Unequal burdens of research in underrepresented contexts: Developing language metrics and morphemising protocols for use in complex multilingual environments

In child language research and clinical speech language pathology settings, transcripts of speech are often used for investigations into language acquisition and development. The existing literature is heavily biased towards monolingual contexts, particularly from a small number of languages, meaning that there are few examples of theory-driven protocols for handling speech and deriving developmental metrics from contexts where multiple typologically diverse languages appear together. In the current report, we address limitations of existing approaches and describe a theory-driven protocol for transcription and morphemizing of multilingual translanguaged speech, as applied to large corpus of parent-child speech from multilingual Singapore. The protocol aims to be language-model neutral, treating all speech as valid input to the child, without excluding or separating by official language labels, and may be applicable to a variety of linguistic contexts. One key finding arising from this research was that applying the protocol created unique research challenges for the typological distinct languages used by parents in the local context (predominantly English, Mandarin Chinese, Malay and Tamil alongside unique lexical items from the local context, known as Red-Dot words), and that structural inequalities make those challenges more difficult to resolve for minoritized and under-resourced languages.