the sumner suite

‘The social weight of spoken words’ is an interesting overview of the phenomenon of social representation and social weight in the context of speech. It summarizes a body of research on the presence of social information in speech, how these representations lead to social weighting, and how such weighting affects our speech development and comprehension. The most intriguing fact about this article to me was that people remember socially idealized pronunciations better than forms with stigmatized dialects, even if they are speakers of that stigmatized dialect! 

I’m curious about the tradeoff between exposure to an accent and the prestige of such an accent; i.e. if a New Yorker had only heard a ‘prestigious’ accent once in their lives, would they still have an easier time remembering some word spoken with that accent in comparison with their own? In other areas of study (e.g. history, biology), I find it very hard to remember particular facts without sufficient exposure to the general context of such facts. Likewise, one might expect that without sufficient exposure to some dialect, it might be harder to recall words from that accent despite its prestige. 

The Kataoka and Sumner experiment addressed the question of social weighting by testing recall rates in listeners across speakers of different access. Consistent with the explanation given in ‘The Social Weight of Spoken Words’ article, words spoken by people with accents of more ‘prestige’ (i.e. Southern Standard British English) were recalled better than ‘stigmatized’ accents (i.e. NYC) regardless of the frequency of exposure.

The King and Sumner article describes two experiments about how associations between speakers and their speech influences understanding of words. The first experiment they conduct is a free association task in which listeners hear speakers from various demographics say various words and they’d say the first word they came to mind upon hearing the utterance. Responses to the same word spoken by different speakers were compared across participants, and there were robust correlations between speakers and the perceived meaning of the word, thus showing that who says some word influences how that word is understood. The second part of the experiment tests whether listeners' reaction times in a lexical decision task were influence by whether a speaker was saying the word that was found to be their ‘top-associate’ in the previous experiment, and they found that it did! Listeners were able to understand the word more quickly if it was the top associate. 

I found both experiments to be convincing, insightful, and intuitive in their results. The second result easily lends itself to an interpretation of Bayesian reasoning; one's prior probability is updated by other contextual clues (i.e. register of speaker voice, dialect, accent), and this influences what is likely to be the meaning of a word. The fact that this happens on a subconscious, reaction time level is suggestive (like much other evidence) of the fact that many sensory systems leverage bayesian inference.