Examining influencing factors in measuring mean length of utterance: A case study of Chinese-speaking older adults with cognitive decline

Abstract

Linguistic changes in later life serve as observable indicators of age-related cognitive decline, with mean length of utterance (MLU) being a widely adopted metric for assessing syntactic complexity in aging populations. However, the reliability of MLU depends on utterance segmentation consistency, a methodological challenge that remains underexplored in Chinese, where standardized guidelines are lacking and phenomena such as pro-drop and cohesive repetition systematically obscure boundary judgments. This study presents a methodological investigation into how segmentation variability influences MLU values. Using narrative and descriptive speech samples from Chinese older adults with differing cognitive statuses, three MLU variants were derived from the transcripts: MLU_Raw, MLU adjusted for pro-drop (MLU_PDP), and MLU adjusted for cohesive repetition (MLU_CHS). Inter-annotator agreement in utterance boundary judgments was assessed via accuracy, while Pearson Correlation and Bland-Altman analysis quantified cross-method agreement. Results showed that without explicit segmentation guidelines, inter-annotator agreement was quite low, indicating high subjectivity. Bland-Altman analysis showed significant MLU variability between calculation methods. Task complexity and cognitive status critically moderated MLU validity. MLU_PDP amplified differences between cognitively impaired and healthy older adults during narrative tasks. In contrast, MLU_Raw and MLU_CHS demonstrated high agreement, supporting their stability for basic language assessment. The findings demonstrate that MLU is not a fixed property of speech but a construct highly sensitive to preprocessing decisions, task demands, and annotator interpretation. This study underscores the necessity for evidence-based, linguistically motivated segmentation protocols for Chinese and provides a methodological framework for evaluating measurement consistency. It argues that systematic methodological investigation including explicit reporting of segmentation criteria and multi-method agreement analyses is a prerequisite for advancing MLU as a valid measure in lifespan linguistics and clinical applications.