Using hierarchical clustering to identify patterns of clause combining in written Hebrew narratives

Combinations of clauses are typically classified as involving coordination, complementation, relativization, or adverbial subordination. However, these categories are often criticized as overly rigid. Many advocate viewing clause combining as a continuous space defined by multiple intersecting parameters, aligning with usage-based views of grammar as fluid and emergent. While alternative classifications have been proposed, these are rarely grounded in empirical corpus analyses. Conversely, corpus studies employing multivariate methods have focused on specific constructions, whereas analyses of the domain’s overall structure remain scarce. The current study used hierarchical clustering to identify patterns of clause combining in a corpus of 100 written Hebrew narratives, aiming to develop a usage-based classification. Focusing on non-embedding cross-clausal relations, every pair of successive ranking clauses (RCs) was analyzed, defined as one main clause plus any clauses embedded within it. Each RC pair was characterized using twelve parameters, including eleven formal parameters (e.g., syndesis, punctuation) and one semantic parameter (the relation between RCs, such as Succession or Result). Hierarchical clustering identified seven diffuse RC pair clusters, representing prototypical clause combining strategies: high connectivity, low connectivity, initial connective, infinitival purpose, embedded conjuncts, entity elaboration, and external linkage. This taxonomy is compared to traditional classifications, and its implications are discussed.