The ability to perceive speech requires first encoding the acoustics of the speech inputs and then transforming them into an internal linguistic space where they map onto familiar words and concepts. A critical stage of processing mediating this transformation is sublexical processing, in which acoustic patterns are mapped onto language-specific speech segments. Understanding how and when sublexical processing occurs is important – but the tools to study this transformation are underdeveloped. Because sublexical processing is an intermediate stage of processing, subjects cannot respond based only on activation at this level. Further, because speech perception is a cascading process, information flows rapidly across all levels. Thus, to study the sublexical level of speech encoding, it is necessary to record these processes directly, and then to understand how patterns in neural activations reveal the cognitive processes that are their corollary. The goal of this dissertation is to understand how one such neural activation measure, the N1 electrophysiological response, relates to acoustic and sublexical processing. The N1 response measured with EEG is the first non-invasive (and therefore widely deployable) neural measure that relates to sublexical level aspects of the speech input. Toscano, McMurray, Dennhardt, and Luck (2010) demonstrated that

the N1 response patterns linearly with the voice onset time (VOT), of the eliciting speech sound. VOT is important because it is a linguistic cue used in many languages (Lisker, 1986), that separates voiced sound categories (e.g., /d/) from unvoiced sound categories (e.g., /t/). It is also the most frequently studied speech contrast. The pattern of the N1 response to VOT has been interpreted as evidence for how the speech perception system represents VOT. Based on the linear pattern of the N1 response to VOT, Toscano et al. (2010) claim support for a pre-categorical neural encoding of phonetic features. However, this claim may be slightly premature. It is not yet known how the N1 relates to acoustic or to sublexical levels of sound processing. That is, we do not yet know how the neural measure the N1 relates to cognitive theoretic levels of processing.

Despite the ambiguity about its cognitive correlate, the N1 has since been used to examine important theoretical questions about early speech sound processing. These have included including whether VOT encoding (as measured by the N1) is sensitive to lexical bias (Noe & Fischer-Baum, 2020) and is sensitive to predictive context (Getz & Toscano, 2019). However, we do not yet fully understand the N1 as a tool. Understanding what type of neural information processing drives the N1-voice-onset-time relationship relates to is foundational to its deployment to study speech processing. A better characterization of the relationship of the N1 to cognitive levels of processing would clarify how these N1-lexical bias and N1- predictive context effects should be interpreted.

The focus of this dissertation will be to improve our understanding of how the N1 relates to cognitive processes. Specifically, we will evaluate whether the N1 is indexing neural representations that are primarily responsive to acoustic level variations in the sound or are responsive to phonetic level variations. I propose several lines of experimentation to examine this question. In the first, I propose a set of neuropsychological experiments where we test the intactness of the N1-VOT response in a pure-word-deaf patient to investigate how the N1 response is disrupted or preserved in a patient with deficits in sublexical levels of encoding. If the N1 is intact in a patient in whom phonetic levels of speech processing are damaged, this is evidence that the N1 is pre-phonetic, and thus indexes acoustic processing. In the second line of work, we investigate how the N1 response changes in aging and in cochlear implant populations. By looking at how the N1 response to VOT varies across populations that systematically vary in how acoustic cues are mapped to language, we can test which acoustic cues drive the N1-VOT relationship. Older adult listeners have intact access to spectral cues but disruption of fine- grained temporal information. Cochlear implant populations have greatly decreased spectral resolution but relatively intact amplitude encoding. Young adult listeners, in whom the N1-VOT is normally studied have access to both these acoustic cues, so by comparing across these atypical populations we can gain some traction on which acoustic cues are necessary for the N1-VOT relationship.