Generalized Adaptation to Novel Foreign Accents

Emily Simon, Lauren Oey, Crystal Lee, T. Florian Jaeger, and Xin Xie

Technology has made the world an increasingly interconnected sphere– one in which conversations can occur seamlessly while speakers sit oceans apart. However, with increasing globalization comes increasing demands on listeners to comprehend extensive variability in speech, particularly that of foreign-accented speakers. Nevertheless, evidence suggests that listeners rapidly adapt to accented speech – across varying speaker background, differing intelligibility and relatively brief exposures (Clarke & Garrett, 2004; Bradlow & Bent, 2008; Sidaras et al., 2009). After further exposure, listeners can generalize such adaptation to novel speakers with whom a listener has not previously interacted (Bradlow & Bent, 2008; Baese-Berk et al., 2013). The scope of this generalization, as well as its underlying mechanism, are still unknown. This is largely due to the inherent difficulties of measuring variability within and across speakers quantitatively. We examined generalizability of adaptation to accented speech in cases of exposure to multiple foreign accents. Using an online crowdsourcing paradigm, we will measure listener’s transcription accuracy after exposure to accented speech to assess generalized adaptation ability. During Exposure, listeners are assigned to one of three listening conditions; either 5 speakers of native English, 5 speakers of Mandarin-accented, or 5 speakers of varying language backgrounds (Korean, Thai, Hindi, Russian and Mandarin). After exposure, all listeners will be tested on a novel speaker of a familiar accent, and critically, a novel speaker of a novel accent. We hypothesize that transcription accuracy of novel foreign-accented utterances will be greatest in the case that listeners are exposed to the most systematic variability in accented speech. Under this assumption, we predict that listeners exposed to multiple foreign accents will perform best when tested on a novel accent.


Learning Adjective Meanings Through Variable Exemplars

Crystal Lee and Chigusa Kurumada

How have we learned the meaning of words like ‘full’ or ‘straight’? As adults, we know ‘full’ means ‘containing as much as possible without spilling over’. However, young learners often observe examples where ‘full’ is used to describe objects or situations that deviate from the prototypical definition, despite being contextually appropriate. e.g., a ‘full’ cup could be 90% full when transporting drinks. In fact, Syrett et al. (2010) found contrasting comprehension of absolute gradable adjectives (e.g., full, straight) between children and adults. When asked to give ‘the full cup’ with one 90% full cup and one 70% full cup present, four-year-olds were more willing to pass the 90% full cup. Contrastingly, adults were more likely to say that neither is ‘full’. We hypothesize that accumulated experiences allow adults to account for contextual contributions to word meaning: a 90% full cup is deemed full in an appropriate context (e.g., transporting drinks); otherwise any deviation from the prototypical meaning invalidates an instance to be judged ‘full’. We test this hypothesis by teaching adult subjects a novel gradable adjective; ‘pelty’ roughly meaning ‘tight-fitting’. 60 Subjects are randomly assigned either to With- or Without-context Condition. In Exposure, subjects watch 12 videos exemplifying the word use of objects that are tight-fitting to a varying degree. Those in the With-context condition receives contextual justification (e.g. A moderately tight-fitting shoe is still ‘pelty’ because it has to be worn with a thick sock) and those in the Without-context condition do not. In Test, participants see two novel objects (one 90% pelty and one 70% pelty) and a ‘neither’ option. We predict that subjects in the With-context condition, just as adult subjects in Syrett et al. (2010), would be more willing to select ‘neither’ than those in the Without-context condition.

Contextual Factors in Child Adjective Comprehension

Wesley Orth, Amanda Pogue, and Chigusa Kurumada

This research investigates preschoolers’ understanding of adjectives (e.g., big, clean, metal). While adjectives are part of very basic vocabulary acquired early in development, it is not clear whether children’s conceptual understanding of them is equivocal to that of adults. In particular, I am interested in how they acquire subtle meaning differences across adjectives. Some adjectives require a listener to reason about other objects in order to verify that they are true (e.g., to say a cat is big, one needs to know how large cats usually are), whereas others do not (e.g., to say a cat is striped, one needs to know if that cat has at least one stripe). I created a guessing game to directly compare young children and adults in their comprehension of various adjective types. 16 preschoolers and 20 adults were asked to match a description of an object a card; either a face-up or face-down card. There were three trial types: 1) adjectives that require a comparison class (e.g., big), 2) adjectives that are binary in meaning and do not require a comparison class (e.g., striped), and 3) adjectives that denote a property of the noun (e.g., metal). Participants may or may not flip the face-down card before making a match, the likelihood of which tells us whether they thought a given adjective requires a comparison class. The results show that adults seek out comparative information for the type 1 adjectives but not other adjectives (60%, 11.25%, 9.16% respectively), whereas children seek out less comparative information when necessary, yet still show a similar pattern (35%, 17.5%, 20%). I conclude that children’s understanding of these types of adjectives is qualitatively similar to adults’ while there is a quantitative difference between them. I am currently running a follow up experiment to investigate the nature of the difference.