Colloquium

Ming Xiang – Reducing uncertainty in contact-induced linguistic change

Colloquium talk by Prof. Ming Xiang (University of Chicago)

Language change could arise when the younger generation of speakers do not fully replicate the linguistic patterns produced by the older generation. The statistical learning literature has shown that children (and adults too under some conditions) tend to regularize the statistical patterns in their input, i.e. they reduce the amount of variation in the input data (Hudson Kam&Newport 2009). From the information-theoretic perspective, this process reflects the reduction of uncertainty (entropy) in the data. The regularization process over multiple generations has also been formalized using Bayesian models (Reali and Griffiths 2009). These studies, however, were based on artificial language learning paradigms, and the question remains whether and how similar regularization processes could arise in natural languages. In this talk I will present some recent work that investigates word order changes in Cantonese in the last 60 years. Two groups of native Cantonese speakers (18-70 years old, highly proficient speakers) were examined, one from Guangzhou and the other one from Hong Kong. Both Cantonese-speaking communities are in close contact with Mandarin Chinese, but the two communities also have substantial differences in their linguistic experiences and linguistic attitudes. Production data collected on ditransitive constructions revealed regularization patterns in Cantonese across generations in both communities, and crucially, this result is not an assimilation to the Mandarin patterns. A control group of Mandarin speakers in Beijing showed no changes in Mandarin on the same task. These results suggest that the effect of reducing uncertainty/variation is specific to bilingual/multilingual environments. A separate set of experiments suggest that it is likely the bilingual production process, instead of the tracking/encoding of the bilingual input, that led to the observed behavior. Finally, I will also show some preliminary data that found the opposite pattern: in some situations Cantonese speakers also introduced more variants into their language, instead of reducing variants. I tentatively suggest that increasing variation could also be interpreted as uncertainty reduction if we distinguish speaker (production) oriented uncertainty vs. listener (comprehension) oriented uncertainty.