Language-specific distributional learning advantages in Singapore English-Mandarin bilingual adults

• Main Author: Goh, Hannah Letitia
• Other Authors: Scott Reid Moisik
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Social Sciences; Cognitive psychology; Psycholinguistics
• Online Access: https://hdl.handle.net/10356/175760
• Institution: Nanyang Technological University

In this thesis, we set out to determine if we could find evidence for language learning advantages in bilingual adults that stem from language-specific transfer effects. The language-specific hypothesis of bilingual learning advantages suggests that language learning advantages are tied to an individual’s linguistic experience. Thus, learning advantages are likely to stem from transfer effects between known languages, and to-be-learnt language stimuli. We focused our examination of this question on English-Mandarin bilingual adults in Singapore, as they consist of a largely homogenous group of bilinguals with intragroup differences in the language balances of their language repertoire (particularly for Mandarin Chinese). We chose to utilise a distributional learning paradigm in this investigation as such paradigms have been shown to be a robust method of assessing language-specific learning effects in adults (e.g., Chládková & Šimáčková, 2021; Chládková et al., 2022). Based on the language-specific hypothesis of bilingual learning advantages, we reasoned that individual variance in Mandarin proficiency would act as a predictive factor in determining individual differences in learning effects following bimodal distributional training on stimuli that share featural similarities with Mandarin Chinese. In essence, individuals with higher understanding abilities in Mandarin Chinese would demonstrate stronger learning effects for novel language stimuli that share phonemic features with Mandarin. In this project, we conducted a total of six studies to thoroughly address this question. In Chapter 2 and Chapter 7, we directly examine this main research question. To fine-tune our investigation, we conducted four studies (Chapter 3, Chapter 4, Chapter 5, Chapter 6) documenting unique aspects of perception and production for the alveolar-retroflex contrast in the under-documented outer-circle variety of Singapore Mandarin. Overall, this project contributes novel insight into the validity of the language-specific learning hypothesis in bilinguals, and our documentations of Singapore Mandarin provide fresh details on the nature of the alveolar-retroflex contrast in Singapore Mandarin with the use of fine-grained acoustic analyses and behavioural measures. Chapter 1 lays the basis for our main research question on the language-specific hypothesis of bilingual learning advantages. In Section 1.1, we conduct a thorough review of existing literature on three main overarching hypotheses on the nature of language learning advantages in bilinguals and multilinguals. We justify our interest in pursuing an investigation on the language-specific hypothesis. In Section 1.2, we discuss the unique linguistic landscape of Singapore, and detail how resident English-Mandarin bilinguals present us with a uniquely suitable group of individuals to examine this question with. Finally, in Section 1.3 we examine literature on distributional learning. We discuss studies on language acquisition via distributional learning in naturalistic settings, as well as in experimental settings and detail our reasoning for choosing a distributional learning paradigm in our investigation of the language-specific learning hypothesis of bilingual learning advantages. Chapter 2 contains our first experimental exploration on the language-specific hypothesis of bilingual learning advantages via transfer effects. In this investigation, we trained English-Mandarin bilingual adults on a Hindi dental-retroflex phoneme contrast with a bimodal distributional learning paradigm. We reasoned that participants with higher Mandarin understanding proficiencies should show larger learning effects due to transfer effects as Hindi and Mandarin ostensibly both feature a dental/alveolar-retroflex place of articulation contrast (Hindi: stop consonants, Mandarin sibilant consonants). We examined overall learning effects in a pilot study (N = 15) and a main study (N = 50) and assess language-specific transfer learning effects in the main study. While we found evidence of overall learning in the pilot study, this was not replicable in our main study with a larger sample size, and we did not observe any language-specific transfer effects. A main confounding factor was identified from studies that were published after this study was completed. While formal Mandarin educational texts suggest that “retroflex” phonemes in Mandarin share a “curled” tongue place of articulation similar to that of Hindi retroflexion (Ou & Guo, 2014), recent lingual imaging studies show that Taiwan and standard Beijing Mandarin speakers are unlikely to produce “curled” retroflexion in speech at all (Luo, 2020; Tiede et al., 2019). We reasoned that this could have affected our participants’ ability to learn the Hindi contrast. This would also mean that the Hindi dental-retroflex stimulus set would be unsuitable for assessing a language-specific hypothesis of transfer effects in Singapore English-Mandarin bilingual adults. To confirm if the “curled” retroflex is indeed absent in Singapore Mandarin, we then conducted an exploratory observational lingual ultrasound imaging study in Chapter 3 (N =5). The results of this study revealed that Singapore Mandarin speakers do not use a “curled” retroflex tongue position – their production of Mandarin retroflex phonemes, instead using a “humped” laminal post-alveolar tongue posture for Mandarin retroflexion. This study showed that we would need a more suitable set of stimuli to assess the language-specific hypothesis of bilingual learning advantages in Singapore Mandarin speakers. Studies (Chung, 2006; Chen et al., 2016) suggest that the phonological nature of the Singapore Mandarin alveolar-retroflex contrast likely differs from Mandarin varieties documented in existing Mandarin corpora (standard Mandarin: Sze et al., 2014; Hong Kong Mandarin: Tse et al., 2017) due to phoneme contrast merger (deretroflexion). In order to obtain a suitable stimulus set aligned to the unique norms of Singapore Mandarin, we thus conducted a detailed documentation of perception and production of the alveolar-retroflex contrast in contemporary Singapore Mandarin. In Chapter 4, we discuss the details of this documentation conducted with novel speech elicitation and word identification tasks (N = 50). We examined the acoustic nature of the contemporary Singapore Mandarin alveolar-retroflex contrast with centre of gravity (CoG) analyses and general additive mixed modelling (GAMM) on long-term averaged spectra (LTAS) of for the first time in known literature. We also examined participants’ word identification accuracy for alveolar-retroflex minimal pairs of words in Singapore and standard Beijing Mandarin. The results of the study revealed that contemporary Singapore Mandarin speakers produce alveolar and retroflex phonemes contrastively, albeit to a lesser extent than is typically reported for standard Beijing Mandarin (e.g., Chang, 2012). We also found that Singapore Mandarin speakers typically demonstrate very high levels of Mandarin word identification accuracy in both the local as well as the standard variety of Beijing Mandarin. Using the acoustic information obtained in Chapter 4, we synthesised a novel 8-step alveolar-retroflex [tsʰuə́n]-[ʈʂʰuə́n] contrast continuum based on the acoustic norms of contemporary Singapore Mandarin. We then conducted a study in Chapter 5 with a novel paradigm (the Spring-Village CROWN Game) to obtain fine-grained measures of perceptual sensitivity in the form of speech sound ambiguity resolution using our novel alveolar-retroflex [tsʰuə́n]-[ʈʂʰuə́n] contrast continuum (N = 62). We also conducted a correlation analysis to determine if higher levels of perceptual contrast sensitivity are linked to higher Mandarin understanding abilities. The results of this study showed that the majority of contemporary Singapore Mandarin speakers perceive the alveolar and retroflex categories of speech sounds contrastively, albeit with a large range of individual differences in sensitivity for speech token ambiguity. We did not find evidence for a link between Mandarin understanding abilities and these individual differences in perception, suggesting that high levels of perceptual sensitivity for this phoneme contrast are not required for high levels of general speech comprehension in contemporary Singapore Mandarin. We then conducted an exploratory study in Chapter 6 (N = 20) to examine links between perception and production of an alveolar-retroflex [tsʰuə́n]-[ʈʂʰuə́n] contrast within our target population of Singapore English-Mandarin bilinguals. We used the speech elicitation task from Chapter 4 to collect measures of individual differences in production of the alveolar-retroflex [tsʰuə́n]-[ʈʂʰuə́n] contrast, and the Spring-Village CROWN Game from Chapter 5 to measure individual differences in perception. The results of this study revealed a significant speech perception-production link tied to the retroflex affricate [ʈʂʰ] on the level of the individual. This link showed that individuals who produced their retroflex frication in a manner more similar to that of alveolar frication tended to have less sensitivity for retroflex cues in ambiguous speech sounds. Finally, we returned to our main investigation of the language-specific hypothesis of bilingual learning advantages with our study in Chapter 7. We trained English-Mandarin bilingual adults on our novel synthesised alveolar-retroflex [tsʰuə́n]-[ʈʂʰuə́n] contrast with a bimodal distributional learning paradigm. We reasoned that participants with higher Mandarin understanding proficiencies should show larger learning effects due to transfer effects between their real-world Mandarin experience and the acoustically complex training stimuli. We examined overall learning effects in a pilot study (N = 20) and a main study (N = 50) and assessed language-specific transfer learning effects in the main study. We found evidence of overall learning effects in both the pilot and the main study, confirming that adults can learn from unattended distributional training if there is sufficient overlap between their languages and the to-be-learnt stimuli. We also found evidence of a transfer learning effect tied to Mandarin understanding proficiency with larger learning effects being significantly associated with higher Mandarin understanding proficiencies. This is the first time evidence has been found to show that language-specific learning transfer effects can be strong enough to be observed on the level of the individual within the linguistic variations of a single group of largely homogenous bilingual adults.