| Description |
Multilevel Phylogenetic Inference of Language Universals and Change
One of the most well-known typological generalizations is that languages tend to maintain consistent ordering between the grammatical head and its dependents (Greenberg 1963; Hawkins 1983; Dryer 1992). Over the past decades, competing theories and new empirical findings have been advanced to explain the word order universals and language change. These theories range from functional explanations focusing on cognitive and learning biases (Hawkins 1983; Culbertson, Smolensky & Legendre 2012; Futrell, Levy, & Gibson 2020), to alternative views, emphasizing the roles of cultural evolution and historical accidents in language change (Bybee 1988; Dunn et al. 2011; Cristofaro 2019). However, there is little corpus-based evolutionary evidence either favoring or against harmonic head-dependent ordering, and it remains unclear whether word order changes follow universal patterns or lineage-specific trends.
To address these open questions, we conducted two case studies: one using large dependency-annotated corpora in Indo-European and another leveraging the Grambank database to examine functional dependencies in word order change throughout the history of language families. We go beyond previous phylogenetic approaches that model the correlated evolution separately for each pair of word orders (Dunn et al. 2011; Jäger & Wahle 2021; Hartung et al. 2022). Instead, we have developed a multilevel phylogenetic Continuous-time Markov Chain model to estimate evolutionary rates for individual word orders while accommodating lineage-specific variation (Nalborczyk et al. 2019; Stan Development Team 2022).
Our results show no apparent difference in the estimated rate ratio for harmony between the observed data and two random baselines. When the distribution of individual word orders is kept constant in a language (second baseline), there is little room left for additional harmonic constraints between word orders in language use. Moreover, compared to the first baseline, the observed data exhibit a lower rate ratio, suggesting a weaker evolutionary bias for harmony. Overall, our findings contradict previous theories that predict a general preference for head-initial or head-final structures (Hawkins 1994; Ferrer i Cancho 2017). Instead, word order changes appear to be more diverse than commonly assumed, and they tend to evolve toward more mixed configurations both within Indo-European and globally. In line with Jäger & Wahle (2021), our global phylogenetic analyses suggest that a universal model better explains word order patterns than lineage-specific models with and without correlation.
|
