Phonological acquisition in multilingual speech

Multilingualism

regular use of two or more languages; experienced by over a billion people worldwide.

L3 (Third Language) Phonological Acquisition

The acquisition of phonological systems in a third language, influenced by previously acquired languages (L1, L2).

Cross-Linguistic Influence (CLI)

Transfer effects from previously known languages (L1/L2) onto the acquisition of another language (L3).

Property-by-Property Transfer

only certain features transfer, not whole systems

L2 Status Factor

The idea that L2 often influences L3 more strongly than L1, especially in early stages of L3 acquisition.

Typological / Structural Similarity

CLI is more likely between languages that are structurally or typologically similar.

Models

• Scalpel Model → CLI is selective, allowing both positive and negative transfer on a property-by-property basis.

• Linguistic Proximity → Transfer depends on perceived similarity between languages rather than acquisition order.

• Speech Learning Model (SLM) → L2/L3 phonetic categories are shaped by L1 perception; similar sounds are hardest to acquire as “new categories”.

Perceptual Delinking

The process of breaking an L2/L3 sound’s perceptual link with an L1 category.

Voice Onset Time (VOT)

The timing between stop release and onset of voicing; key for distinguishing stop consonants.

RVA (Regressive Voicing Assimilation)

consonant matches voicing of following consonant

• Example: English /s/ → [z] before voiced /b/ (baseball)

PSV (Presonorant Voicing)

Voicing of an obstruent before a sonorant consonant.

Sandhi

phonological processes across word boundaries (e.g., between two words)

Separate phonetic categories

The ability to maintain distinct phonetic systems for each language.

Language-Specific Patterns

• Hungarian: Voicing language, RVA mandatory, PSV absent, no word-final devoicing

• Spanish: /s/ can voice before voiced consonants; no phonemic /z/, PSV dialectally variable

• English: Stop contrast via aspiration; RVA & PSV absent

• Implications: Hungarian learners must suppress RVA/PSV when speaking English

Cognates

Similar words across languages

• Facilitative: L1/L2 pattern matches L3 → helps pronunciation

• Inhibitory: L1/L2 conflicts with L3 → hinders pronunciation

• Triple cognates: exist in L1, L2, L3

Experiments

• Production → actual voicing in speech

• Perception → ability to perceive voicing contrasts

• Contexts: presonorant, pre-obstruent, word-internal, sandhi

Hypotheses

1. Inhibitory cognates → show less target-like voicing than non-cognates.

2. Facilitative cognates → show more target-like voicing than non-cognates.

3. L1 dominates if L1/L2 conflict

4. Sonorants do not trigger PSV across word boundaries in English or Spanish.

5. Obstruents trigger RVA across word boundaries in both English and Spanish.

Production Experiment Methods

• English & Spanish sessions separated; half started with English, half with Spanish.

• Recorded in soundproof booth; sentences shown on screen, read 4× each.

• Total: 83 sentences × 4 repetitions × 14 speakers = 4648 recordings.

• Acoustic analysis (Praat): measured fricative length & % voicing

  • Example: 33 ms /s/, 29 ms voiced → 88% voicing → perceived as [z]

• Perception threshold: ≥30% voicing → likely perceived as [z]; <30% → [s]

• Contexts tested: presonorant (before sonorants), pre-obstruent (before stops), word-internal, across word boundaries (sandhi)

Production & Cognate Findings

• English /s/ before sonorants: mostly <30% voicing → voiceless target achieved; slight increase in SP−HU− (e.g., Yasmin 35%) → small inhibitory cognate effect

• English /z/ before sonorants: high voicing (>30%), except SP+HU− (Bosnia 13.9%) → L1 Hungarian dominates

• Spanish /s/ before sonorants (PSV): mostly 12–18% voicing → participants did not acquire PSV

• English /s/ before voiced stops (RVA): 50–60% voicing → failed to suppress Hungarian RVA

• Spanish /s/ before voiced stops: 44–60% voicing → L1 RVA dominates; no cognate effect

• Speaker variation: both inter- & intra-speaker variation observed

Word-boundary & Perception Results

• Word-final /s/:

  • Before sonorants → low voicing (<30%) → PSV mostly absent

  • Before voiced stops → high voicing (English ~50%, Spanish ~81%) → RVA persists

• Perception:

  • English: participants noticed non-native RVA → rated worse than native-like

  • Spanish: participants did not reliably detect absence of PSV

• Implications:

  • Dynamic processes (RVA, PSV) harder to acquire than phonemic contrasts

  • L1 Hungarian dominates production; L2 Spanish has little effect

  • Accurate perception does not always transfer to production (SLM prediction)