Help Privacy Policy Disclaimer
  Advanced SearchBrowse





All by myself or Obama's elf? The influence of social network size on speech perception


Lev-Ari,  Shiri
Psychology of Language Department, MPI for Psycholinguistics, Max Planck Society;

External Resource
No external resources are shared
Fulltext (public)
There are no public fulltexts stored in PuRe
Supplementary Material (public)
There is no public supplementary material available

Lev-Ari, S. (2016). All by myself or Obama's elf? The influence of social network size on speech perception. Poster presented at the 29th CUNY Conference on Human Sentence Processing, Gainesville, FL, USA.

Cite as: http://hdl.handle.net/11858/00-001M-0000-002B-9C02-1
Infants and adults learn new phonological varieties better when exposed to multiple speakers rather than a single speaker during learning (Bradlow & Bent, 2008; Rost & McMurray, 2009, 2010;Lively, Logan & Pisoni, 1993), presumably because such input is more varied. The studies here test (1) whether having a larger social network similarly facilitates phonological performance, and (2) what the underlying mechanism of this effect is. In Study 1, 60 native Dutch speakers reported all their interactions for one typical week. They were then tested on transcription of nonwords in noise, on talker normalization, and on a host of cognitive measures (WM, auditory STM, selective attention, task switching). Results showed that, as predicted, participants with larger social networks were significantly better at speech perception in noise, but they were not better at talker normalization. Crucially, these findings were obtained despite controlling for cognitive skills and for amount of talk, indicating that the effect of social network size on speech perception is not due to differences in cognitive skills among people with different network sizes or to amount of linguistic input. Study 2 used computational simulations with agent-based models to explore the mechanism underlying the effect of social network size on speech perception. Networks were created by randomly selecting people from a population speaking 12 Dutch vowels with a mean and SD as described in Pols, Tromp & Plomp (1973). During interactions, the agent met with a random member of her network and received one labeled set of formants for each vowel, and stored it. At test, the agent received unlabeled sets of formants from members of the population that are not in her network, and needed to recognize them. Results showed that having a larger social network led to greater accuracy in phoneme categorization. Interestingly, even though larger networks were also associated with greater variability, as reflected in larger category SDs, variability did not predict performance. Instead, as Figure 1 illustrates, the benefit of having a larger network was fully explained by a novel measure of Smooth Sampling, which calculated coverage of central areas and penalized for vowel overlap. Further simulations that manipulated network properties orthogonally showed that the positive effect of social network size on speech perception is independent of amount of input received but is modulated by the ratio of int ra- to inter-individual variability, such that having larger social networks is most helpful when speakers are consistent within themselves, and the population is varied. These results held whether phoneme categorization was carried out by matching to the closest similar stored token or by calculating the token’s probability of belonging to each category according to the category’s distribution. Together, these studies show how having a larger social network leads to better speech perception by influencing the distribution of the input, and thus show how aspects of our life- style can influence our linguistic performance