capitalization fell between the two predicted patterns of results. These students capitalized two‐cue words more often than one‐cue sentence‐initial words, but not more often than one‐cue proper nouns. Although the differences were not significant between the one‐cue word types, the limited evidence for no difference from the Bayesian analysis rendered the one‐cue comparison inconclusive. Indeed, when accounting for the rate of overcapitalization (discriminability), no differences between cued conditions were meaningful for either high school or university. Taken together, these results suggest that any benefit of accumulating cues over the sole proper noun cue is marginal. Further, the proper‐noun cue is more salient than the sentence‐initial cue, but this discrepancy may reduce with age.

Capital use for no‐cue words (which should not be capitalized) did not differ between primary and high school, nor between high school and university. This suggests that the older students' greater sensitivity to cues when they appear (cf. better avoidance of capitals where they do not belong) is the mechanism behind the apparent increase in discriminability with age. Evans (2015) also found that adults and children overcapitalized common nouns, although twice as often as those in the current study. Perhaps by writing a sentence‐initial and a sentence‐medial word per sentence (cf. just one sentence‐medial word, as in Evans's study), our participants were motivated to differentiate the two positions and consider sentence structure to a greater extent than Evans's participants. If so, primary school students' capitalization decision‐making might improve further with stimuli better resembling naturalistic sentences.

Indeed, the structure of our cloze sentences may have contributed to primary school students' lesser capitalization of sentence‐initial words than of proper nouns. The cloze format was chosen to draw attention to the target words' grammatical context. However, this format may not achieve its purpose if spellers only listen for a single missing word and ignore the sentence context (Evans 2015). Our single‐word cloze task may thus have encouraged developing spellers to consider word‐level information (here, proper‐noun identity, which is specific to the word) more than sentence‐level information (sentence‐initial‐word identity, which requires consideration of grammatical context). One way to address this potential problem is to have participants write consecutive words to encourage more attention to the sentence context, while minimizing the increased load on writing capacity. This format may more closely resemble naturalistic sentence‐writing and thus facilitate syntactic processing (leading to greater capitalization of both sentence‐initial words and proper nouns). However, if the increased number of words to spell increases cognitive load, then working memory constraints may reduce the capitalization of sentence‐medial proper nouns. High school students' strong performance suggests that there is considerable improvement through late primary school, in line with Odom's (1962) results. Thus, it is important to examine whether capitalization decision‐making may develop gradually between Grades 3 and 6, or whether there is a more marked change between adjacent grades.

Therefore, the second study aimed to determine whether primary school children would rely on one capitalization cue or benefit from accumulating cues when writing consecutive words. (We did not test this stimulus format with high school or university students, who performed near ceiling in Study 1.) It also examined whether capitalization decisions would differ between grades. If a significant effect of format were observed (between Studies 1 and 2), we planned to use only the consecutive‐word spelling data (Study 2, hereafter referred to as “two‐word”) to test the following hypotheses:

1. Students would use capitals more often, and better discriminate where capitals are required, for two‐cue words than for one‐cue words. Capitalization rates would be greater for one‐cue words than for no‐cue words. However, there would be no difference between one‐cue word types in discriminability nor in capitalization rate.

2. Capitalization discriminability scores would increase, and response bias scores would decrease, with grade. The use of capitalization for cued words would increase with grade, but there would be no difference between grades for no‐cue words.

3. Given the lack of previous research on primary school children's use of capitalization, we made no predictions about an interaction between word type and grade for the rate‐based analyses, nor for discriminability. If an interaction were to arise, we intended to undertake planned comparisons in the same way as for the main effects.

5. Study 2

5.1. Method

5.1.1. Participants

Initial participants were 105 students in Grades 3–6, recruited from three South‐Eastern Australian primary schools not involved in Study 1. We excluded one participant for apparently deliberately misspelling all words (a reason unanticipated at preregistration), and three participants who were literate in alphabetic L1s whose capitalization rules differ from those of English. The remaining 101 participants (53 female, 47 male, 1 no answer; 93 White and monolingual) had a mean age of 10.6 years (SD = 1.1, range = 8.9–12.9). The distribution of participants across grades, as well as their mean raw (WRAT‐5, M = 34.1, SD = 6.1) and standardized spelling ability scores (M = 105.7, SD = 14.9, within the normal range) was matched to those of the primary school students in the previous study.

5.1.2. Materials and Procedure

The participants completed the same target sentences as in Study 1. However, in this two‐word version of the cloze task, participants were asked to spell not only the two target words per sentence, but also the word immediately following the first target and the word immediately preceding the second. Thus, participants spelled four words per sentence, or sometimes three when the post‐ and pre‐target words coincided. To reduce our young participants' writing fatigue, we omitted Study 1's filler sentences. The procedure was otherwise identical to that of Study 1. Testing sessions ran for approximately 50 min, with between 18 and 39 participants per testing group.

5.2. Results

Linear mixed models were run with word type and grade (Grades 3–6) as fixed factors, and random intercepts for participants. School could not be included as a blocking variable because we could not assess all grades in every school. Capitalization rates (Table 4) were converted to d' and c scores as in Study 1. As in Study 1, we ran planned comparisons for the rate‐based analyses. For signal detection analyses, post hoc tests used Tukey's HSD method for comparing a family of three (word type) or four estimates (grade). Capitalization was generally greater in higher grades and was consistently used more often by primary school students in the two‐ than those in the one‐word format, across grades and word types.

TABLE 4.

Capitalization rates (and standard deviations) by format, word type, and grade, Studies 1 and 2.

Grade	Proper nouns		Common nouns
	Sentence‐initial	Sentence‐medial	Sentence‐initial	Sentence‐medial
One‐word (Study 1)
3 (n = 30)	0.56 (0.28)	0.54 (0.22)	0.30 (0.27)	0.11 (0.12)
4 (n = 21)	0.72 (0.28)	0.62 (0.27)	0.46 (0.38)	0.15 (0.18)
5 (n = 28)	0.70 (0.31)	0.65 (0.28)	0.49 (0.37)	0.10 (0.13)
6 (n = 20)	0.74 (0.30)	0.75 (0.28)	0.62 (0.41)	0.12 (0.16)
Overall	0.67 (0.30)	0.63 (0.27)	0.45 (0.37)	0.12 (0.14)
Two‐word (Study 2)
3 (n = 37)	0.80 (0.25)	0.71 (0.25)	0.68 (0.35)	0.09 (0.13)
4 (n = 25)	0.90 (0.22)	0.78 (0.25)	0.81 (0.32)	0.09 (0.14)
5 (n = 22)	0.94 (0.16)	0.87 (0.16)	0.88 (0.25)	0.10 (0.11)
6 (n = 17)	1.00 (0.00)	0.94 (0.11)	0.98 (0.05)	0.06 (0.15)
Overall	0.89 (0.21)	0.80 (0.23)	0.81 (0.30)	0.09 (0.13)

5.2.1. Comparison Between Studies 1 and 2

Square‐root‐transformed scores reduced skew for discriminability and response bias analyses, but not for capitalization rate analyses. Mann–Whitney U tests showed that students in the two‐word format capitalized cued words more often overall than did those in the one‐word format (see Table 4), U = 24,687, p < 0.001, δ = −0.45. However, there was negligible difference between formats in the capitalization of no‐cue words, U = 5711.5, p = 0.064, δ = 0.14. Linear mixed models showed that students in the two‐word format were significantly better discriminators (M = 2.40, SD = 0.94) than students in the one‐word format (M = 1.49, SD = 1.13), χ²(1) = 43.43, p < 0.001, ${\eta }_p^2$ = 0.18. Students in the two‐word format were also significantly less conservative in using capitals overall (M = 0.12, SD = 0.46) than those in the one‐word format (M = 0.46, SD = 0.58), χ²(1) = 23.49, p < 0.001, ${\eta }_p^2$ = 0.11. Guided by our Study 2 research questions, the remaining analyses used only the two‐word format data.

5.2.2. Capitalization Rates

In the capitalization rate analyses, main effects of word type (χ²(3) = 451.06, p < 0.001, ${\eta }_p^2$ = 0.85) and grade (χ²(3) = 12.26, p = 0.007, ${\eta }_p^2$ = 0.14) were subsumed by their interaction, χ²(9) = 30.67, p < 0.001, ${\eta }_p^2$ = 0.10. Overall, there was a clear effect of number of cues: two‐cue words were capitalized significantly more often than each one‐cue word type (ps < 0.026, ds > 0.30), and one‐cue word types significantly more often than no‐cue words (ps < 0.001, ds > 3.10). There was no difference between one‐cue word types, neither overall (p = 0.425, d = 0.04), or in any grade (ps > 0.125, ds < 0.48). As in Study 1, to better understand these null results, we conducted Bayesian paired‐samples t‐tests to assess the evidence in favor of no difference between means. These analyses yielded moderate evidence of no difference in capitalization rates between one‐cue word types for Grade 3 (BF₀₁ = 4.68) and Grade 5 students (BF₀₁ = 4.40), and anecdotal evidence of no difference for Grade 4 students (BF₀₁ = 2.65). For Grade 6 students, there was anecdotal evidence of a difference in capitalization rates (BF₀₁ = 0.59). Thus, overall, the data suggest that capitalization rates did not differ between one‐cue word types.

Grade 3 students capitalized two‐cue words more often than one‐cue sentence‐initial words, and Grade 4 and 6 students capitalized two‐cue words more often than one‐cue proper nouns (ps < 0.036, ds > 0.38). No other differences between one‐ and two‐cue word types were significant (ps > 0.054, ds < 0.58). All grades capitalized one‐cue word types significantly more often than no‐cue words (ps < 0.001, ds > 2.22). For cued word types, no comparisons between adjacent grades (3–4, 4–5, or 5–6) were significant (ps > 0.058, ds < 0.56). However, Grade 6 students capitalized all cued word types significantly more often than Grade 3 students did (ps < 0.001, ds > 1.12). There were no differences between adjacent grades, nor between Grades 3 and 6, for no‐cue words (ps > 0.464, ds < 0.31).

5.2.3. Discriminability

Inspection of d' values (presented in Table 5) suggested that, for each word type, discriminability scores increased as grade increased. Across grades, discriminability was consistently and substantially higher for two‐cue words than for either of the one‐cue word types, with smaller differences in discriminability between one‐cue word types for Grades 3–5. Inferential analyses (reported below) provided evidence supporting these patterns.

TABLE 5.

Mean discriminability scores (and standard deviations) by word type and grade, Study 2.

Grade	Proper nouns		Common nouns	Overall
	Sentence‐initial	Sentence‐medial	Sentence‐initial
3	2.27 (0.91)	1.99 (0.86)	1.91 (1.24)	2.06 (1.02)
4	2.61 (0.74)	2.18 (0.85)	2.33 (1.06)	2.37 (0.90)
5	2.71 (0.80)	2.47 (0.75)	2.55 (1.03)	2.58 (0.86)
6	3.07 (0.49)	2.82 (0.56)	3.00 (0.53)	2.96 (0.53)
Overall	2.58 (0.83)	2.28 (0.84)	2.33 (1.17)	—

Note: As in Study 1, where participants missed a word (< 0.001% of trials), a hit rate or false‐alarm rate was calculated out of the total number of trials completed for that word type.

The word type‐grade interaction was not significant (p = 0.103, ${\eta }_p^2$ = 0.05), so a model was run with these variables as independent predictors. Residual variance was negatively skewed for both models; square‐root transformations improved normality, and were applied for the analyses. The transformed model produced a significant main effect of word type, χ²(2) = 24.84, p < 0.001, ${\eta }_p^2$ = 0.11. Participants were significantly better at discriminating where to use capitals for two‐cue words than they were either of the one‐cue word types (ps < 0.005, ds > 0.32; see Table 5). The difference between one‐cue word types was not significant (p = 0.208, d = 0.13). There was also a significant main effect of grade, χ²(3) = 13.21, p = 0.004, ${\eta }_p^2$ = 0.12. Discriminability increased numerically with grade (see Table 5); post hoc tests showed that this increase was significant between Grade 3 and Grade 6 (p = 0.004, d = 1.00). No comparisons between adjacent grade levels were significant (ps > 0.378, ds < 0.51).

5.2.4. Response Bias

The model for response bias contained only grade as a predictor. Residual variance was positively skewed; a square‐root transformation improved normality and was applied for the analysis. The linear mixed model produced a significant main effect of grade, χ²(3) = 9.88, p = 0.020, ${\eta }_p^2$ = 0.09. Inspection of response bias scores suggested that Grade 3 (M = 0.28, SD = 0.50) and Grade 4 students (M = 0.12, SD = 0.52) underutilized capitalization overall, whereas there was little bias evident in scores for Grade 5 (M = <−0.01, SD = 0.31) and Grade 6 students (M = −0.06, SD = 0.29). A pairwise comparison of Grade 3 and 6 students was of medium effect, but was not statistically significant (p = 0.057, d = 0.74), nor were any comparisons between adjacent grades statistically significant (ps > 0.265, ds < 0.41).

The analyses demonstrating that spelling ability mediated the effect of grade on discriminability are shown in Supporting Information.

5.3. Discussion

The aim of the second study was to determine whether primary school students (a) benefit from accumulating capitalization cues or (b) rely on a single cue when they write consecutive words, and whether this pattern is consistent across grades. Participants using this two‐word format were less conservative in their use of capitals overall than were primary school participants in the one‐word format of Study 1. Although there was no difference between studies in the use of capitals for no‐cue words, participants in the two‐word format were also better at identifying words that needed capitals than were those in the one‐word format, with a considerable increase in capitalization rates of all cued word types. This suggests that the two‐word format encouraged more engagement with sentence‐level information than the one‐word format did, and that any additional cognitive load did not affect the capitalization of sentence‐medial proper nouns. Therefore, the rest of the analyses addressed only the data collected for Study 2: the two‐word data.

As hypothesized, there was an effect of the number of cues in the capitalization rate analysis when collapsing across grades: two cues were capitalized more than one, and one more than none. As with all education levels in Study 1, each individual grade capitalized one‐cue words more often than no‐cue words. In Grades 3–5, there was evidence for a lack of difference between the one‐cue word types, suggesting that there was no discrepancy here in the strength of individual cues. At this level, the capitalization rate comparison between two‐cue words and one‐cue sentence‐initial words was significant only for Grade 3, and that of two‐cue words and one‐cue proper nouns was significant only for Grades 4 and 6. However, when accounting for the rate of overgeneralization through discriminability, the small discrepancies between grades were diminished, leaving only a main effect aligning with the number of cues. These results may suggest that, when promoting the use of both sentence‐ and word‐level information, two‐cue words provided the most consistent evidence, but one‐cue words were not far behind. Indeed, the considerable increase to sentence‐initial capitalization in Study 2 supports the idea that Study 1's one‐word format did not encourage engagement with sentence‐level information. This provides some further specification to IMP theory: it is important to consider that spelling cues' relative strength may be vulnerable to differences in how spelling is tested, especially for grammatical patterns. Thus, overall, the effect of accumulating cues provides evidence for IMP (Treiman and Kessler 2014), while illustrating (similarly to Treiman et al. 2021, when comparing production and choice tasks) that methodological factors (here, the placement of target words within the task format) can affect how well a developing speller uses a spelling cue.

As with Study 1, there was no meaningful difference in students' overgeneralization of capitals to no‐cue words between grades, but the use of capitals for cued words did increase as grade increased. There were only limited differences between adjacent grades; it appears that the sensitivity to cues may improve very gradually throughout the later primary school years. Main effects of grade for discriminability and response bias also supported this conclusion: despite response bias‐based pairwise comparisons being non‐significant, capitalization use became less conservative as grade increased, which manifested in better discriminability of cued from no‐cue words in Grade 6.

6. General Discussion

The use of grammatical patterns relies on word‐level and sentence‐level syntactic information, but little research has considered whether it is the identity or number of cues that improves spelling decision‐making (de Bree et al. 2016). Here, we tested whether capitalization performance was better with one type of cue than another, or with accumulating cues, in line with the predictions of Integration of Multiple Patterns (IMP, Treiman and Kessler 2014). We also investigated how performance differs across age groups.

In Study 1, writing one word at a time, all education levels (primary school, high school, and university) capitalized one‐cue words (sentence‐initial proper nouns, sentence‐initial common nouns) substantially more often than no‐cue words (sentence‐medial common nouns), as hypothesized. However, in a result more complex than predicted, students generally capitalized two‐cue (sentence‐initial) and one‐cue proper nouns better than one‐cue sentence‐initial words. This effect was more pronounced for primary school students than for older students, whose discriminability was near ceiling across word types. As conservative capitalizers, primary school students were considerably poorer at discriminating cued from no‐cue words than high school students were, as predicted, but high schoolers did just as well as the university students. However, the rate of incorrect capitalization of no‐cue words was consistent across education levels.

In Study 2, primary school students wrote consecutive words (two‐word cloze format) and were no better at avoiding overgeneralization than those in Study 1 (one‐word format). However, those writing consecutive words were significantly less conservative and significantly better at discriminating overall than those who wrote one word at a time. In the two‐word format, primary school students capitalized two‐cue words better overall than one‐cue words, aligning with our predictions, and those of IMP. Although meaningful differences between one‐ and two‐cue word types varied between grades, the use of capitals for one‐cue proper nouns was generally no different from that of one‐cue sentence‐initial words. As predicted, discriminability increased as grade increased (between Grades 3 and 6) and as conservative bias reduced, but the use of capitals for no‐cue words did not vary with age.

6.1. Relative Cue Strength

Here, we make the case that although IMP (Treiman and Kessler 2014) provides a strong foundation for understanding spelling acquisition, the theory should be extended to recognize the relative strength of different cues (e.g., de Bree et al. 2016). Our studies provide some evidence that developing spellers do better with two capitalization cues than one. However, in the one‐word format, it was the identity of the proper‐noun cue that most influenced students' decision‐making. All education levels capitalized one‐cue proper nouns just as well as two‐cue words, but were poorer at capitalizing one‐cue sentence‐initial words, especially in primary school. Thus, Study 1's findings suggested that the proper‐noun cue to capitalization may be stronger than the sentence‐initial word cue.

A speller must use word‐level information to apply the rule that proper nouns require capitalization. They must first use semantic or syntactic knowledge to identify that the target is a noun. Then, they must establish whether it is a common or proper noun, using semantic knowledge (common nouns can be multiple entities; proper nouns are normally a single entity) or syntactic knowledge (e.g., a determiner such as “the” is more likely to precede common than proper nouns). To apply the rule that sentence‐initial words require capitalization, a speller must use sentence‐level syntactic knowledge both to recognize the relevance of sentence structure and to identify that the target word is sentence‐initial. Thus, the sentence‐initial rule is purely grammatical; semantic information cannot contribute to sentence‐initial capitalization. However, spellers could correctly capitalize proper nouns with limited use of semantic or syntactic processing: given that proper nouns are (almost) always capitalized, spellers may hold in memory only that capitalized (cf. all lowercase) form. In contrast, orthography is ambiguous for one‐cue sentence‐initial words because the uncapitalized form occurs most often for words other than proper nouns (see word frequency corpora, e.g., van Heuven et al. 2014). Thus, there is a potential orthographic advantage to the capitalization of proper nouns over sentence‐initial words (Peressotti et al. 2003).

Primary school students must have used enough sentence‐level information to capitalize common nouns considerably more often in sentence‐initial position than in sentence‐medial position. However, given that they capitalized proper nouns no better in sentence‐initial position (where there were two cues) than in sentence‐medial position, it seems that participants relied on word‐level processing more than sentence‐level processing in Study 1. Indeed, younger participants may not have spent additional cognitive resources on syntactic processing when half the words—the proper nouns—could be spelled using only shallow orthographic processing (e.g., Peressotti et al. 2003, Experiment 5). For the one‐cue word types, orthographic representation and sentence position provide conflicting information, which, according to IMP, should make the use of capitals more variable for these than for two‐cue words (Treiman and Kessler 2014). Here, if participants did indeed prioritize orthographic processing, this may have increased the likelihood of capitalization for one‐cue proper nouns, but not for one‐cue sentence‐initial words.

However, this discrepancy between cue strengths was not apparent when students wrote consecutive words. In the two‐word format, primary school students' capitalization of cued words was considerably greater than that of students in the one‐word format, especially for sentence‐initial words. Thus, it appears that there was no advantage to the proper‐noun cue in Study 2, not because proper‐noun capitalization declined, but because that of sentence‐initial words increased. We speculate that the two‐word format might have better resembled normal sentence writing than the one‐word format did, facilitating students' consideration of sentence structure. Given that syntactic processing is relevant to both sentence‐initial and proper‐noun capitalization, this may contribute to why primary school students' capitalization was better in the two‐word format than in the one‐word format.

Overall, students' capitalization in the two‐word format was influenced more strongly by the number of capitalization cues than by either cue identity. The benefit of accumulating cues aligns with the general claims of IMP (Treiman and Kessler 2014), that spellers do better when patterns converge on the same spelling choice (here, using a capital), and showed that developing spellers may not yet use capitalization rules consistently (Evans 2015). As a deterministic pattern, a single cue should be sufficient for applying capitalization; however, all grades made some errors with single‐cue words. The strength of this “number‐of‐cues” effect must be interpreted with caution. In the rate‐based analyses, the effect of word type was qualified by an interaction with grade, at which level the differences between one‐ and two‐cue word types were not consistent. This may suggest that some students still prioritized either word‐level processing (leading to greater capitalization of proper nouns) or sentence‐level processing (promoting capitalization of sentence‐initial words). The ability to consider sentence‐level information is increasingly important for communication (Hauerwas and Walker 2003). Thus, this study indicates that it is important for developing spellers to practice generalizing word‐level rules to sentence contexts.

Therefore, it appears that IMP provides a strong theoretical base, but the role of cues in spelling performance may not always be simply additive. Spelling performance may also vary with the strength of the cue (de Bree et al. 2016), especially if one cue can be identified through more types of cognitive processing than another cue. Indeed, the format of the task itself may even influence which types of processing spellers use most, which may in turn affect the extent to which participants notice and use cues.

6.2. Capitalization Development With Age

Taken together, our studies suggest that capitalization may gradually improve through primary school and into high school. By adolescence, students reach near‐ceiling performance, leaving little room to improve in adulthood. Indeed, high school and university students were largely not affected by the identity or number of cues: although they capitalized two‐cue words more often than one‐cue sentence‐initial words, this difference was no longer meaningful when statistically accounting for their rate of overgeneralization to no‐cue words (through discriminability). Thus, although skilled spellers may accumulate cues for patterns learned implicitly (Kemp and Bryant 2003), our studies suggest that spellers may reduce their dependence on cue accumulation for explicitly taught skills, like capitalization, as their fluency develops. Indeed, as discussed in the Supporting Information, our results suggest that absolute spelling ability mediates the effects of age on students' ability to discriminate cued from no‐cue words. However, since we did not pit this variable against any others, it is difficult to draw strong conclusions here; this could be followed up in future research.

Our results cannot be directly compared to those of naturalistic studies, which identify capitalization errors as some of the most common mistakes in adolescents' writing (e.g., Lunsford and Lunsford 2008). However, our studies align with Evans' (2015) findings that adults capitalize proper nouns at near‐ceiling levels in English. It also extends this finding to sentence‐initial words, for both adolescents and adults, for the first time.

In both current studies, younger spellers underused capitalization overall. In primary school, discriminability appears to increase gradually as grade increases, with students showing incrementally better identification of cued words between Grades 3 and 6. The conflicting information between orthographic representation and sentence position may have made the younger grades' capitalization of one‐cue words more variable. These results also concur with Evans (2015), whose Grade 5 and 6 students used capitals correctly more often than students in Grades 3 and 4. However, Evans did not separately analyze correct application and correct omission. Our studies extend Evans's work by differentiating capitalization rates from discriminability to examine the mechanisms underlying age‐related changes to capitalization.

6.3. Mechanisms of Capitalization Use

In the current studies, the apparent age‐based improvement to capitalization could be taken to arise from older students' greater ability to identify the cues when they appear (cf. identify when words do not need capitalizing). Words in prose that need capitalizing are statistically infrequent; most words are not proper nouns, and there is only one initial word per sentence. Therefore, it makes sense not to use up cognitive resources looking for patterns to avoid when they are encountered relatively infrequently. In terms of orthographic representation, the role of the first‐letter capital is specific to proper nouns (Peressotti et al. 2003); a speller cannot statistically learn to avoid capitalization for other words quite as easily.

Given the lack of difference in the use of capitalization for no‐cue words between studies, it seems unlikely that the overgeneralization of capitals was primarily due to insufficient sentence‐level processing. Indeed, with the rate also remaining reasonably steady across education levels, it suggests that some common nouns may possess word‐level qualities that increase their likelihood of being capitalized. For example, a speller's syntactic understanding of noun classification may be blurred if they are not sure if a noun is a general term for multiple possible individual referents (e.g., one of many games of “football”) or a specific term within a broader category (e.g., “Football” must need a capital because it is a specific type of sport). From an orthographic perspective, words that appear in proper noun titles (e.g., National Football League) may be seen disproportionately often in capitalized form. Therefore, although the proper noun rule has virtually no exceptions, people may not use it deterministically, because they find the boundaries difficult to understand.

6.4. Limitations and Future Directions

Our studies were subject to some limitations. The cross‐sectional nature of the research means that any interpretations of improvements across age groups must be considered with caution, and that longitudinal data would help to support the current conclusions. The dictation format meant we could control the target words that our participants wrote. However, as was clear from the contrasting results of Studies 1 and 2, children may not always prioritize sentence‐level information when making spelling decisions. Thus, although there is no reason for our participants' spellings to differ from those of other populations, caution should be exercised in assuming our results are fully representative of the use of capitalization in children's everyday writing. In naturalistic writing, children may capitalize better if they are mostly writing self‐chosen words, and may attend well to context because they are forming sentences themselves. Conversely, devising and writing their ideas may consume more cognitive resources than filling in cloze tasks to dictation, and may reduce their capacity to capitalize (Sweller 1994). Further work on both experimental and naturalistic tasks is required.

7. Conclusions

We integrate clues from writing inputs to make good spelling decisions. The current studies showed that even deterministic patterns are not always used deterministically, but may be influenced by statistical learning. Adolescents and adults are skilled capitalizers, but still overgeneralize capitalization. Primary school students tend to capitalize better with two cues than one, in line with IMP (Treiman and Kessler 2014). However, where the format of a task is less naturalistic (single‐word spelling), spellers' attention to sentence‐level cues may be reduced. Thus, our research suggests that, with developing spellers, it is important to promote both word‐ and sentence‐level processing for grammatical patterns. Targeted training could help younger spellers utilize syntactic information to inform their spelling.

Supporting information

Data S1: cdev70035‐sup‐0001‐Appendix.docx.

Data Availability Statement

The data and code necessary to reproduce the analyses presented here are publicly accessible, as are the materials necessary to attempt to replicate the findings. Analyses were also pre‐registered. Data, code, and materials for this research can be accessed through the Open Science Framework (OSF) project URL (https://osf.io/j2mdw/), and the preregistration through its own OSF URL (https://osf.io/vdz9r).