Eleanor Chodroff is Distinguished Science of Learning Pre-doctoral Fellow in the Departments of Cognitive Science and Electrical and Computer Engineering. Her research broadly addresses the cognitive mechanisms essential for mapping acoustic-phonetic instantiations of language to its intended phonological representation. Listeners face tremendous variability in the signal, yet demonstrate remarkable ability in identifying the intended speech sound. Acoustic variability arises from many sources including, but not limited to a given speaker’s vocal tract physiology, dialect, and idiolect. Nevertheless, restrictions on acoustic variation must also exist for mutual intelligibility, suggesting some structured variability in the speech signal. Various aspects of this dynamic, but structured mapping are explored in her research through phonetic cue weighting, phonetic accounts of perceptual assimilation, and most recently, speaker variability and systematicity.
Together, these lines of research provide complementary perspectives on the cognitive mechanisms and representations converting the speech signal to linguistic representation. Eleanor's research on phonetic cue weighting focuses on the constellation of acoustic properties or cues that constitutes a perceptual phonetic category, in particular for American English (AE) stop consonants (/ptkbdg/). She has also examined a case of perceptual assimilation, where English listeners report Hebrew word-initial /tl/ and /dl/ clusters as beginning with /k/ and /g/, respectively. The seeming distortion in this mapping appears to reflect the well-learned acoustic-phonetic structure specific to AE stop consonants, but more generally, it provides insight into the listener’s language-specific representations in speech perception. Finally, she has begun to explore talker variability and systematicity in the acoustic realization of stop consonants using large-scale corpus approaches. The knowledge of structured variability across a population of speakers may be exploited in the rapid adaptation observed in human listeners. Through the Science of Learning Fellowship, she plans on furthering this line of research, understanding perceptual learning of speaker variability with techniques from both cognitive science and automatic speech recognition.