Vol. 41:1 (2024) ► pp.46–98
Disentangling Ancestral State Reconstruction in historical linguistics
Comparing classic approaches and new methods using Oceanic grammar
Ancestral State Reconstruction (ASR) is an essential part of historical linguistics (HL). Conventional ASR in HL relies on three core principles: fewest changes on the tree, plausibility of changes and plausibility of the resulting combinations of features in proto-languages. This approach has some problems, in particular the definition of what is plausible and the disregard for branch lengths. This study compares the classic approach of ASR to computational tools (Maximum Parsimony and Maximum Likelihood), conceptually and practically. Computational models have the advantage of being more transparent, consistent and replicable, and the disadvantage of lacking nuanced knowledge and context. Using data from the structural database Grambank, I compare reconstructions of the grammar of ancestral Oceanic languages from the HL literature to those achieved by computational means. The results show that there is a high degree of agreement between manual and computational approaches, with a tendency for classical HL to ignore branch lengths. Explicitly taking branch lengths into account is more conceptually sound; as such the field of HL should engage in improving methods in this direction. A combination of computational methods and qualitative knowledge is possible in the future and would be of great benefit.
Article outline
- 1.Introduction
- 2.Background
- 2.1The methods of ancestral state reconstruction in traditional historical linguistics
- 2.1.1Disagreements in HL
- 2.2Evaluating if the data are valid for phylogenetic analysis: The Double Cognacy Condition and phylogenetic signal
- 2.3Computational phylogenetic methods
- 2.1The methods of ancestral state reconstruction in traditional historical linguistics
- 3.Materials and methods
- 3.1Methods: Maximum parsimony, maximum likelihood and most common
- 3.2Calculation of similarity between predictions from conventional HL and computational approaches
- 3.3Data
- 3.3.1The Grambank dataset
- 3.3.2Data coverage
- 3.3.3The trees
- 3.3.4Data from historical linguistics on Oceanic proto-language grammar
- 4.Results
- 4.1Concordance between traditional HL and computational methods
- 4.2New predictions
- 4.3Where the conflicts are: Ergativity
- 5.Conclusions
- Acknowledgements
- Notes
- Abbreviations
- Supplementary material
-
References
For any use beyond this license, please contact the publisher at [email protected].
https://doi.org/10.1075/dia.22022.ski