Recent publications
Vacancies
MPI colloquium series
IMPRS for language sciences
This content is archived, it could be outdated.
Weak evidence for word-order universals
On April 13, 2011, MPI researchers Michael Dunn and Stephen Levinson's paper, 'Evolved structure of language shows lineage-specific trends in word-order universals', has been published online in Nature. Using Bayesian phylogenetic methods, the authors show that the 'rules' governing word order change are different in different language families, casting doubt on universal theories of grammar.
Michael Dunn
Languages dazzle with different possible orderings of their elements – in some languages verbs come at the beginning of a sentence, in others at the end, or in the middle; sometimes prepositions come before the noun, sometimes after (in which case we talk of ‘postpositions’). Greenberg pointed out that these possible orderings are not randomly distributed within or across languages – for example, languages with verbs at the end tend to have postpositions (see Figure). These observations triggered the birth of the field of language typology, and encouraged generative linguists to think in terms of innate ‘parameters’ which a child might set on the basis of superficial evidence.
Correlations between preposition/postposition and verb-initial/verb-final final word order features in part of the Indo-European family of languages.
Blonde hair, blue eyes
But Michael Dunn and Steve Levinson (MPI Nijmegen) with their New Zealand colleagues Russell Gray and Simon Greenhill (University of Auckland) doubted that these word order correlations were firmly grounded, because they were based on world-wide correlations without enough allowance for relatedness. For example, suppose we observe that most people with blonde hair have blue eyes – should we assume that the genes are linked, or could it be that most such people share a distant common ancestor who had both traits? 'To control for common ancestry you should look within the family tree: if descendants who have dark hair nearly always have dark eyes and vice versa, that suggests gene linkage.'
Using Bayesian phylogenetic methods to reconstruct evolutionary processes, the researchers therefore looked at four large language families from different continents to see whether word-order traits are linked when common inheritance is controlled for (see map).
The locations of the languages analysed in the study.
The results are surprising. Pairs of word order features (like ‘verb at the end’ and ‘postpositions’) do not have universal correlations across language families in the way that has been expected. Instead, Dunn and his colleagues found that each language family had its own evolutionary tendencies. The way in which languages change and develop seems to be most strongly constrained by the historical starting point. Evolutionary contingencies trump human cognitive-linguistic universals as the driving factor in the evolution of language structure.
Link to the publication.
Contact: Michael.Dunn@mpi.nl
