On the Feasibility of Dependency Parsing of Non-Human Sequences Without a Gold Standard (arXiv 2507.06818)

Paper Overview

• Field: NLP
• Authors: Ramon Ferrer-i-Cancho, Catherine Hobaiter, Thore Bergman
• Published: 2025-07-09
• arXiv: 2507.06818

Summary

Dependency parsing consists of finding a tree representation for a sequence. Unsupervised dependency parsing aims to develop parsing methods without a gold standard during model training. In human languages, an unsupervised parser can be evaluated because some gold standard is usually available or can be created. For other species, a gold standard is unknown. Thus one may conclude that it is impossible to determine the accuracy of an unsupervised parser and, consequently, dependency parsing is unfeasible in other species. However, the authors apply recent advances in network science to demonstrate that the proportion of correct edges retrieved by a parser must be high for the sequences of vocalizations or gestures that non-human primates produce, due to the fast decay of the sequence length distribution. In contrast, human language sequences lack this property. Therefore, evaluation without a gold standard is feasible in non-human primates but is a challenging problem in humans.

Key Takeaways

• Unsupervised dependency parsing is normally hard to evaluate without a gold standard.
• For non-human primate vocal and gesture sequences, network science results guarantee a high proportion of correct edges retrieved by parsers.
• Human language sequences lack this favorable length-distribution property, so the same guarantee does not hold.
• This reverses the usual assumption: parser evaluation without ground truth is more tractable for non-human primates than for human languages.