The construct of second language (L2) utterance fluency is typically operationalized through various individual temporal features. However, in natural speech, fluency (or disfluency) is often characterized by the clustering of multiple temporal features, collectively revealing the speaker’s effort in speech production or disfluency recovery. In this study, we explore the co-occurrence patterns of disfluency features in L2 speech and their associations with speakers’ L2 oral proficiency. We initially segmented all speech samples into analysis of speech (AS)-units. Within each AS-unit, six individual fluency features were manually coded, standardized, and subsequently subjected to a hierarchical-based k-means cluster analysis to examine their co-occurrence patterns. The results revealed four distinct disfluency clusters. A subsequent qualitative analysis of disfluencies in each cluster revealed distinct distributional patterns, disfluency makeup, and communicative functions. Additionally, the proportions of different disfluency clusters were significantly influenced by speakers’ proficiency level, first language background, and their interaction. These findings carry implications for L2 speaking research in general, shedding light on the intricate nature of speech fluency and presenting an alternative approach to the operationalization of this multidimensional construct.

中文翻译：

---

不流畅性不是孤立发生的

第二语言 （L2） 话语流利度的结构通常通过各种单独的时间特征来作。然而，在自然语音中，流利度（或不流畅度）的特征通常是多个时间特征的聚集，共同揭示了说话者在语音产生或不流畅恢复方面的努力。在这项研究中，我们探讨了 L2 语音中不流畅特征的共现模式及其与说话者 L2 口语熟练度的关联。我们最初将所有语音样本细分为语音 （AS） 单元分析。在每个 AS 单元中，对六个单独的流畅性特征进行手动编码、标准化，然后进行基于分层的 k-means 聚类分析，以检查它们的共现模式。结果揭示了四个不同的不流畅集群。随后对每个集群中 Disfluness 的定性分析揭示了不同的分布模式、不流畅性构成和交际功能。此外，不同不流畅聚类的比例受说话人的熟练程度、第一语言背景及其互动的显著影响。这些发现对总体上 L2 口语研究具有意义，阐明了语音流利度的复杂性，并提出了一种作这种多维结构的替代方法。