Collaboratively Mitigating Racial Disparities in Automated Speech Recognition and Language Technologies with African American English Speakers

Abstract

Automated speech recognition (ASR) systems that rely on natural language processing (NLP) techniques are becoming increasingly prevalent within people's everyday lives. From virtual assistants integrated into mobile devices (e.g. Apple's Siri), smart home assistants (e.g. Google's Nest; Amazon's Alexa/Echo), and vehicles (e.g. Apple's CarPlay; Android Auto); to software tasks such as automatic translation, automatic captioning, automatic subtitling and even hands-free computing, ASR systems are core components of IoT (Internet of things) devices and applications. However, scholars have begun to show that with these increasing innovations and system capabilities, emerges fairness-related harms of user experiences and racial disparities that negatively impact African American speakers of African American Vernacular English (AAVE). As users of ASR, AAVE speakers’ language is less accurately recognized and processed, leading to inequitable interactions among this ethnolect community. My graduate research seeks to address this challenge by developing and validating community-collaborative methods to innovate responsible approaches toward designing more representative and equitable ASR language technologies that accommodate African American speakers of AAVE.