Literature Review on Forced Alignment Challenges in Diverse

Abstract

Forced alignment (FA) is a foundational technique in speech science that enables the automatic temporal alignment of transcriptions with speech signals. Although FA systems perform reliably for high-resource languages, their accuracy degrades when applied to typologically diverse and under-resourced languages. This literature review suggests that the current constraints of forced alignment are not merely technical issues but are rooted in fundamental representational issues. By synthesizing research from multilingual processing, language documentation, and articulatory phonology, this review illustrates how traditional FA systems often mask critical cross-linguistic differences in articulation and temporal coordination. Through an analysis of empirical studies involving WebMAUS and cross-language alignment alongside theoretical work on gestural models, this review argues for the integration of gestural scores into computational frameworks. Such an integration provides a principled solution to cross-linguistic variability by capturing the physical realities of speech production that symbolic and purely acoustic models miss.