The Effect of Code Switching (Somali–Arabic) on Narrative Coherence in Contemporary Somali Short Stories: A Study of Mogadishu Based Literary Works

1. Introduction

This study investigates the impact of Somali–Arabic code-switching on narrative coherence in contemporary Mogadishu short stories, treating code-switching as a communicative medium rather than a deficiency (Schächinger Tenés et al., 2025). The urban corpus, shaped by civil wars, diaspora return, and Islamic revival, raises a key question: Does increased switching reduce referential clarity, temporal sequencing, and causal links or enhance coherence by signaling authenticity, social hierarchy, and psychological interiority? This switching is neither random nor gap-filling but a measurable coherence strategy (Zhou & Fu, 2025).

A thousand years of trade and religion created diglossia, where Arabic is sacred and Somali nationalistic, actively cultivated by Siad Barre’s 1972 Latin script reform. Since the state collapse in 1991, Gulf-sponsored Arabic teachers have reintroduced Arabic into Mogadishu’s literary production, making switching more common and foreshadowing contemporary short stories as palimpsests of changing language policies, religious authority, and urban life (Lu et al., 2026). Thus, coherence is contingent and historical.

Global code-switching studies favor written corpora and European–indigenous pairs (Spanglish, Hindi–English), assuming real-time processing capabilities. Arabic–vernacular research focuses on Maghrebi or Sudanese diglossia, where Arabic is a colloquial sibling rather than Somali’s typologically distant (Cushitic vs. Semitic) liturgical second language. This study challenges generalizations from European pairs to non-colonial liturgical contact situations (Chapwanya & Nel, 2025). Horn of Africa research prioritizes Amharic–English and Swahili–English, while Arabic–vernacular code-switching remains under-researched (Rizq et al., 2026). Mogadishu, the most active Somali-language publishing center since 2015, provides abundant corpora for frequency-to-coherence analyses.

Somalia’s unique ecology includes near-universal Somali L1 acquisition, widespread Arabic L2 oral competence, and script-switching ability (Latin-script Somali literacy, Arabic-script religious literacy) without regional parallels due to the 1972 Latin orthography reform. A literary renaissance began post-2000, with Somali-Latin writers and Gulf-returnees producing short stories where Arabic serves bilingual readers but is assessed by oral poetry standards (Maanso/Gabay), valuing associative over Western linear coherence (Salig et al., 2025).

Four narrative coherence dimensions (referential, temporal, causal, and thematic) were operationalized (Graesser et al., 2004), with code-switching measured as continuous frequency (per 1,000 words). The switching types include realism, characterization, metalepsis and foregrounding. Switching at moral climaxes may repair, rather than disrupt, coherence (Ray, 2025). This study combines Gumperz’s (1982) situational/metaphorical switching distinction with Berman and Nir’s (2010) Narrative Coherence Index and the Frequency Hypothesis, which posits that low and high frequencies detract from coherence, while moderate frequencies improve rhetorical variation (Waluyo & Khan, 2026). Myers-Scotton’s Matrix Language Frame model, adapted for written narratives, shows that lexical-level Arabic switches because little coherence drop, but clause-level switches exceeding 12 per 1,000 words cause significant drop unless quotation-bound.

Preliminary data indicate that high-frequency unrehearsed switching may disrupt narrative continuity (Lachemat et al., 2025). Systematic measurement of switching frequency correlated with coherence scores is absent, causing editorial uncertainty, limited audience reach and ideology-driven contestation. This study establishes the first empirical baseline using 30 purposively selected stories (2015–2025).

The novelty of this study includes the following: (i) the first continuous frequency measurement of Arabic code-switching in the Somali literary corpus; (ii) the first four-dimensional coherence disaggregation comparing high/low frequency with matched pairs; and (iii) the first taxonomy of literary functions for Cushitic liturgical-vernacular switching. The Frequency-Adjusted Coherence Model (FACM) predicts an optimum frequency window (12–18 switches/1,000 words) beyond which coherence drops. An annotated Somali–Arabic short story corpus was provided (Maskat et al., 2024), along with an editorial coherence audit checklist, pedagogical rubric, and evidence for Somali language policymakers (Singh et al., 2025; Lian, 2026).

This study focuses on Somali-Latin short stories (1,500–5,000 words) by Mogadishu-based authors (2015–2025), excluding poetry, novels, diaspora works, spoken genres, reader-response factors, and Somali–English switching. The five specific objectives were as follows: (1) measure code-switching frequency; (2) measure coherence levels; (3) assess frequency effects on coherence; (4) compare high vs. low frequency coherence; and (5) identify literary functions. This empirical research program brings the abstract “code-switching effect” into measurable, empirical territory, setting a methodological standard for African literary multilingualism research.

2. Literature Review

This review presents theoretical and empirical research on code-switching, coherence, and their intersection in multilingual literary contexts to develop a testable model for Somali–Arabic short stories (Rekun and Meir, 2024). With no prior Somali–Arabic literary coherence studies, the review moves from foundational theories to objective alignment, relational analysis, hypothesis development, gap identification, conceptual explications of frequency and coherence, and a synthesized framework. Code-switching and narrative coherence have received much isolated and combined research, but no systematic investigation exists for typologically different literary language pairs, and measurements are absent.

3. Research Methodology

The methodological framework employed in this section to examine the impact of Somali–Arabic code switching on the coherence of Somali short stories in contemporary Mogadishu is presented. This section includes a description of the study area, research design, target population, sample size, sampling techniques, data collection methods, variable measurement, pilot testing, data analysis, validity, reliability, ethical considerations, and limitations. This methodology aims to empirically generate a baseline for Somali–Arabic literary code-switching for the first time.

This study was carried out in Somalia’s most active Somali-language publishing hub, Mogadishu, since 2015. The linguistic ecology of Mogadishu is unique in that it has near-universal L1 Somali and high L2 Arabic proficiency from religious education, and the ability to switch between Somali and Arabic scripts. This study focuses on the literary short story in the Latin script written by authors in Mogadishu between 2015 and 2025, marking the beginning of the post-civil war literary renaissance, during which Arabic code switching became widespread and of stylistic importance. The total number of stories is estimated to be around 3,000 based on published records in literary magazines, online publications, and anthologies.

This study was descriptive, quantitative, and qualitative. The quantitative method involves measuring the level of code switching and calculating the multidimensional coherence scores before conducting regression, t-tests, and correlation analysis on the data. The qualitative part classifies switch functions for moderation analysis. This design examines whether the frequency-coherence relationship is curvilinear, as opposed to linear, and examines the moderating effects of switch type and position in the narrative.

The target population comprises 3,000 contemporary Somali short stories written in Somali but with the use of Latin script, written by residents of Mogadishu (at least two years), published from 2015 to 2025, prose narrative form (1,500–5,000 words), and with at least one non-nativized Arabic entry. This is based on a bibliographic survey of Qarawii magazine, Somali PEN anthologies, stand-alone collections and online sites. With 95% confidence and a 5% margin of error, Yamane’s formula yielded a sample size of at least 353 stories. To overcome the problems associated with corpus cleaning, the sample was expanded to a final sample of 400 stories, or 15% above the originally planned number. This size provides sufficient statistical power (≥ 0.80) to test moderate effects (Cohen’s d ≥ 0.4) in subgroups and regression.

The research frame constitutes the full list of all 3,000 stories that fulfilled the inclusion criteria and was created using systematic bibliographic identification, metadata extraction, author residency verification, and deduplication. The table was created in Excel using unique identifiers (ST001 to ST3000), author names, titles, sources, dates, and word counts. A story was included if it met the following criteria: written in Somali in the Latin script; authored by an individual who lived in Mogadishu for ≥2 years while composing the text; published between 2015–2025; was a prose short story of 1,500–5,000 words; included at least one non-nativized Arabic insertion; and the entire text was available for coding. A story was excluded from consideration if it was a novel excerpt, a poetry piece, written by a diaspora writer, lacked Arabic switches, was in Arabic script, was only nativized loans (e.g., kitaab), was under or over 1,500 words, was published prior to or subsequent to 2015, or was incomplete or corrupted.

A probability selection type of simple random sampling was used. The 3,000 stories were assigned individual numbers. A computer-generated random sequence generated 400 subjects. The corresponding stories were retrieved from their source. This method avoids selection bias, permits generalization to the 3000 stories in the population, and permits inferential statistics without complicated weighting.

A pair of complementary instruments was used: First, the frequency data were recorded on a coding sheet developed by the researchers that included information on story identifier, word count, absolute number of Arabic switches, switch type, switch function, quotation-bound status, and narrative position. The sheet was designed using Microsoft Excel with dropdown menus. Second, a rubric for narrative coherence was adapted from Berman and Nir (2010) and used 5-point Likert scales for four dimensions. Five trained bilingual raters independently rated each story as follows: four sub-scores (each 0–25) were added together to give a composite score of coherence (0–100). The pilot stories yielded Fleiss’ κ = 0.78 for training.

The number of non-nativized Arabic insertions in 1,000 words (ratio scale) was used to measure code-switching frequency. The frequency was also grouped as low (<8), medium (8–22), or high (>22) switches/1,000 words. The switch type and function were nominal (categorical) variables. The status of being in quotation marks was either on or off. The narrative coherence score was an interval variable (0–100 composite). The four sub-scores (referential, temporal, causal, and thematic) were interval variables (0–25 per sub score). The following covariates were included in the model: story length (ratio), publication year (interval), and author (nominal random effects).

Primary source data were employed throughout because there was no existing dataset for Somali–Arabic literary code switching. Three sources of data were used: the original short story texts drawn from archives, coding sheets developed by the researcher, and coherence scores assigned by the raters. All data were entered into an Excel document, transferred into the SPSS v31 software, and then stored in a secure encrypted document that was password-protected. The pilot test was conducted on 30 randomly selected stories (not included in the main sample) and was completed four weeks prior to the main data collection. The results showed that (1) there was ambiguity in nativized versus non-nativized switches, which was resolved by a decision tree with 15 examples; (2) stories longer than 4,000 words resulted in rater fatigue (split into two sessions); (3) 12% of the switches were unclassifiable (added “other” category and clarified the definition of metalepsis); and (4) rating time was reduced from 18 to 12 minutes per story after rubric revision and retraining. The pilot confirmed the feasibility, reliability (κ ≥ 0.78), and validity of the tool.

Written informed consent was obtained from all five bilingual raters prior to their participation in the coherence evaluation. Each rater received a detailed consent form explaining the study’s purpose, the nature of the rating task, the estimated time commitment (approximately 12 minutes per story across multiple sessions), and their right to withdraw at any time without penalty. The consent form also specified that all ratings would be anonymized, that no individual rater’s scores would be identifiable in any publication, and that data would be used solely for academic research purposes. Consent was documented via signed paper forms, which were stored separately from the rating data in a locked cabinet accessible only to the corresponding author. Written consent was chosen over verbal consent to ensure a clear, auditable record of voluntary participation, to comply with Salaam University Research Ethics Committee (SU-REC) requirements for studies involving human judgment tasks with potential for rater fatigue or perceived evaluative pressure, and to provide legal documentation of the terms agreed upon by both parties. No vulnerable populations were involved, and no compensation was provided to raters beyond acknowledgment in the manuscript.

SPSS v31 and NVivo 14 were used for data analysis. Based on descriptive statistics (mean, SDs, range, and frequencies), frequency distributions and coherence profiles were determined. The means for the dimensional coherence were compared using paired t-tests. To test the curvilinear relationship between frequency and coherence (H3), polynomial regression was used, followed by an F-test of ΔR². Matched paired t-tests compared high frequency (>22/1,000 words) and low frequency (<8/1,000 words) stories (matched by length ± 500 words and by year ±2 years) and Cohen’s d effect size. Moderation analysis (Model 1 of PROCESS v4.3) was used to determine whether the switch function and quotation-bound status moderated the frequency–coherence effect (H5). Other procedures used were the Shapiro–Wilk normality test, Levene’s test, winsorizing outliers (>3 SD), and Bonferroni (α = 0.0125 for dimensional tests) and α = 0.05 with 95% CIs for primary hypotheses.

For internal validity, controlled sampling, operational definitions, and rater blinding were used. Full-period and probability sampling was used to ensure external validity, which led to generalization to the 3,000-story population. Construct validity was assessed using standard measures (switches/1000 words; adapted Berman & Nir rubric). The four dimensions of content validity were assessed by expert review. Face validity was achieved through pilot testing. Interrater reliability (Fleiss’ κ) = 0.78–0.84 (threshold ≥0.70). The intra-rater reliability varied from 0.88 to 0.92 (thresholds ≥0.85). The internal consistency (Cronbach’s α) was 0.86 (≥0.80). The split-half reliability (Spearman-Brown) was 0.83 (threshold ≥0.80). The agreement between the coding on the test and test-retest was 97.2% (threshold ≥95%).

4. Results and Discussion

This section presents the findings of this study on the effect of Somali–Arabic code-switching on narrative coherence in contemporary Somali short stories from Mogadishu. The results are organized according to five specific objectives and their corresponding hypotheses. Quantitative data were analyzed using SPSS Version 31 and NVivo 14, based on a sample of 400 short stories randomly selected from a population of 3,000 stories published between 2015 and 2025.

5. Discussion

This study provides the first empirical evidence of Somali–Arabic code-switching and narrative coherence in Mogadishu short stories. The mean frequency was 14.8 switches per 1,000 words, with intra-sentential switches (62.4%) supporting Myers-Scotton’s Matrix Language Frame model. Referential coherence (16.2) was significantly lower than temporal (17.4) and thematic coherence (16.9), partially supporting H2. The inverted U-shaped relationship supports the Frequency Hypothesis, with optimum coherence at 13–14 switches per 1,000 words. Coherence increased at moderate frequencies but declined sharply beyond 22 switches.

Low-frequency stories (68.4) scored 14.2 points higher than high-frequency stories (54.2, Cohen’s d = 1.65, p < 0.001). Foregrounding (β = 6.42) and characterization (β = 3.15) were positively associated with coherence. Quotation-bound status moderated the frequency–coherence effect (β = 4.28, p < 0.001), with quotation-bound religious switches showing 7.6 points higher coherence. Authors should aim for 12–18 switches per 1,000 words and avoid exceeding 22.

6. Conclusion

This study provides the first empirical baseline data on Somali–Arabic code-switching (14.8 switches/1,000 words) and validates the curvilinear (inverted U-shaped) relationship between code-switching and narrative coherence, with the optimal number of switches being 13–14 per 1,000 words. The coherence rises at moderate frequencies but falls precipitously beyond 22 switches. Referential coherence was the most sensitive, and temporal cohesion was the most robust. Low-frequency stories (68.4) earned a higher score than high-frequency stories (54.2, Cohen’s d = 1.65). The coherence–frequency interaction was positively moderated by foregrounding and characterization switches with quotation-bound status. Moderate and functional code-switching allows for coherence to be maintained, while excessive code-switching causes it to be lost.

7. Recommendations

The authors are expected to include 12–18 Arabic switches in 1000 words, not include more than 22 switches, and focus primarily on quotation-bound and foregrounding switches. The four-dimensional coherence rubric should be used, and stories with more than 22 switches should be flagged by the editor. Publishers must set standards for frequency and educate editors on evidence-based assessments. The Frequency-Adjusted Coherence Model should be incorporated into creative writing education. Policymakers should move from ideologies to functional coherence criteria that help support and promote authentic bilingual literary practices.

Contributions of the Study

Empirical Contributions

This study offers the first frequency baseline for Somali–Arabic literary switching (mean = 14.8), the first multidimensional coherence profile for Somali short stories, and the first annotated corpus of 400 stories with replicable metadata.

Methodological Contributions

This study provides an operationalization of the concept of frequency (low <8, medium 8–22, high >22); is based on a four-dimensional rubric (Fleiss’ κ = 0.78–0.84); illustrates polynomial regression; and shows matched-pair t tests.

Policy Contributions

This study offers quantitative benchmarks for assessing multilingual works and considers code switching as a valid literary approach in non-formal education.

Practical Contributions

This study provides a coherence audit checklist, a pedagogical rubric for creative writing, and a validated questionnaire and coding sheet that can be adapted for other multilingual settings.

Recommendations for Future Research

Future studies should include other genres of literature, other Somali regions, and diaspora literature. The model should be evaluated on other language pairs (Somali–English and Swahili–Arabic). Experimental research should be conducted to establish causation. Studies focusing on readers should assess their comprehension. Longitudinal studies should measure changes over time. Typological distance effects should be discussed in cross-language comparison studies.

Limitations of the Study

Findings are limited to short stories (1,500–5,000 words) written by authors from within the Mogadishu community between 2015–2025, and do not apply to poetry, novels, or diaspora literature. Coherence was rated by five bilingual experts but not assessed through reader-response and/or eye tracking. Nativized loans (e.g., kitaab) were omitted, perhaps underestimating the Arabic influence. Typologically distant language pairs (Somali–Cushitic and Arabic–Semitic) may not be applicable to closer pairs. The 13.3% sample (400 of 3,000 stories) had a 5% margin of error. Raters were over-represented in the formal literacy and university-educated categories. The observational design precludes the ability to make clear causal inferences.

Ethical Approval Statement

This study was conducted in accordance with the internationally recognized ethical principles for research involving human participants, as outlined in the Declaration of Helsinki. Ethical approval was obtained from the Salaam University Research Ethics Committee (SU-REC), Center for Research and Development. The approval details are as follows: Approval Reference Number: 2025/SU-REC/AMSHS/P0394; Ethical Approval Date: August 17, 2025. The research protocol, including the questionnaire, informed consent procedures, data handling protocols, and participant protection measures, was reviewed and approved by the committee prior to commencing data collection. All research activities were conducted in strict adherence to approved protocols.

Informed Consent

Informed consent was obtained from all participants in this study. Before completing the survey, each participant was presented with a detailed digital consent form outlining the study’s purpose, procedures, potential risks and benefits, and the voluntary nature of participation. The consent form explicitly stated that the data would be used solely for academic research and that all responses would remain anonymous and confidential. Participants were informed of their right to withdraw from the study at any time, without penalty. Consent was obtained electronically before proceeding to the questionnaire, ensuring that participation was transparent, informed, and voluntary, in accordance with the ethical guidelines of the Salaam University Research Ethics Committee (SU-REC).

Approval to Publish

The author confirms that this work is original, has not been published elsewhere, and is not currently under consideration by any other journal. The author grants the publisher a license to publish this study in any form. The author has reviewed and approved the final version of this manuscript for submission to this journal.

Consent to Publish

Not applicable.

Clinical Trial Registration

Not applicable.