“It didn't just feel like translation”: Saudi EFL learners’ perceptions of translanguaging vs. English-only written corrective feedback

Although pedagogical translanguaging has gained increasing attention in multilingual education, little is known about how learners perceive teacher-provided translanguaging written corrective feedback (WCF) compared with conventional English-only WCF. This mixed-methods study investigated the perceptions and experiences of 60 first-year Saudi EFL university students who received either translanguaging or English-only WCF in their English for Specific Purposes writing classes. Data were collected through post-intervention questionnaires (n = 55) and focus-group interviews (n = 8) and analysed using descriptive statistics, Mann-Whitney U tests, and thematic analysis. Students who received translanguaging WCF perceived it as significantly easier to understand and apply than English-only WCF. Qualitative findings showed that strategically integrating Arabic and English enhanced comprehension of metalinguistic explanations, reduced cognitive effort, and helped learners understand crosslinguistic differences. In contrast, students receiving English-only WCF frequently relied on dictionaries and machine translation to interpret feedback. While most participants preferred strategically integrated bilingual feedback, some continued to favour English-only feedback despite acknowledging its greater difficulty. The reliance of English-only participants on dictionaries and machine translation suggests that excluding the first language merely shifted the burden of language mediation to learners rather than removing it. The persistence of a preference for English-only feedback among learners who acknowledged its difficulty reflects the durability of monolingual beliefs. These findings support the pedagogical value of strategically incorporating learners' first language into teacher-provided written corrective feedback in EFL higher education.

Feedback can only work if learners can understand it and are willing to use it (Winstone et al., 2017). This condition is not trivial in English as a foreign language (EFL) higher education, where written corrective feedback (WCF) is typically delivered exclusively in English, even to learners whose proficiency makes English-only metalinguistic commentary difficult to process (Lee, 2019; Saeli et al., 2023). In Saudi higher education this default is a matter of stated policy as well as convention. Alomaim and Altameemi (2022) document that the language education policy of the preparatory year programme in Saudi universities prohibits the use of Arabic in English classes, that many students were uncertain whether such a ban existed, and that teachers were often unaware of it or declined to implement it. Their study concludes by urging policy-makers to replace the English-only policy with an “English-mainly” policy that treats students as bilingual speakers of Arabic and English. Learners who cannot decode a comment on article usage or clause cohesion cannot act on it, and feedback that is not acted upon has no instructional value.

The monolingual default in EFL writing classrooms is sustained less by evidence than by institutional language policies and by widely held beliefs, among teachers and learners alike, that any use of the first language (L1) impedes second language (L2) development (Menken and Sánchez, 2019; Sah and Li, 2024; Sun, 2025; Yu and Lee, 2014). This is the problem the present study addresses. English-only WCF is the standard practice in Saudi EFL programmes, yet the assumption underpinning it, that withholding Arabic maximises English learning, has not been tested against learners’ own accounts of what happens when they receive feedback they cannot fully understand.

Translanguaging scholarship offers a principled alternative. Conceiving of bilinguals as owners of a single, integrated linguistic repertoire (García and Wei, 2014; Otheguy et al., 2015; Wei, 2018), pedagogical translanguaging denotes the planned, strategic mobilisation of learners’ languages for learning (Cenoz and Gorter, 2020, 2022). Studies of translanguaging in feedback have found that it improves comprehension, engagement, and writing performance (Saeli and Rahmati, 2023; Sun and Zhang, 2022; Wang and Li, 2022; Yu, 2016; Zhao, 2010). That evidence base, however, is concentrated on peer and oral feedback. Remarkably little is known about how learners perceive and experience teacher-provided WCF delivered through translanguaging, and still less in Arabic-speaking Gulf contexts, where monolingual classroom policies remain the norm.

Learner perceptions matter for two reasons. First, they mediate engagement: feedback perceived as incomprehensible or illegitimate is unlikely to be used (Winstone et al., 2017; Zhao, 2010). Second, they reveal the ideological terrain on which any pedagogical innovation must operate: learners socialised into monolingual norms may resist L1-inclusive practices even when those practices demonstrably help them (Kim and Chang, 2022; Yeh, 2018; Yu and Lee, 2014).

The significance of the study is therefore threefold. Theoretically, it extends pedagogical translanguaging into teacher-provided WCF, a feedback mode in which the teacher controls language choice and in which the distinction between translanguaging and translation becomes consequential. Methodologically, it compares two parallel groups within a single intervention, allowing perceptions of translanguaging WCF to be read against a contemporaneous English-only baseline rather than against learners’ recollections. Practically, it provides learner-side evidence for institutions in Saudi Arabia and comparable Gulf settings that maintain English-only feedback policies without having examined their effects on comprehension and uptake.

The study is guided by two research questions:

RQ1. How do Saudi EFL learners receiving translanguaging WCF and those receiving English-only WCF differ in their reported comprehension, application, and perceived benefit of the feedback?

RQ2. How do learners experience and evaluate the two feedback conditions, and how do their evaluations interact with prevailing monolingual beliefs about language learning?

Background

Multilingualism, pedagogical translanguaging, and the EFL language classroom

Translanguaging belongs to a broader reorientation in the study of multilingualism. Where earlier accounts treated multilingual speakers as possessing separate, parallel language systems, the multilingual turn in applied linguistics reconceived multilingualism as the normal human condition and the multilingual speaker, rather than the idealised monolingual, as the appropriate reference point for language education (Conteh and Meier, 2014; May, 2014). Within that reorientation, a heteroglossic view of multilingual competence holds that speakers draw on one repertoire of linguistic features rather than switching between bounded languages, and it is this view that underpins translanguaging as both a theory and a pedagogy.

Translanguaging theory accordingly rejects the view of the bilingual as “two monolinguals in one” in favour of a heteroglossic account in which speakers deploy features from an integrated repertoire (García and Wei, 2014; Wei, 2018). Recent work has continued to develop this position and to test it in classroom settings: Sah and Li (2024) examine how selectively applied translanguaging can reproduce rather than redress inequities among multilingual learners, and Sun (2025) traces how multilingual students seek and use translingual feedback in academic writing, extending the framework into feedback processes specifically. Pedagogical translanguaging is distinguished from spontaneous code-switching, and, importantly for feedback, from translation, by its planned and purposeful character (Cenoz and Gorter, 2020, 2022). Classroom research has associated it with deeper comprehension of complex content, heightened crosslinguistic awareness, and fuller participation, particularly for learners with limited L2 proficiency (Adamson and Coulson, 2015; Duarte, 2019; Liu et al., 2020; Makalela, 2015; Vogel and García, 2017). In writing instruction specifically, García and Kano (2014) found that translanguaging helped Japanese learners become more aware of differences in the construction of Japanese and English texts, and Parmegiani (2022) showed the mother tongue functioning as a resource for academic success in a bilingual writing programme.

A distinction that matters for the present study concerns the instructional setting. Much recent translanguaging research has been conducted in English-medium instruction (EMI) content classrooms, where English is the vehicle for learning disciplinary subject matter and the L1 is mobilised to support content comprehension. The present study is situated differently: in an EFL language classroom, where English is itself the object of study. This distinction sharpens rather than dissolves the theoretical question. In EMI settings, the argument for translanguaging is largely one of content access; in language classrooms, it must contend with the counter-argument that L1 use displaces target-language input and reduces opportunities for L2 processing. Translanguaging in a language classroom must therefore be justified on the grounds that it supports the learning of English itself, for instance by rendering metalinguistic explanation comprehensible, or by making crosslinguistic contrasts explicit, rather than merely by making content accessible.

Why translanguaging bears on written corrective feedback

Translanguaging is a general account of multilingual language use, and written corrective feedback is one specific classroom practice. The connection between them is not automatic, and this section sets out the four reasons the framework applies to feedback with particular force, since the argument of the study depends on them.

The first concerns what kind of language metalinguistic feedback is. A comment such as “the article is required here because the noun has already been introduced” is not an instance of the target language being used for communication; it is commentary about the target language, and its function is explanatory rather than exemplary. This distinction matters because the standard objection to first-language use in a language classroom is that it displaces target-language input. That objection has real force where the L1 replaces English the learner would otherwise have processed as language. It has much less force in the case of metalinguistic commentary, whose value lies in whether the explanation is understood, not in whether the learner has been exposed to more English while reading it. An explanation the learner cannot decode delivers neither the explanation nor usable input; it delivers nothing.

The second concerns where the difficulty is concentrated. Metalinguistic feedback is dense in exactly the linguistic material that lower-proficiency learners find hardest: technical grammatical terminology, abstract nouns, conditional and causal constructions, and hedged formulations of rules and exceptions. It is, in other words, systematically more demanding than the writing it comments on. A learner producing text at B1 level may routinely receive commentary pitched well above that level, so that the feedback is harder to read than the essay that prompted it. This inversion is specific to feedback and does not arise in most other classroom language, and it is what makes the language of feedback a consequential variable rather than an incidental one.

The third follows from translanguaging's central theoretical claim. If bilinguals draw on a single integrated repertoire rather than two separate systems (García and Wei, 2014; Otheguy et al., 2015; Wei, 2018), then a learner reading English-only feedback does not process it monolingually. The learner's Arabic remains available and is drawn on whether or not the teacher sanctions it. An English-only policy therefore cannot make feedback processing monolingual; it can only determine whether the first language enters the process under the teacher's control or the learner's. The unitary-repertoire premise thus makes a specific and testable prediction about feedback: withholding the L1 will relocate L1 mediation rather than eliminate it. Section 4.2.3 reports what that relocation looked like for these participants.

The fourth concerns the risk that makes feedback a demanding test case for the framework. Because WCF is textual and asynchronous, it is the mode of L1-inclusive practice most easily reduced to translation, and a teacher who wishes to help can simply render the English comment in Arabic beside it. Cenoz and Gorter (2020, 2022) distinguish pedagogical translanguaging from translation by its planned and purposeful character, and in feedback that distinction becomes practically consequential rather than merely definitional: the two produce visibly different artefacts, one in which the languages duplicate each other and one in which they divide the explanatory labour. The present study exploits this, delivering direct corrections in English while reserving Arabic for grammatical terminology, complex explanation, and crosslinguistic contrast, so that each language does work the other does not. Whether learners themselves perceive this distinction, and whether it matters to them, is an empirical question this study is able to answer (Section 4.2.4).

Taken together, these four points explain why feedback is not simply one more site at which translanguaging might be applied. It is a site where the case for translanguaging is unusually strong, where the standard counter-argument is unusually weak, and where the theoretical distinction between translanguaging and translation becomes visible in the material the learner receives.

Nevertheless, monolingual (“English-only”) ideologies remain deeply institutionalised in EFL settings. Menken and Sánchez (2019) describe how such policies position the L1 as an obstacle rather than a resource, and research with learners repeatedly finds internalised versions of this stance (Kim and Chang, 2022; Yeh, 2018; Yu and Lee, 2014). In Saudi higher education these norms are institutionalised at the level of written policy. Alomaim and Altameemi (2022) report that the preparatory year programme in the university they studied bans Arabic in English classes, and that this ban operates largely as a de facto policy: most students were unsure whether it existed, and teachers who were aware of it did not consistently observe it. The effect is an expectation, rarely stated but widely assumed, that all classroom communication, including feedback, will be conducted in English. Learner attitudes in this setting are not uniformly opposed to the first language. Alqahtani (2022) found that Saudi students broadly supported translanguaging in EFL classes while remaining concerned that it might not carry them to the proficiency they sought, a combination of endorsement and anxiety that the present study encounters again in Section 4.2.6. Where translanguaging is permitted selectively rather than systematically, questions of linguistic justice and equal access arise for lower-proficiency learners (Sah and Li, 2024, in an EMI context whose ideological dynamics are comparable).

Learner preferences for written corrective feedback

Research on WCF preferences has established that learners hold strong and specific views about the feedback they receive, and that these views vary systematically with learner characteristics. Studies have examined preferences for feedback scope (comprehensive versus focused), explicitness (direct versus indirect), and source, and have consistently found mismatches between what learners want and what teachers provide.

Individual differences have emerged as an important moderator. Farangi et al. (2025) investigated WCF preferences among 135 Iranian EFL learners classified as field-dependent or field-independent using the Group Embedded Figures Test, and found that cognitive style shaped preferences for the amount and type of WCF learners considered useful. Their findings indicate that WCF preference is not a uniform property of learner populations but is patterned by stable individual characteristics, and that mixed-methods designs combining preference questionnaires with qualitative accounts can reveal reasoning that quantitative preference rankings alone conceal.

What this literature has not systematically examined is the language in which feedback is delivered as a dimension of preference. Studies of WCF preference typically hold language constant, implicitly assuming an L2-only default, and investigate variation in scope, explicitness, or focus instead. Yet for learners operating below advanced proficiency, the language of the commentary plausibly conditions whether any of these other dimensions can take effect: an elaborated, well-focused metalinguistic explanation is of no use if it cannot be decoded. The present study treats the language of feedback as a variable in its own right, and asks how learners evaluate it.

The language of written corrective feedback and learner perceptions

The present review concentrates on written feedback, since it is written commentary that learners must decode unaided, without the interactional repair available in spoken exchanges. A consistent finding across written feedback research is that comprehensibility conditions use. Chen (2018) found that a majority of feedback on grammar and discourse errors went uncorrected when learners had difficulty understanding it, and Winstone et al. (2017) documented that feedback perceived as hard to process is often simply abandoned.

Studies of L1 use in written feedback speak directly to this problem. Although several of these examine peer rather than teacher commentary, they establish the mechanisms, clarity, usability, and learner control over language choice, that the present study tests in teacher-provided WCF. Zhao (2010) found that learners used and understood more of peers’ L1 feedback than teachers’ L2 feedback, and that prior learning experiences shaped their responses. Yu and Lee (2014) showed Chinese learners strategically using the L1 in peer feedback to explain comments against English writing criteria, and Yu (2016) found that L1 use made comments clearer and more usable. Yeh (2018) reported that learners valued the L1 for ease of communication in peer review, though some preferred the L2 for practice. Kim and Chang (2022) found Japanese learners calibrating language choice to their peers’ proficiency, with some preferring L2 feedback “for the sake of L2 learning per se ” (p. 1372).

Turning to teacher feedback, Saeli et al. (2023) documented Iranian teachers’ strategic use of translanguaging in providing WCF, especially in metalinguistic information on grammar, and found the practice particularly supportive of low-proficiency learners’ affective engagement; Saeli and Rahmati (2023) likewise found that in situ translanguaging enhanced affective engagement with written peer feedback. Gomez (2020) reported that learners valued feedback delivered through translanguaging in an EFL course, and Sun and Zhang (2022) and Sun et al. (2023) linked translingual feedback practices to improved writing performance and feedback literacy.

Three gaps remain. First, the teacher-feedback studies describe teachers’ practices rather than learners’ perceptions of receiving them. Second, no study has compared learners receiving teacher-provided translanguaging WCF with a parallel group receiving English-only WCF within the same intervention, leaving open whether perceived advantages reflect the feedback language or the elaborated quality of intervention feedback generally. Third, the interaction between learners’ evaluations and the monolingual beliefs they bring to the classroom has not been examined in a Gulf EFL setting. The present study addresses all three.

Design, context, and participants

This study employed a two-condition, class-level randomised, post-intervention sequential mixed-methods comparative design. It examined Saudi EFL students’ perceptions of two forms of WCF: translanguaging WCF and English-only WCF.

Each element of this design follows from the research questions. A two-condition comparative design was necessary because the study asks how learners evaluate translanguaging WCF relative to the English-only feedback that is standard in this setting. Asking a single group to judge translanguaging feedback in the abstract would have produced evaluations anchored to whatever those learners happened to remember of previous feedback; running the two conditions in parallel within one intervention supplies a contemporaneous baseline, so that the comparison holds constant the teacher, the writing tasks, the target features, and the elaborateness of the commentary, leaving the language of the feedback as the difference between groups. Class-level randomisation was used because the intervention had to be delivered to whole classes: feedback is given in the course of teaching, and it is not possible to give some students in a class bilingual commentary while their neighbours receive English-only commentary without the two conditions becoming visible to each other. Randomising which class received which condition removes the researcher's judgement from the assignment, though it does not equate the groups, and the consequence of that limitation is addressed below. A post-intervention design was appropriate because the object of study is learners’ evaluation of feedback they have received, which cannot be measured before they have received it. Finally, a sequential mixed-methods design was chosen because the questionnaire and the focus groups answer different questions and the second depends on the first. The questionnaire establishes whether the two groups differed in their reported experience and by how much, but cannot say what the difference consisted of or why learners judged as they did; the focus groups, conducted afterwards, were designed to explain the questionnaire pattern rather than to duplicate it. This ordering also allowed learners’ reasoning to be explored in cases where their ratings and their accounts diverged, which proved analytically productive (Section 4.2.6). Although English is used as the medium of instruction in some disciplinary courses at the university, the present study was conducted in English for Specific Purposes (ESP) language classes. It therefore investigates translanguaging as a pedagogical practice in an EFL writing classroom, rather than translanguaging in an EMI content course.

An intact-class sampling approach was used. Two existing first-year ESP classes at a Saudi University participated, and the two classes were randomly assigned at class level to one of the feedback conditions.

Intact classes were used for three reasons. The first is practical: students at the institution are enrolled in fixed class groups by the registry, and reassigning individuals to new groups for research purposes was not permitted and would have disrupted their timetabled instruction. The second is methodological. Written corrective feedback is embedded in ongoing classroom teaching, and studying it in a group assembled artificially for the purpose would have removed it from the context that gives it meaning. Because the feedback was delivered as part of the students’ actual coursework, on tasks that formed part of their normal writing instruction, the study observes the practice as it would occur rather than as a laboratory approximation, which strengthens the ecological validity of the findings and their relevance to teachers working under the same conditions. The third concerns the integrity of the conditions themselves, noted above: within a single class, students compare feedback, and a mixed-condition class would have exposed each group to the treatment the other received.

This choice has a cost, which the study reports rather than minimises. Because each condition maps onto one class, condition is confounded with class, and no analysis can fully separate the effect of the feedback language from any characteristic of the class that received it. The comparison of placement-test scores reported below addresses the most obvious of these characteristics, English proficiency, but cannot address others such as group dynamics or scheduling. This is why the questionnaire comparison is presented as exploratory, and why the study's central claims rest on the convergence of questionnaire and interview evidence rather than on the statistical comparison alone. Each class comprised 30 students, giving an intervention sample of N = 60. Participants were male Saudi students aged 19–22 years and in the first year of their undergraduate studies. Arabic was their first language, and their English proficiency was approximately CEFR B1 based on the Oxford Placement Test. All had begun studying English in the fourth grade of school and had therefore received formal English instruction for approximately nine years before entering university. Despite this length of study, their productive proficiency remained limited, a pattern the participants themselves attributed to the near-total absence of opportunities to use English outside the classroom and to the quality of English instruction they had received at school. This profile, in which extended formal exposure is combined with restricted functional proficiency, is characteristic of EFL learners in settings where English has no community presence, and it is directly relevant to the present study, since it is precisely such learners for whom English-only metalinguistic commentary is most likely to exceed available processing resources.

Students at the institution are assigned to classes on the basis of an English placement test administered on entry, and both participating classes were drawn from the same placement band.

Because the two conditions were delivered to intact classes, the comparability of the groups before the intervention was examined using their Oxford Placement Test scores. The two groups were closely matched: the translanguaging group had a mean score of 68.40 (SD = 8.15) and the English-only group a mean of 67.88 (SD = 8.42). The difference was negligible and did not approach statistical significance, t(58) = 0.24, p =.81, d = 0.06. Differences in perception reported below are therefore unlikely to be attributable to pre-existing differences in English proficiency between the two classes.

Owing to gender segregation in the Saudi higher education system, only male students could be recruited.

Following the intervention, 55 students completed the questionnaire: all 30 from the translanguaging condition (100%) and 25 of the 30 from the English-only condition (83.3%), giving an overall response rate of 91.7%. Because the study involved only two intact classes, with one class assigned to each condition, comparisons between individual questionnaire responses should be interpreted as exploratory comparisons of reported perceptions rather than as definitive causal estimates of the effects of the feedback conditions.

The study received approval from the Institutional Review Board of the Saudi University that the present study took place. All participants provided written informed consent.

Feedback intervention

The intervention comprised three writing sessions held in three consecutive weeks, in each of which participants completed one writing task, giving three writing tasks in total. The full research procedure, including the testing sequence described below, extended over approximately two months. Each task received one round of written corrective feedback, after which participants revised their texts in light of the feedback they had received. Feedback was therefore provided once per task, and revision followed once; there were no further cycles of feedback and revision within a task. Revised texts were collected as part of the procedure but are not analysed here, for the same reason: the present article reports perceptions rather than performance. Because participants revised after each feedback round, the questionnaire items concerning ease of applying the feedback in revision refer to an activity participants had actually carried out rather than to a hypothetical judgement.

The feedback targeted three recurrent areas of difficulty in students’ writing: subject–verb agreement, the English article system, and cohesive markers. In both conditions, the researcher provided direct and metalinguistic WCF in the margins and in end comments. The researcher was a native speaker of Arabic and an English-language instructor who shared the participants’ first language and cultural background.

In the translanguaging condition, direct corrections were provided in English, while Arabic was used strategically in metalinguistic explanations, particularly for grammatical terminology, complex explanations, and crosslinguistic contrasts. The purpose was not to translate every English explanation into Arabic, but to draw on students’ full linguistic repertoire to make feedback more comprehensible and usable. In the English-only condition, direct corrections and metalinguistic explanations were provided exclusively in English.

The intervention also generated writing-performance data, collected through a pre-test, an immediate post-test, and a delayed post-test. These measures are not analysed here, because the present article addresses learners’ perceptions of the two feedback conditions rather than their effects on writing accuracy. They are described because they form part of the procedure and determine when the questionnaire was administered. The questionnaire was administered after the delayed post-test, at the end of the two-month research procedure. Participants therefore evaluated the feedback having completed the whole intervention, rather than while any single feedback episode was still fresh. This placement was deliberate: it allowed learners to report on the feedback as a sustained experience across three tasks rather than on a single instance, which is the level at which the research questions are pitched. It does mean that several weeks separated the final feedback session from the point of evaluation, and the perceptions reported here are accordingly retrospective. The focus-group interviews, conducted after the questionnaire, provided an opportunity to probe the reasoning behind those retrospective judgements.

Questionnaire

After completing the feedback procedure, participants completed an online questionnaire administered through Qualtrics. The questionnaire comprised 14 closed items on a six-point Likert scale (1 = strongly agree; 6 = strongly disagree) and two open-ended questions. The closed items examined: (a) the overall comprehensibility and usefulness of the feedback; (b) perceived understanding, application, and improvement in relation to the three targeted linguistic features; and (c) perceptions of margin and end comments. The open-ended questions enabled participants to explain their responses and feedback preferences in greater detail.

Two condition-specific versions were created, identical except that items referred to the feedback condition the relevant group had received. Items were presented in both Arabic and English to ensure that students could fully understand them.

The questionnaire was purpose-built for this study rather than adapted from an existing instrument, since no available measure addressed learners’ perceptions of the language in which written corrective feedback is delivered.

The items were written against a specification derived from two sources. The first was the study's research questions, which identify three dimensions of learners’ response to feedback: whether they understood it, whether they were able to apply it in revision, and whether they believed it had improved their writing. The second was the design of the intervention, which determined the scope the items had to cover. Because feedback was delivered on three target linguistic features and in two comment formats, the specification required items addressing the feedback as a whole, items addressing each target feature, and items addressing each format, within each of the three dimensions. Fourteen items were written to fill this specification in a single draft, distributed as six items on understanding, four on application, and four on perceived improvement. No item pool was generated and subsequently reduced; the item set follows directly from the specification, and the only revisions made were those recommended by the expert panels described below.

A six-point agreement scale was used, a format that provides no neutral midpoint. Midpoints on agreement scales attract genuine ambivalence and disengaged responding alike, and with groups of this size a substantial cluster of midpoint responses would have left the two conditions difficult to distinguish. Requiring participants to fall on one side of the scale yields a directional judgement from every respondent.

The questionnaire was then submitted to two independent panels of experts before administration.

The first panel comprised three Saudi specialists holding doctoral degrees in English and experienced in English language teaching, who conducted a translation check. They compared the Arabic and English versions item by item and judged the semantic equivalence of the two wordings, the clarity and naturalness of the Arabic, and the cultural appropriateness of the phrasing for the target population. The second panel comprised three further Saudi specialists, also holding doctoral degrees in English and experienced in English language teaching, who conducted a content review. They evaluated each item for clarity, for relevance to the constructs under investigation, and for alignment with the study's research questions and feedback design.

Modifications suggested by both panels were incorporated before administration. Together these procedures provided translation-equivalence and content-validity evidence for the instrument. The 14 closed items were designed to tap three dimensions of learners’ response to the feedback, and each item was assigned to one of them: understanding the feedback (6 items), applying it in revision (4 items), and perceived improvement in writing (4 items). Within each dimension, items varied in scope, some addressing the feedback as a whole and others addressing feedback on a particular target feature or in a particular comment format. Responses were averaged within each dimension to yield three subscale scores per participant, which formed the basis of the between-condition comparison reported in Section 4.1.

Internal consistency was assessed for the instrument as a whole and for each subscale separately, since the subscale composites rather than the individual items form the basis of the analysis. Across all 14 items, computed on the full questionnaire sample (N = 55), internal consistency was good (Cronbach's α =.84). The three subscales were also internally consistent: understanding the feedback, α =.82 (6 items); applying it in revision, α =.79 (4 items); and perceived improvement in writing, α =.80 (4 items). The subscale values fall slightly below the overall figure, as is expected of shorter scales, and all exceed the conventional threshold of.70, supporting the aggregation of items within each dimension. No pilot study was conducted; this is acknowledged as a limitation in Section 7.

Focus-group interviews

Two condition-specific focus-group interviews were conducted after the questionnaire to obtain a richer understanding of participants’ experiences of receiving WCF. Focus groups were chosen in response to calls in feedback research to cultivate natural conversation among students about their views of feedback (Best et al., 2015) and because group interaction can elicit more fully articulated accounts than individual interviews (Galloway, 2020). Four volunteers from each condition participated, four being recommended as a size that limits overdominating voices (Onwuegbuzie et al., 2009), resulting in two focus groups with eight participants in total. Interviews were conducted online through Zoom, used Arabic and English as needed, were audio-recorded with participants’ consent, and lasted approximately 50 min each.

The focus groups explored participants’ perceptions of feedback comprehensibility, usefulness, preferences, and experiences of Arabic and English in feedback explanations.

The interview questions were written by the researcher for the present study, derived from the research questions and from the design of the two feedback conditions, and were then reviewed by three Saudi specialists holding doctoral degrees in English and experienced in English language teaching. The reviewers evaluated the clarity and neutrality of the wording, the relevance of each question to the study's aims, and the suitability of the questions for group discussion; their suggestions were incorporated into the final protocol. Interviews and open-ended questionnaire responses were conducted and written in Arabic and English. All quotations presented in the Results are the researcher's own translations of the Arabic originals, rendered into idiomatic English while preserving the content and emphasis of what participants said.

Quotations are identified by condition, instrument, and participant number. Focus-group participants are labelled TL-1 to TL-4 and EO-1 to EO-4. Open-ended questionnaire responses are labelled TL-Q n and EO-Q n, where n indexes the questionnaire respondent. The two numbering series are independent of each other. Questionnaire respondents were anonymous and cannot be matched to focus-group participants, so TL-3 and TL-Q3, for example, are different individuals, as are EO-2 and EO-Q2. Readers should not treat a shared numeral across the two series as indicating a single participant reporting through both instruments.

Data analysis

Closed questionnaire responses were analysed using descriptive statistics (means and standard deviations). Item responses were averaged within each of the three subscales described in Section 3.3 to produce composite scores, and the two conditions were compared on these composites using Mann–Whitney U tests, selected because the responses were ordinal and the group sizes modest. The alpha level was set at.05, two-tailed. Because the comparison was conducted on three subscale composites rather than on individual items, no correction for multiple comparisons was applied. Open-ended questionnaire responses and focus-group transcripts were entered into NVivo and analysed thematically.

Thematic analysis proceeded through six phases. In the first, the researcher read the full corpus of open-ended responses and focus-group transcripts twice without coding, recording initial impressions in memos, in order to become familiar with the data as a whole. In the second, the data were coded inductively line by line in NVivo, with codes generated from the participants’ own formulations rather than from an a priori framework. Codes stayed close to what participants said and named the action, difficulty, or judgement being reported. Examples include the need to utilize a dictionary, it's more than translation, the difficulty of understanding grammatical terminology, relying on Arabic for the explanation, the time and effort of translating, the fear of becoming dependent on Arabic, and encountering unfamiliar connectives. To illustrate how coding was applied, the comment “The first problem I faced was that I did not understand some words, so I had to use a dictionary” (EO-2) was coded as the need to utilize a dictionary, and “It explained the idea using both English and Arabic together, so I felt like I was reading one language” (TL-2) was coded as it's more than translation. In the third, codes were sorted into candidate themes by grouping those that addressed a common aspect of participants’ experience, and codes were compared across the two conditions to establish whether a pattern was condition-specific or shared. In the fourth, candidate themes were reviewed against the coded extracts and then against the full dataset, and themes with insufficient supporting data were collapsed or discarded. In the fifth, each retained theme was defined and named in a short written definition specifying its scope and boundary. In the sixth, representative extracts were selected to illustrate each theme, with attention to extracts that qualified as well as those that supported the pattern.

During interpretation, attention was given to both shared patterns and differences between the translanguaging and English-only groups, and to points at which participants’ survey ratings and interview accounts diverged.

As this is a single-authored study, the researcher was the only person who coded the data. It was therefore not possible to calculate intercoder agreement, which is normally reported when two or more analysts code the same material independently. This is a limitation, and it is stated again in Section 7.

Four steps were taken to reduce the risk that the themes reflected the researcher's expectations rather than what participants actually said. First, codes were taken from participants’ own wording rather than from a list of categories prepared in advance, so that the analysis began from the data. Second, every coding decision was recorded in NVivo at the time it was made, which means each theme can be traced back to the specific extracts it came from and the analysis can be audited by another researcher. Third, once a theme had been formed, it was checked against the whole dataset rather than only against the extracts that had suggested it, and any theme that fewer than two participants supported was removed. Fourth, extracts that contradicted a theme were kept rather than set aside, and they appear in the Results. The clearest instance is Section 4.2.6, where participants argue in favour of English-only feedback, which is the opposite of the study's overall conclusion.

Finally, the qualitative findings were compared with the questionnaire results, so that the interpretation of each theme was checked against the pattern of ratings from the same participants rather than resting on the interview data alone.

Because the researcher delivered the feedback and conducted the interviews, positionality was considered throughout analysis. As a bilingual Saudi EFL instructor who has himself learned English as an additional language, the researcher's insider position supported rapport and an informed understanding of participants’ linguistic context; interpretations were nonetheless made cautiously to avoid assuming that participants would necessarily favour translanguaging. Survey and interview evidence was compared to strengthen the interpretation of the findings.

Questionnaire results

The 14 closed items were organised into three subscales corresponding to the three dimensions the questionnaire was designed to capture: understanding the feedback (6 items), applying it in revision (4 items), and perceived improvement in writing (4 items). Each participant's responses were averaged within each subscale to produce three composite scores, and the two conditions were compared on these composites using Mann–Whitney U tests. Aggregating to the subscale level rather than testing each item separately avoids the inflation of familywise error that fourteen individual comparisons would entail, and gives each construct a more stable estimate than any single item could provide at this sample size.

Overall perceptions. Both groups evaluated their feedback positively; no composite score in either group fell in the disagreement half of the scale. The strength of agreement, however, differed by condition. On the understanding subscale, translanguaging participants rated the feedback as easier to understand (M = 1.20, SD = 0.48) than English-only participants (M = 1.60, SD = 0.71), a significant difference, U = 256.00, p =.013. The same pattern held on the applying subscale (M = 1.30, SD = 0.53 versus M = 1.72, SD = 0.79), U = 266.50, p =.034. On the improvement subscale, the difference (M = 1.30, SD = 0.53 versus M = 1.60, SD = 0.71) did not reach significance, U = 289.00, p =.088. Table 1 summarises these results.

Two features of this pattern bear on the interpretation. First, the advantage of translanguaging appeared on the two subscales that ask participants to report on something they had done (understanding the feedback and using it in revision) but not on the subscale asking them to judge how much their writing had improved. Learners were able to report on the experience of processing feedback but were, on this evidence, less well placed to assess what they had gained from it. Second, the difference is one of degree rather than direction: both groups evaluated their feedback positively, and the English-only group's ratings were favourable in absolute terms. The comparison is therefore between a condition learners found straightforward and one they found manageable but harder, not between a workable and an unworkable form of feedback. The qualitative data in Section 4.2 give this difference its texture, showing what the additional difficulty consisted of and what learners did in response to it.

Because the analysis operates at subscale level, the data cannot establish which of the individual linguistic features, or which of the two comment formats, drove the differences observed. Participants addressed both questions directly in the focus groups, and those accounts are reported in Section 4.2; they should be read as indicating where the effect may be concentrated rather than as demonstrating it statistically.

Qualitative findings

Comprehensibility and usefulness

Translanguaging-group participants described the feedback as useful, clear, less time-consuming, and comprehensible, and no participant in this condition dissented from that view. One participant situated it in his learning history:

I think explaining in Arabic is more useful than explaining in English. Students have studied in Arabic for many years, so it can be difficult for them when they go to university and everything is in English. They may not understand the teacher's explanation or feedback very well. Also, if they have studied English for only one year, getting feedback in Arabic can help them understand their mistakes better than getting it in English. (TL-1)

Another explained the mechanism: “If you use Arabic and English, the student can understand the meaning better. The student can know why to use one word and not another word” (TL-2). A third made the same point prospectively: “Using their first language can make it easier and simpler for them to use it in their future writing” (TL-Q3).

Two features of these accounts are worth drawing out. First, participants located the difficulty not in English as such but in the mismatch between many years of Arabic-medium schooling and an abrupt shift to English-only academic communication on entering university. The first language is framed as continuity with prior learning rather than as a concession to weakness. Second, the mechanism they describe is explanatory rather than merely facilitative: Arabic is valued because it conveys why one form is required instead of another, which is precisely the function of metalinguistic feedback.

English-only participants presented a more divided picture. Several acknowledged real comprehension costs while still valuing the condition: “Using English in the feedback was harder to understand, but it was better and more helpful” (EO-Q2); “It was difficult to understand, and sometimes I had to use a translator, but it was more helpful” (EO-Q3). Others openly endorsed a bilingual alternative: “It is better to use both languages. The explanation should be mostly in English, but using some Arabic words can help students understand the ideas better” (EO-Q1). Another concluded simply that “It is good to use them both” (EO-2). A third specified a condition: “If the rule is very difficult, I need someone to explain it in an easy way. Using Arabic is the easiest way to explain it to me, but it is not always the best way” (EO-3).

The internal structure of these English-only responses is notable. Where translanguaging participants described a single, coherent experience, English-only participants characteristically produced a concessive construction in which difficulty is acknowledged and then overridden by an assertion of benefit. The final quotation makes the tension explicit, separating what is easiest from what is assumed to be best, and treating the two as necessarily in opposition. This structure recurs throughout the English-only data and is analysed further in Section 4.2.6.

Compensating for limited English proficiency

Participants in both groups framed the first language as compensation for limited proficiency. A translanguaging participant reported reading selectively: “I focused more on Arabic because my English was at an intermediate level. I used English mainly for reading” (TL-4). English-only participants converged on the same point from the other side: “When a person is weak in English, as is the case with us as first-year students, it is preferable to use both languages” (EO-2); “Both Arabic and English comments could help because we are not that fluent in English” (EO-Q2).

Both groups thus shared the view that translanguaging feedback would be well suited to learners with limited proficiency, a proficiency-contingent framing that is consistent with findings that low-proficiency learners benefit most from L1-inclusive feedback (Saeli et al., 2023). This convergence is significant for interpreting the study as a whole: it is not an artefact of the condition participants happened to receive, since learners in both groups arrived at it independently. It does, however, carry an implicit assumption that translanguaging is a temporary support for learners who have not yet reached a threshold, rather than a resource for bilingual writers at any level. This assumption reflects the deficit framing of the first language discussed in Section 5.

Forced translation and its costs

Participants were asked what they did, or would do, when English-only feedback was not understood. Every participant who had encountered this situation described turning to Arabic through translation: “First, I translate what I do not understand into Arabic. If I still do not understand it, I search for it in an Arabic-English dictionary” (EO-1). English-only participants reported that translation was not optional but forced on them: “I had to translate many words and phrases into Arabic” (EO-Q3); “The first problem I faced was that I did not understand some words, so I had to use a dictionary” (EO-2); “I did not understand some grammar words, so I preferred to translate them into Arabic” (EO-Q4).

Participants described this iterative reading-and-translating as consuming considerable time and effort, and reported that the resulting word-for-word renderings of metalinguistic terminology were often unclear: “Although I translated the words I did not understand into Arabic, I think I did not correct my mistake well because it was difficult to understand” (EO-3).

The implication is that withholding the first language did not remove it from the feedback process. It displaced first-language mediation onto learners themselves, in a slower and less reliable form. The learner in the final extract completed the translation and still failed to correct the error, which locates the failure not in access to Arabic but in the quality of the mediation available to him. Machine and dictionary translation of grammatical terminology is precisely the case in which decontextualised word-for-word equivalence is least adequate, since the terms name categories that are configured differently across the two languages. The English-only condition therefore did not produce monolingual feedback processing. It produced unsupported bilingual feedback processing.

Translanguaging versus translation

Participants who received translanguaging feedback drew a sharp and unprompted distinction between the strategic interweaving of the two languages and translation. One explained:

What I liked about this feedback was that it was not just translating from English to Arabic. It explained the idea using both English and Arabic together, so I felt like I was reading one language. (TL-2)

Another appreciated the manner of the mixing rather than its presence: “Using Arabic and English in your own way was very good. I think this is very important because it made the feedback clear and easy to understand” (TL-3). A participant who had previously received translated feedback judged it unhelpful precisely because it was duplicative:

One of my teachers gave me feedback in English and Arabic, but the Arabic was just a direct translation of the English. This was not useful for me because I could do that myself. But your feedback was different because you did not translate every English word into Arabic. (TL-2)

Participants stated that they would welcome bilingual feedback in future provided that it is not merely a translation but a planned movement between the two languages.

These accounts show learners perceiving and articulating, from the receiving end, the defining property of pedagogical translanguaging: its strategic, non-redundant character (Cenoz and Gorter, 2020). The reasoning in the final extract is particularly telling. The participant rejects translated feedback on grounds of redundancy, namely that he could perform the translation himself, so the teacher adds nothing by performing it for him. What he values in translanguaged feedback is that the two languages are distributed across the explanation rather than duplicated, so that each contributes something the other does not. This is a learner-side statement of the theoretical distinction between translanguaging and translation, and it indicates that the distinction is not merely analytic but experientially available to learners.

Past experiences with feedback

Participants contrasted the intervention feedback with the feedback of their prior schooling, which they characterised as general, undirected, and unexplained: “He would just write something like use conjunctions here” (TL-1); “You have a mistake but the teacher does not specify it for you, and you do not know where the mistake is or what he meant” (EO-3). One participant recalled comprehension without usability:

I was able to understand the feedback, but when I tried to use it, I could not. He did not give me enough details, and he only used English. If he used English and Arabic together, I think I could understand it better and use it correctly. (TL-2)

Another reported that “I was surprised by how much detail there was in each point” (TL-4), identifying the metalinguistic explanation as the decisive difference from past feedback. Nearly all participants had received English-only feedback in the past; two had received translated feedback, which they judged ineffective.

These accounts serve two interpretive functions. First, they identify elaboration, not language choice alone, as a salient dimension of the intervention: participants in both conditions valued the specificity of the feedback relative to what they had previously received, which cautions against attributing all positive evaluation to translanguaging. Second, and more importantly, they establish the experiential baseline against which learners judged L1-inclusive feedback. Because almost all prior exposure to bilingual feedback had taken the form of literal translation, learners’ scepticism about Arabic in feedback was grounded in a reasonable inference from an unrepresentative sample. This matters for interpreting the monolingual preferences reported below, which may rest less on ideological commitment than on the absence of any experience of well-designed translanguaging.

Preference and the persistence of monolingual beliefs

Asked about future preferences, most participants in both groups favoured continuing with combined Arabic and English feedback, with two refinements. Some proposed a staged model, with translanguaging feedback early in the academic journey shifting to English-only later once proficiency allows: “Because all my courses will be in English in the future, I think it is better to get feedback in English only later” (TL-4). Others wanted translanguaging specifically for grammar. One participant worried that future teachers who do not share Arabic simply could not provide such feedback, resigning him to translation.

A minority of English-only participants, however, maintained a principled preference for English-only feedback despite describing its costs. One stated simply: “I did not feel a need for Arabic. English was easy and it improved my writing” (EO-Q3). Another set out the reasoning at greater length:

Getting feedback in English can help us learn more new words and improve our vocabulary. It may be difficult at first, but it can help us improve our English over time. If we get feedback in both Arabic and English, it will be easier for us to understand. However, we may depend too much on Arabic, and this may not help us improve our English. I think using English more in the feedback can help us become better and understand more English words. (EO-1)

The second extract is worth examining closely, because its structure reveals the reasoning at work. The participant concedes that bilingual feedback “will be easier” and nonetheless rejects it, not despite the ease but because of it. Ease is treated as evidence of insufficient learning, and reliance on Arabic as a form of dependence that forecloses progress. The premise is that second-language development is a function of second-language exposure alone, such that any Arabic in the feedback represents English displaced. This is a coherent position, and it is the position that institutional English-only policy transmits (Alomaim and Altameemi, 2022). It is simply not one supported by the evidence on feedback comprehensibility.

A similar discrepancy is visible within a single participant's account. EO-1, who defended English-only feedback in the extract above, also reported in the same interview that “I had difficulty understanding the feedback on my grammar mistakes.” The endorsement and the difficulty are voiced by the same speaker, minutes apart, without any sense of contradiction. The same tension appears at group level: the English-only condition was rated favourably across all three questionnaire subscales, yet English-only participants were also the ones who described using dictionaries, translating repeatedly, and failing to correct errors even after translating (Section 4.2.3).

The two are reconcilable if positive ratings are read as endorsements of the condition's legitimacy rather than as reports of how easy it was to use. This suggests that learner satisfaction measures in feedback research may index ideological approval as much as experienced usability, and that they should not be treated as straightforward evidence that feedback was accessible.

Feedback on cohesive markers

Participants in both groups singled out feedback on cohesive markers as especially valuable, because much of it was new to them. TL-3 explained:

The feedback I received about my mistakes was very useful, especially the parts about using words like “in addition,” “subsequently,” and “following that.” These words were new to me, and I had never used them before. I did not know that I could use these words in my writing to connect my ideas and make my writing better. The feedback helped me learn new words and understand how I can use them in my future writing.

An English-only participant reported the same benefit: “The comments I received helped me connect one sentence to another by using important words that I did not know before. They helped me understand how to move from one idea to another in my writing. These words were new to me, and I think they will help me make my writing better” (EO-4). Participants also valued explanation of how articles and cohesive devices are used differently in Arabic and English writing.

That this theme emerged in both conditions indicates that it is attributable to the focus of the feedback rather than to its language. Its interest lies in what it reveals about which features benefit most from crosslinguistic explanation. Articles and cohesive markers are precisely the features that either lack functional equivalents in Arabic or are deployed according to different conventions. They are therefore features for which an English-only explanation must establish a category the learner does not yet have, while a translanguaged explanation can proceed by contrast with a category the learner already possesses. Participants’ enthusiasm for this feedback, combined with their explicit appreciation of Arabic and English contrasts, suggests that the value of translanguaging is greatest not for features that map straightforwardly between languages but for those that do not.

Three findings organise the discussion.

First, translanguaging WCF was perceived as significantly easier to understand and to apply than English-only WCF, corroborating perception studies of translanguaged feedback (Gomez, 2020; Saeli and Rahmati, 2023; Sun and Zhang, 2022) and consistent with the heteroglossic claim that mobilising the full repertoire reduces the cognitive load of processing complex information in the L2 alone (Goodman and Tastanbek, 2020). The qualitative data specify the mechanism: Arabic-mediated metalinguistic explanation allowed learners to cross-reference explanations with direct corrections, engage in meaning-making, and grasp crosslinguistic contrasts (articles, cohesion) that were otherwise opaque, providing scaffolding within the learners’ zone of proximal development (Saeli and Rahmati, 2023). Notably, the survey found no significant group difference in perceived improvement. The divergence between the two comprehension-related subscales, which separated the groups, and the improvement subscale, which did not, suggests that learners can report reliably on the experience of processing feedback but are poorly placed to judge what they have learned from it. This is a caution that applies to the wide use of learner satisfaction measures as proxies for feedback effectiveness. The qualitative data support this reading: participants described the process of understanding and applying feedback in specific and consistent terms, but spoke about their own improvement only in general ones.

Second, the English-only condition did not remove the L1 from the feedback process; it privatised it. All participants described translating incomprehensible feedback via dictionaries, Google, or YouTube, at considerable cost in time and with poor results for metalinguistic terminology. Given that feedback which is difficult to process often goes unused (Chen, 2018; Winstone et al., 2017), the practical choice facing teachers in shared-L1 contexts is not between monolingual and bilingual feedback but between principled, teacher-controlled translanguaging and unprincipled, learner-improvised translation. This reframing matters for the language-classroom setting specifically. The standard objection to L1 use in EFL classrooms is that it displaces target-language processing; the present data suggest that in the case of metalinguistic WCF, an English-only policy does not secure additional English processing but instead generates a private, low-quality translation episode that the learner conducts alone. Learners themselves marked the distinction with striking precision, praising feedback that “didn't just feel like translation” but read as “one explanation using English and Arabic together”. This is an experiential validation of the theoretical line drawn between pedagogical translanguaging and translation (Cenoz and Gorter, 2020, 2022) and of translanguaging's unitary-repertoire premise (García and Wei, 2014; Wei, 2018).

Third, the findings expose the resilience of monolingual ideology. A subset of English-only participants insisted that English-only feedback was optimal even while recounting the difficulties it caused them, reasoning that Arabic would breed dependence. This stance parallels learners elsewhere who prefer L2 feedback “for the sake of L2 learning per se ” (Kim and Chang, 2022, p. 1372; see also Saeli et al., 2023; Sun et al., 2023; Yu and Lee, 2014) and is best understood as internalised institutional policy: these students learn English in a language institute where English has always been the required classroom code, an environment that teaches that mastery is achieved only through English (Menken and Sánchez, 2019; Yeh, 2018). Past experience compounds ideology: learners’ only prior encounters with L1-inclusive feedback were literal translations, which they reasonably judged useless, and such experiences are known to shape feedback perceptions (Zhao, 2010). The contrast with the translanguaging group is instructive: its members, having experienced strategic translanguaging, overwhelmingly wished to continue with it. This suggests that exposure to well-designed translanguaging can shift evaluations where argument alone may not. Learners’ own proposal of a proficiency-staged model indicates a sophisticated, non-dogmatic stance that policy could accommodate, though it also preserves the assumption that the L1 is a scaffold to be dismantled rather than a permanent feature of bilingual writing competence.

Saudi EFL learners found teacher-provided translanguaging WCF significantly easier to understand and apply than English-only WCF, experienced it as clarifying rather than crutch-like, and could distinguish strategic translanguaging from the literal translation they had encountered and rejected in the past. At the same time, monolingual ideologies absorbed from prior schooling and from English-only institutional policy led some learners to champion English-only feedback in principle even as they struggled with it in practice. The way forward is not to argue learners out of these beliefs but to give them the experience these participants had: feedback in which their whole linguistic repertoire is treated as a resource, “used in your own style and your own way,” in the service of their English writing.