L2 difficulties in the perception of Mandarin tones: Phonological universals or domain-general aptitude? (Zhou & Veríssimo, 2025)
Citation
Zhou C, Veríssimo J. L2 difficulties in the perception of Mandarin tones: Phonological universals or domain-general aptitude? Bilingualism: Language and Cognition. Published online 2025:1-15. doi:10.1017/S1366728925100114
My thoughts
- OCP operates as a language-specific phonotactic constraint and is acquired on the basis of input distribution (Boll-Avetisyan, 2012; Boll-Avetisyan & Kager, 2014).
- OCP: Identical vs non-identical
- TMS: accuracies in rising, falling, and level tone pairs.
- Their use of Lextale is a recognition of form, and does not measure tone perception.
- The statistical models were fit in a Bayesian Framework, and combined prior information with evidence from the data to produce a posterior distribution for each parameter, which is the probability distribution over a parameter’s possible values. Importantly, and in contrast to frequentist analyses, Bayes factors can provide support for the null hypothesis, that is, for the equality between conditions or groups. They calculated the natural logarithm of the Bayes factor in favour of the alternative hypothesis using the Savage-Dickey method. Values of InBF10 greater than 1 support the hypothesis that an effect is different from zero (H1), while negative values smaller than -1 support the hypothesis that the effect is absent (H0), values of InBF10 between -1 and 1 are essentially inconclusive.
- can accuracy proportion = production proportion?
- Start with stats models with OCP and TMS, and then combine them with Bayesian mixed-effects models to analyze the factors, not sure if I want to include vocabulary measures, maybe use age as a grouping factor, and run separate models with children and adults.
- can do analysis of rates in each syllabic position (across identical and non-identical pairs)?
Tones in Mandarin
- High-level (T1)
- Rising (T2)
- Low-level or low-dipping (T3)
- Falling (T4)
- Around 70% of the words in Mandarin are disyllabic
Backgorund
-
Tones are essential for distinguishing lexical meanings in Mandarin, which makes them one of the primary linguistic components for learners from the very beginning. But given enough training, the learners are able to identify and produce the tone categories (pitch differences, which is also the primary acoustic cues for lexical tones in L1 English children, L2 Chinese speakers, etc).
- Mandarin lexical tones are very challenging for novice learners, especially for those from a non-tonal background.
-
Disyllabic tone sequences are more challenging for L2 learners, due to higher short-term memory load, tone sandhi, coarticulatory influence, and more extensive lexical access.
- Learning disyllabic tone sequences is further influenced by some universal processes common to language learning and humans, which include two phonological universals, the Obligatory Contour Principle and the Tonal Markedness Scale, that may constrain the L2 acquisition of Mandarin lexical tones, regardless of learners’ first language.
- Obligatory Contour Principle
- Restriction on co-occurrence of identical or homorganic sound structures, when there are two identical tones that occur on adjacent vowels, the rightmost one is deleted.
- Conceptualise it as a violable constraint under the framework of OT or relativise it to certain phonological structures in a language-specific manner, and evidence supporting this comes from large-scale quantitative typological reports, which show nearly all languages are restricted by some kind of similarity avoidance.
- Easier to learn in artificial language learning experiments, compared to a place harmony process and to an arbitrary pattern.
- Example: Mende tonal system, no adjacent identical tones are allowed (e.g., *HHL or *LLH)
- Tonal Markedness Scale
- Level tones mainly involve F0 height, and contour tones (e.g., rising and falling) additionally entail an F0 slope.
- Production: Rising tones take longer to produce than falling tones, and falling tones are phonetically less demanding.
- *Rising » *Falling » *Level
- which used to explain the tone sandhi pattern in Hakha Lai.
- shown in acquisition and typological studies
- acquisition: stabilisation of lexical tones generally followed the TMS (e.g., T2 < T4 < T1)
- typological: if a language has contour tones, it also employs level tones, and if it has rising tones, it also employs falling tones.
Two phonological universals and L2 acquisition
- OCP only affected rising and falling tone pairs, not level pairs, and the production accuracy of both identical tone sequences and individual tones followed the TMS.
- YouTube videos, Song observed the production accuracy of individual tones followed TMS, but not the tone sequences, for the OCP effect, they found lower accuracy for identical than non-identical sequences for T4-T4. Conclusion:
- OCP: There is some evidence for OCP effects in L2 tonal production, but they might be tone-specific.
- TMS: Consistently found on the production of individual tones, but shown conflicting evidence for its operation on tonal pairs.
Research Questions
- Is L2 tonal perception subject to OCP effects? We expected to find OCP effects in L2 tonal perception, at least for some tone pairs (e.g., T4-T4, Song, 2021). OCP effects would be borne out if the accuracy rate in the identical condition was lower than in the non-identical condition, because the OCP should force identical tone pairs into non-identical ones, but not vice versa.
- Is L2 tonal perception subject to TMS effects? We hypothesised that learners’ identification accuracy of tone pairs (H. Zhang, 2016) and that of individual tones (Song, 2021; H. Zhang, 2016) would both conform to the TMS (i.e., T2 < T4 < T1).
- Are the effects of the OCP and the TMS modulated by L2 speech proficiency? We predicted that the effects of OCP and TMS would be more pronounced in learners with lower Mandarin proficiency. Given that both the OCP (except for T3-T3) and the TMS are not active in Mandarin phonology, an increase in L2 speech proficiency should lead learners to gradually overcome the influences exerted by these two universals.
Hypothesis
- To assess evidence for null hypotheses for two phonology universals
- Whether mastering isolated tones will lead to good performance in identifying tone sequences in disyllabic words, such as individual tones, can be regarded as the building blocks of tonal sequences.
- Learners with more advanced Mandarin proficiency are better at overcoming the interference of these two universals.
- Learners with good pitch acuity show advantages in the acquistion of Mandarin lexical tones.
Method
- Participants: 59 speakers of L2 Chinese
- Material: 96 Mandarin syllables from the lexical database to creat 48 disyllabic pseudowords items carry two identical lexical tones.
- Predictors of L2 speech proficiency
- Vocabulary size: a larger vocabulary size can better distinguish between confusable L2 sound categories due to learning phonological neighbours, many contribute to the consolidation and refinement of existing phonological representations in the lexicon. Phonotactic generalisations are made across lexical entries.
- Lextale - CH vocabulary test
- Pitch discrimination task from auditory processing test batteries
- Auditory processing ability: individual differences in auditory acuity, which is the ability to detect subtle differences in various aspects of acoustic input at a fine-grained level (auditory precision hypothesis). Learners with better auditory acuity can increase the precision of their auditory representations and contribute to the increase of L2 speech proficiency. Domain-specific auditory acuity is strongly linked to the acquisition of phonological contrasts that are robustly distinguished by that particular dimension.
- Perceptual testing
- Perceptual identification task with pseudo-words, with potential influence of imprecise lexical encoding and articulation.
- Identify two lexical tones by typing 1, 2, 3, 4
- T3 never occurs in this task
- Bayesian mixed-effects models
Analysis
- Mixed-effects binomial (logistic) regression with maximal random-effects structure
- The statistical models were fit in a Bayesian Framework, and combined prior information with evidence from the data to produce a posterior distribution for each parameter, which is the probability distribution over a parameter’s possible values.
- OCP: compare the identification accuracies of identical vs non-identical tone pairs (T1T1 vs. T2T1)
- TMS: compare within identical tone pairs, the identification accuracies of rising, falling, and level tones
- combination of speech proficiency measures on participants’ accuracies and assessed whether the OCP and TMS effects were modulated by L2 proficiency
Results
- Provide null effects of both phonological universals, but also provide a clear determinant of tonal identification accuracy in the participants’ pitch acuity, and suggest that domain-general auditory processing (pitch acuity) underlies the learning of the L2 phonological categories.
- Learners’ lexical knowledge of Chinese characters did not predict L2 tonal development.
- OCP: Bayes factor analysis showed evidence against a difference between conditions and supports the null hypothesis that tonal perception is equally accurate in identical and non-identical pairs.
- TMS: Rising pairs are indeed harder to identify correctly, as predicted, but falling tones cannot be distinguished from level tones in their identification accuracy.
- Vocabulary size did not have an effect on tone identification.
- Participants with better pitch acuity were more accurate at identifying tones, across identical and non-identical pairs.