IMS at SemEval-2020 Task 1: How low can you go? Dimensionality in Lexical Semantic Change Detection

We present the results of our system for SemEval-2020 Task 1 that exploits a\ncommonly used lexical semantic change detection model based on Skip-Gram with\nNegative Sampling. Our system focuses on Vector Initialization (VI) alignment,\ncompares VI to the currently top-ranking models for Subtask 2 and demonstrates\nthat these can be outperformed if we optimize VI dimensionality. We demonstrate\nthat differences in performance can largely be attributed to model-specific\nsources of noise, and we reveal a strong relationship between dimensionality\nand frequency-induced noise in VI alignment. Our results suggest that lexical\nsemantic change models integrating vector space alignment should pay more\nattention to the role of the dimensionality parameter.\n