Developing a bilingual speech neuroprosthesis - PROJECT SUMMARY/ABSTRACT Anarthria, a loss of the ability to articulate speech, can be a severe symptom of neurological conditions, including paralysis, stroke, and motor neuron disease. Invasive speech neuroprostheses that decode cortical activity into intended speech have potential to restore naturalistic and rapid communication to patients with anarthria and paralysis by bypassing diseased motor pathways. Speech decoding has expanded rapidly in recent years. Multiple studies have demonstrated that English phrases may be decoded from cortical activity in participants with anarthria. While these are critical milestones, an ideal speech neuroprosthesis would fully restore communication to any person living with paralysis, regardless of their language background. Currently, over half the world is bilingual with proficiency in at least two languages. Extensive qualitative and quantitative research has demonstrated that bilingual speakers use each language in distinct social contexts and report that their two languages contribute unique dimensions to their overall personality and worldview. Despite this, prior speech decoding studies have exclusively focused on monolingual decoding. A key challenge in developing a bilingual speech neuroprosthesis is decoding the intended language of the user in addition to specific phrase content. Prior neuroscientific studies of bilingual speech have not provided a clear consensus on the representation and localization of cortical activity patterns that are language-specific or shared across languages. Thus, to develop a bilingual neuroprosthesis, we must first understand the basic cortical mechanisms that underlie bilingual speech. Based on our preliminary data and prior literature, we hypothesize that lexical access is language-specific and supported by higher-order speech regions (i.e., inferior frontal gyrus) whereas vocal-tract articulation is shared and supported by the speech sensorimotor cortex (SMC). We hypothesize that these representations will be sufficient to decode both intended language and specific phrase content in participants with anarthria. SMC articulatory activity, specifically, may facilitate rapid generalizability of decoders across languages. To test these hypotheses, we propose studying bilingual speakers and bilingual participants with anarthria, using high spatio-temporal resolution electrocorticography (ECoG). In aim 1, we will record cortical activity while bilingual speakers produce a large set of unique English and Spanish phrases, allowing us to assess whether shared or language-specific cortical activity underlies bilingual lexical access and articulation. In aim 2, we will develop algorithms, in bilingual participants with anarthria, to decode cortical activity into intended language and phrases. Finally, in aim 3, we will explore which cortical representations and regions allow content decoders to rapidly generalize across languages, reducing training times for system users.
