Sentence length and mean dependency distance across varieties, genres, and diachrony: a tri-dimensional corpus study

Abstract

Dependency Distance Minimization (DDM) predicts that languages tend to reduce the linear distance between syntactically related words, a cognitive constraint operationalized as Mean Dependency Distance (MDD). Yet the relationship between sentence length (SL) and MDD, while well attested, has rarely been examined across language variety, genre, and diachrony within a single unified design. This study fills this gap using the entire Brown Family of corpora (six million words, six corpora, AmE/BrE, 15 genres, 1960s–2000s), based on 331,573 sentences parsed with spaCy. Linear Mixed-Effects Modeling with random intercepts and random slopes for genre reveals three main results. First, the positive SL–MDD effect is robust across varieties, with no baseline difference between AmE and BrE. Second, genre exerts a powerful moderating effect: narrative fiction shows the steepest slopes and specialized texts the flattest, revealing a systematic trade-off between cognitive accessibility and informational density. Third, MDD has increased significantly over time, and the SL–MDD slope has steepened in the 2000s, somewhat more markedly in AmE. These findings refine DDM by demonstrating that the cognitive constraint is universal yet dynamically modulated by genre and diachronic context, pointing to a probabilistic rather than deterministic interpretation of the principle.