Swap distance minimization shapes the order of subject, object and verb in languages of the world

We systematically compare word order preferences in decoder-only language models across 192 artificial languages and typologically diverse natural languages. On artificial languages, models exhibit a left-branching preference that aligns with neither natural language universals nor human word order learning biases. On natural languages, monolingual models show no clear base word order bias at small scales, but as data grows, a preference for right-branching subject-verb-object (SVO) languages emerges while SOV falls behind despite being the most frequent order cross-linguistically. This SVO advantage extends to multilingual models and correlates with language resource level and data quality rather than word order. Thus, the same architecture exhibits opposite preferences on artificial and natural languages, establishing that word order biases observed in practice are data-driven. Since highly-resourced languages are overwhelmingly SVO, these biases risk gradually reducing word order diversity, particularly in languages that productively use multiple word orders, with the widespread adoption of LLMs.

Dependency length minimization (DLM) is a well-documented processing universal, but previous studies report a single mean dependency distance (MDD) per language, obscuring variation across syntactic relation types. We analyze 122 languages in UD and SUD (version 2.17), showing that DLM operates on two distinct levels. Grammar-driven optimization targets functional dependencies (det, case, aux), which are universally short (mean 1.71, $σ$ = 0.33) and invariant across typologically diverse languages. Processing-driven optimization operates on lexical dependencies (nsubj, obj, obl), which are longer (mean 2.87), highly variable ($σ$ = 0.63), and constrained by word-order typology. This asymmetry holds in SUD despite reversed head direction (r = 0.92). We conclude that ''the grammar does the work'' of minimization by scaffolding sentences with local functional attachments, leaving processing pressures to determine the ordering of lexical heads.

The syntactic structure of a sentence can be represented as a tree where edges indicate syntactic dependencies between words. When that structure is a star, it has been demonstrated that the head should be placed in the middle of the linear arrangement according to the principle of syntactic dependency distance minimization. However, hubs of stars tend to be put at one of the ends, against that principle. Here we address two questions: (1) How difficult is it to minimize dependency distance? (2) Why anti dependency distance minimization effects have been found in star structures but not in path structures? The ease of optimization is determined by the shape of the optimization landscape. It was demonstrated that the landscape of star structures is quasiconvex (Ferrer-i-Cancho 2015, Language Dynamics and Change). As for (1), here we show that it is indeed convex (a particular case of quasiconvexity) both for star trees and quasistar trees and thus the distance-based optimization problem is simpler than previously believed. As for (2), we argue that (a) competing principles, rather than the difficulty of optimization, must be the actual reason for anti-dependency distance minimization effects and that (b) dependency distance minimization on star-like structures is less rewarding compared to other structures.