1 of 29

Social meaning in syntax: Finite vs. non-finite complements in Serbo-Croatian

Predrag Kovačević, Stefan Grondelaers, Marko Simonović,

Stefan Ivanović and Martina Podboj

2 of 29

Aim

Identify and quantify social meanings associated with the use of infinitives and finite clausal complements in Serbo-Croatian

Problem: social meaning and identity considerations are known to stratify the use of phonetic and lexical but not syntactic variants:

I can stylize myself as cool by using English lexis or as “Dutch” by imitating the local accent, but not, or not as easily, through syntax
Syntactic variation not accessible to self-stylization because it is situated too deeply in the linguistic motor [Labov 1993]

3 of 29

Some counter evidence

Grondelaers, S., van Hout, R., van Halteren, H., & Veerbeek, E. (2023). Why do we say them when we know it should be they? Twitter as a resource for investigating nonstandard syntactic variation in The Netherlands. Language Variation and Change, 35(2), 223-245.

Rapid diffusion of stigmatized subject-hun variant in Netherlandic Dutch is boosted by dynamic prestige associations (urban, cool, dissident,…)

But: is subject-hun a syntactic variant? Labov (1993): many allegedly syntactic variables are intrinsically lexical (e. g. negative concord: anybody vs. nobody)

4 of 29

The phenomenon

Serbo-Croatian (SC) allows two different kinds of complement clauses after volitional, modal, aspectual verbs

Petar želi da pobedi na takmičenju. ‘da+present’ clauses (DPC)�Peter wants da win.3.sg on competition �‘Peter wants to win the competition.’
Petar želi pobediti na takmičenju. infinitive (INF)�Peter wants win.inf on competition�‘Peter wants to win the competition.’

5 of 29

Regional/ethnic factors that effect the choice

Normative grammars of all varieties of SC accept both INF and DPC

But: DPC discouraged in Croatian

Double unofficial norm in Serbia:

DPC as a “Balkanism”
INF as Croatian

Regional

(INF → Northwest; DPC→Southeast)

Ethnic [Kovačević and Milićev 2018]

Croats → INF; Serbs → DPC

6 of 29

Linguistic factors

Syntax:

Future tense constructions: INF strongly preferred [DPC for future - ‘futuroid’; Tanasić 2015]

Clausal subjects: INF preferred [Belić 2005]

Complement size: Larger (TP-sized) → DPC; smaller (VP-sized) → INF [Wurmbrand et al. 2020]

Semantic Factors:

Modal flavor: Epistemic → INF; Root/Deontic → DPC [Kovačević & Milićev 2018]

Subject type: Abstract/inanimate → INF; Concrete/animate → DPC [Arsenijević et al. 2024; Kovačević in press]

Stylistic Pressure:

Avoidance of da-stacking in complex clauses

7 of 29

To make life even more complicated: social meanings

In Croatia, DPC-constructions are perceived as Serbian

In Bosnia, the constructions strongly index ethnicity

In Serbia, INF is associated with Croatians, while DPC is considered a Balkanism

Anecdotal: for young educated speakers from Zagreb, DPC is considered a sign of urbanity and “being above ethnic divisions in the region”

8 of 29

Questions

Can a core syntactic variable like complement choice be linked to identifiable social meanings?

How do we extract these social meanings in a responsible way in a community which is so ethnically polarized?

We want to dig deeper than the most public stereotypes

To go beyond nation-based standard language ideologies and
To pinpoint the social meaning correlates / boosters of the actual language dynamics (and language change)

9 of 29

Method: Stimuli & design

Written materials
Serbian or Croatian (as similar as possible)
6 x 2 x 2 (24 target sentences)
6 different syntactic contexts which correspond to the internal predictors of complement choice (slide 6)
2 complement types: INF or DPC
2 sentence frames per context
Between-design: respondents evaluated only 1/24 sentences

10 of 29

Method: Measures

Acceptability judgments

Rate the acceptability of the sentence based on scale 1-100 (with slider)

Scaled evaluation of social meanings

12 scales targeting Superiority, Dynamism, Personal Integrity, Traditionalism (uses Cyrillic / celebrates Slava / is opposed to EU integration)

Keyword associations to extract social meanings

Write down first three adjectives in reaction to the target sentence (focus of this talk!)

Regional identification

A person who would utter this sentence is most likely from …

10 cities (Belgrade, Novi Sad, Niš, Zagreb, Osijek, Rijeka, Sarajevo, Banja Luka, Mostar, Podgorica)

“Palate cleansers” (covertly extract participants’ (inter)national and social status orientations)

Food choice: high/low status national; high/low status international

Music choice: high/low status national (turbo folk / tavern style); high/low status international

11 of 29

Method: Respondents

Region

Vojvodina (N= 95)
(Central and South) Serbia (N= 103)
Croatia (N = 103)

Age (mean = 27.13)

150 Younger (<39)
151 Older (>39)

Gender

228 female
73 male

12 of 29

Regression model with scaled social meaning ratings

Serbia

No main effect of complement type on acceptability
But INF better with clausal subjects
Superiority strongly predicts acceptability

Croatia

Superiority strongly predicts acceptability
INF deemed (much) more acceptable
Older respondents award higher acceptability
Preference for prestigious international music has negative effect on acceptability

So: social evaluations do not stratify specific complement preferences

13 of 29

Focus on the valence of the keyword returns

Ghyselen, A.-S., Grondelaers, S., Doerga Misier-Patadien, S., & Balesar, U. (2022). Standard language dynamics in postcolonial Suriname: Measuring language attitudes and ideologies in Paramaribo. Lingua, 273

Scaled responses & free response returns to unveil standard language dynamics in Suriname

For 83 % of the keyword returns, we had valences (is this a positive or negative word?) from the psycholinguistic literature (Moors et al. 2013)

Regression analysis on these valences

14 of 29

Alternative approach

Grondelaers, S., Speelman, D., Lybaert, C., & van Gent, P. (2020). Getting a (big) data-based grip on ideological change: Evidence from Belgian Dutch. Journal of Linguistic Geography, 8(1), 49-65

Automatic clustering of responses on the basis of semantic similarity (through vector-based modelling)

Correspondence analysis to map associations of evaluation clusters and investigated varieties

15 of 29

This study

456 response types, 958 tokens

Crucially: no literal repetitions from stimulus materials

Manual clustering (three independent annotators), independently of the stimuli in return to which the keywords were produced

Valence: 1 - positive, 0 - neutral, -1 - negative

10 categories of semantically / conceptually related keywords

16 of 29

17 of 29

Prescriptivism: correctness & prestige

Positive:

učen 'learned’
kulturan 'cultured’,
pismen 'literate’

Negative:

pokondiren 'snobbish’
nenačitan 'awkwardly phrased’

Neutral:

svakodnevan 'everyday (speech)’
običan 'regular’
neformalan 'casual’

Note: one word can be annotated in more than one category (e.g. professional - competence/intelligence; formality/register; prescriptivism; social distance)

18 of 29

Regression

Method:

Separate models fitted for each category (prescriptivism, character, etc.)

Grouping:

Regional grouping based on respondent’s place of residence (Vojvodina, South/Central Serbia, Croatia)

Modelling (cumulative-link model - clm; package ordinal in R)

Dependent variable: valence

Only 2 independent variables:

Region
Complement type

Main effects and two-way interaction

19 of 29

Results

Models did not converge for Age, Physical Traits (N low)

The model for Social Distance did not reach significance

Successful models:

Prescriptivism: (in)correct
Formality/Register: official/casual
Character Traits: honest/arrogant
Emotional Tone: angry/disgusting
Competence/Intelligence: educated, dumb

20 of 29

Results: Prescriptivism / correctness

Interaction Region x CompType (p = .014):

DPC rated higher in Serbia (β= 1.79, p=0.01)

21 of 29

Results: Formality/Register

Interaction Region x CompType (p = .03):

DPC rated higher in Serbia

22 of 29

Results: Character

Main effect of CompType:

DPC rated lower across regions (p = .006)

Interaction Region x CompType (p = .009)

DPC rated higher in Vojvodina (β= 1.09, p=0.03)

23 of 29

Results: Emotional Tone

Main effect of CompType:

DPC rated lower (p = .02)

Interaction Region x CompType (p = .04):

DPC rated higher in Serbia (β=1.14, p=0.04)

24 of 29

Results: Competence

Main effect of CompType:

DPC rated lower (p = .0001)

Interaction Region x CompType (p = .01)

DPC rated higher in Serbia

Interaction Region x CompType (p = .06)

DPC rated higher in Vojvodina

25 of 29

Results: Background associations

26 of 29

Key finding

Many more indexicalities for the Croats

Strong standard language ideology
Obsession with linguistic purity

27 of 29

Discussion: Theoretical Implications

Challenges the claim that syntactic variation lacks social meaning [Labov 1993]: Grammar, too, is a site of identity, ideology, and evaluation

INF vs. DPC alternation perceived as socially meaningful across multiple dimensions (competence, character, tone)

INF/DPC does not carry one fixed stereotype → instead, evokes multiple traits varying by region

28 of 29

Discussion: Regional/National Ideologies

INF = Central European, prescriptively preferred (especially in Croatian standard)

DPC = Balkanism, colloquial default in Serbian, stigmatized in Croatia

Macro-level ideology projected onto syntax:

INF indexes a “modern, cosmopolitan” persona; DPC a “local, traditional” one

29 of 29

Methodology

Preference for a multi-method approach:

Scaled ratings typically underspecify the social evaluation of a construction, but they can typically access deeper, newer indexicalities
Keyword evaluations are much richer but more restricted to older, persistent stereotypes
Keyword evaluations typically return standard language ideologies

Acceptability ratings are not only a reflection of how (grammatically) acceptable a construction is: packed with indexicalities