Predictors of dual-language literacy attainment in Irish-English bilinguals

ABSTRACT

The aim of this study is to examine which predictor variables are related to literacy attainment in Irish (Gaelic)-English bilinguals. The participants were in their second (n = 115) and third (n = 125) year of schooling in Ireland and were drawn from both native speaker and new speaker backgrounds. The constructs of phonemic awareness, verbal fluency (RAN) and verbal short-term memory were investigated as predictors of Irish word reading accuracy and spelling. The results indicate that the predictors of Irish and English literacy attainment differ, though phonemic awareness is the most effective predictor of literacy attainment in each language. Notably, the Irish predictor tasks were significant predictors of English literacy attainment, while the English predictor tasks were not significant predictors of Irish literacy attainment, supporting the unidirectional transfer of skills from Irish to English in this cohort. The implications of this study highlight the importance of developing bilingual screening and diagnostic assessments for literacy in the Irish-English context, something which is currently lacking.

KEYWORDS:

• Literacy predictors
• bilingual
• biliteracy
• Irish language
• phonemic awareness

Introduction

This study concerns the predictors of literacy attainment in Irish-English bilinguals. Ireland has two national languages, Irish – a minority language – and English. Approximately 7% of children in Ireland attend Irish-medium schools (Department of Education and Skills: DES 2022). For a minority of the students in Irish-medium schools, Irish is their first language. These students typically attend schools in Gaeltacht areas, where Irish is the language of the community. The majority of students attending Irish-medium schools, however, are native speakers of English or other languages. They typically live in English-speaking areas and learn Irish in immersion schools which are often situated in urban areas.

Despite the growing interest in Irish-medium education (DES 2022), at present, there are no standardised assessments for dyslexia for Irish-English bilinguals, and students in Irish-medium education are typically assessed in English only (Nic Aindriú, Ó Duibhir, and Travers 2021). This is a possible contributing factor to the lack of early identification of dyslexia in Irish-medium schools (Nic Aindriú, Ó Duibhir, and Travers 2020). The present study aims to contribute to a research base on which a bilingual assessment for dyslexia could be designed.

Irish linguistic structures

Irish is a Celtic language which has three main dialects and no spoken Standard, though there is a written Standard. Irish has a more transparent orthography than English, though it is high in complexity (Stenson and Hickey 2016). There are approximately 49 phonemes in Irish and 177 graphemes which represent them (Ó Raghallaigh 2014). 71% of the most Irish frequent words are consistent (Stenson and Hickey, 2016), compared to 52% of the most frequent English words (Stuart et al. 2003).

Irish has a series of velarised-palatalised consonant contrasts which distinguish between minimal pairs (e.g. /lˠo:nˠ/ <lón > meaning ‘lunch’ and /lʲo:nˠ/ <leon > meaning ‘lion’) as well as between the singular and plural forms of a noun (e.g. /bˠa:dˠ/ <bád > meaning ‘boat’ and /bˠa:dʲ/ <báid>, meaning ‘boats’). Velarised phonemes are neighboured by back vowel graphemes <a>, <o > and <u>, while palatalised phonemes are neighboured by front vowel graphemes <e > and <i>. Vowel graphemes have two different functions in Irish, illustrated in a word such as <buíon > /bˠi:nˠ/, in which the <u > and <o > only mark consonant quality while /i:/ is articulated as the syllable nucleus (Ní Chasaide 1999).

Irish is a morphologically-rich language, and has two types of initial mutations. Lenition involves the replacement of stop consonants with homorganic fricatives (e.g. /cˠu:/ <cú > meaning ‘dog’ changes to /chú/ <xu:>). Eclipsis on the other hand, involves replacing voiceless stops with their voiced counterparts (e.g. /cˠu:/ changes to /gˠu:/) and replacing voiced stops with homorganic nasals (e.g. /bˠa:dˠ/ changes to /mˠa:dˠ/). The word’s base form is preserved in the Irish orthography (i.e.<cú > changes to <gcú>, <bád > to <mbád>), contrasting with Welsh in which the initial consonant is omitted.

There are a number of frequently-used homographic words common to both Irish and English which are pronounced differently in both languages (e.g.<teach > /tʲax/ ‘house’). While this has the potential to lead to confusion in the early acquisition of literacy, it can also be used as a method to discuss the different spelling systems (e.g. Stenson and Hickey 2018). Despite the regularity of the Irish orthography, students in Irish-medium schools have higher English literacy attainment overall (Parsons and Lyddy 2016). This may be due to the dominance of English, but may also be attributable to literacy instruction methods. Traditionally, a whole-word method was used in Irish (Stenson and Hickey 2014), though there is more emphasis on phonics-based instructional methods in recent years with the introduction of additional resources for teachers (e.g. Stenson and Hickey 2018).

Predictors of literacy attainment across languages

This is an exploratory study and examines phonemic awareness (PA), rapid automatised naming (RAN) and verbal-short term memory (VSTM) as predictors of literacy. These constructs were chosen to allow for comparisons with previous crosslinguistic studies (e.g. Caravolas et al. 2012; Georgiou, Parrila, and Papadopoulos 2008; Mann and Wimmer 2002; Vaessen et al. 2010; Ziegler et al. 2010).

PA can be defined as an awareness of the phonological structure of speech (Liberman 1991), though there is evidence that orthographic knowledge influences PA (Castles et al. 2003; Castles, Wilson, and Coltheart 2011). Two PA tasks were chosen for this study following Saeigh-Haddad (2007; 2019); the phoneme matching task is intended to measure the epilinguistic component of PA, which involves phonological judgements on the similarity of phonemes, while the phoneme deletion task is intended to measure the metalinguistic component of PA which involves implementing an operation on a phonological unit.

Conceptually speaking, PA is considered to have an important role in learning to read in alphabetic languages, where language is represented at the phoneme level. In order to become literate, children must segment the speech stream into words, and words into phonemes. A The relationship between PA and literacy attainment is most likely reciprocal (e.g. Peterson et al. 2018): explicit awareness of the sound structure of the language facilitates initial literacy development, and later experience with literacy facilitates further PA development.

RAN tasks examine the speed with which an individual can name a series of objects. Bilinguals are faster at completing RAN tasks in their dominant language than in their non-dominant language, which is most likely a frequency effect (Gollan et al. 2008; Pae, Sevcik, and Morris 2010). RAN has been conceptualised as a microcosm of the reading process as it involves the same mechanism involved in skilled reading; processing a visual stimulus and linking it to a verbal stimulus (Norton and Wolf 2012). Performance on RAN tasks is considered to involve three subcomponents: the quality of phonological representations, visual representations, and of connection between orthographic and visual representations (Manis, Seidenberg, and Doi 1999).

VSTM tasks typically involve repeating a series of lexical items after the assessor. VSTM relies on the phonological loop component of short-term memory (Baddeley and Hitch 1974), which allows for temporary store of phonemes for their integration into phoneme-grapheme correspondences in long-term memory (Gathercole, Willis, and Baddeley 1991). The episodic buffer has been proposed as the construct which allows for the binding of multimodal information (Baddeley and Hitch 2019). Previous research has established no significant difference between scores on Irish and English versions of a VSTM task of children in Irish immersion schools (McVeigh, Wylie, and Mulhern 2019). This study used an object span task, which is also used in the present study as it allows for control of the articulatory length of words (e.g. single syllable words can be used in each language), which is not the case with other VSTM tasks (e.g. digit recall tasks).

The majority of the literature in relation to predictors of literacy attainment indicates that the efficacy of predictors of literacy attainment differs across languages (Furnes and Samuelsson 2011; Georgiou, Parrila, and Papadopoulos 2008; Landerl et al. 2013; López-Escribano, Ivanova, and Shtereva 2018; Mann and Wimmer 2002; Vaessen et al. 2010; Ziegler et al. 2010; though see Caravolas et al. 2012; Caravolas, Volín, and Hulme 2005). Cross-linguistic differences appear to be primarily quantitative in nature. PA appears to be a more effective predictor of reading attainment in more opaque orthographies for a longer period of time (Landerl et al. 2018; Mann and Wimmer 2002; Vaessen et al. 2010; Ziegler et al. 2010) while the efficacy of RAN tasks as a predictor increases with reading expertise (Vaessen et al. 2010),

This study examines the contribution of PA, RAN and VSTM when the variables of home language and gender are accounted for. Previous research on Irish reading comprehension found that higher exposure to Irish at home is associated with higher reading scores (Harris et al. 2006) and that girls had higher Irish reading comprehension scores than boys (Harris et al. 2006).

Cross-language transfer of reading subskills to reading skills

Cummins’s Underlying Proficiency (1981) model states that given adequate exposure to each language, there is a shared proficiency between each language which means that learning experience in L1 can result in the development of L2 proficiency and vice versa, though the extent of cross-language transfer depends on the degree to which they are typologically related (Cummins 2005). The level of transfer is also skill-dependent, with more transfer occurring in PA and reading than in oral language (Melby-Lervåg and Lervåg 2011).

The present study is concerned with cross-linguistic transfer from a reading subskill to a reading skill as opposed to cross-linguistic transfer within the same construct. A number of previous studies examine the role of L1 predictors in L2 reading. Jared et al. (2011) found that English subskills predicted later French reading attainment. Similarly, Lindsey, Manis, and Bailey (2003) found that Spanish subskills predicted later English word reading. Similar cross-linguistic findings are established in studies which examine predictor variables in both L1 and L2; both French and English phonological awareness predict both French and English reading (Comeau et al. 1999; Lafrance and Gottardo 2005).

A contrasting finding was obtained in a study of Spanish-English bilinguals (Swanson et al. 2008) that found that when both within and cross-language predictors were entered into a single regression model, the only measures to account for unique variance were within-language predictors. A study by Pasquarella et al. (2015) sheds light on why contrasting findings may occur. They found unidirectional transfer from English to Spanish word reading accuracy in children who initially learned to read in English, and suggest that more advanced English literacy skills facilitated this. Interestingly, the same was not true for the Chinese-English bilinguals, possibly reflecting the greater typological differences between English and Chinese compared to English and Spanish.

The present study

The present study examines the following research questions:

1. To what extent does phonemic awareness predict reading and spelling in Irish-English bilinguals when demographic variables are accounted for?

2. To what extent does RAN predict reading and spelling in Irish-English bilinguals when demographic variables are accounted for?

3. To what extent does verbal short-term memory predict reading and spelling in Irish-English bilingual when demographic variables are accounted for?

4. Do cross-language predictors account for any additional variance over and above that accounted for by within-language predictors and demographic variables?

Materials and methods

This study was approved by the Ethics Committee of Trinity College Dublin under the application identifier HT42.

Sampling

This study employed non-probability sampling. A total of fourteen schools were selected based on their location – within one large urban area or one Gaeltacht region – and their student enrolment exceeding 100 pupils. Each school received an invitation, and the first four immersion schools to respond and two Gaeltacht schools that responded were included. All schools were mixed sex, and introduced Irish literacy instruction in Grade 1 (referred to as Junior Infants in Ireland) before introducing English literacy instruction at the end of Grade 2 (referred to as Senior Infants) or the commencement of Grade 3 (referred to as First Class). Moreover, the socioeconomic status (SES) of the six schools varied, with two schools categorised as serving affluent catchment areas, one as serving a disadvantaged catchment area, two exhibiting mixed SES profiles, and one serving a catchment area which is marginally above the average SES.

Participants

The study included a total of 240 participants (55% girls), comprising 115 Grade 2 pupils and 125 Grade 3 pupils. The Grade 2 cohort encompassed 94 students from Irish immersion schools and 21 students from Gaeltacht schools, while the Grade 3 cohort comprised 105 students in Irish immersion schools and 20 students from Gaeltacht schools. Irish immersion schools displayed a higher participation rate (73%) compared to Gaeltacht schools (42%). The mean age of Grade 2 students was 6 years and 3 months, while Grade 3 students had a mean age of 7 years and 2 months for immersion schools and 7 years and 3 months for Gaeltacht schools.

Parents completed surveys pertaining to their child's linguistic and educational backgrounds. No participant reported any diagnosis which would affect literacy (though note dyslexia diagnoses in this context typically occur after Grade 3: Nic Aindriú, Ó Duibhir, and Travers 2020). All study participants attended some form of pre-school education which is typical in Ireland. The majority of students in Irish immersion schools had parents who identified as native speakers of English, though 10% had a parent who was a native Irish speaker. In Gaeltacht schools, 92% of participating students had at least one native Irish-speaking parent. Correspondingly, parents in Irish immersion schools predominantly reported that they always or frequently communicated in English with their child, while parents in Gaeltacht schools commonly reported that they always or frequently conversed in Irish.

Procedure

The data collection was carried out in schools with individual children by an Irish-speaking researcher between January and April of the school year. Recorded stimuli from a young female native Irish speaker were used in the Phoneme Matching, Phoneme Deletion, Object Span and Spelling tasks to ensure consistency across participants. Note that the data reported is part of a larger study which included Grade 1 pupils. Written consent was obtained from all parents/guardians and oral assent was obtained from all of the participants. This data was collected at a single timepoint, meaning that this study examines concurrent predictors, rather than longitudinal predictors.

Each task was framed as a game and was implemented on a Samsung Galaxy tablet. The participants were given instructions (in the language of the task) and each task included a number of unscored practice items. Each of the language versions were carried out with the participant during the same school day at least 2 h apart. Approximately half of the participants undertook the English language version of the task battery first, while half undertook the Irish version first.

Research instruments

The research instruments used were designed for this study, with the aim of developing equivalent measures in Irish and English, guided by the International Test Commission (Muñiz, Elosua, and Hambleton 2013). Tasks were pre-tested on two occasions and subsequently refined based on the findings.

Predictor tasks

Phoneme matching task

The participant was asked to indicate whether a pair of words start with the same phoneme (matching), or with different phonemes (mismatching). The task contained a total of 24 stimuli pairs in each language; 16 were consonant-initial pairs and 8 were vowel-initial pairs. The Irish and English versions were matched for the number of syllables in each pair, and words were either monosyllabic or disyllabic. The vast majority of items are concrete nouns which are found in school books commonly used by Grade 2 (Senior Infant) and 3 (First Class) pupils in Irish-medium schools. Guttman’s λ 4 split-half reliability coefficient was .74 for Irish Phoneme Matching task and .74 for English.

Phoneme deletion task

Participants were asked to delete a phoneme from initial or final position in a word. There are three conditions in this task, and a total of 12 stimuli. In the first condition, a single initial phoneme is deleted; in the second condition, an initial phoneme is deleted from a consonant cluster; in the final condition, a phoneme is deleted from final position. The majority of the Irish and English stimuli are matched for the number of letters and phonemes, though a small number are only matched for the number of letters. The number of stimuli in which the target word is a real word – as opposed to a nonword – is matched in the Irish and English version. Familiar words were chosen and each version contains 11 concrete nouns and 1 pronoun. Guttman’s λ 4 split-half reliability coefficient was .82 for the Irish version of the task and .84 for the English version of the task.

Rapid automatised naming task

A RAN Objects task was chosen to measure verbal fluency in order to maximise the level of equivalence between the Irish and English versions of the task. The differing articulatory length of digits in Irish and English precluded the use of RAN Digits. RAN letters was also unsuitable as typically the same letter names are used in Irish and English.

Participants were required to name a series of objects on an A4 page as quickly as possible. The same images (bád, clog, bó, ubh, bróg in Irish, boat, clock, cow, egg, shoe in English) are included on Irish and English versions. The five objects were chosen for their similar articulatory length in Irish and English, and because they are all possible to depict in an unambiguous way in an image.

Forward span task (VSTM)

A Forward Span task was chosen to measure VSTM. As with RAN tasks, various versions of span tasks exist. An object span task was chosen to allow for equivalence in the articulatory length of items in Irish and English. The participant was asked to listen and recall a list of common objects. The task begins with 2 words and increases by a single word each time up to a maximum of 8 words. In an effort to match the articulatory length of words in Irish and English, each language version is matched for the number of syllables and phonemes, the syllable type (open/closed) and the presence of consonant clusters. All of the stimuli are concrete nouns which are found in age-level schoolbooks.

Outcome measures

Single word Reading

Participants were asked to read a list of single words. The word reading task for Grade 2 pupils contained 20 items and was untimed. It was only implemented in Irish as they had not yet formally began English reading instruction, The word reading task for Grade 3 pupils contained 40 items and was timed; participants had two minutes to read as many words as possible. The Grade 3 Irish and English version were matched for number of letters and number of syllables. Each language version of the task contained six 3-letter words, eight 4-letter words, thirteen 5-letter words, six 6-letter words, three 7-letter words three, 8-letter words and one 9-letter word. In both languages, the stimuli selected reflect a range of syllable structures and grapheme-phoneme rules. The words were ordered by orthographic complexity (number of letters and complexity of grapheme-phoneme rules).

Single word spelling

This task was administered to Grade 3 pupils only. Participants were asked to spell eight words in each language in untimed conditions. The Irish and English version of this task were matched for the number of letters, syllables and consonant clusters in each item. All but one item is matched for the number of phonemes. The stimuli were selected to reflect a range of age-level phoneme-grapheme rules (Cronbach’s α = .72 for Irish version; .70 for English version).

Data analysis

Five separate analyses were conducted in IBM Statistics 27. One outlier was removed from the Grade 2 sample and one from the Grade 3 sample, based on an analysis of Cook’s D using a cut-off of 0.1. Data for Gaeltacht and immersion school pupils are included in the same analysis as the sample size of the Gaeltacht group does not allow for a robust analysis. In a pre-analysis, entering the variables of ‘school’, ‘school type’ and ‘SES’ introduced collinearity into the model. It was decided to enter the ‘school’ variable only as it captured more variance than ‘school type’ and ‘SES’ together.

The frequency of parental communication in Irish or English (matching the language of the dependent variable in the analysis), was entered in the first block. This was a self-reported measure between 1–5 (1 = always speaking Irish/English; 5 = never speaking Irish/English). Gender (boys/girls) was also entered in first block with boy as the reference category. The second block is the school variable (as a dummy variable). Beta coefficients for individual schools are not reported as we are not interested in comparing individual schools and in any case, the coefficients of dummy variables cannot be interpreted in the same way as those of continuous variables. The third block includes the predictor variables in the language of the dependent variable. The fourth block includes predictor variables in the second language. Note that Phoneme Deletion is only included in the language of the dependent variable; this is due to the fact that Phoneme Deletion scores in Irish and English are highly correlated and including it in both languages introduced multicollinearity into the model.

Results

Descriptive Statistics are provided in Table 1, while Pearson correlations between scores on the predictor and outcome tasks are provided in Tables 2 and 3. The results of the regression analyses are reported in Tables 4–6.

Table 1. Descriptive statistics.

	Language	Grade	N	Minimum	Maximum	Mean	Std. Deviation
Phoneme Matching	English	3	124	10	24	19.60	2.515
		2	114	11	24	18.95	2.732
Phoneme Deletion	English	3	124	1	13	9.79	3.124
		2	114	0.0	13.0	7.772	4.0963
RAN	English	3	124	24	140	49.06	16.494
		2	114	29	127	57.44	18.819
Object Span	English	3	124	3	8	5.89	1.380
		2	114	4	8	5.41	1.151
Word Reading Accuracy	English	3	124	1	40	23.54	13.489
Spelling	English	3	124	0	8	3.93	2.325
Phoneme Matching	Irish	2	124	9	22	17.95	2.277
				9	22	17.26	2.510
Phoneme Deletion	Irish	3	124	1	13	10.19	3.016
		2	114	0	13	7.96	4.508
RAN	Irish	3	124	29	123	55.79	17.557
		2	114	31	148	60.55	21.674
Object Span	Irish	3	124	3	8	5.60	1.330
		2	114	3	8	5.32	1.359
Word Reading Accuracy	Irish	3	124	0	40	24.44	12.571
		2	114	0	20	10.77	6.452
Spelling	Irish	3	124	0	9	4.05	2.323

Table 2. Pearson correlations for Grade 2 (senior infant) predictor and outcome variables.

	1	2	3	4	5	6	7	8	9
1. Phoneme Matching (EN)	1	.449**	-0.105	0.156	.594**	.439**	-.346**	0.155	.432**
2. Phoneme Deletion (EN)	.449**	1	-.335**	.317**	.357**	.863**	-.474**	.237*	.666**
3. RAN (EN)	-0.105	-.335**	1	-.210*	-0.079	-.397**	.454**	-0.158	-.226*
4. Object Span (EN)	0.156	.317**	-.210*	1	0.066	.230*	-0.165	.397**	.274**
5. Phoneme Matching (GA)	.594**	.357**	-0.079	0.066	1	.345**	-.316**	0.121	.306**
6. Phoneme Deletion (GA)	.439**	.863**	-.397**	.230*	.345**	1	-.428**	.206*	.733**
7. RAN (GA)	-.346**	-.474**	.454**	-0.165	-.316**	-.428**	1	-0.122	-.492**
8. Object Span (GA)	0.155	.237*	-0.158	.397**	0.121	.206*	-0.122	1	.300**
9. Word Reading Accuracy (GA)	.432**	.666**	-.226*	.274**	.306**	.733**	-.492**	.300**	1

Table 3. Pearson correlations for Grade 3 (First Class) predictor and outcome variables.

	1	2	3	4	5	6	7	8	9	10	11	12
1. Phoneme Matching (EN)	1	.324**	-0.128	0.114	0.126	0.164	.430**	.252**	-0.028	0.091	0.141	0.174
2. Phoneme Deletion (EN)	.324**	1	-.264**	.236**	.682**	.593**	.380**	.821**	-.455**	.186*	.719**	.626**
3. RAN (EN)	-0.128	-.264**	1	-.254**	-.304**	-.294**	-0.068	-.322**	.445**	-.179*	-.326**	-.277**
4. Object Span (EN)	0.114	.236**	-.254**	1	.259**	.302**	0.170	.252**	-.189*	.510**	.271**	.264**
5. Word Reading Accuracy (EN)	0.126	.682**	-.304**	.259**	1	.829**	.368**	.717**	-.552**	.383**	.932**	.797**
6. Spelling (EN)	0.164	.593**	-.294**	.302**	.829**	1	.405**	.707**	-.493**	.366**	.813**	.764**
7. Phoneme Matching (GA)	.430**	.380**	-0.068	0.170	.368**	.405**	1	.398**	-.213*	.251**	.356**	.395**
8. Phoneme Deletion (GA)	.252**	.821**	-.322**	.252**	.717**	.707**	.398**	1	-.530**	.247**	.782**	.675**
9. RAN (GA)	-0.028	-.455**	.445**	-.189*	-.552**	-.493**	-.213*	-.530**	1	-.274**	-.581**	-.511**
10. Object Span (GA)	0.091	.186*	-.179*	.510**	.383**	.366**	.251**	.247**	-.274**	1	.390**	.411**
11. Word Reading Accuracy (GA)	0.141	.719**	-.326**	.271**	.932**	.813**	.356**	.782**	-.581**	.390**	1	.821**
12. Spelling (GA)	0.174	.626**	-.277**	.264**	.797**	.764**	.395**	.675**	-.511**	.411**	.821**	1

Table 4. Hierarchical regression analysis.

Grade 2 (Senior Infants) Word Reading Accuracy
Model	R Square	R Square change	Adjusted R Square	β
1. GA comm	0.068		0.051	−0.217*
G = boy				.138
2. School	0.209	.141	0.157	NA
3.	0.710	.501	0.679
PM:GA				−0.037
PD:GA				0.614**
RAN:GA				−0.166*
Span: GA				0.140*
4.	0.721	.011	0.681
PM: EN				0.108
RAN:EN				0.055
Sanden				0.049

Note: GA comm = frequency of communication in Irish at home; G = boy = gender with ‘boy’ as the reference category; PM = Phoneme Matching; PD = Phoneme Deletion; RAN = Rapid Automatised Naming; Span = Forward Object Span; GA = Irish; EN = English.

Table 5. Hierarchical regression analysis.

Grade 3 (First Class)
Irish Word Reading Accuracy					Irish Spelling
Model	R Square	R square change	Adjusted R Square	β	R Square	R square change	Adjusted R Square	β
1. GA comm.	0.024		0.008	−0.156	0.019		.002	−0.136
G = boy			-.010					-.004
2. School	0.159	.135	0.108	NA	0.210		0.162	NA
3.	0.713	.554	0.684		0.628		0.592
PM:GA				0.021				0.107
PD:GA				0.607**				0.423**
RAN:GA				−0.183*				−0.182*
Span: GA				0.159*				0.202*
4.	0.716	.003	0.680		0.631		0.584
PM: EN				−0.046				−0.024
RAN:EN				−0.058				−0.061
Sanden				−0.029				−0.037

Note: GA comm = frequency of communication in Irish at home; G = boy = gender with ‘boy’ as the reference category; PM = Phoneme Matching; PD = Phoneme Deletion; RAN = Rapid Automatised Naming; Span = Forward Object Span; GA = Irish; EN = English.

Table 6. Hierarchical regression analysis.

Grade 3 (First Class)
English Word Reading Accuracy					English Spelling
Model	R Square	R square change	Adjusted R Square	β	R Square	R square change	Adjusted R Square	β
1. EN comm.	0.064		.049	.243*	0.086		.071	.280*
G = boy				.070				.086
2. School	0.156	.092	0.105	NA	0.199		0.151
3.	0.571	.415	0.529		0.523		0.476
PM:EN				−0.104				−0.023
PD:EN				0.583**				0.409**
RAN:EN				−0.220*				−0.242*
Span:EN				0.107				0.183*
4.	0.643	.072	0.597		0.608		0.557
PM: GA				0.120				0.237*
RAN:GA				−0.196*				−0.169*
Span:GA				0.180*				0.121

Note: EN comm = frequency of communication in English at home; G = boy = gender with ‘boy’ as the reference category; PM = Phoneme Matching; PD = Phoneme Deletion; RAN = Rapid Automatised Naming; Span = Forward Object Span; GA = Irish; EN = English.

Grade 2 (Senior Infant) Irish Word Reading Accuracy (WRA). Model 1, which included the frequency of parental communication in Irish, as well as gender, accounted for 7% of the variance in Grade 2 Irish word reading (p = .019). Model 2, which included the school variable, accounted for 21% of the variance in Grade 2 Irish Word Reading, (p < .001). Model 3 which included scores on Irish predictor tasks accounted for 71% of the variance(p < .001). Model 4 accounts for 72% of the variance in Grade 2 Irish Word Reading (p < .001).

Note that frequency of communication in Irish at home is a significant predictor of Irish WRA in Model 1 and this measure involves a scale from 1 (always) to 5 (never). Perhaps unintuitively, the Beta coefficient indicates that a unit decrease in frequency of communication in Irish (e.g. from ‘always’/1 to ‘sometimes’/2) is accompanied with an decrease in reading scores, meaning that more frequent communication in Irish predicts higher reading scores. Gender did not emerge as a significant predictor of WRA for Grade 2 Pupils.

The Beta coefficients indicate that Irish Phoneme Deletion was the most effective predictor of Grade 2 WRA, (p < .001) followed by Irish RAN (p = .008) and Irish Object Span (p = .013), and that the English tasks were not significant predictors of Irish Word Reading when the Irish predictor tasks are accounted for.

Grade 3 (First Class) Irish Word Reading Accuracy. Model 1, which included the frequency of parental communication in Irish and gender accounted for a non-significant (p = .225) 2% of the variance in Grade 3 Irish word reading. Model 2, which included the school variable, accounted for 16% of the variance in Grade 2 Irish Word Reading, (p = .005). Model 3 which included scores on Irish predictor tasks accounted for the most amount of variance, at 71% (p < .001). Model 4 which included scores on the English predictor tasks, did not account for a significant amount of additional variance in Grade 2 Irish Word Reading.

In contrast to the results for Grade 2 pupils, frequency of communication in Irish at home did not emerge as a significant predictor of Grade 3 WRA. In line with the Grade 2 analysis, gender did not emerge as a significant predictor. Irish Phoneme Deletion was the most effective predictor of Grade 3 Irish WRA (p > .001), followed by Irish RAN (p = .004) and Irish Object Span (p = .004). In line with the Grade 2 analysis, the English tasks were not significant predictors of Irish Word Reading when the Irish predictor tasks are accounted for.

Grade 3 (First Class) Irish Spelling. Model 1, which included the frequency of parental communication in Irish and gender accounted for a non-significant (p = .322) amount of the variance in Grade 3 Irish word reading. Model 2, which included the school variable, accounted for 21% of the variance in Grade 2 Irish Word Reading, (p < .001). Model 3 which included scores on Irish predictor tasks accounted for the most amount of variance, at 63% (p < .001). Model 4, which included scores on the English predictor tasks, did not account for any additional variance in Grade 2 Irish Word Reading.

In keeping with the results in relation to Grade 3 Irish WRA, frequency of communication in Irish at home did not emerge as a significant predictor of Grade 3 WRA. In line with the previous analyses, gender did not emerge as a significant predictor. Irish Phoneme Deletion was the most effective predictor of Grade 3 Irish Spelling (p > .001), followed by Irish RAN (p = .011) and Irish Object Span (p = .001). The English tasks were not significant predictors of Irish Word Reading when the Irish predictor tasks are accounted for.

Grade 3 (First Class) English Word Reading Accuracy. Model 1, which included the frequency of parental communication in English and gender accounted for a significant 6% of variance (p = .018) in Grade 3 English word reading. Model 2, which included the school variable, accounted for 16% of the variance in Grade 2 Irish Word Reading, (p = .005). Model 3 which included scores on English predictor tasks accounted for 57% of the variance (p < .001), while Model 4 – which included the Irish predictor tasks – accounted for 64% of the variance in English WRA (p > .001).

Note that frequency of communication in Irish at home is a significant predictor of English WRA¹ in Model 1 and this measure involves a scale from 1 (always) to 5 (never). Perhaps unintuitively, the Beta coefficient indicates that a unit decrease in frequency of communication in English (e.g. from ‘always’/1 to ‘sometimes’/2) is accompanied with an increase in reading scores, meaning that less frequent communication in English predicts higher reading scores. Gender did not emerge as a significant predictor of WRA for Grade 3 Pupils. English Phoneme Deletion was the most effective predictor of Grade 3 English Word Reading (p < .001), followed by English RAN (p = .003), Irish RAN (p = .013) and Irish Object Span (p = .014).

Grade 3 (First Class) English Word Spelling. Model 1, which included the frequency of parental communication in English and gender accounted for 9% of the variance (p = .004) 2 in Grade 3 English word reading. Model 2, which included the school variable, accounted for 20% of the variance in Grade 2 Irish Word Reading, (p < .001). Model 3 which included scores on English predictor tasks accounted for 52% of the variance (p < .001), while Model 4 – which included the Irish predictor tasks – accounted for 61% of the variance in English WRA (p > .001).

In line with the results for Grade 3 English Word Reading, frequency of communication in English at home is a significant predictor of English Spelling in Model 1, meaning that less frequent communication in English predicts higher spelling scores. As in previous analyses, gender did not emerge as a significant predictor. English Phoneme Deletion was the most effective predictor of Grade 3 English Spelling (p < .001), followed by English RAN (p = .002), Irish Phoneme Matching (p = .002), English Object Span (p = .010) Irish RAN (p = .041).

Discussion

This is the first study to examine the predictors of literacy attainment in Irish-English bilinguals. In this study, the pattern of predictors of reading and spelling differed in Irish and English, in keeping with findings from other dual-language environments (e.g. Pasquarella et al. 2015; Swanson et al. 2008).

Predictors of Irish literacy attainment

Overall, Irish Phoneme Deletion was the most effective predictor of Irish literacy attainment in both Grade 2 and Grade 3 pupils. It was more effective in predicting Irish word reading than Irish spelling. In contrast, Irish Phoneme Matching contributed more to Irish spelling than to Irish word reading, however it failed to reach significance as a predictor. The finding that PA is the strongest predictor of Irish literacy attainment is in keeping with findings in other European languages (e.g. Ziegler et al. 2010).

Irish RAN contributed to both Grade 2 and 3 reading and spelling, but was a more effective predictor of Grade 3 reading and spelling. This is in keeping with previous research that has found RAN to be a more effective predictor of literacy attainment as literacy develops (Vaessen et al. 2010). In the present study, RAN contributes to word reading and spelling to a similar extent. Previous research indicates that RAN contributes more to literacy skills which involve whole-word orthographic representations, as opposed to skills for which grapheme-phoneme knowledge is important (Manis, Seidenberg, and Doi 1999). It is possible that whole-word orthographic representations are utilised more in Irish than in other languages, as Irish literacy instruction has traditionally focussed on whole-word recognition methods despite changes in recent years (Stenson and Hickey 2018).

Irish Object Span (measuring VSTM) predicted Grade 2 and Grade 3 Word Reading and Spelling; however it is a less effective predictor overall than Phoneme Deletion or RAN. A number of studies have found that VSTM is not a significant predictor of literacy attainment across a range of languages (Caravolas et al. 2012; Mann and Wimmer 2002; Parilla, Kirby, and McQuarrie 2004; Vaessen et al. 2010). However, the findings of the present study is in keeping with those of Ziegler and colleagues (2010) who found that VSTM was a predictor in some languages, albeit a relatively weak one.

Predictors of English literacy attainment

In keeping with the findings for Irish and for other European languages (e.g. Ziegler et al. 2010), English Phoneme Deletion was the strongest predictor of English Word Reading and Spelling. It is of interest that Irish Phoneme Matching emerges as a unique predictor of Grade 3 English Spelling, though it did not emerge as a predictor of Irish Spelling.

English and Irish RAN were both unique predictors of English reading and spelling. English Object Span was a significant predictor of English spelling, but not English word reading. In contrast, Irish Object Span was a significant predictor of English Word Reading but not English Spelling. It is striking that scores on Irish language tasks were predictive of English literacy attainment, while none of the scores on English language tasks were predictive of Irish literacy attainment. One explanation is that these literacy-related skills were originally developed in Irish, the language of initial literacy acquisition and subsequently contribute to the development of English literacy. This would be in keeping with the Common Underlying Proficiency model (Cummins1981).

This finding aligns with those of Pasquarella et al. (2015), indicating a unidirectional transfer of skills from the initial language of reading instruction. Note that the present study examined whether PA, RAN and VSTM predicted cross-linguistic literacy attainment only when within-language predictors were accounted for; this precludes precise comparisons with studies which only examine predictors in a single language (Jared et al. 2011; Lindsey, Manis, and Bailey 2003) and studies which do not enter L1 and L2 predictor tasks into the same analysis (e.g. Lafrance and Gottardo 2005). Swanson et al. (2008) found that cross-language predictors were not unique when within-language variables were accounted for, contrasting with the present study; however the predictor variables in this study differed from those of the present one.

The frequency of parental communication in Irish was a significant positive predictor of Irish literacy in Grade 2, but not in Grade 3. It may be that home language of communication plays a role in the initial stages of literacy and diminishes thereafter. Contrastingly, the frequency of parental communication in English at home was a significant negative predictor of English literacy in Grade 3. Note that frequency of communication in Irish and English were inversely correlated, and we consider that a higher frequency of communication in Irish could support initial literacy acquisition, which later transfers to English. Longitudinal research in this area could elucidate these matters further. Gender did not emerge as a significant predictor of literacy attainment in this cohort of early readers, in contrast with Harris and colleagues (2006).

Implications for literacy assessment

This study echoes findings by Nic Aindriú and colleagues (2021) emphasising that it is necessary to examine literacy and related skills in both languages for children in dual-language. The fact that English language predictors were found not to significantly contribute to Irish literacy attainment underscores the fact that the current practice of assessment for dyslexia in Ireland – which involves diagnostic assessment in English only – is fundamentally flawed. These findings highlight the urgency and necessity of developing dual-language screeners and diagnostic assessments for pupils in Irish-medium and Gaeltacht schools.

Though Phoneme Matching did not emerge as a significant predictor in this study, we believe that it would be useful to include both a Phoneme Matching and Phoneme Deletion task on assessments. Information from both tasks could be used to distinguish between students who need support of a more fundamentally phonological nature and those who need support which is more metalinguistic in nature. The more basic ability to distinguish between phonemes is particularly important in a dual-language context (Saiegh-Haddad 2019). This is especially true for Irish, given that many of the phonemic contrasts relevant for understanding the writing system are allophones in English, a context in which PA might be most challenging to develop (Flege 1995).

Limitations

The data for this study was collected at a single timepoint, meaning that these predictors are concurrent rather than longitudinal. A small number of constructs were examined as predictors, and given that Irish is a morphologically-rich language, it would be worth exploring the role of morphological awareness in literacy attainment especially for more experienced readers. In addition, the response rate was relatively low in Gaeltacht schools and only a small proportion of the participants are from Gaeltacht schools, limiting the extent to which the findings of this study are generalisable. The low response rate may occur due to research fatigue as people in Gaeltacht areas are frequently invited to participate in linguistic research.

Conclusion

This study examined the predictors of Irish and English literacy attainment in children in Irish-medium immersion and Gaeltacht schools. Though PA was the strongest predictor in each language, overall the pattern of constructs which predicted Irish and English literacy attainment differed, underscoring the importance of investigating the predictors of literacy attainment in each language. It is notable that while the English predictor tasks were not effective in predicting Irish literacy attainment, the Irish predictor tasks were effective in predicting English literacy attainment. Higher exposure to Irish at home was associated with higher scores in Irish and higher exposure to English associated with lower scores on English literacy attainment tasks. The research emphasises the urgent need for bilingual literacy screeners and diagnostic tests in Ireland.

Authors’ contributions

Study conception and design is the work of EB in consultation with ANC and NNC. Material preparation, data collection and analysis were performed by EB and reviewed by ANC. Manuscript was written by EB and reviewed by NNC and ANC. All authors read and approved the final manuscript.

Acknowledgements

We would like to thank the participating children and their parents, and teachers, principals and staff of the participating schools. We would also like to thank the reviewers of this paper for their insightful comments.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by Irish Research Council [Grant Number GOIPG/2018/2425].

Notes on contributors

Emily Barnes

Emily Barnes is Assistant Professor of Language Education in the School of Education, Trinity College Dublin (TCD) and works in collaboration with the Phonetics and Speech Laboratory TCD.

Neasa Ní Chiaráin

Neasa Ní Chiaráin is Ussher Assistant Professor in Irish Speech and Language Technology, and a Prinicpal Investigator on the ABAIR project in the Phonetics and Speech Laboratory, TCD.

Ailbhe Ní Chasaide

Ailbhe Ní Chasaide established the Phonetics and Speech Laboratory, TCD, and is a Principal Investigator on the ABAIR project.

Notes

1 Note that Frequency of Irish communication at home and Frequency of English communication at home are inversely correlated, r =  -.68, p <.001.