Tamil Speakers in Switzerland: An Intergenerational and Typological Perspective

Abstract

Since the mid-1980s, many Tamils left their homeland because of the civil war in Sri Lanka (1983–2009) and for other reasons and settled in different countries. More than 40,000 Tamil migrants have come to Switzerland since then, and Tamil is spoken as a heritage language by second- and third-generation speakers who were born and raised in Switzerland. Within this context, it is the aim of the current study to shed light on the difference between Tamil spoken in the first generation (migrant language) and the second generation (heritage language) in the Swiss German and Swiss French parts of Switzerland. We therefore study Tamil, which is part of the Dravidian language family, in different majority language contexts, i.e., a Germanic language and a Romance language, respectively. While some research on Tamil in a diaspora setting already exists on migrated Tamil communities in Lancaster, California (US), East London (UK) and Toronto (Canada), the focus on Switzerland and contact with German and French has not previously been investigated. The data under investigation, which stems from 20 speakers in total (i.e., 5 first-generation and 5 second-generation speakers from the Swiss German and the Swiss French parts respectively), was collected in 2024 by way of a semi-structured interview based on a sociolinguistic questionnaire and a linguistic test. The data serves as the basis for the intergenerational and typological comparison. The analysis reveals systematic intergenerational differences across several morphosyntactic domains, including agreement, negation pattern, case marking, and subject pro-drop. While first-generation speakers retain greater access to dialect-specific and register-sensitive patterns, second-generation speakers show increased reliance on discourse-pragmatic cues and reduced sensitivity to morphologically encoded distinctions. These findings highlight the role of register, input conditions, and discourse context in shaping heritage Tamil across generations in Switzerland.

1. Introduction

The topic of intergenerational language transmission in diasporic communities has already received a good amount of attention in the field of heritage linguistics (see for instance Montrul, 2015, 2022; Purkarthofer, 2020; Filipović & Jovanović, 2025). A community and their language in the diaspora that has not been studied much to date is that of the Tamils (studies available to date are by Polinsky (1997) and Canagarajah (2008, 2013, 2019) on migrated Tamil communities in Lancaster, California (US), East London (UK) and Toronto (Canada)). Due to the civil war in Sri Lanka (1983–2009), as well as other reasons, many Tamils left their homeland and settled in different countries. One of these countries is Switzerland, where more than 40,000 Tamil speakers migrated to since the early 1980s. Today, Tamil, which is part of the Dravidian language family, is therefore spoken by different generations in Switzerland, i.e., the original migrants (first generation) and by second- and third-generation speakers that were born and raised in Switzerland as a heritage language (cf. Rothman, 2009, p. 159). Due to the fairly recent arrival of Tamil speakers in Switzerland, it is possible to collect data from first-generation speakers and the second generation in a systematic manner. Moreover, the multilingual nature of the host country Switzerland with four national languages (German, French, Italian, and Romansh) that are geographically distributed allows for comparisons of second-generation Tamil data produced in typologically different language regions.

Within this context, the aim of the current study is to investigate Tamil spoken in the first generation (migrant language) and the second generation (heritage language) in two linguistic regions of Switzerland, notably the Swiss German (CHG) and Swiss French (CHF) regions. From a typological perspective, we study a Dravidian language in different majority language contexts, i.e., a Germanic language and a Romance language, respectively, across two generations. More precisely, this study is based on data collected in 2024 by way of a semi-structured interview that consisted of a sociolinguistic questionnaire and a linguistic test. In this paper, we report our findings from 20 speakers in total, notably 5 first-generation and 5 second-generation speakers from the Swiss German and the Swiss French parts respectively, as this allows for an intergenerational and a typological comparison. This study presents differences in language use as well as factors that may explain the differences between migrant and heritage language. In this study, the term Swiss Tamil refers to different varieties of Tamil spoken among Tamil speakers living in Switzerland. The speakers discussed in this study are primarily of Sri Lankan Tamil origin. The term is used descriptively and does not imply the emergence of a distinct dialectal variety comparable to established regional varieties in Sri Lanka or India, nor to Tamil spoken in other diasporic communities1.

This article is structured as follows: Section 2 provides background information on relevant studies from the field of heritage linguistics, particularly with regard to different generations. It will also review existing work on Tamil spoken in the diaspora and introduce the Swiss Tamil project in which the current study is couched. Section 3 provides detailed information on the data collected and the methods applied to carry out the current study. Section 4 presents the results by focusing on linguistic differences between the first and the second generations in the first instance, followed by a comparison of the second-generation data in the Swiss German and Swiss French parts of Switzerland that allows for the consideration of typological differences. Finally, Section 5 concludes the article by presenting the main findings, by discussing the limitations of this study, and by providing an outlook for further research.

2. Theoretical Background

2.1. Intergenerational Language Change

Linguistic change in a migration context, particularly the difference between the language used by the migrants (first generation) and their children, the second generation, has already been studied in different settings with different language combinations and with different theoretical aims. One of the key concepts that is used in these types of studies is ‘heritage speaker’, which Wiese et al. (2022) describe as follows:

[…] [S]peakers who grow up in a bi- or multilingual home with a minority language in addition to the majority language(s) dominant in the larger society (see, e.g., Montrul & Polinsky, 2011). Accordingly, these are speakers who acquire their heritage language early and naturally in a home environment as a first language but who also acquire another language early on, which is more dominant in the larger society and will often be the only language supported in the formal context of schooling.

We adopt this definition in our study to refer to second-generation speakers. As regards theoretical queries linked to cross-generational language change, this can concern the aim to define a ‘native speaker’ and their grammar (see for instance Polinsky, 2018; Wiese et al., 2022), the acquisition of heritage languages (cf. Montrul, 2015), language maintenance and shift (see for instance Fishman, 1964; Potowski, 2013; Pauwels, 2016; Brown, 2022), as well as language policy research (see for instance Seals & Shah, 2017; Curdt-Christiansen, 2018 on family language policy specifically).

Concerning existing research on Tamil speakers in the diaspora, to our knowledge, few studies exist to date. In a study on cross-linguistic parallels in language loss, Polinsky (1997) compared selected linguistic features (agreement constructions, constructions without presumptive pronoun, conditional forms, null copying under coreference, relatives clauses, prepositional oblique nominals) of two speakers of so-called ‘reduced Tamil’ in the US to speakers of a range of other languages, notably Eastern Armenian, Lithuanian, Polish, Russian, Kabardian or Circassian. Canagarajah (2008, 2013, 2019) focussed more extensively on the Tamil community in the diaspora. In the first two studies, the focus is on language maintenance and shift in heritage Tamil spoken in Lancaster, California (US), East London (UK) and Toronto (Canada), whereby the 2008 article sheds light on language shift viewed from a generational perspective and the role that social stratification (caste system) and colonisation in the homeland may have played particularly, and 2013 article focuses on language maintenance linked to attitudes and identity construction. The 2019 article, which takes a qualitative approach, draws on the previously mentioned Tamil data from three different cities and investigates language ideologies of community members, particularly the alignment of “Tamil verbal resources […] with multimodal semiotic resources and spatial repertoires to accomplish social and cultural communicative activities” (Canagarajah, 2019, p. 9). As all studies to date that we are aware of are concerned with Tamil data in English-dominant contexts, the current study, which is couched in a larger project on Tamil spoken in Switzerland, is able to provide new perspectives on intergenerational language use and change by focusing on different dominant languages and different cultures.

2.2. Swiss Tamil Project

The current study is part of a larger Indo-Swiss funded research project entitled “Mapping Heritage Language Structure Through Sociolinguistic Cues: A Case Study of Swiss Tamil” (2023–2026) that aims at testing the so-called simplification hypothesis, i.e., the claim that the linguistic structure of heritage languages is less complex, thus simpler, than that of the homeland language variety due to limited input, restricted domains of use, and influence from dominant languages (cf. Polinsky, 2018; Scontras et al., 2015). To do so, a cross-linguistic perspective is taken in the systematic (socio)linguistic comparison of Tamil natively spoken in Northern Sri Lanka and South India and Tamil produced by particularly second-generation speakers (but also first-generation speakers) in the German- and French-speaking parts of Switzerland.

This study has three main objectives. First, it explores the varieties of Tamil spoken across generations and regions, specifically within the French-speaking and German-speaking areas. Second, it investigates how Tamil is maintained and transmitted within the diaspora context, including the role of Tamil schools, cultural networks, and community activities. Third, it examines a set of grammatical features such as plural marking in quantified contexts, differential object marking, case alternation, subject–verb agreement, pronominal forms (including medial pronouns), negation marking, subject drop, anaphor and binding interpretation, and coordination structures to assess whether the changes observed in Swiss Tamil reflect simplification, restructuring, or patterns already present in the heritage input.

Data Collection in Switzerland

In order to examine the existing claims about the simplification hypothesis that the linguistic structure of heritage languages is simpler than the homeland language variety and to test its validity cross-linguistically in the Indo-Swiss project, multiple types of data from different generations (including baseline/homeland data) have been collected since 2024, including a sociolinguistic questionnaire administered through semi-structured interviews (conducted in Tamil), a structured linguistic task, and narrative data based on the picture storybook Frog, Where Are You?.

The SNSF study has progressed in two phases. The pilot phase was conducted in two locations. In December 2023, the Indian research team carried out five pilot interviews in Colombo, Sri Lanka, with speakers representing Negombo Tamil (NT), Upcountry Tamil (UT), and Jaffna Tamil (JT) varieties. Between May and August 2024, the Tamil-speaking postdoctoral researcher based in Switzerland conducted four pilot interviews on the Swiss side, which included two first-generation and two second-generation speakers across both the French- and German-speaking regions. This initial phase helped refine the questionnaires and observe dialectal variation to understand the difference between the features of migrant and heritage language.

The main phase of this study was led by the postdoctoral researcher and took place across Switzerland, Sri Lanka, and India. Between June and December 2024, data were collected from 18 second-generation and 6 first-generation participants in the French-speaking cantons (Vaud, Neuchâtel), and 11 second-generation and 6 first-generation participants in the German-speaking cantons (Zurich, Bern, Aargau, Uri, St. Gallen, Basel). In February 2025, the fieldwork in Sri Lanka and India was conducted, yielding interviews with 3 first-generation and 9 second-generation speakers from Colombo, Jaffna, Batticaloa, and Kandy. Additionally, one interview was conducted in Tamil Nadu, India, with a speaker whose age profile matched that of Swiss-born second-generation participants.

Participant recruitment initially posed challenges, particularly due to the researcher’s unfamiliarity with local community networks. The first major breakthrough came through contact with the Lausanne University Tamil Students’ Association (LUTSA), whose members showed a keen interest in the study and actively supported the project. This led to a snowball sampling process through which most participants were recruited. In the German-speaking region, access was more limited, but valuable connections were established through research participants from the French-speaking region and peers who shared contacts from their prior anthropological work with Sri Lankan Tamil communities.

All interviews were conducted in Tamil, which functioned both as the participants’ heritage language and as a practical medium of communication, given that the postdoctoral researcher does not speak French or German. Interviews were conducted either in person or, when necessary, online via Zoom. Most first-generation participants were interviewed in their homes, allowing the researcher to engage with them in familiar settings. Participants extended warm hospitality, and this personal setting often led to rich, reflective responses that included intergenerational commentary about their children’s linguistic and cultural attitudes.

Second-generation participants were generally interviewed in cafés or the researcher’s office at the University of Lausanne, based on each participant’s choice. The interviews were audio-recorded using a Zoom H1N handheld recorder or Zoom’s built-in platform, depending on the format. Building trust with participants was a central part of the fieldwork process, and though time-intensive, it was crucial for eliciting open and detailed responses given the sensitive migration context.

This study consists of two components: a sociolinguistic component and a linguistic component. Sociolinguistic data were collected using two questionnaires: (i) a Sociolinguistic Questionnaire adapted from Hulsen (2000) and Dagamseh (2020), with additional questions on Tamil use in religious practices, Tamil associations, and even interactions with pets; and (ii) a Subjective Vitality Questionnaire, adapted from Hulson and originally based on Giles et al. (1977), which probes perceptions of both Swiss-origin and Tamil-origin communities in Switzerland.

The linguistic data collection component included three tasks: (i) a narration task using the illustrated book Frog, Where Are You? (Mayer, 1969); (ii) a structured linguistic task targeting morphosyntactic features such as plural marking, case, agreement, negation, coordination, binding, and word order; and (iii) a translation task, in which participants translated ten short sentences from their dominant language (French or German) into Tamil. The tasks were designed using insights from typological and diachronic comparisons across Tamil varieties, as well as French and German.

The sociolinguistic questionnaire responses were entered into LimeSurvey for systematic analysis, which is currently ongoing. Responses from the linguistic tasks analysed in this paper were entered and coded in Microsoft Excel. Each participant was assigned a unique identifier to pseudonymize their identity and ensure confidentiality. The coding system includes the speaker’s gender, age, region, and the date and time of their first interview. For example, a code like M26CHG240720231115 represents a 26-year-old male participant from German-speaking Switzerland, first interviewed on 24 July 2023 at 11:15 AM. These codes were used to label all metadata and audio files, in compliance with data protection guidelines.

Overall, the SNSF project currently comprises approximately 58 full interviews across Switzerland (45), Sri Lanka (12), and India (1), combining sociolinguistic, linguistic tasks, and narrative data.

3. Data and Method of the Current Case Study

3.1. Data Used for the Case Study

This paper is based on data collected in 2024 as part of the above-mentioned research project. As previously pointed out, the fieldwork was carried out in both French-speaking and German-speaking cantons of Switzerland. The current case study focuses on first-generation migrants from the Sri Lankan Tamil community and their second-generation children, with the aim of comparing the language of the migrant generation with the heritage variety spoken by their children. We focus on a curated subset of 20 speakers drawn from a larger corpus of 45 interviews carried out in Switzerland. The selected subset includes 10 first-generation and 10 second-generation speakers, balanced across Switzerland’s two major linguistic regions: 5 from French-speaking cantons and 5 from German-speaking cantons in each generational group.

All participants are of Sri Lankan Tamil origin, and most trace their roots to the Jaffna region in northern Sri Lanka. Within each generational and regional group, the sample includes both male and female speakers. In several cases, second-generation participants are direct relatives of the first-generation speakers (e.g., their children), and others were chosen based on close peer relationships. This pairing structure was intentional as it allows for fine-grained comparison between migrant Tamil and heritage Tamil within shared family or social networks.

While demographic variation was taken into account, the main aim was to create a balanced and focused data sample that enables a comparative analysis of intergenerational linguistic patterns. Though the larger dataset continues to expand in the project, the sample used in this paper provides an overview for evaluating how migration, multilingualism, and intergenerational change shape Tamil language use in Switzerland.

Speaker Profiles

This subsection introduces speaker profiles to contextualise the linguistic data and to indicate the sociolinguistic cues to understand potential intergenerational differences. The current study is based on data from 20 speakers—10 first-generation migrants and 10 second-generation heritage speakers—drawn from both the French- and German-speaking regions of Switzerland. To contextualise the linguistic findings of this study, speaker metadata has been grouped by generation, whereby

Table 1. First-generation speaker profiles.

Code	Canton	Homeland Dialect	Age	Year of Migration	Languages Known
M61CHF17072024	Vaud (CHF)	Jaffna	61	1988	Tamil, Basic French
F60CHF17072024	Vaud (CHF)	Jaffna	60	1990	Tamil, Sinhala, English
F49CHF31072024	Vaud (CHF)	Jaffna	49	1993	Tamil, Basic French
F57CHF31072024	Vaud (CHF)	Jaffna	57	1991	Tamil, English, French
F26CHF23072024	Vaud (CHF)	Jaffna	26	2015 (18 years old)	Tamil, French, English
M27CHG17082024	Uri (CHG)	Jaffna	27	2015 (19 years old)	Tamil, German
F54CHG06112024	Aargau (CHG)	Upcountry	54	1995	Tamil, German
F55CHG11112024	Bern (CHG)	Colombo	55	1998	Tamil, English, German
F45CHG11112024	Bern (CHG)	Jaffna	45	2000	Tamil, German
M27CHG26082024	Bern (CHG)	Batticoala	27	2005 (8 years old)	German, Tamil, English

shows the first generation and

Table 2. Second-generation speaker profiles.

Code	Canton	Homeland Dialect	Age	Attended Heritage Tamil School	Languages Known
M21CHF17072024	Vaud (CHF)	Jaffna	21	No	French, Tamil, English
F27CHF21062024	Vaud (CHF)	Jaffna	27	KG to 5	French, Tamil, English
F24CHF11072024	Vaud (CHF)	Jaffna/ Colombo	24	KG to 12	French, Tamil, English, Sinhala, German and Spanish
F31CHF22072024	Vaud (CHF)	Jaffna	31	Yes, but stopped around 11 years old	French, Tamil, English
M25CHF25062024	Vaud (CHF)	Jaffna	25	KG to 10	French, Tamil, English and German
F23CHG06112024	Aargau (CHG)	Upcountry	23	KG to 10	(Swiss) German, Tamil, English
F23CHG07102024	Bern (CHG)	Colombo	23	KG to 12	(Swiss) German, Tamil, English
F25CHG17102024	Aargau (CHG)	Jaffna	25	KG to 12	(Swiss) German, Tamil, English
M28CHG26112024	Bern (CHG)	Jaffna/ Tricanmolee	28	KG to 10	(Swiss) German, Tamil, English
M27CHG06082024	Basel (CHG)	Jaffna	27	KG to 12	(Swiss) German, Tamil, English

shows the second-generation speaker profile. Each participant is identified through a unique speaker code. The metadata includes variables such as canton (Swiss-French or Swiss-German region), parental origin (Jaffna, Upcountry, Batticaloa), migration history, age, heritage language schooling (if they attended from Kindergarten, KG to Class 12), and other relevant sociolinguistic details. These profiles assist in understanding both the type of linguistic input available to each speaker and their broader exposure to formal and informal Tamil.

3.2. Methods Applied in This Case Study

In order to shed light on morphosyntactic variation between migrant and heritage Tamil, this study adopts a mixed methodological approach combining structured linguistic elicitation with qualitative sociolinguistic observation. The core dataset analysed for this paper is based on a structured linguistic task administered orally in Tamil. Participants responded to controlled prompts designed to target morphosyntactic features (listed in bold in

Table 3. Comparison of morphosyntactic features across two generations.

Feature	Homeland Variety	1st Gen (Migrant)	2nd Gen (Heritage)
Plural marking in quantified contexts	Plural suffix (-kal/gal) regularly omitted (in spoken form) after numerals or quantifiers but obligatory in written form	Consistent with spoken form	Mix of spoken and written
3PL agreement on verb for rational nouns	-inam suffix used (e.g., varuv-inam, they will come)	Consistent with homeland	Consistent with homeland; elicited forms vary by register
Case alternation (instrumental vs. sociative)	Instrumental is marked by suffix -a:l or postposition ‘vechu’	Consistent with homeland by all speakers	Consistent with homeland by most speakers but some marked with sociative -o:da
Genitive marking	Obligatory (-nda)	Obligatory	Obligatory but dropped by a few
Differential object marking (DOM) on irrational nouns	Based on specificity and definiteness	Not marked	Mostly consistent
Oblique marker on Noun	Example: puthagam > puthaga-n-kal puthagam > puthaga-t-ai	Consistent with homeland	A few do overgeneralisation. Example: puthagam > puthagam-kal puthagam > puthagam-ai
Demonstrative pronouns (3-way distinction)	ivan/uvan/avan (prox./medial/distal)	Medial used contextually (e.g., gossip or disrespect)	Medial absent; only proximal/distal used
Negation patterns (illa vs. alla)	illa = existential, alla = equative	alla used in different context; replaced by illa in both distributions	alla absent; replaced by illa in both distribution
Agreement in dative subject negated verb	Agreement marked (e.g., teriy-a:-du know-neg-3sgn)	Consistent with homeland	Agreement omitted by a few speakers
Binding of reflexive and 3SG pronoun	ta:n and aval/avan	Ambiguous interpretation is not available	Ambiguous interpretation is available

) such as case marking, agreement, plural marking, binding, definiteness, and related grammatical domains. Table 3 summarises the key morphosyntactic features examined across generations.

The linguistic component of this study comprised four complementary task types, each targeting different aspects of morphosyntactic competence. First, a production task (PT) elicited sentence-level responses in Tamil, focusing on features such as plural marking in quantified contexts, case marking and alternation (including oblique allomorphy), verbal negation with illai and a:thu2 and agreement morphology (e.g., third-person plural agreement and agreement in dative-subject constructions). Second, an interpretation task (IT) probed referential possibilities in binding contexts, specifically contrasting the third-person reflexive anaphor (ta:n ‘self’) with third-person pronouns (aval/avan ‘she/he’). Third, a grammaticality judgement task (GJT) tested acceptability contrasts related to word order, subject omission, and negation patterns. Fourth, a translation task required participants to translate controlled sentences from their dominant societal language (French or Swiss German) into Tamil, targeting typological contrasts across domains such as definiteness, agreement, word order, negation, and argument structure.

This task design enables a distinction between productive use, interpretive flexibility, metalinguistic judgments, and contact-sensitive patterns across generations.

Although the same prompts were used for all participants, responses varied due to dialectal background, register choice, and generational status. To capture this variation systematically, responses were organised into two parallel datasets. The first dataset consists of full sentence responses entered in Romanised Tamil. This version allows for the identification of register choice (spoken, written-like, or mixed) and enables the detection of broader structural patterns beyond the targeted feature. The second dataset isolates the relevant target form for each prompt (for example, a specific case marker such as -ukku), allowing focused comparison of how particular grammatical elements are realised across speakers. Spoken and written Tamil forms were both treated as valid outcomes; register differences were tracked through qualitative annotations and visual markers within the dataset rather than being treated as errors.

In addition, a separate table was prepared to document metalinguistic comments3 and spontaneous explanations provided by participants during elicitation, thus serving as data for a qualitative analysis. This file records the speaker code, the relevant response, accompanying commentary, and precise timestamps, enabling direct cross-referencing with the audio data. These comments proved particularly informative for interpreting intraspeaker variation. For instance, some first-generation speakers explicitly reflected on the restricted discourse use of medial pronouns such as uva, while others commented on regional or caste-associated variants linked to their home varieties.

Such explicit metalinguistic reflection on variation and register choice is consistent with findings from other heritage language contexts, where metalinguistic awareness enables speakers to manage register variation and task-specific language use without implying reduced grammatical competence (Riehl, 2021). Together, these qualitative observations (discussed in detail in Section 4.2.1) enrich the quantitative patterns by revealing speakers’ metalinguistic awareness and sensitivity to register and social context.

3.3. Analytical Framework

The analysis proceeded in two complementary stages designed to capture both distributional tendencies and structured sources of variation in the dataset. This staged approach allowed us to separate item-level variability from broader qualitative patterns and to identify which morphosyntactic domains warranted closer examination in the discussion.

In the first stage, responses from the linguistic elicitation task were subjected to a descriptive, distribution-based analysis. Each response was coded binarily as either a match (i.e., the speaker’s response matched the expected response) or a mismatch (i.e., the speaker’s response did not match the expected response) relative to the expected target form. A response was coded as a match if it conformed to the target structure identified on the basis of homeland Tamil usage (Jaffna Tamil and/or widely attested spoken Tamil norms) and as a mismatch if it deviated from this expected response. This process could be better understood if we examine this with an example: For the sentence in (1) (see Appendix A, item 4a in

Table A1. Linguistic task items showing item-based variation.

Item.No	Questions	English Translation
39a	idu tanni illa	This is not water
39b	idu tanni alla	This is not water
40a	ba:la: oru samayalka:ran illa	Bala is not a cook
40b	ba:la: oru samayalka:ran alla	Bala is not a cook
32a	mi:na [ba:nu tan-ai pathi pesi-n-a:l-nnu] so-n-a:l	Meena said Banu talked about herself.
14b	…ribbon kattirukka	She (Medial) is wearing a ribbon
18a	avalgal mi:na:kku parisu…	They (3PlF) gave Meena a gift
18b	avangal mi:na:kku parisu…	They (3PlM) gave Meena a gift
4a	ne:tu, ra:mu vajala oru … pa:ta:n	Yesterday, Ramu saw a karugam in the field.
31b	mi:na : [aval iDli-ai sa:pi-TT-a:l-nnu] son-n-a:l.	Meena said she ate idli
31a	mi:na aval-ai kanaDi-le pa:r-t-a :l.	Meena saw her in the mirror
11	Meri … pidithu dinamum vilayaduva:	Mary catches butterflies and plays everyday
34b	mi:na:-ukku pe:na: parisa: koDu-tt-e:n	(I) gave meena a pen as a gift
34c	pe:na: parisa: koDu-tt-e:n	gave a pen as a gift
35b	tamil paLLikku pojirukka:	(She) went to Tamil school
10b	avan ippo : oru … veTTura:n.	Now, he is cutting a tree
4b	innaikku ka:laila, irandu … pa:ta:n	Today, he saw two karugams
5	sivappu toppi ja:r talajila irukku?	Who is wearing the red cap?
8b	mi :na : ra :mu-vai … aDitta:L	Meena hit Ramu with the bat
10a	kuma:r dinamum ka:TTukku … veTTa po:va:n	Kumar goes to the forest daily to cut trees.

), the expected answer (shown in bold) is with DOM marked (accusative case) on the noun.

1.	ne:tu	vajal-la	ra:mu	oru	karuga-th-ai	pa:-t-a:n
	Yesterday	field-loc	Ramu.3sgm.nom	one	karugam.3sgn-obl-acc	see-pst-3sgm
	Yesterday, Ramu saw a karugam in the field.

If in response to (1) the speaker had answered karuga-th-ai (karugam-obl-acc), then in the response cell for item (4a) ‘1’ will be marked to indicate a match to expected response, and any other response will be marked ‘0’.

The full dataset comprised 1640 observations, derived from 82 linguistic items with responses from 20 speakers per item. Of these, 327 responses were coded as mismatches, corresponding to an overall variability rate of 19.9% across the dataset. To identify how this variability was distributed across linguistic items, mismatch rates were calculated for each item by dividing the number of mismatches by the total number of responses for that item. These item-level rates were then visualised using a histogram, which revealed that variability was not evenly distributed across the task but clustered around a subset of items.

On the basis of this distribution, items exhibiting comparatively elevated levels of variability were identified. Rather than treating all deviations as equally informative, this step allowed us to distinguish between incidental variation and item-based variability that may reflect systematic differences in linguistic behaviour. A conservative threshold was adopted to select items for further analysis, focusing on those that showed substantial divergence across speakers rather than sporadic or idiosyncratic responses.

The second stage of the analysis consisted of a qualitative, domain-based examination of the morphosyntactic features targeted by these high-variability items. Items were grouped according to the linguistic domains they probed (e.g., pronominal systems, agreement morphology, case marking, negation), and patterns were examined across speakers and generations. This stage draws not only on task responses but also on speakers’ metalinguistic comments and self-reports during elicitation, allowing us to interpret variability in light of register awareness, dialectal knowledge, and contact with written Tamil.

Importantly, not all linguistic domains tested in the task displayed elevated variability. Domains that showed low mismatch rates or consistent convergence on spoken Tamil norms were not treated as sites of variation but are instead discussed in relation to register choice and dialectal awareness in Section 4.2. The combined two-stage approach thus ensures that the discussion in Section 4 is grounded in empirically observable patterns while avoiding overinterpretation of isolated or marginal deviations.

The resulting distribution exhibited a pronounced right skew (Figure 1): the majority of items clustered at low mismatch (variability) values, while a smaller subset showed substantially higher rates of mismatch (variability). A clear inflexion point was observed at approximately 30%, beyond which the number of items dropped sharply (marked by the red bar indicating the 30% threshold in Figure 1). Items above this threshold formed a sparse but distinct tail of elevated variability.

On this basis, a variability rate of 30% was adopted as a heuristic threshold for identifying items that warrant closer qualitative analysis. This threshold was not treated as a statistical cutoff, but rather as a data-driven criterion distinguishing routine performance variability from patterns suggestive of structural instability.

Items exceeding this threshold were therefore selected for in-depth qualitative analysis. Their elevated variability rates indicate recurrent divergence across speakers rather than sporadic error, making them particularly informative for examining the nature of linguistic variation and potential contact-induced change. By focusing on this subset (given in

Table 4. Selected item based on variation.

Item No	No of Expected Responses	No of Varying Responses	Sample Size	Rate of Variability	Task Type	Target Structure
39a	0	20	20	1	GJT	Existential negation
40a	0	20	20	1	GJT	Existential negation
32a	2	18	20	0.9	IT	Long distance and local binding
14b	2	18	20	0.9	PT	Medial pronoun
18a	3	17	20	0.85	PT	3plf agreement
39b	4	16	20	0.8	GJT	Equative negation
40b	4	16	20	0.8	GJT	Equative negation
4a	7	13	20	0.65	WT	dom marked in definite noun
31b	8	12	20	0.6	PT	Pronoun binding with matrix subject or outside the clause
18b	8	12	20	0.6	PT	3plm agreement
31a	2	11	20	0.55	IT	Pronoun binding with reflexive argument (mirror)
11	10	10	20	0.5	PT	dom marked in unmarked
34b	12	8	20	0.4	GJT	Subject pronoun drop
34c	13	7	20	0.35	GJT	Subject pronoun drop
35b	13	7	20	0.35	GJT	Subject pronoun drop
10b	13	7	20	0.35	PT	dom in irrational noun
4b	12	6	20	0.3	PT	Plural marking in quantified noun phrases
5	14	6	20	0.3	PT	gen unmarked in marked
8b	13	6	20	0.3	PT	Instrumental marked with sociative
10a	14	6	20	0.3	PT	dom marked in irrational noun

), the analysis targets forms where structural reanalysis or competing grammatical representations are most likely to be observable.

Alongside the structured elicitation tasks, the dataset also includes a translation task in which participants were asked to translate ten controlled sentences from their dominant societal language, French or (Swiss) German, into Tamil. The translation stimuli were designed to probe typological contrasts between Tamil and the two majority languages across a range of morphosyntactic domains, including definiteness and article use, adjective–noun order, agreement, plural marking, tense and aspect, argument structure, negation, subordination strategies, and clause structure. Because first-generation speakers typically use French or German as a second or foreign language, while second-generation speakers are dominant in these languages, the translation task provides a complementary window onto potential contact-induced effects under different language dominance configurations. However, a systematic analysis of speakers’ performance on the translation task, and of the linguistic knowledge reflected in their translations, raises distinct methodological and theoretical questions that go beyond the scope of the present paper. For this reason, the translation data are not analysed here; a dedicated study focusing specifically on translation performance and cross-linguistic influence in Swiss Tamil is currently in preparation and will be reported separately.

This paper follows the step-by-step analytical procedure discussed above for both a quantitative and qualitative investigation of morphosyntactic variation, and it also enables the researchers to relate observed patterns to dialect contact, register competition, and generational change in Swiss Tamil.

The selection of morphosyntactic domains tested in the linguistic task was informed by three complementary considerations. First, previous work on heritage grammars (e.g., Polinsky, 1997, 2018; Scontras et al., 2015; Montrul & Polinsky, 2019; Montrul & Silva-Corvalán, 2019) has identified certain domains such as pronominal systems, case marking, agreement, and binding as particularly sensitive to variation under reduced or variable input. Second, the task design took into account typological contrasts between Tamil (Schiffman, 1999; Wilden, 2018) and the dominant contact languages in Switzerland, namely (Swiss) French (Fagyal et al., 2006) and (Swiss) German (König & van der Auwera, 1995), in order to identify domains where cross-linguistic influence or transfer could plausibly arise. While the translation task directly targeted cross-linguistic effects, as mentioned earlier, it is not discussed in the present paper. Third, domain selection was guided by descriptive work on Sri Lankan Tamil (Suseendirarajah, 1967, 1979; Thananjayarajasingham, 1977; Gair & Suseendirarajah, 1981), in particular Jaffna Tamil (homeland spoken form as baseline), which constitutes the dominant input variety for both first- and second-generation speakers in the present study and is known to preserve several morphosyntactic features that have undergone change or reduction in other Tamil varieties. Together, these considerations motivated the choice of domains examined in Section 4, without presupposing direct baseline comparison with homeland varieties.

4. Preliminary Results and Discussion

The discussion in this section is organised into three parts. Section 4.1 presents a set of group-level comparisons, using chi-square (χ²) tests as a diagnostic tool to explore whether the rate of variability in responses differs across socio-demographic categories such as generation, region, and gender. This exploratory scan offers a distributional overview of potential associations between social groupings and linguistic behaviour. Section 4.2 examines specific morphosyntactic domains that display elevated variability, focusing on items that exceed the 30% threshold established in the item-level analysis (see Section 3.3). These patterns suggest areas of structural instability or competing grammatical strategies, warranting deeper structural investigation. Section 4.3 summarises the main interpretive implications of these findings and relates them to other work on migrant and heritage languages.

As previously indicated, the linguistic task was administered orally in Tamil, and responses were audio-recorded. Although full transcription and coding are ongoing, a subset of morphosyntactic features has already been systematically coded for this group, revealing early patterns of variation across generations. These patterns suggest only minimal intergenerational restructuring, with many features retained in the heritage variety. At the same time, the data also reflects the effects of migration, contact with Indian Tamil varieties, and the diglossic nature of Tamil, particularly in heritage schooling and formal domains.

4.1. Group-Level Comparisons

This section presents a diagnostic overview of variability patterns across three socio-demographic groups: generation, region, and gender (see

Table 5. Summary of group comparisons.

Comparison	Groups Compared	χ2 Value	p-Value	Interpretation
Generation	First vs. Second	0.53	0.47	No significant difference in mismatch rates
Region	CHF vs. CHG	3.60	0.059	Marginal trend; not statistically significant
Gender	Female vs. Male	0.20	0.65	No meaningful difference

). These group-level comparisons are exploratory in nature, serving to flag broad distributional tendencies in the data. The statistical procedure employed here is the chi-square test of independence, a descriptive, non-parametric method used to examine whether the distribution of two categorical variables is independent.

For each of the 1640 items under consideration, a contingency table was constructed with group membership (e.g., first vs. second generation, CHF vs. CHG, female vs. male) on one axis and response type (match vs. mismatch) on the other. The chi-square (χ²) statistic evaluates whether the observed frequencies deviate from what would be expected under the assumption of independence between these variables. The corresponding p-value gives a descriptive measure of how pronounced any deviation is—not a measure of effect size or predictive strength, but rather a tool to identify trends that may merit closer qualitative attention. While this procedure is not designed to test formal hypotheses, it allows for the identification of broad patterns of variability that may correlate with participant attributes.

• Generation

Across the dataset, first-generation speakers showed a rate of variability of 13.1% (74 out of 566 items), while second-generation speakers exhibited a slightly higher rate at 14.6% (83 out of 570). The chi-square statistic (≈0.53, p ≈ 0.47) indicated no meaningful distributional difference across generations. This suggests that generational status, by itself, does not account for the observed variability in responses.

• Region

Speakers from the French-speaking region (CHF) displayed a higher rate of variability (15.8%, 90/570) compared to those from the German-speaking region (CHG) (11.8%, 67/570). While the chi-square result (≈3.6, p ≈ 0.059) suggests a weak trend, this pattern does not meet conventional thresholds for statistical significance. Nevertheless, it points to a possible area for closer examination in the qualitative analysis.

• Gender

Female speakers exhibited a rate of variability of 13.4%, while male speakers showed a comparable rate of 14.5%. The chi-square value (≈0.20, p ≈ 0.65) revealed no notable group-level distinction by gender. These findings indicate that gender does not materially influence the distribution of responses in this dataset.

These group-level comparisons do not suggest any strong or systematic differences in response variability by generation, region, or gender. The regional contrast shows a slight but inconclusive tendency. These results serve as a diagnostic scan, informing the shift toward a Morphosyntactic analysis of linguistic variables, where variability patterns will be examined in greater contextual and structural detail.

4.2. Morphosyntactic Features—Data Analysis and Discussion

The observations below are based on coded responses from the linguistic task. In the tables below, boldface highlights the numerical values that reflect differences in the distribution of responses between first- and second-generation speakers.

Generally, the linguistic task reveals limited but systematic variation across generations, with most differences emerging at the level of specific morphosyntactic domains (these domains include pronominal systems, agreement morphology, negation patterns, binding and case marking, each of which is discussed below) rather than across the grammar as a whole. Importantly, linked to the research question of the larger Indo-Swiss project, the patterns observed do not straightforwardly support the simplification theory; thus, the heritage grammars do not tend to exhibit structurally reduced or reorganised morphosyntax (cf. Polinsky, 2018; Scontras et al., 2015). Instead, they point to functional reanalysis, register effects, and changes that parallel developments already attested in contemporary Tamil varieties (India and Sri Lankan spoken varieties).

This also supports Silva-Corvalán’s (1994) observation that structural changes in contact settings are not always the direct result of language contact. Rather, contact may accelerate changes already underway in a language’s natural diachronic development. In the case of Tamil, understanding the historical trajectory of the language is essential for interpreting changes observed in Swiss Tamil. Wilden (2018) points out that Tamil has undergone considerable structural shifts across its historical stages from Old Tamil (pre-700 CE) and Middle Tamil (700–1600 CE) to Modern Tamil (post-1600 CE). For example, the tense system evolved from a basic past/non-past distinction in Old Tamil to an addition of present and future markers in Modern Tamil. The negation system shifted from participial forms (-a) to more complex constructions using negative copulas such as illa (existential negation) and alla (equative negation). Morphological case marking also became more fixed, with increased influence from Sanskrit and therefore flexible word order. These diachronic developments provide an essential backdrop for evaluating whether observed changes in Swiss Tamil reflect innovation, retention, or simplification.

In what follows, we examine in more detail variation in selected morphosyntactic domains4 that show elevated rate of variation in the quantitative analysis, specifically (1) pronominal systems (including medial demonstratives), (2) plural morphology (plural marking in quantified contexts and third-person plural agreement), (3) negation patterns (including illa/alla alternation and agreement reduction in negated predicates), (4) case marking and differential object marking, and (5) binding relations involving reflexives (ta:n) and third-person pronouns.

4.2.1. Pronominal Systems

One salient morphosyntactic domain examined in this study concerns the demonstrative and pronominal system of Tamil, which encodes a three-way deictic distinction between proximal (i-), medial (u-), and distal (a-) forms, as shown in

Table 6. Three-way deictic distinction in Old Tamil.

Reference Type	Pronoun 3sgf	Pronoun 3sgm	Pronoun 3sgn	Pronoun 3pl
Proximal	ival She	ivan he	ithu this	ivai they
Medial	uval She	uvan he	uthu that	uvai they
Distal	aval She	avan he	athu that	avai they

. Proximal pronouns typically refer to entities close to the speaker; medial pronouns refer to entities that are familiar or jointly accessible within the shared perceptual space of both speaker and hearer; and distal pronouns refer to entities located outside the immediate space of both interlocutors.

The medial demonstrative base u- has long been identified as a conservative feature of Jaffna Tamil. As noted by Suseendirarajah (1979), u- occurs frequently in Sangam literature (roughly 300 BCE–300 CE) and is retained productively in Jaffna Tamil, while it has gradually disappeared from mainland Tamil varieties. Unlike Indian Spoken Tamil, which has largely reduced the deictic system to a proximal–distal contrast, Jaffna Tamil has been consistently described as maintaining a three-way distinction among a-, i-, and u- demonstrative bases. This description aligns with early grammatical traditions, including the categories enumerated in the earliest Tamil grammatical treatise Tolkāppiyam, and with later descriptive work on northern Sri Lankan Tamil.

Given its dialect-specific status and its reported vulnerability even in homeland varieties, the medial pronoun system constitutes a particularly suitable testing ground for examining intergenerational change and attrition in Swiss Tamil. On the basis of previous descriptions and the known reduction of the medial pronoun in Indian Spoken Tamil, we expected medial pronouns to be particularly unstable in the Swiss Tamil data, especially among second-generation speakers. This expectation is borne out by the distribution of responses in the elicitation task targeting medial pronouns.

As shown in

Table 7. Distribution of medial pronoun use (Item 11).

Category	Subgroup	Responses with Medial Pronoun	Responses Without Medial Pronoun
Gender	Male	0	7
	Female	2	11
Generation	First	2	8
	Second	0	10
Region	CHF	2	8
	CHG	0	10

, medial forms are virtually absent in the Swiss Tamil dataset. No second-generation speaker produced a medial pronoun, and medial forms are unattested in the German-speaking region. The medial pronoun use is restricted to two first-generation female speakers from the French-speaking region. In contrast, proximal and distal pronouns are robustly retained across generations and regions, indicating that the observed pattern reflects selective attrition of the medial category rather than a general erosion or innovation of the pronominal system.

This distribution suggests that the lack of medial pronouns in Swiss Tamil reflects the continuation and acceleration of an ongoing change already observable in both Jaffna Tamil and Standard Spoken Tamil (SST), rather than a case of uniform heritage-language simplification. The data discussed in this study further suggest that this dialect-specific category undergoes functional narrowing in diaspora contexts, with limited use among first-generation speakers and near-complete absence among the second generation, a pattern that is further supported by the metalinguistic commentary presented below.

The lack of use of medial pronouns suggests that morphosyntactic change in Swiss Tamil targets domains that are already unstable across homeland and contact varieties, rather than reflecting a general erosion of the pronominal system.

Crucially, the quantitative pattern is corroborated by metalinguistic commentary from first-generation speakers, which provides insight into the functional semantics of medial forms and the mechanisms underlying their change. One first-generation speaker (F57CHF31072024) explicitly described u- forms as encoding shared visual access and familiarity, contrasting them with a- and i- forms that rely on discourse anaphora or speaker-oriented deixis. Her examples are given in (2) and (3).

2.	uva-kkum	theriy.um
	she.med-dat	know.hab.3sn
	‘She also knows.’

3.	uva-ta:n	ena-kku	sonn-a-val
	she.med-foc	I-dat	tell.pst-rel.3sgf
	‘She is the one who told me.’

At the same time, she reported consistent comprehension breakdowns in interactions with her second-generation daughter, who fails to interpret medial forms even in directive contexts, instead requesting clarification (e.g., “Which one?”). This indicates that the medial category is no longer part of the interpretive system of second-generation speakers. Her examples for this context are given in (4a–b).

4.a.	uthu	edu
	that.med	take.imp
	‘Take this one.’ (1st Gen)

b.	ethu	eduk-an-um
	which	take-pres-2sg
	‘Which one should I take?’ (2nd Gen)

Additional first-generation speakers corroborate this pattern from different perspectives. Another speaker (F49CHF31072024) noted increasing interchangeability between medial (uva) and proximal (iva) forms in everyday interaction, suggesting ongoing levelling within the community. A younger first-generation speaker (M27CHG17082024) reported that he avoids u- forms altogether, associating them with non-respectful or emotionally marked contexts, although older speakers continue to use them. In his interpretation, u-forms tend to occur when referring to a third person in critical discourse (for instance when commenting on or complaining about someone), rather than in their earlier deictic function of referring to entities accessible to speaker and hearer. He gave an example for its use in (5) where the medial form uva does not refer to an entity physically near both the speaker and the hearer, but instead to a referent that is mutually familiar to both interlocutors.

5.	uvai-ya	sonn-a-vai	na:n	e:n	ke:k-an-um
	they.med-emph	say.pst-rel-3pl	I.sg	why	listen-pres-1sg
	‘Why should I listen to what they said?’

Second-generation speakers consistently report having heard medial pronouns in the speech of first-generation speakers, particularly parents, but also report an inability to interpret their meaning or use them productively. Together, these observations point to functional narrowing and reanalysis of the medial pronoun system among first-generation speakers, alongside near-complete absence among the second generation.

The convergence of quantitative distributional evidence and speaker-based metalinguistic commentary indicates that while proximal and distal pronouns remain stable in Swiss Tamil, the medial category has undergone extensive attrition. This pattern reflects structured intergenerational change affecting a historically conservative and dialect-specific feature, rather than random variability or a grammatical change.

4.2.2. Plural Morphology

Plural morphology constitutes a particularly informative domain in the present data because it involves two structurally distinct phenomena that arise from different sources of variation. On the one hand, third-person plural agreement on the verb reflects a dialect-specific property of Jaffna Tamil, realised through the plural agreement -inam marked on the verb for the rational5 subjects, which is not shared with other Tamil varieties. Variation in this domain therefore bears directly on the maintenance of regional morphosyntactic features under migration. On the other hand, plural marking in quantified noun phrases is a property of Tamil that is closely tied to register differences between spoken and written varieties, with plural suffixes being optional or omitted in numeral contexts in spoken variety (this applies across all dialects). These two contexts of plural marking thus involve different grammatical mechanisms and sociolinguistic influence and must be analysed separately. Accordingly, this section first examines verbal plural agreement in rational contexts (Section Third-Person Plural Agreement on the Verb), followed by plural marking in quantified noun phrases (Section Gender-Marked Third-Person Plural Pronoun and Verbal Agreement).

Third-Person Plural Agreement on the Verb

In Jaffna Tamil, agreement with third-person plural rational subjects is realised by a verbal suffix typically transcribed as -inam (with phonological variants such as -yinam depending on stem shape). This marker is a well-documented feature of northern Sri Lankan Tamil varieties and does not occur in other Tamil varieties in Sri Lanka and India, which instead employ forms such as -a:ŋga/-a:rga(l) (Suseendirarajah, 1979).

Importantly, the -inam suffix is not merely a plural marker but is also associated with politeness and neutralisation of gender distinctions in third-person plural contexts (Suseendirarajah, 1979). Descriptive grammars note that while gender-specific agreement forms exist in Jaffna Tamil (Suseendirarajah, 1979; Thananjayarajasingham, 1977), -inam functions as a polite plural form for both masculine and feminine rational referents.

The contrast between Jaffna Tamil and Standard Spoken Tamil third-person plural agreement morphology is summarised in

Table 8. Third-person plural agreement for rational nouns in Tamil varieties.

Tense	Jaffna Tamil Form	Standard Spoken Tamil Form
Present	seyr-inam/sey-yinam do.pres-3pl	seyr-a:ŋga do.pres-3pl
Past	seyt-inam do.pst-3pl	seyt-a:ŋga do.pst-3pl
Future	seyv-inam do.fut-3pl	seyv-a:ŋga do.fut-3pl

, which illustrates the distribution of -inam versus -a:ŋga(l)/-a:rgal across tense categories.

The grammatical distinction targeted in the elicitation task is illustrated schematically in examples (6a–b), where the same propositional content is realised with Jaffna Tamil agreement morphology in (6a) and Standard Spoken Tamil morphology in (6b). At this stage, these examples serve to establish the descriptive contrast reported in the literature and do not yet reflect speaker preferences in the present dataset.

6.a.	kuma:r-nda	snehidar-gal	chennai-yil	ve:lai	sey-yinam
b.	kuma:r-in	snehidar-gal	chennai-yil	ve:lai	seyr-a:ŋga
	Kumar.3sgm-gen	friend-3pl	Chennai-loc	work	do.pres.3pl.rat
	‘Kumar’s friends work in Chennai.’

Turning to the elicitation results, five speakers frequently alternated between -inam and -a:ŋga forms. And 11 out of 20 speakers initially produced -a:ŋga forms, which are characteristic of Standard Spoken Tamil, even when responding to prompts designed to elicit Jaffna Tamil agreement. However, one first-generation speaker explicitly reported using -inam in everyday speech, and five first-generation speakers switched to this form once the elicitation context was clarified. This behaviour suggests that the alternation observed during elicitation reflects register sensitivity and stylistic choice (likely linked to the interview situation) rather than categorical change in agreement morphology.

The quantitative distribution of responses is presented in

Table 9. Distribution of third person plural marker for rational nouns (e.g., 6a–b).

Category	Subgroup	Responses with JT Form	Responses with SST Form	Responses with Both JT and SST Forms	Responses with Invariant Form
		seyy-inam	seyr-a:ŋga
Gender	Male	2	3	1	1
	Female	1	8	4	0
Generation	First	1	4	5	0
	Second	2	7	0	1
Region	CHF	2	4	3	1
	CHG	1	7	2	0

, which differentiates between productions of the Jaffna Tamil form (-inam), the Standard Spoken Tamil form (-a:ŋga), mixed responses where both forms were accepted or produced, and invariant or non-target forms. The results show a clear generational asymmetry. While first-generation speakers frequently show access to both agreement forms, second-generation speakers show a stronger preference for -a:ŋga forms and are less likely to produce or accept -inam.

To explain sociolinguistic cues further, a closer inspection of the distribution in Table 9 shows that the observed variation is systematically aligned with the speakers’ dialectal background and acquisition trajectories. Among first-generation speakers who produced only the Standard Spoken Tamil form (-a:ŋga), the majority are non-Jaffna Tamil speakers (Batticaloa, Up-country, or Colombo Tamil), for whom -inam is not the form in their spoken form. The remaining one speaker in this group consistently used a formal Tamil across tasks, which plausibly accounts for the avoidance of dialect-specific spoken forms. Conversely, the limited use of -inam among first-generation speakers is restricted to those with a clear Jaffna Tamil background, consistent with descriptive accounts of this marker as dialect-specific rather than across Tamil varieties.

Among second-generation speakers, reliance on -a:ŋga is even more pronounced. Although several second-generation speakers have Jaffna Tamil as their dominant parental input, most nevertheless produced only the Standard Spoken Tamil form. This pattern suggests an ongoing shift away from the dialect-specific agreement marker even among speakers with Jaffna Tamil backgrounds. At the same time, the presence of a small number of second-generation speakers who produced -inam indicates that access to the marker has not been categorically lost but is unevenly retained and highly sensitive to individual acquisition histories and input conditions.

The occurrence of an invariant agreement form (-o:m, 1pl instead of -inam or -a:ŋga, 3pl forms) in one second-generation speaker further supports this interpretation. This speaker (M21CHF17072024), a male participant from the French-speaking region, has a documented delayed and interrupted acquisition trajectory, having begun active Tamil use relatively late and primarily through media exposure rather than sustained home interaction. The invariant form thus plausibly reflects incomplete acquisition of the agreement paradigm (see for instance Putnam & Sánchez, 2013; Montrul & Silva-Corvalán, 2019) rather than dialect levelling or restructuring of the system as a whole.

Finally, speakers who accepted or produced both agreement forms are predominantly found among the first generation, indicating metalinguistic awareness of competing agreement systems rather than instability. In contrast, such dual access is markedly reduced in the second generation, where speakers tend to converge on one form which is a widely shared spoken form. Based on the observations above, these patterns point to dialectal realignment toward a supra-regional spoken norm rather than uniform simplification or contact-induced restructuring.

Gender-Marked Third-Person Plural Pronoun and Verbal Agreement

In addition to the neutral plural agreement marker -inam, descriptive work on Jaffna Tamil has documented a system of gender-marked third-person plural forms, both in pronouns and in verbal agreement. As noted by Thananjayarajasingham (1977) and Suseendirarajah (1979), Jaffna Tamil distinguishes masculine and feminine plural referents through forms such as avangal (3plm) vs. avalgal (3plf), with corresponding verbal agreement morphology. This system contrasts with Standard Spoken Tamil, which employs a gender-neutral plural form (avaŋal and avargal) in the spoken and written varieties respectively.

Table 10. Gender-marked third-person plural pronouns and agreement in JT.

Gender	3pl Pronoun (Jaffna Tamil)	Standard Spoken Tamil
Masculine	avangal kodut-a:ngal 3plm   give.pst-3plm	Not available
Feminine	avalgal kodut-a:lgal/kodut-ave 3plf    give.pst-3plf	Not available
Neutral	avargal kodut-inam 3pl     give.pst-3pl.rat.pol	avaŋgal kodut-a:ŋgal 3pl     give.pst-3pl.rat.pol

summarises this contrast for the sentence ‘They gave’.

Importantly, the gender-marked plural system in Jaffna Tamil is not obligatory across all registers. Descriptive grammars note that gender-marked plural agreement is more likely to occur in careful or emphatic contexts, while neutral forms are increasingly used in everyday speech. Nevertheless, the availability of gender distinction in the plural paradigm is treated as a structural property of the dialect, setting Jaffna Tamil apart from other Tamil varieties.

To examine whether this dialect-specific distinction is maintained in Swiss Tamil, the elicitation task included two items targeting gender-marked plural agreement: one with a feminine plural referent (Item 18a) and one with a masculine plural referent (Item 18b). Speakers were expected to produce a gender-marked plural agreement form; responses using a gender-neutral plural form or no response were coded as mismatches. The quantitative distribution of responses is presented in

Table 11. Distribution of gender-marked third-person plural agreement (Items 18a–b).

Category	Subgroup	Responses with Gender-Marked Form	Responses with Neutral Form
Gender	Male	0	7
	Female	3	10
Generation	First	1	9
	Second	2	8
Region	CHF	2	8
	CHG	1	9

.

Across both items, the results (shown in bold in Table 11) show a near-categorical avoidance of gender-marked plural agreement. For the feminine plural context (Item 18a), no male speakers and only three female speakers produced the expected gender-marked form. Similarly, for the masculine plural context (Item 18b), responses overwhelmingly favoured gender-neutral agreement, with the same three female speakers producing gender-marked forms. Crucially, many speakers explicitly reported that they “do not use” plural forms such as avalgal or avangal with gender distinction, instead defaulting to a neutral plural (avangal, avargal).

This pattern differs qualitatively from the variation observed for the -inam marker. In the present case, speakers do not alternate between two productive options; rather, the gender-marked plural forms appear to be largely absent in both first- and second-generation speakers. At the same time, the limited number of speakers who produced gender-marked plural forms indicates that the system has not been categorically lost. These responses are scattered across generations and regions, pointing to residual availability rather than productive use.

Based on the findings, one could observe that the gender distinction in third-person plural agreement represents a highly vulnerable morphosyntactic feature in Swiss Tamil. Unlike -inam, which remains variably available through register competition, gender-differentiated plural agreement appears to be undergoing advanced attrition. This attrition is plausibly linked to its marginal status even in contemporary Jaffna Tamil and to the dominance of gender-neutral plural forms across Tamil varieties more generally.

Plural Marking in Quantified Noun Phrases

Plural marking in quantified noun phrases represents a morphosyntactic domain that is not dialect-specific but is instead sensitive to register and rationality (human vs. non-human). Across Tamil varieties, plural suffixes are obligatory on human nouns but optional on non-human nouns, particularly when plurality is independently expressed by a numeral or quantifier (Lehmann, 1989). This optionality is especially characteristic of contemporary spoken Tamil, whereas written Tamil typically retains overt plural marking in the same contexts. Variation in this domain therefore reflects register choice rather than grammatical restructuring. Examples (7a–b) illustrate this contrast between spoken (7a) and written (7b) strategies in quantified noun phrases.

7.a.	na:n	na:lu	puthagam-∅	vech-irukk-e:n-nu	kuma:r	so-nn-a:n
	I.1sg	four	book-∅	keep.adv.ptcp-be.pres-1sg-comp	kumar.3sgm.nom	say-pst-3sgm
	‘Kumar said I have four books’

b.	na:n	na:lu	puthagan-kal	vaith-iru.kkir-e:n	endru	kuma:r	so-nn-a:n
	I.1sg	four	book-3pl	keep.adv.ptcp-be.pres-1sg	comp	kumar.3sgm.nom	say-pst-3sgm
	‘Kumar said I have four books’

To test this, the linguistic task (wug-test) included sentences with quantified noun phrases, including wug-test items and nouns requiring oblique stems for plural and case suffixation. Across both generations in the present dataset, responses show variation rather than categorical behaviour: some speakers omitted plural suffixes in quantified contexts, while others included them. This mixed pattern suggests that speakers have access to both spoken and written strategies for plural marking and alternate between them.

The quantified noun phrases were examined across four lexical items: a nonce English item (wug), a nonce Tamil item (karugam), and two established Tamil nouns (puthagam ‘book’ and mi:n ‘fish’). Responses were coded according to whether they matched spoken Tamil norms (plural omission) or written Tamil norms (overt plural marking).

Although this alternation is well attested in homeland Tamil and is not specific to migrant or heritage grammars, it remains relevant to examine how speakers in a heritage context distribute spoken- and written-aligned strategies. In the present study, this domain did not appear in the ≥30% item-based variability table (see Table 4; Section 3.3) because both spoken and written plural forms were coded as acceptable responses. However, given the register-sensitive nature of the phenomenon, responses were re-analysed using a secondary coding distinguishing the plural marking in spoken and written variety.

Table 12. Plural marking in quantified noun phrases.

Noun	Generation	Spoken Variety-Aligned	Written Variety-Aligned
wug	First	10	0
	Second	4	6
karugam	First	7	3
	Second	2	8
puthagam	First	6	4
	Second	3	7
mi:n	First	5	5
	Second	3	7

presents the distribution of spoken- and written-aligned responses by generation.

The results reveal a systematic generational asymmetry. First-generation speakers overwhelmingly favour spoken-aligned strategies across all lexical items, particularly with nonce forms (wug and karugam), consistent with homeland spoken Tamil norms. Second-generation speakers (shown in bold in Table 12), while still permitting spoken-aligned forms, show a stronger tendency toward written-aligned plural marking, especially with established Tamil nouns (puthagam and mi:n).

Crucially, this pattern does not indicate loss of grammatical knowledge. Rather, it reflects differential sensitivity to register and lexical familiarity. Spoken-aligned strategies are most robust with non-lexicalised or novel items, whereas written-aligned strategies are more likely to surface with familiar Tamil nouns, particularly among second-generation speakers. This suggests that plural marking choices are mediated by interaction between register awareness and lexical status, rather than by categorical grammatical restructuring.

The findings from this test indicate that plural marking in quantified noun phrases remains structurally intact in Swiss Tamil. Variation in this domain reflects selective redistribution of spoken and written strategies across generations, reinforcing the broader conclusion that morphosyntactic variation in Swiss Tamil is domain-specific and register-sensitive.

4.2.3. Negation Patterns

Negation in Swiss Tamil shows two related patterns of variation: (i) the distribution of the existential negator illa versus the equative negator alla, and (ii) the weakening of agreement marking in negated dative-subject constructions. This section discusses these two phenomena and their observation in Swiss Tamil in detail.

Existential vs. Equative Negation

In Old Tamil, negation traditionally distinguishes between illa, used for existential negation (‘there is no…’), and alla, used for equative predication (‘X is not Y’). While this distinction (given in example 8a and 8ba) is reported to be robust in older descriptions of Jaffna Tamil (Gair & Suseendirarajah, 1981), it is known to be weakened or lost in many contemporary spoken varieties, where illa (example 8c) is increasingly used across contexts (Wilden, 2018).

8.a.	aŋga	tanni	illa
	there	water	be.ext.neg
	‘There is no water there.’

b.	bala	oru	samajalka:ran	alla
	Bala.3sgm.nom	one	cook.nom.3sgm	be.equ.neg
	‘Bala is not a cook.’ (Used in Old Tamil and Jaffna Tamil)

c.	bala	oru	samajalka:ran	illa
	Bala.3sgm.nom	one	cook.nom.3sgm	be.equ.neg
	‘Bala is not a cook.’ (Used in Modern Tamil)

The elicitation task included items targeting this distinction (Items 39–40). In contexts where alla would be expected under a strict Jaffna Tamil analysis, most speakers instead accepted or produced illa. Quantitative results show that responses matching the expected alla form are rare and restricted almost entirely to first-generation speakers. In contrast, second-generation speakers overwhelmingly reject alla and rely exclusively on illa, even in equative environments.

Importantly, even among first-generation speakers who accepted alla, its use appears highly restricted. Several speakers commented that alla sounds formal, emphatic, or contrastive, and that illa would be preferred in everyday speech. This suggests that alla remains available in the grammar of some first-generation speakers but is no longer part of their default negation strategy. Among second-generation speakers, the distinction is effectively absent.

Table 13. Responses for Equative Negation by illa vs. alla.

Category	Subgroup	Equative Negation by illa		Equative Negation by alla
		Yes	No	Yes	No
Gender	Male	7	0	1	6
	Female	13	0	3	10
Generation	First	10	0	3	7
	Second	10	0	1	9
Region	CHF	10	0	2	8
	CHG	10	0	2	8

(Yes = accepted as grammatical in the judgement task) shows the distribution of preferences for illa vs. alla across speaker groups elicited for item numbers (39) and (40) (the responses are the same for both items).

Agreement in Negated Dative-Subject Constructions

A related pattern emerges in negated experiencer constructions with dative subjects. In Tamil, such constructions may involve an infixal negation strategy in which the verb carries a neuter agreement suffix -thu, as given in example (9).

9.	ena-kku	avan-ai	teriy-a:-thu
	1sg-dat	he.3sg-acc	know-neg-n
	‘I don’t know him.’

The other pattern of marking negation is with the complex predicate combination: infinitive verb + illa pattern, which lacks tense and agreement morphology as illustrated in example (10).

10.	na:n	un	kanna:diy-ai	ka:n-a-yilla
	1sg.nom	your	spectacles-acc	see-inf-neg
	‘I didn’t see your spectacles.’

In the present dataset, speakers across both generations produced the pattern in example (10) without any variation, but there is a minor variation in negated dative-subject constructions (example 9). The data indicates that two male second-generation speakers (each one from CHF and CHG regions) omit the default agreement suffix (-thu) in these contexts, resulting in forms that align more closely with the general infinitive + illa pattern. But in contrast the first-generation speakers more consistently retain (-thu) marking.

The negation data show that Swiss Tamil exhibits levelling toward illa as a general negator in copular negation contexts, alongside reduced sensitivity to agreement distinctions in negated dative-subject constructions. These changes parallel patterns reported for contemporary spoken Tamil, while also revealing an emerging pattern involving agreement marking. Importantly, both phenomena differ in degree across generations: first-generation speakers retain greater residual access to older distinctions, whereas second-generation speakers show more advanced levelling and a tendency toward innovative agreement patterns.

4.2.4. Binding Interpretation

This section examines variation in the interpretation of third-person pronouns with respect to Chomsky’s (1981) Principle B of the binding theory, focusing on two elicitation items (31b and 31a) that yielded systematic intergenerational differences in the Swiss Tamil data.

As per the classical definition of Binding Theory, anaphors are contrasted with pronouns in terms of how they pick out reference within a sentence. While anaphors are required to have a local antecedent within the same sentence, pronouns are not required to have a local antecedent. For instance, in the sentence ‘John likes himself’, ‘John’ serves as the local antecedent for the anaphor ‘himself’. This can be contrasted with pronouns in a sentence such as ‘John likes him’, where ‘John’ does not serve as a local antecedent for the pronoun ‘him’. Chomsky characterises these distributional properties of anaphors and pronouns as Principle A and Principle B, respectively. Extending this observation to Tamil, a third-person pronoun such as aval ‘she’ is not expected to take local antecedent within the same clause and must instead refer to a discourse-antecedent or higher-clausal antecedent.

However, Tamil is known to permit a wider range of pronominal interpretations than strictly configurational languages like English, particularly in embedded contexts and in sentences involving reflexive or “mirror” predicates. Testing Principle B interpretations therefore provide a window into how anaphoric constraints are acquired, maintained, or reanalysed in heritage grammar.

The present study targeted two interpretive environments in which Principles A and B of binding theory are expected to apply but may be defeasible. Of the eight Items designed to test these environments, only two (item 31a and 31b) showed systematic variation: (i) third-person pronouns in embedded clauses (Item 31b), and (ii) third-person pronouns in simple clauses with reflexive argument structure (Item 31a).

Consider example (11) from Item (31b), where Principle B predicts that aval can either refer to matrix subject Meena or any other referentially salient individual.

11.	mi:na	[aval	idli-ai	sa:pi-tt-a:l-nnu]	so-nn-a:l.
	Meena.3sgf.nom	she.3sgf.nom	idli.3sgn-acc	eat-pst-3sgf-comp	tell-pst-3sgf
	‘Meena said she ate the idli.’

The results (

Table 14. Distribution of third-person pronoun Principle B interpretation (Item 31b).

Category	Subgroup	Responses with Meena/Someone Else	Responses with Meena	Responses with Someone Else
Gender	Male	3	3	1
	Female	5	7	1
Generation	First	1	7	2
	Second	7	3	0
Region	CHF	3	6	1
	CHG	5	4	1

) show a clear generational asymmetry. Second-generation speakers overwhelmingly allow aval to refer to Meena/someone else. In contrast, first-generation speakers show a stronger preference for the referentially salient individual (Meena). This pattern suggests that there is a move from the potential ambiguity to disambiguation strategy by consistently referring the pronoun to the matrix subject.

This strategy is expanded and regularised among second-generation speakers, particularly in the German-speaking region. This shift plausibly reflects the influence of heritage language schooling and prescriptive grammatical instruction, where pronouns in reported speech are often taught as coreferential by default. Rather than indicating gradual loss, the pattern points to reanalysis of an existing ambiguity, with one interpretation becoming dominant.

However, the disambiguation preference employed in the complex clause is absent in the simple ‘mirror’-predicate clause like the following.

12.	mi:na	aval-ai	kanna:Di-le	pa:r-t-a:l.
	Meena.3sgf.nom	she-acc	mirror-loc	see-pst-3sgf
	‘Meena saw her in the mirror.’

In this sentence (12), Principle B would predict that aval should refer to someone other than Meena. However, the presence of a reflexive-like argument structure (‘mirror’) introduces an additional source of ambiguity. In Tamil, mirror predicates allow interpretations where the pronoun may refer either to the subject (Meena) or to a salient referent in the discourse.

As shown in

Table 15. Distribution of third-person pronoun Principle B interpretation (Item 31a).

Category	Subgroup	Responses with Someone Else	Responses with Meena	Responses with Meena/Someone Else
Gender	Male	1	3	3
	Female	1	8	4
Generation	First	1	9	0
	Second	1	2	7
Region	CHF	1	7	3
	CHG	1	4	4

, responses cluster into three types: Meena, someone else, and an explicitly ambiguous reading Meena/someone else. Importantly, the ambiguous response type is dominant among second-generation speakers, whereas first-generation speakers show a stronger bias toward the Principle B-compliant interpretation.

From the above observations, items (31a) and (31b) reveal that variation in binding interpretation in Swiss Tamil reflects systematic differences in how ambiguity is resolved across generations. In embedded contexts (31b), second-generation speakers strongly favour Meena, the matrix subject interpretation, suggesting a reanalysis of pronoun interpretation under reported speech. In reflexive environments (31a), second-generation speakers are more likely to preserve ambiguity rather than enforce a single Principle B reading.

These findings suggest that ambiguity by itself is not problematic for heritage speakers; rather, the grammatical context in which the ambiguity arises is crucial. While simple mirror predicates retain ambiguity, complex sentence structures lead to disambiguation. Therefore, Principle B itself is not lost, but its interaction with discourse structure and reflexivity is reweighted in heritage grammars. The increased intuition for ambiguity among second-generation speakers is consistent with the influence of explicit instruction and reduced exposure to subtle discourse-conditioned contrasts in the input.

Crucially, this pattern supports the broader claim of the paper that morphosyntactic variation in Swiss Tamil is domain-specific and interpretive, involving restructuring of constraint application rather than a whole grammatical simplification in heritage grammar.

4.2.5. Subject pro-Drop

Tamil is a canonical pro-drop language, where subject information can be recovered from verbal agreement morphology (Lehmann, 1989). Previous work on heritage Tamil (Polinsky, 1997) has suggested that heritage speakers may show a reduced use of pro-drop, potentially favouring overt subjects. In the present study, however, the pattern does not support a decline in the use of pro-drop. Instead, the data point to a preference-based asymmetry: speakers retain the grammatical option of subject omission but favour overt subjects in elicited contexts. This preference is best understood in terms of discourse pragmatics and elicitation effects.

Subject pro-drop was tested (GJT) using short discourse contexts designed to licence omission of the subject. In each item, the referent was made salient in the preceding context, such that a response without an overt subject pronoun would be pragmatically appropriate. One of the contexts (Item 35a-b) used is given below in example (13) in which the participant has to choose the most acceptable answer, i.e., either Item 35a or Item 35b.

13.	Context: A father is looking for his daughter Meena and asks his wife where she is. Meena enga irukka? (‘Where is Meena?’)
a.	ava pallikku poiyirukka ‘She has gone to school.’
b.	pallikku poiyirukka ‘(She) has gone to school.’

Table 16. Subject pro-drop responses by item.

Category	Subgroup	(34b)		(34c)		(35b)
		Yes	No	Yes	No	Yes	No
Gender	Male	3	4	4	3	4	3
	Female	9	4	9	4	9	4
Generation	First	6	4	6	4	7	3
	Second	6	4	7	3	6	4
Region	CHF	5	5	5	5	5	5
	CHG	7	3	8	2	8	2

presents responses to three elicitation items (34b, 34c, 35b) testing subject pro-drop in discourse-licenced contexts. The response ‘Yes’ refers to pro-drop being possible and ‘No’ refers to pro-drop not being possible. Each item is reported separately to preserve item-specific effects while allowing comparison across gender, generation, and region.

Across all three items (34b, 34c and 35c), both first- and second-generation speakers (shown in bold in Table 16) show a consistent preference for overt subjects, despite the discourse contexts licencing pro-drop. Crucially, pro-drop responses are attested across all groups and items, indicating that subject omission remains grammatically available but is pragmatically dispreferred. The similarity of response patterns across items suggests a stable preference rather than item-specific effects.

4.2.6. Case Marking Variation

Case marking in Tamil encodes a range of semantic and pragmatic distinctions, including animacy, definiteness and argument structure. Previous work has shown that these distinctions are particularly sensitive to contact and reduced input in heritage and migrant grammars (Polinsky, 1997; Montrul & Sánchez-Walker, 2013). This section examines patterns of case marking in Swiss Tamil, focusing on differential object marking (DOM) and argument case alternations, with particular attention to generational differences.

Differential Object Marking

Tamil exhibits differential object marking whereby accusative case marking is conditioned by factors such as rationality (humans vs. non-humans), specificity, and discourse prominence. In homeland varieties (both India and Sri Lanka), human objects (14) and definite objects (example 15b) are more likely to be marked with accusative case, whereas non-human objects that are not marked for definiteness or specificity may surface without overt case marking (15a).

14.	me:ri	tan	to:zhiy-ai	pa:r-t-a:l
	Mary.3sgf.nom	self.3sg	friend-acc	see-pst-3sgf
	‘Mary saw her friend.’

Let us consider example (15a). Here, accusative case marking is not realised because the object ‘butterfly’ is non-human and not specified as definite or discourse prominent. In contrast, in (15b), accusative marking is present on ‘butterfly’ due to the presence of a demonstrative, which encodes specificity.

15.a.	me:ri	vanna:thupoochi-∅	pidi-tu	vilaja:du-r-aa:
	Mary.3sgf.nom	butterfly-∅	catch-adv.ptcp	play-pres-3sgf
	‘Mary is catching butterfly and playing.’

b.	me:ri	antha	vanna:thupoochi-ai	pidi-tu	vilaja:du-r-a:
	Mary.3sgf.nom	that.dem	butterfly-acc	catch-adv.ptcp	play-pres-3sgf
	‘Mary is catching that butterfly and playing.’

In heritage language contexts, DOM has been shown to be particularly vulnerable. Montrul and Sánchez-Walker (2013, p. 118), for instance, report systematic omission of DOM in English-dominant heritage Spanish and attribute this pattern to reduced and inconsistent input during acquisition. Similar effects have been reported in other contact settings (cf. Scontras et al., 2015).

The present study tested DOM using elicited sentences with irrational nouns and definiteness of the object. The results reveal low overall rates of accusative marking across both generations, but with distinct conditioning factors.

Table 17. Differential object marking: definite vs. discourse-salient contexts.

Category	Subgroup	DOM Present in Definite Marked (Item 4a)		DOM Marked in Discourse Salient (Item 11)
		Yes	No	Yes	No
Gender	Male	2	5	2	5
	Female	5	8	8	5
Generation	First	3	7	4	6
	Second	4	6	6	4
Region	CHF	2	8	5	5
	CHG	5	5	5	5

(In the table, ‘Yes’ indicates the presence of DOM and ‘No’ for the absence) summarises this distribution and highlights a generational shift in the cues guiding DOM use.

The contrast between Items (4a) and (11) reveals a clear difference between morphologically encoded definiteness and discourse-based salience in conditioning differential object marking. In Item 4a, where definiteness was explicitly marked on the noun phrase, DOM was expected; however, only a few speakers produced DOM in this context. This holds for both generations and indicates that overt morphological definiteness alone does not strongly trigger DOM in elicited production.

In contrast, Item (11) involved a bare noun without morphological definiteness, but the object was discourse-salient due to the picture-based context, which rendered a specific referent prominent. In this condition, second-generation speakers show higher rates of accusative marking than in Item (4a). This suggests that discourse salience introduced through context plays a stronger role in triggering DOM than overt definiteness marking for this group.

First-generation speakers, by contrast, show relatively similar behaviour across the two items, with no strong increase in DOM in the discourse-salient condition. Taken together, these results point to a generational difference in how cues for object marking are weighted: second-generation speakers appear to rely more on discourse-pragmatic salience than on morphological definiteness, whereas first-generation speakers show more stable but overall lower rates of DOM across both contexts (see the results in bold in Table 17).

At first glance, this pattern appears to diverge from previous findings on heritage DOM, which report increased omission of accusative marking among heritage speakers (e.g., Montrul & Sánchez-Walker, 2013). However, the difference is only apparent. While heritage Spanish speakers show reduced sensitivity to DOM-triggering features, the Swiss Tamil data suggest that second-generation speakers retain DOM but specific to the conditions under which it is provided. Rather than responding primarily to morphologically encoded definiteness, second-generation speakers are more responsive to discourse-based salience. In this respect, the Swiss Tamil data extend heritage DOM findings by showing that reduced input does not necessarily lead to uniform omission but can result in selective reorganisation of the factors conditioning DOM.

Case Marking: Genitive and Instrumental Case

This section examines two case-related domains that show limited but systematic variation in the dataset: (i) genitive case marking, and (ii) instrumental case marking and its alternation with the associative marker. These patterns are discussed together because both involve optionality or substitution in case marking because neither pattern is evenly distributed across speaker groups.

• Genitive case marking

In Jaffna Tamil, genitive case marking is generally obligatory and realised through the suffix -nda. In several other spoken Tamil varieties -o:da suffix is used; however, genitive marking may be optionally omitted. Item 5 tested whether speakers retain or drop genitive marking in contexts where Jaffna Tamil would require it.

16.	ba:lu-(nda)	talayi-la	thoppi	iru-kku
	Balu.3sgm.gen	head-loc	hat.3sgn	be-pres.3sgn
	‘The hat is on Balu’s head.’

Responses were coded according to whether speakers produced the genitive marker (matching the expected Jaffna Tamil form) or omitted it.

Table 18. Genitive case marking—Item (5).

Category	Subgroup	Genitive Present	Genitive Dropped
Gender	Male	5	2
	Female	9	4
Generation	First	8	2
	Second	6	4
Region	CHF	5	5
	CHG	9	1

(“Genitive present” indicates production of the expected genitive marker; “Genitive dropped” indicates omission) summarises the distribution of genitive marking across gender, generation, and region.

The results show that genitive marking is retained by a majority of speakers, but omission is attested across both generations. Importantly, genitive omission is more frequent among the second-generation speakers.

• Instrumental vs. associative case marking

The instrumental case marking is marked by -a:l in Tamil (example 17), while most speakers produced the instrumental form, a small number instead used the associative (sociative) case marker (-o:da). These responses were coded as mismatches, as they involve substitution of one case marker for another rather than omission.

17.	me:ri	bae:tt-a:l	avan-ai	adi-t-a:l
	Mary.3sgf.nom	bat-inst	he-acc	hit-pst-3sgf
	‘Mary hit him with a bat.’

Table 19. Instrumental vs. associative case marking—Item (8b).

Category	Subgroup	Instrumental Used	Associative Used
Gender	Male	6	1
	Female	10	3
Generation	First	10	0
	Second	6	4
Region	CHF	7	3
	CHG	9	1

(“Instrumental used” indicates the expected instrumental case marker; “Associative used” indicates substitution with the associative case marker) presents the distribution of instrumental versus associative marking across speaker groups.

A striking pattern emerges from the distribution: all instances of associative marking in this context were produced by second-generation speakers (four speakers in total). First-generation speakers uniformly retained the instrumental marker. This indicates that the instrumental–associative alternation is not evenly distributed across the sample but is restricted to a small subset of second-generation speakers.

The use of the associative marker does not indicate loss of case marking, since the relation is still overtly marked. Rather, it suggests a limited reanalysis in which the associative marker is extended into contexts typically associated with instrumental semantics. Given the small number of speakers involved, this pattern should be interpreted cautiously; nevertheless, it points to a generationally restricted innovation rather than broad restructuring of the case system.

From the findings with genitive case and instrumental case marking, genitive optionality and instrumental–associative alternation shows that case marking in Swiss Tamil remains largely intact, with variation concentrated in specific domains. Genitive marking variability aligns with dialectal input differences, while the instrumental–associative alternation appears as a limited, second-generation-specific pattern. Neither domain shows evidence of categorical loss; instead, the data point to selective variation shaped by input diversity and generational transmission.

4.3. General Discussion

This study examined morphosyntactic variation in Swiss Tamil with a focus on differences between first- and second-generation speakers across several morphosyntactic domains. Rather than showing uniform change across the grammar, the results reveal domain-specific patterns shaped by register, discourse context, and generation.

Across domains, register sensitivity emerges as a central factor. Speakers frequently alternate between spoken Tamil patterns and more written-like forms in elicited tasks. Second-generation speakers who attended Tamil heritage schools show clearer awareness of this distinction and maintain sharper register separation, while others produce mixed forms. Importantly, this alternation is systematic and task-dependent rather than random, indicating that speakers possess multiple grammatical options and select among them based on context.

Research on heritage languages in diglossic settings provides a useful point of comparison. Studies on heritage Arabic show that speakers raised outside the homeland context maintain core morphosyntactic properties while displaying systematic variation linked to register, task type, and literacy exposure rather than unsystematic change (Albirini & Benmamoun, 2022). Similar patterns are observed in Swiss Tamil, where speakers shift between spoken and written-like forms depending on the elicitation context, with clearer register separation among second-generation speakers who attended heritage schools. These patterns suggest that variation in Swiss Tamil reflects differences in register use shaped by literacy practices and sociolinguistic context, rather than uniform change in the heritage grammar. This interpretation aligns with Brown and Bousquette’s (2018) argument that heritage language grammars must be analysed within their specific sociolinguistic ecologies, where patterns of language use across contexts are central to understanding observed variation.

In domains such as plural marking in quantified noun phrases, variation aligns closely with spoken versus written norms. Both generations demonstrate access to spoken strategies that allow omission of plural marking under quantification, alongside written-style forms with overt plural suffixes. This suggests continuity with homeland spoken Tamil rather than restructuring specific to heritage grammars.

In contrast, dialect-specific features associated with Jaffna Tamil (homeland), such as gender-marked plural agreement and the illa–alla negation distinction, show clearer generational asymmetries. Gender-marked plural agreement is largely avoided by both generations, with only limited residual use. Similarly, the equative negator alla is restricted to a small subset of first-generation speakers and is effectively absent from second-generation usage. These patterns indicate that features with low frequency and limited functional load in contemporary input are particularly vulnerable across generations.

Case marking presents a more nuanced picture. Differential object marking (DOM) remains available in both generations but is conditioned by different cues. First-generation speakers show relatively stable, though overall low, use of accusative marking across contexts. Second-generation speakers, however, are more sensitive to discourse-based salience than to morphologically encoded definiteness, marking accusative cases more readily when the referent is contextually prominent. This shift suggests a reweighting of cues rather than loss of the case system itself. Other case-related phenomena, such as optional genitive marking and instrumental constructions, further reinforce the importance of spoken usage patterns. Here, variation reflects broader Tamil dialectal and colloquial norms rather than categorical innovation, with some second-generation speakers extending spoken strategies more consistently.

Taken together, the findings indicate that morphosyntactic variation in Swiss Tamil is shaped by an interaction of generation, register, and discourse context. First-generation speakers tend to retain a wider range of older or dialect-specific distinctions, though often in restricted or non-default ways. Second-generation speakers show greater alignment with spoken, discourse-driven strategies and reduced reliance on morphologically explicit cues, particularly in elicited contexts. Importantly, these patterns cannot be straightforwardly attributed to transfer from the dominant contact languages, as no systematic differences emerge between the Swiss German- and Swiss French-speaking regions across the domains examined. Crucially, these patterns do not reflect whole grammatical loss but rather selective shifts in how existing options are available. Similar conclusions have been drawn in relation to other heritage language contexts, where generational differences are analysed as outcomes of contact-induced diachronic change and reorganisation of grammatical options rather than incomplete acquisition (e.g., Raña-Risso & Barrera-Tobón, 2018).

While the present study focuses on morphosyntactic patterns observed in elicited task data, several sociolinguistic factors may contribute to the distribution of these features among second-generation speakers. In the Swiss diaspora context, Tamil is primarily acquired through family transmission, which means that speakers often inherit the dialect of their parents. At the same time, speakers may encounter other Tamil varieties through community interactions in temples, cultural associations, and Tamil heritage schools. In addition, exposure to Tamil through digital media and cinema may introduce speakers to other regional varieties, including Indian Tamil. These sociolinguistic conditions may influence the frequency and distribution of particular morphosyntactic features, although the present study does not attempt to determine the relative contribution of each factor.

Future work may also benefit from comparing these patterns with other contact settings such as Sri Lanka Malay (e.g., Ansaldo, 2008; Nordhoff, 2009, 2013), where contact-induced change has likewise resulted in complex restructuring rather than straightforward simplification6.

5. Conclusions

This paper has shown that morphosyntactic variation in Swiss Tamil is best understood through a comparative generational lens. While both first- and second-generation speakers retain core grammatical structures of Tamil, they differ in how specific features are maintained, reduced, or reweighted.

Across domains, the results highlight the importance of register and task effects. Many patterns observed in the data reflect well-attested differences between spoken (both Jaffna and other Tamil varieties) and written Tamil rather than changes unique to heritage grammars. Second-generation speakers, especially those with exposure to Tamil literacy, demonstrate strong metalinguistic awareness of register distinctions and actively navigate between them.

At the same time, generation plays a clear role in the retention of low-frequency or dialect-specific features. Elements such as gender-marked plural agreement and equative negation with alla are largely confined to first-generation speakers and are absent or marginal in second-generation usage. In contrast, more general grammatical mechanisms, including case marking and subject pro-drop, remain available across generations but are conditioned by discourse and elicitation context.

Overall, the findings support an account in which Swiss Tamil shows selective reorganisation rather than uniform change in heritage grammar. Differences between first- and second-generation speakers reflect shifts in cue sensitivity, register preference, and discourse contexts shaped by bilingual input conditions. By examining multiple morphosyntactic domains together, this study demonstrates that heritage language outcomes are heterogeneous and domain-specific, underscoring the need for fine-grained, usage-sensitive analyses of grammatical variation in diaspora contexts.

At the same time, the present analysis is necessarily limited by its primary focus on elicited production interpretation and grammatical judgement data, which tend to favour careful, register-aware responses and may underrepresent phenomena such as dominant-language transfer or morphosyntactic simplification. Although narration and translation tasks were also collected, these data were not analysed in the present paper; future work incorporating systematic analysis of narrative speech and translation behaviour is likely to provide further insight into contact-induced effects. In addition, the second-generation speakers examined here are largely early adults (31 years to 21 years old), and more pronounced innovation or restructuring may emerge in younger cohorts or third-generation speakers over time. Extending the analysis to these populations will allow a clearer evaluation of how register, literacy, discourse, and intergenerational change (between the migrant and heritage speakers) interact in shaping heritage grammars.