Cheng, Yuan: GAN (Generative Adversarial Networks) Method Applied for Quality Analysis of 3D Printed Objects
Title: GAN (Generative Adversarial Networks) Method Applied for Quality Analysis of 3D Printed Objects
Name: Yuan Cheng
Major: Electricl Engineering
School affiliation: School of Engineering
Programs: Aresty – RA Program
Other contributors: Weihong Guo
Abstract: As children learn about the world around them, they turn to familiar people within their own community in order to obtain the most accurate information. When learning about object labels in particular, children recruit both linguistic and extralinguistic information. Previous research has shown that some of this extralinguistic information includes the trustworthiness and accuracy of the speaker, and the speaker’s accent as an indication of in-group/out-group status (Corriveau et al. 2013). In this study, we ask whether these external markers of speaker status can also be recruited for learning about properties of objects which go beyond objective truth-conditional meaning. Here, we extend the findings of Corriveau, Kinzler, and Harris (2013), beyond simple object labels to objective and subjective adjectives, in particular, predicates of personal taste (PPTs) such as fun or yummy. While assertions such as “That is a strawberry” and “That flower is yellow” are objective statements with distinct truth values, being either true or not, assertions such as “That is fun” and “That is not fun” can both be true, if ‘anchored’ to two different speakers, resulting in a so-called faultless disagreement. In an experimental study with monolingual English preschoolers (ages three to five) and adults, we presented participants with two speakers: one with a familiar, American English accent and the other with an unfamiliar, Spanish accent. At the same time, we strategically manipulated speaker accuracy through the labeling of familiar objects and their properties. We find that while adults and children may initially prefer to hear from the familiar speaker, they trust the accurate speaker to learn the label of a novel object, regardless of accent. However, in instances of faultless disagreement, adults are moderately influenced by accent, while children are influenced by the positive valence of the adjective, appearing to prefer the speaker who said something nice about the object. Thus, we find that accent, accuracy, and social norms all play a role as children rely on speakers to learn about the world around them.