Transparent Semantic Change Detection with Dependency-Based Profiles

TL;DR

This work presents a transparent, dependency-based approach to lexical semantic change detection that uses dependency-slot co-occurrence profiles and measures change with $JSD$. By applying frequency filtering and POS removal, it produces interpretable, slot-level attributions of semantic shift across English, German, Swedish, and Latin on SemEval-2020 Task 1 data, achieving competitive performance in several languages and tasks. The results demonstrate that explicit linguistic signals can rival embedding-based methods in interpretability and, in some cases, in performance, offering a bridge between NLP and diachronic linguistics. The work suggests a viable, theory-grounded alternative or complement to purely statistical embedding approaches, with potential for further refinement through adaptive slot weighting, targeted slot selection, and hybrid methods.

Abstract

Most modern computational approaches to lexical semantic change detection (LSC) rely on embedding-based distributional word representations with neural networks. Despite the strong performance on LSC benchmarks, they are often opaque. We investigate an alternative method which relies purely on dependency co-occurrence patterns of words. We demonstrate that it is effective for semantic change detection and even outperforms a number of distributional semantic models. We provide an in-depth quantitative and qualitative analysis of the predictions, showing that they are plausible and interpretable.

Figures (2)

• Figure 1: Changes in the frequencies of 6 adjectival modifiers of the English noun plane between the 2 sub-corpora for English of the SemEval 2020 shared task 1 schlechtweg_semeval-2020_2020. Plane is annotated as semantically changed in the dataset
• Figure 2: The JSD contributions of different slot-fillers in the slot chi_compound of the English noun graft. Green bars indicate an increase in relative frequency; red bars indicate a decrease.