Language ideology and lexical borrowing: attitudes toward Arabic loanwords among Swahili-speaking academics

1 Introduction

Hundreds of years of rich contact between Arabs and East Africans have led to enormous changes, not only in the ethnicity and culture of the Bantu peoples, but also in their languages and linguistic ideologies. Despite the fact that Arabic and Swahili belong to two distinct language families, a lexical borrowing relationship is evident between the two languages. Based on several independent studies, Arabic loanwords constitute from 18% (Hurskainen, 2004) to 40% (Johnson, 1939) of Swahili's lexical inventory. Regardless of the precise figure, both studies confirm that Arabic has contributed a significant portion of Swahili's lexicon. Such long-standing contact between the two languages has prompted Swahili speakers to borrow a large number of words from Arabic and adapt them phonologically and morpho-syntactically so that they function as Bantu words.

A number of studies have been carried out on Arabic loanwords in Swahili, investigating various linguistic aspects, among them the phonological and morphological adaptation of borrowed forms, their distribution across semantic fields, and their share of the Swahili lexicon (Bosha, 1993; Hurskainen, 2004; Mwita, 2009; Tsvetkov et al., 2015). However, little is known about any empirical study that has investigated Swahili speakers' covert and overt attitudes toward Arabic loanwords. An understanding of speakers' attitudes can help predict the outcomes of linguistic contact. Studying linguistic ideologies further enhances our understanding of how and why such contact outcomes have come about (Rodríguez-Ordóñez, 2019; Thomason, 2001). Extensive research has also shown that studying speakers’ attitudes has several implications for the fields of bilingualism, language contact, and sociolinguistics (Baker, 1992; Garrett, 2010). As Garrett (2010) notes, attitudes constitute an essential component of sociolinguistic competence, influencing not only language use but also language change and maintenance. Speakers' attitudes toward a donor language naturally affect borrowing (Rüdiger, 2018).

The current study aims to determine how far Arabic loanwords are conventionalized in participants' Swahili, and to examine the covert and overt attitudes speakers hold toward them. Because attitudes toward a loanword are best interpreted in light of how entrenched it is, the study addresses conventionalization first and then tackles speakers' attitudes. Accordingly, it puts forth the following three research questions:

• To what extent are the target Arabic loanwords conventionalized in participants' Swahili?
• What covert attitudes are linked to Arabic loanwords among speakers of Swahili?
• What overt attitudes are linked to Arabic loanwords among speakers of Swahili?

2 Literature review

2.1 Swahili–Arabic contact

Swahili is a well-known Bantu language of Eastern and Central Africa. It is spoken by more than 90 million people as a second language (Eberhard et al., 2023). According to Rothman (2002, p. 79), most Swahili speakers today are located in Tanzania, Kenya, and Uganda. The language serves a wide range of functions; it is spoken by Bantus at home, work, school, and at religious and political meetings. It is the daily language of radio, television, newspapers, and journals in the East African Community (Schadeberg, 2009, p. 76). Although Swahili is the national language of Tanzania and is spoken by a large number of Bantus in other countries, it does not actually belong to a particular country. While Ekwelie (1971) characterized Swahili as a lingua franca with no native speakers, this characterization is now considered outdated. Contemporary estimates indicate approximately five million first-language speakers (Eberhard et al., 2023), with native Swahili-speaking communities concentrated along the East African coast and on islands such as Zanzibar. Ekwelie's broader point still holds, however: the term Swahili describes the cultural group that belongs to the coastal civilization of East Africa rather than a particular ethnic group.

Arabic, by contrast, is a Semitic language spoken mainly in the Middle East. According to Aoun et al. (2010, p. 1), Arabic emerged in the Arabian Peninsula and later spread widely with Islam to cover most of North Africa, the Middle East, and other Muslim territories. According to Ethnologue, Arabic is spoken natively by over 300 million people, placing it among the most spoken languages in the world (Eberhard et al., 2023). It is the official language of 20 Middle Eastern and African countries (Holes, 2004, p. 2) and the language of the Quran (the holy book of Islam); thus, all Muslims, regardless of their origins and mother tongues, are in consistent contact with Arabic (Holes, 2004, p. 1). In most Arabic-speaking countries, classical Arabic is mainly used in formal settings (e.g., writing or reading a sacred or formal text, or narrating religious stories), whereas local varieties are used for everyday communication (Soliman, 2008, p. 12).

Swahili has historically been in intense contact with many external languages, such as English, Hindi, and Portuguese. Nevertheless, Arabic has been the most prominent donor language throughout the history of Swahili. Even the term “Swahili” is derived from the Arabic word swahil, the plural form of sahil, “coast” (Polomé, 1967, p. 8). Swahili people were traders on a global scale. The intensive contact between Swahili coastal towns and foreign visitors and traders affected the language enormously, and among all donor languages in the Indian Ocean trading network, Arabic was the most prominent (Rothman, 2002, p. 79). The map in Figure 1 shows the geographical setting of Swahili along the East African coast and the surrounding Indian Ocean trade network. The coastal positioning of the major Swahili-speaking centers, from Mogadishu and Lamu in the north to Kilwa and Sofala in the south, illustrates why these communities were so heavily exposed to Arab maritime traders and helps explain the depth of Arabic influence on the language.

Arabs were the most frequent visitors during this period of Indian Ocean trade (c. 800–1920); indeed, some settled in East Africa for the rest of their lives (Polomé, 1967, p. 9). After the Omani forced the Portuguese to leave Swahili towns in the middle of the 17th century, Arabs took over all Swahili territories and became a major power on the East African coast (Polomé, 1967, p. 12). Importantly, in the early 19th century, Sultan Syyid Said, the ruler of Oman, moved the capital from Muscat to Zanzibar, a Swahili town. During this period, the variety of Swahili spoken in Zanzibar, “Kiunguja”, was in intense contact with Arabic, consequently, a tremendous number of Arabic loanwords leaked into Swahili (Schadeberg, 2009, p. 83).

Religion was a further channel of contact. Thomason (2001) highlights that languages may come into contact through the spread of religion. Islam has spread widely across the world; as Holes (2004, p. 4) reports, more than 90% of Arabs are Muslims. Islam first arrived in Swahili society through Muslim visitors and immigrants, and by the 12th or 13th century it had become the dominant religion in many Swahili-speaking areas (Schadeberg, 2009, p. 82). The dominance of Islam and the political power of Arabs in East Africa together established Arabic as the donor language for Swahili and many other Bantu languages (Polomé, 1967). Because the Quran was written in classical Arabic, Muslim Swahili speakers found themselves needing to learn Arabic in order to practice their religion—to recite the Quran, perform prayers, and read the prophet's sacred words.

2.2 Arabic loanwords in Swahili

Although Arabic and Swahili are typologically distinct languages with different phonological and morpho-syntactic systems, the intense, long-standing contact between them led to a substantial number of borrowings in the replica language, Swahili. Several studies have investigated these loanwords. As reported in Tsvetkov et al. (2015), Johnson (1939) carried out an independent study and found that Arabic loans represent about 40% of Swahili vocabulary, while Hurskainen (2004), using the Helsinki Corpus, found that they constitute 18% of the Swahili lexical inventory. Because Hurskainen (2004) examined only nine text types, mostly from formal or religious contexts, its limited scope may not capture the full range of Arabic loanword usage, which could explain the disparity with Johnson (1939), whose estimate came from a full dictionary count. In his detailed investigation, Schadeberg (2009, p. 88) explored Arabic loanwords in Swahili and found that borrowing was concentrated in particular semantic domains. As Table 1 shows, the most borrowable meanings cluster around religion and belief, possession, and law, domains closely tied to the historical channels of Arab–Swahili contact described above.

Table 1. Top five most borrowable meanings in Swahili (Schadeberg, 2009, p. 88).

Each value indicates the percentage of words within that semantic field that are borrowed from Arabic.

Table 1 aligns with expectations given the historical role of Islam and trade in this contact situation, with religion and belief predictably leading as the most borrowable semantic field. The table also shows that Swahili has borrowed from Arabic both cultural items (words pertaining to religion, possession, and law) and non-cultural items (words related to cognition and emotions). In terms of word class, Schadeberg (2009, p. 88) illustrates that content words (nouns and adjectives) constitute most Arabic loanwords in Swahili, as shown in Table 2. This point was also made by Thomason (2001) and Field (2002): non-basic words and content items are more subject to borrowing than function or basic words. The presence of Arabic function words in the Swahili lexical inventory is nonetheless a significant indicator of the intensity of contact between the two languages, according to Thomason's model of borrowability.

Table 2. Arabic loanwords class distribution (Schadeberg, 2009, p. 88).

Each value indicates the percentage of words within that word class that are borrowed from Arabic.

Loanwords in any replica language often undergo phonological adjustment to be adapted as native words. Arabic loanwords likewise underwent both phonological and morphological reshaping to sound more like other Bantu words. As Mwita (2009) and Tsvetkov et al. (2015) note, Arabic loanwords in Swahili were subject to four phonological constraints: (1) substitution of Arabic phonemes absent from Swahili with similar Swahili phonemes; (2) vowel epenthesis to break up consonant clusters; (3) consonant degemination, since geminates are not permitted; and (4) final vowel substitution, since syllables must end in a vowel. Table 3 provides an illustrative derivation, replicated from Bosha (1993), showing how these rules apply to three Arabic forms to yield their Swahili surface representations.

Table 3. Phonological adaptation of selected Arabic loanwords (Bosha, 1993).

UR, underlying representation; SR, surface representation; C, consonant; V, vowel; Ø, null (deletion or epenthesis site); #, word boundary; Cα, a consonant identical to the following consonant.

Arabic loanwords were also morphologically integrated. As Schadeberg (2009, p. 91) notes, borrowed nouns in Swahili take additional morphemes, such as the prefix –ma, added to singular nouns to form the plural, while other bound morphemes attach to borrowed verbs, such as the causative suffix -ish and the verbalizing final-vowel suffix –a. Table 4 presents representative examples of these morphological adaptations.

Table 4. Morphological adaptation of selected Arabic loanwords (Schadeberg, 2009).

N.SG, singular noun; N.PL, plural noun; V, verb; V ACC, verb (here, an action/eventive verb); ADJ, adjective.

According to Schadeberg (2009, p. 91), borrowed adjectives are morphologically less integrated than nouns and verbs. Native adjectives in Swahili take a nominal prefix agreeing with the class of the head noun, as in “kitambaa kichafu”/“vitambaa vichafu” for “a clean cloth”/“clean cloths”, respectively, whereas such agreement does not hold for Arabic borrowed adjectives, as in “kitambaa safi”/“vitambaa safi” (Schadeberg, 2009, p. 91). Furthermore, as Mpiranya (2015, p. 84) reports, Swahili adds the marker -ni for “you all” to some Arabic verbs ending in/e, i, u/, as in the imperative karibu(ni) “welcome (you all)”, from the Arabic imperative karibu. Taken together, these phonological and morphological adaptations demonstrate that Arabic loanwords have been thoroughly nativized, reshaped to conform to Swahili's sound system and grammatical structure, which is itself a strong indicator of their deep integration into the language.

2.3 Language ideology and speaker attitudes

This study is grounded in the theoretical framework of language ideology developed by Woolard and Schieffelin (1994) and Kroskrity (2004). Language ideology, broadly defined, refers to the sets of beliefs and conceptualizations about language that users articulate to rationalize or justify perceived language structure and use (Silverstein, 1979). According to Woolard and Schieffelin (1994), language ideologies mediate between social structures and forms of talk. In the context of language contact and borrowing, ideological orientations toward donor languages can shape the degree to which loanwords are accepted, resisted, or conventionalized within the recipient community. Backus (2014) describes loanword integration as a gradual process in which borrowings shift from enrichment, where they are recognized as foreign additions to the lexicon, toward conventionalization, where they become so entrenched that speakers perceive them as native vocabulary. Speakers' attitudes play a crucial role in this process, shaping whether borrowings are resisted or accepted on their way to conventionalization.

Preston (2010) argues for the importance of studying what he terms “language regard”, encompassing both attitudes and ideology, because it yields detailed insights into linguistic variation and change. Baker (1992) similarly maintains that understanding speakers' attitudes is pivotal to questions of language maintenance, preservation, and shift, serving as an indicator of community beliefs, preferences, and desires. To capture both the conscious and unconscious dimensions of these orientations, the present study employs the distinction between covert and overt attitudes. Covert attitudes, usually elicited through indirect methods such as the matched guise technique (Lambert, 1967), reveal implicit evaluative reactions that speakers may not openly acknowledge. In this paradigm, such reactions have been shown to cluster around two dimensions: competence (e.g., intelligence and education) and social attractiveness or solidarity (e.g., kindness, sincerity, and religiosity) (Giles and Billings, 2004; Zahn and Hopper, 1985). Overt attitudes, by contrast, are assessed through direct self-report and reflect speakers' conscious positions. Any discrepancy between the two can be particularly informative about the sociolinguistic dynamics of a community (Garrett, 2010).

Empirical work on loanword attitudes illustrates how these constructs operate in practice. In a questionnaire study of South Korean students' attitudes toward English loanwords, Rüdiger (2018) found mixed, often negative evaluations, alongside a range of perceived reasons for loanword use. Because that study relied on direct, self-report questionnaire measures, it captured speakers' overt attitudes but not their covert ones, a gap the present study addresses by pairing direct questioning with an indirect matched guise task. In his analysis of Catalonian Spanish, Davidson (2015) examined speakers' attitudes toward intervocalic/s/ voicing and velarization of/l/. He found that the positive attitudes associated with linguistic cosmopolitanism were not equally distributed across variants. Unlike Rüdiger, Davidson investigated both covert and overt attitudes by combining a matched guise questionnaire with debriefing interviews. Such ideologies are rarely uniform across a speech community; Field (2010), for example, shows that younger, more fluent speakers of Navajo code-mix with English more frequently than older speakers, indicating that age is a significant factor in shaping attitudes and usage. These observations align with Rodríguez-Ordóñez's (2019) argument that linguistic ideologies should be brought to the foreground of contact linguistics, since they bear not only on “what” contact outcomes occur but on “how” and “why”.

3 The current study

Building on the theoretical foundations outlined above, the present exploratory study examines Swahili speakers’ covert and overt attitudes toward Arabic loanwords. Although a considerable body of work has documented the linguistic properties of Arabic loanwords in Swahili, far less attention has been paid to how speakers evaluate these loanwords or what attitudes are associated with their use. The present study aims to address this gap by answering the following three research questions: (1) To what extent are the target Arabic loanwords conventionalized in participants' Swahili? (2) What covert attitudes are linked to Arabic loanwords among speakers of Swahili? and (3) What overt attitudes are linked to Arabic loanwords among speakers of Swahili?

To address these questions, the present study adopts the combined covert–overt design pioneered by Davidson (2015). Davidson combined an indirect instrument with direct questioning on the same participants: the first captures evaluations speakers may not state openly, the second the reasons they give for them. Convergence or divergence between the two then becomes part of the evidence. The research first establishes the degree of conventionalization of the target loanwords through a set of usage-based tasks adapted from Backus (2014). It then combines a matched guise experiment (for covert attitudes) with direct elicitation and open-ended metapragmatic questioning (for overt attitudes). Together, these various methods provide a more comprehensive account of speakers' attitudes than any single method could offer alone. In doing so, the study aims to contribute much needed empirical evidence on Swahili speakers' attitudes toward Arabic loanwords and, more broadly, to deepen our understanding of how speaker attitudes shape the outcomes of language contact.

4 Methods

The data for this study were collected through an online questionnaire administered in English. English was chosen because participants were originally from three countries and different regional varieties, and because it was the only language they shared with the researcher. The choice carries a cost as all participants were graduate students working routinely in English, so comprehension is unlikely to have been a limiting factor, but responding in a second language may still have shortened the open-ended answers. The target items themselves were presented in Swahili, so the loanword manipulation was unaffected. The questionnaire was organized into three sections, each serving a distinct purpose. The first section addressed the first research question (conventionalization) through a set of usage-based tasks (three translation tasks and one judgment test), establishing a necessary precondition for the analysis. This section taps into the extent to which the target Arabic loanwords are already conventionalized in participants' Swahili. Establishing this baseline matters because attitudes toward a loanword can only be meaningfully interpreted once its degree of entrenchment is known. The second section addressed the second research question (covert attitudes) by means of a text-based adapted matched guise experiment. This indirect technique allows speakers' implicit evaluations to surface through their judgments of the two text versions, rather than through self-report. The third section addressed the third research question (overt attitudes) through three open-ended metapragmatic questions. These open-ended questions allowed participants to express their conscious, openly stated attitudes in their own words. Each of these three components is described in detail in Section 4.2.

4.1 Participants

Of the 18 respondents who started the survey, only nine completed it. Seven of the nine were originally from Kenya, while the other two were from Tanzania and Ghana. Slightly more male (5) than female (4) participants took part, and most were graduate students or university academics, aged 27–45. Regarding their Swahili background, seven had acquired the language before the age of 7, while the remaining two acquired it between 15 and 20. The participants varied in age, gender, and nationality, but the sample was small and weighted toward highly educated Kenyans. The study is therefore best understood as exploratory, since nine participants cannot support generalizable conclusions or provide adequate statistical power. Nonetheless, small-sample, exploratory designs of this kind have an established place in sociolinguistic and applied linguistics research (cf. Duff, 2008).

4.2 Instruments

The questionnaire comprised three main components, corresponding to three experimental techniques. The first section adopted a usage-based approach (Backus, 2014) to gauge the degree of entrenchment and conventionalization of Arabic loanwords; the second elicited speakers' covert attitudes through a text-based adapted matched guise experiment; and the third tapped into speakers' overt attitudes through three open-ended questions.

The same set of ten Arabic loanwords was used in both the usage-based tasks (Section 4.2.1) and the adopted matched guise experiment (Section 4.2.2) so that the analysis of conventionalization (RQ1) and the analysis of covert attitudes (RQ2) would be based on the same lexical items. These ten words were chosen for two reasons. First, each has at least one non-Arabic variant in Swahili, which makes it possible to contrast the two variants. This is essential for both the translation tasks, where participants may select either variant, and the matched guise texts, where one variant is substituted for the other. Second, according to the World Loanword Database (Haspelmath and Tadmor, 2009), five of the words represent the most borrowed meanings in Swahili and five the least borrowed, providing a spread of high and low-borrowability items against which entrenchment can be examined. Table 5 lists these ten loanwords together with their Swahili variants and English glosses.

Table 5. Semantic fields of the 10 target Arabic loanwords.

The first five rows are the most borrowed meanings in Swahili; the last five are the least borrowed meanings.

4.2.1 Usage-based approach

The usage-based section was designed to establish how far Arabic loanwords are conventionalized before speakers' attitudes were examined. It combined three translation tasks with one judgment test. The three translation tasks formed a graded sequence. In Task 1, participants translated the 10 English target words into their variety of Swahili, providing the first equivalent that came to mind. In Task 2, they were asked to provide any additional Swahili words they know for the same 10 meanings, which made it possible to see whether an Arabic loanword and its Swahili variant were both available to a speaker. In Task 3, participants translated the same 10 meanings again, this time embedded in full sentences so that loanword knowledge could be assessed in context rather than in isolation. Comparing the three tasks allows the analysis to distinguish a speaker's first-choice variant from their full productive repertoire. Finally, the judgment test asked participants, in a single multiple-choice item, to indicate whether each of the 10 Arabic loanwords was borrowed and, if so, from which language, thereby probing their metalinguistic awareness of the words' origins. A loanword that speakers no longer recognize as borrowed can be taken as more deeply entrenched in the recipient lexicon.

4.2.2 Matched guise task (covert attitudes)

To elicit speakers' covert attitudes indirectly, a text-based adapted version of the matched guise experiment (Lambert, 1967) was implemented. The classic matched guise technique employs audio recordings of the same speaker using different language varieties; the present study adapted this design to a written format in which participants read rather than listened to the stimuli. Although audio can be embedded in online surveys, a written format was chosen deliberately. Because the variable under study is lexical rather than phonetic, a written medium isolates the loanword contrast while avoiding confounds from voice quality or accent. It also kept the survey relatively short and limited participant fatigue. This adaptation has precedent in literature (e.g., Garrett, 2010). The author nonetheless acknowledges that reading permits more conscious processing than listening and may partly blur the covert/overt distinction. The adaptation is therefore not equivalent to the classical paradigm, in which auditory stimuli are presented at the speaker's pace and elicit more automatic reactions. The evaluations reported in Section 5.2 are accordingly best described as indirect rather than fully covert, and the term should be understood in that weaker sense throughout.

Participants read two short Swahili texts (one at a time) and rated each on a series of Likert and semantic-differential scales. The two text versions were short (around 60 words) and almost identical in content and structure; the only difference was that the first version contained 10 Arabic loanwords (see Table 5), which were replaced with their Swahili variants in the second. After reading Text 1 (the version containing the 10 Arabic loanwords), participants rated three statements on a 1–7 scale: (1) The person who wrote this text looks educated and knowledgeable, (2) The person who wrote this text looks religious, and (3) I identify with the person who wrote this text. They then rated the same items again for Text 2 (the version in which the loanwords were replaced with their native Swahili variants).

The two texts were written for this study rather than drawn from existing material. A single passage was composed around contexts in which all ten target meanings could occur; the second version was then produced from it by substituting each Arabic loanword with its Swahili variant, leaving everything else unchanged so that the loanword contrast is the only systematic difference between the guises. Both text versions were checked by a Swahili-speaking professor to confirm that they were natural, grammatically accurate, and equivalent in meaning apart from the target loanword substitutions (see Appendix A). The three evaluative statements and semantic-differential scales were adapted from established matched guise instruments (Lambert, 1967; Zahn and Hopper, 1985). The section closed with two comparison statements, each rated on the same 1–7 scale: (1) The writer of Text 1 is more likely to be a native speaker of Swahili than the writer of Text 2, and (2) The writer of Text 2 is more likely to be a native speaker of Swahili than the writer of Text 1.

All participants read Text 1 before Text 2, so the presentation sequence was not counterbalanced, and any difference between the two guises may reflect an order effect in addition to the loanword manipulation. This bears directly on the internal validity of the comparison. Having already rated one passage, participants met the second with criteria they had just applied, and the close similarity of the two texts makes it likely that the manipulation became apparent on the second reading; ratings of Text 2 may therefore reflect a comparison with Text 1 rather than an independent judgment, with fatigue and practice running in the same direction. Because the design cannot separate these effects from the loanword manipulation, the difference reported in Section 5.2 cannot be attributed to the loanwords alone.

4.2.3 Metapragmatic commentary (overt attitudes)

The third section followed the direct approach used by Rüdiger (2018) and Davidson (2015), in which speakers are asked to comment explicitly about their language attitudes towards Arabic loanwords. This approach elicits overt attitudes, the views speakers can consciously articulate, together with the metapragmatic commentary through which they rationalize and evaluate loanword use. Three open-ended questions were posed: Each question was designed to tap a distinct facet of overt attitude. The first gauges participants' immediate affective reaction once the scale of Arabic borrowing is made explicit, providing a direct measure of acceptance or resistance. The second invites participants to articulate the perceived reasons behind loanword use, yielding metapragmatic commentary on the social and historical motivations they associate with Arabic loanwords. The third addresses the relationship between loanwords and group identity, probing whether speakers construe Arabic borrowing as a threat to, or an enrichment of, Swahili. Because the questions were open-ended, participants could respond freely and at whatever length they wished.

• Are you aware that Arabic loanwords are estimated to represent between 18% and 40% of Swahili words? How do you feel about that?
• Why do you think many Swahili speakers use Arabic loanwords?
• Do you think Arabic loanwords might threaten Swahili identity?

4.3 Procedures

An online survey was distributed in English through the researcher's network of Swahili-speaking colleagues, who are mostly academics, along with a brief description of the study. Recruitment therefore proceeded by convenience sampling, with an element of snowball referral as initial contacts passed the link to other Swahili speakers known to them. Participants were able to finish the survey in approximately 20 min and were given the choice to withdraw at any time without penalty. Since one of the study's aims is to examine covert attitudes, respondents were not informed of the study's actual purpose. At the end of the survey, participants were debriefed about this deception and given the freedom to share or withhold their data. The study received ethical approval from the King Abdullah International Medical Research Center (KAIMRC) Institutional Review Board, and all participants provided informed consent prior to participation.

4.4 Data analysis

The three instruments were analyzed separately and then read against one another. Responses to the translation tasks were coded for each of the ten target meanings as either the Arabic loanword or a Swahili variant and aggregated across participants. The judgment task was tallied by the language each participant named as the source. Matched guise ratings were averaged across participants for each of the five items. Because nine participants cannot support inferential testing, no significance tests were run, and all quantitative statements in this paper are descriptive.

The open-ended responses were analyzed through inductive thematic categorization. The author read the full set of responses to each question, assigned provisional labels to recurring content, and grouped those labels into the categories reported in Section 5.3. These categories were derived from the responses rather than set in advance. Coding was carried out twice by the author, with an interval between passes. The few discrepancies concerned borderline cases between neutral and positive evaluation and were resolved by returning to the wording of the response. As the study has a single author, no second coder was available and no inter-rater reliability statistic was computed. Trustworthiness therefore rests on the coding frame being reported in full, the categories being few and close to the surface content of the responses, and participant quotations being reproduced verbatim in Section 5.3.

5 Results

5.1 Usage-based results

The three translation tasks together indicate that the target Arabic loanwords are, to varying degrees, conventionalized among the participants in this study. It should be recalled that each of the 10 meanings could be rendered either with its Arabic loanword or with a native Swahili variant. Figure 2 reports, for each meaning, how many of the nine participants used the Arabic loanword in each task; because every participant supplied one variant per word, the number using the Swahili variant is simply the complement of each value (9−the Arabic count shown).

A clear and consistent pattern emerges from Figure 2. In Task 1, where participants gave the first word that came to mind, the Arabic loanword was rarely the first choice: for most meanings, only two to four of the nine participants used it. This points to a general preference for native Swahili variants. In Task 2, however, where participants were asked for any further words they knew for the same meanings, the Arabic loanword was produced by a clear majority, typically six or seven of nine. This contrast between the two tasks is the central finding of this section: the Arabic loanwords are evidently familiar to participants, yet they are not the preferred everyday choice. Task 3, which embedded the same meanings in full sentences, closely mirrors Task 1, confirming that the preference for Swahili variants holds in connected speech as well as in isolation.

At the level of individual words, the most conventionalized loanwords were drawn from the most borrowable semantic fields. The word for “judge” hakimu was the clearest exception to the overall pattern, used by the majority of participants even as a first choice in Tasks 1 and 3, and the word for “tax” ushuru was likewise among the most consistently used Arabic forms. At the other extreme, the word for “tool” ala was the least-used Arabic loanword across all three tasks, suggesting that it is far less entrenched in participants' mental representations than the other items. The ten target words were evenly split, by design, between the five most-borrowed meanings (God, tax, judge, king, and citizen) and the five least-borrowed meanings (world, light, family, disease, and tool), allowing borrowability and entrenchment to be compared. The results do not show a straightforward relationship. Consistent with expectation, ala “tool” (least-borrowed set) was the least conventionalized and least entrenched item. Against expectation, however, hakimu “judge” (most-borrowed set) was among the most frequently used loanwords yet scored low on the awareness measure, and dunia “world” and nuru “light”—from the same least-borrowed field—differed markedly in entrenchment.

The entrenchment of Arabic loanwords was further assessed through a multiple-choice question in which a set of 10 Arabic loanwords was presented as Swahili words, and participants were asked whether each was borrowed and, if so, from which language. (This set partially overlapped with, but was not identical to, the 10 items used in the translation tasks.)

This multiple-choice task targets speakers’ awareness directly: if speakers are unaware that a word was borrowed from Arabic, the word may be deeply entrenched in their mental representation. The variation in Figure 3 indicates that most of the Arabic loanwords are considerably entrenched, though, since entrenchment is a property of the individual language user's mind (Backus, 2014), not all words are equally entrenched. On this criterion the ranking runs opposite to recognition: the fewer participants who identified a word as Arabic, the more entrenched it is taken to be. For example, dunia “world” appears less entrenched than nuru “light”, even though both belong to the same semantic field. Notably, hakimu “judge” appears highly entrenched, which is consistent with its being among the most frequently used words in Tasks 1 and 3. The two most entrenched words, hakimu and sultani, both carry political meanings. Both measures used in this section are indirect. Translation preference records which form a participant produced on one occasion, and the judgment task records what a participant reports about a word's origin; neither observes mental representation itself. Each can diverge from entrenchment for unrelated reasons, since a participant may supply a Swahili variant because the task invited a single word, or identify a word as Arabic from orthographic or religious association rather than any felt unfamiliarity. The rankings reported here are therefore proxies consistent with entrenchment rather than direct measures of it.

5.2 Matched guise results (covert attitudes)

For this task, responses were examined to compare speakers' covert judgments of the two texts. Because of the small sample, only descriptive statistics (means) are reported; no inferential tests were conducted, as a sample of nine lacks the power for meaningful parametric or non-parametric analysis. All comparisons below should therefore be read as descriptive trends rather than statistically significant differences. Mean values for each of the five items (three statements for each text plus two comparison questions) were calculated on the 1–7 scale. Figure 4 presents the means for the first three statements across both texts.

The three statements were grouped into two dimensions: competence (“educated and knowledgeable”) and social attractiveness or solidarity (“religious” and “I identify with the writer”). On both dimensions, the writer using Arabic loanwords (Text 1) was rated marginally higher than the writer using Swahili variants (Text 2). On the competence item, the Arabic-loanword guise scored slightly higher (M = 5.5) than the Swahili-variant guise (M = 5.1), and the same direction held on the social-attractiveness items, most clearly for identification with the writer (M = 3.8 vs. M = 3.0). These differences are small, and because no inferential statistics were computed, it cannot be determined whether they are meaningful; the disparity between the two texts is not large enough to establish that participants definitively favored Arabic loanwords.

The two comparison questions were more revealing than the rating statements. Most participants judged the writer who used Arabic loanwords (Text 1) more likely to be a native Swahili speaker than the writer who used the Swahili variants (Figure 5), a pattern reinforced by the open-ended comments, where the Text 2 writer was often described as proficient but not necessarily native. Overall, the Arabic-loanword writer was evaluated more favorably.

5.3 Metapragmatic commentary results (overt attitudes)

The third section in the questionnaire addressed RQ3 by eliciting participants' overt attitudes through three open-ended questions, each targeting a distinct facet of how speakers consciously evaluate Arabic loanwords. As described in Section 4.4, all responses were analyzed through inductive thematic categorization, with recurring themes identified across participants.

The first question asked whether participants were aware that Arabic loanwords are estimated to represent between 18% and 40% of Swahili words, and how they felt about this. All nine participants reported being aware of the prevalence of Arabic borrowing. Their evaluations fell into three groups (Table 6): none were negative, seven were neutral or unmarked, and two were positive. A representative neutral response was, “I know. It is fine because most of them have Bantu pronunciation, (Participant 3)” while a positive response linked the loanwords to cultural affinity: “I like this because I am able to follow some conversations from the Arab world… I embrace art, food, drinks, literature, music from the Arab world because I feel it is close to me due to growing up with Swahili. (Participant 5)” The complete absence of negative reactions is the central result of this section: even when the scale of Arabic borrowing was made explicit, no participant treated it as objectionable.

Table 6. Overt attitudes toward Arabic loanwords.

The second question asked why Swahili speakers use Arabic loanwords. Most participants offered more than one reason, the majority relating to the long history of Swahili–Arabic contact, the Indian Ocean trading network, the influence of Islam, and political dominance (Table 7). One participant captured several of these at once: “Because of the historical contact of the Swahili coast and Arabic. Because most Swahili speakers, especially at the coast and the various Islands, are Muslims and can relate to these words”. (Participant 6) Notably, another participant reframed loanword use in terms of prestige rather than history, echoing the matched guise comparison result: “Because of the status that one gets when they use Arabic loanwords. One is seen to be a better speaker of Swahili when they use and stress these loanwords correctly”. (Participant 5) This convergence between the overt commentary and the covert matched guise data, both pointing to Arabic loanwords as a marker of native-like competence — is one of the clearest cross-method patterns in the study.

Table 7. Perceived reasons for using Arabic loanwords.

Total codings exceed N = 9 because some responses were coded into more than one category.

The third question asked whether participants thought Arabic loanwords might threaten Swahili identity. Most answers were short and emphatic, such as “No, not at all”, with two responses framing the loanwords as enriching rather than threatening Swahili identity:

“Arabic loanwords enrich Swahili identity, not take away from it. It spices up Swahili at all levels—dressing, food, music, religion, name it—thus making Swahili attractive and acceptable”. (Participant 7)

Although a few participants expressed positive attitudes, the majority held no marked attitudes, which again suggests that these loanwords are conventionalized within the Swahili community. Taken together with the first two questions, the responses indicate that participants not only tolerate Arabic loanwords but, in several cases, regard them as integral to Swahili.

6 Discussion

This study addressed three research questions concerning Swahili speakers' language ideologies. The first research question concerned the degree of conventionalization of Arabic loanwords. The translation tasks and judgment test indicated that the target loanwords are conventionalized to varying degrees in participants' Swahili, with most being well established and only a few (notably the word for “tool”) appearing less entrenched. At the same time, a meaning's overall borrowability did not reliably predict how conventionalized or entrenched its specific Arabic loanword was. This is consistent with Backus's (2014) view that entrenchment is a property of the individual language user's mind rather than a fixed property of the word, and it cautions against treating semantic-field borrow ability as a proxy for the status of any single loanword. A further caution applies to the direction of the relationship between attitudes and borrowing. This study measured present-day attitudes toward words that entered Swahili over several centuries, so it cannot show that accommodating attitudes brought that borrowing about. Words in continuous use for generations, reshaped phonologically and morphologically to resemble Bantu words, are unlikely to attract strong evaluation, so the neutrality observed may follow from conventionalization rather than cause it. The two are not mutually exclusive, but no synchronic design can separate them.

The entrenchment findings align with the frequency-based account. The word hakimu “judge” was among the most frequently used loanwords and was also among those least often recognized as borrowed, which is what Backus's (2014) account predicts if frequency of use drives entrenchment. The awareness-based measure remains indirect and should be read as broadly consistent with that account rather than as a precise index. It is also notable that the two most entrenched words, hakimu and sultani, both carry political meanings. Only two of nine participants identified either as Arabic, so any reading of this pattern rests on very few responses; one possibility is that these words entered everyday Swahili early through administrative use and have since lost any perceptible foreign character.

The second research question concerned covert attitudes, examined through a text-based adapted matched guise experiment. Overall, participants expressed relatively more positive attitudes toward the text containing Arabic loans. Given their clear tendency to use Swahili variants over Arabic loanwords in the translation tasks, this result was unexpected, yet it was supported by participants' own commentaries in the final section. Following Giles and Billings (2004), the matched guise items can be read along two dimensions—competence and social attractiveness—and on both, the Arabic-loanword guise was rated at least as favorably as the Swahili-variant guise. Although these descriptive differences are modest and should not be over-interpreted given the small sample, the absence of any competence penalty for the Arabic-loanword guise is itself noteworthy: it contrasts with the more purist attitudes documented in some other loanword contexts (Rüdiger, 2018), where borrowed forms are evaluated negatively. This pattern suggests that, for these speakers, Arabic loanwords carry no covert stigma and may even index native-like command of Swahili.

The third research question concerned overt attitudes, examined through three open-ended questions. Apart from a few positive comments, the majority of participants expressed unmarked attitudes toward Arabic loanwords. This outcome was anticipated, given the intensity of loanword use and the degree of phonological and morpho-syntactic adaptation involved, and it aligns with Schadeberg's (2009, p. 91) prediction that Swahili speakers hold no marked attitudes toward loanwords that are part of their active vocabulary. Such results again imply that Arabic loanwords have been conventionalized among the participants in this study.

Bringing the three strands of evidence together clarifies the overall picture. A behavioral preference for Swahili variants in production coexists with positive-to-neutral covert and overt evaluations of Arabic loanwords, and this convergence across the three instruments strengthens confidence in the central finding: these speakers harbor no negative attitudes toward Arabic loanwords. The divergence between production and evaluation accords with what Garrett (2010) describes as the gap between habitual usage and evaluative reaction, and it resembles the mixed covert–overt patterns Rüdiger (2018) reported for Korean speakers toward English loanwords. The implication is that, although Swahili speakers may not always select Arabic variants in production, they do not hold negative evaluative attitudes toward them—an asymmetry that is itself sociolinguistically informative. More broadly, this pattern is consistent with the theoretical expectation that conventionalized borrowings tend to lose their ideological markedness over time (Backus, 2014; Woolard and Schieffelin, 1994); as Arabic loanwords have been integrated into the active Swahili lexicon over centuries of contact, they appear to have been largely divested of marked ideological associations for these speakers.

6.1 Limitations and future directions

Several limitations must be acknowledged. First and most importantly, the study involved only nine participants, which severely limits the generalizability of the findings and precluded the use of inferential statistics; the observed patterns, while suggestive, cannot be taken as representative of the wider Swahili-speaking community. Second, the sample's demographic composition introduces potential bias, as seven of the nine participants were Kenyan and most were highly educated graduate students aged 27–45; attitudes may differ across nationalities, educational backgrounds, social classes, ages, and genders. Third, the adaptation of the matched guise technique from an auditory to a written format departs from the classic design (Lambert, 1967) and may have affected the covert nature of the elicitation. A follow-up study should recruit a larger, more stratified sample and, where the sample permits, employ appropriate non-parametric analyses.

7 Conclusion

In sum, this exploratory study contributes to our understanding of how speakers' attitudes relate to the outcomes of linguistic contact. The literature shows that Swahili has borrowed extensively from Arabic, and this paper examined, through speakers' attitudes, how and why such borrowing has come about. The examination of covert and overt attitudes revealed that the participants held no negative attitudes toward Arabic loanwords, suggesting that such attitudes are compatible with the extensive lexical borrowing and conventionalization documented for these words. The present design cannot determine whether such attitudes helped bring that borrowing about or emerged as a consequence of it. These preliminary findings, though requiring verification with larger and more diverse samples, offer initial empirical support for Schadeberg's (2009) observation and contribute to our broader understanding of the relationship between language ideologies and contact outcomes (cf. Rodríguez-Ordóñez, 2019; Woolard and Schieffelin, 1994).

Generative AI statement

The author declares that generative AI was used in the creation of this manuscript. During the preparation of this work, the author used AI-assisted tool (Claude, Anthropic) to improve language clarity, grammar, and readability. The final interpretation, intellectual content, and all research findings remain solely the responsibility of the author.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Wenge Chen, Huaqiao University, China

Reviewed by: Ahmet Erdost Yastibaş, Gazi University, Türkiye

Maha S. Yaseen, Al-Ahliyya Amman University, Jordan