Research Shows Treebank Choice Significantly Alters Dependency-Distance Estimates

A new scientific preprint examined how corpus selection impacts dependency-distance estimates across languages. Researchers analyzed 38 same-language treebank pairs from Universal Dependencies v2.18 to test whether single-corpus measurements reflect stable language-level properties.

The findings showed only moderate agreement across different treebanks. Replacing one treebank with another reversed nearly 40 percent of pairwise language rankings, with treebank choice explaining roughly 29 percent of between-group variance across twelve preprocessing specifications.

Despite these variations, all analyzed treebanks confirmed the principle of dependency-length minimization. The authors concluded that mean dependency distance reflects a combination of grammar, register, and annotation factors rather than a standalone linguistic parameter.