The evolution of relative clauses in the IcePaHC treebank

Anton Ingason, Johanna MechlerNAACL 2025

TL;DR

This paper investigates the historical evolution of relative clauses in Icelandic, specifically how the use of relative complementizers (sem vs. er) and relative pronouns has changed over centuries. By analyzing the meticulously annotated IcePaHC treebank, the authors reveal a regular, S-shaped shift from er to sem and explore the factors influencing the rarer use of relative pronouns, offering new insights into syntactic change and the Constant Rate Effect.

What Problem Does This Paper Address?

The paper tackles the question of how grammatical structures change over time within a language. Specifically, it focuses on the historical development of relative clauses in Icelandic, a type of clause that modifies a noun (e.g., "the girl that went away"). Icelandic has two primary ways to introduce these clauses: relative complementizers (like sem and er, analogous to English "that") and relative pronouns (like hverja, analogous to English "whom"). While sem and er are common, relative pronouns are much rarer. The core problem is to understand the diachronic (historical) trajectory of these elements, how their usage has shifted, and what factors might have driven these changes. This is particularly interesting because the entire historical record of this change in Icelandic is preserved, allowing for a detailed analysis of a linguistic evolution.

Methods

The study employs a phrase structure analysis of the IcePaHC (Icelandic Parsed Historical Corpus) treebank. This treebank is a manually annotated collection of Icelandic texts spanning from the 12th to the 21st century, with syntactic structures meticulously parsed. The authors used a Parsed Corpus Query Language (PaCQL) to extract relevant examples of relative clauses. Statistical analysis was performed using R, specifically employing mixed-effects regression models. These models were used to investigate the relationship between the usage of sem and relative pronouns, and predictors such as century, type of gap (subject, object, other), and genre. The analysis also considers the Constant Rate Effect, a linguistic theory suggesting that syntactic changes tend to propagate at a consistent rate across different contexts.

Datasets

The primary dataset used is the IcePaHC (Icelandic Parsed Historical Corpus). This is a manually annotated phrase structure treebank containing one million words of Icelandic text from the 12th to the 21st century, covering five genres: narrative, religious, biographical, legal, and scientific texts. The corpus is described as a valuable resource for digital humanities and historical linguistics research. No new benchmarks were introduced; rather, the study leverages the existing rich annotation of the IcePaHC.

Key Insights and Ideas

  • Regularity of Change: The shift from the relative complementizer er to sem in Icelandic follows a remarkably regular, S-shaped curve over time, a pattern that is more consistent than some previously studied linguistic changes (like do-support in English).
  • Gap Type Influence: The type of "gap" (the missing element in the relative clause, e.g., subject or object) influences the choice between sem and er. Interestingly, this influence changes over time, with sem being favored for subject gaps in later periods when it becomes dominant.
  • Genre Matters for Pronouns, Not Complementizers: While genre was not a significant predictor for the choice of relative complementizer (sem vs. er), it played a role in the usage of relative pronouns. This suggests different drivers for the evolution of these two types of relative clause introducers.
  • Constant Rate Effect Tested: The study provides a test case for the Constant Rate Effect. The interaction between century and gap type in the models suggests that the rate of change might not be perfectly constant across all contexts, hinting at potential complexities or the influence of other factors.
  • Translation Influence: The increased use of relative pronouns in certain periods might be partly attributed to translation from German, where such constructions might have been more common.

Key Results

  • Complementizer Shift: The empirical rate of sem (relative complementizer) shows a clear increase over centuries, largely replacing er. This shift is described as a "very regular S-shaped curve."
  • Relative Pronoun Usage: Relative pronouns are significantly rarer than complementizers. Their usage peaked between the 16th and 18th centuries, reaching about 20% of relative clauses in the 17th century.
  • Statistical Significance: Mixed-effects regression models showed that "century" was a highly significant predictor for both relative complementizers and pronouns. "Gap type" was also significant, particularly in interaction with "century" for complementizers. For relative pronouns, "genre" was a significant predictor, with narrative texts differing from religious and biographical ones.
  • Model Fit: Adding the interaction between "century" and "gap type" improved the model fit for both complementizers and pronouns, suggesting that the rate of change is not entirely uniform across all contexts.

Limitations and Future Work

  • Unaccounted Predictors: The authors acknowledge that other predictors might influence the distribution of relative complementizers and pronouns that were not included in their analysis.
  • Rare Elements: The study focused on the most common elements introducing relative clauses and did not account for rarer alternatives.
  • Written vs. Spoken Language: The analysis is based on written language, and the authors suggest that spoken language might exhibit different patterns or nuances.
  • Corpus Annotation Errors: While the IcePaHC corpus is considered accurate, the analysis relies on its annotations, which could potentially contain errors.
  • Future Work: The authors suggest that further interpretation of the observed effects, particularly the interaction between century and gap type, requires more research. Comparisons with other languages and phenomena are also recommended. The study's findings also contribute to the ongoing research on the Constant Rate Effect, providing another test case for this theory.

Summary updated

BibTeX
@inproceedings{ingason2025evolution,
  author = {Anton Ingason and Johanna Mechler},
  title = {{The evolution of relative clauses in the IcePaHC treebank}},
  booktitle = {Conference of the North American Chapter of the Association for Computational Linguistics},
  year = {2025},
}
Figure