Reconstructing variation in Indo-European word order. A treebank-based quantitative study

Word order is a central issue in the reconstruction of Proto-Indo-European syntax. Categorical approaches have proved to be inadequate because they postulate for the protolanguage a typological consistency which is absent in any of the attested daughter languages. Following recent research, we adopt a gradient approach to word order, which treats word order preferences as a continuous variable. We analyze four word order patterns based on data extracted from treebanks of ancient Indo-European languages. After presenting our results for AdpN/NAdp, GN/NG, AN/NA, and OV/VO, we draw a number of conclusions concerning variation within individual languages, crosslinguistic variation, and variation in diachrony that support the claim that variability should be taken as the normal state across languages, including reconstructed stages. We conclude that a non-discrete approach has the advantage of leading to a reconstruction that better conforms to the situation known from real languages, with variation as a key feature.