Time-Out: Temporal Referencing for Robust Modeling of Lexical Semantic Change
This addresses a specific bottleneck in computational linguistics for researchers studying language evolution, though it appears incremental as it builds on existing skip-gram architectures.
The paper tackles the problem of noise in lexical semantic change detection models caused by vector space alignment, showing that the Temporal Referencing method avoids alignment and outperforms alignment models on synthetic and manual testsets.
State-of-the-art models of lexical semantic change detection suffer from noise stemming from vector space alignment. We have empirically tested the Temporal Referencing method for lexical semantic change and show that, by avoiding alignment, it is less affected by this noise. We show that, trained on a diachronic corpus, the skip-gram with negative sampling architecture with temporal referencing outperforms alignment models on a synthetic task as well as a manual testset. We introduce a principled way to simulate lexical semantic change and systematically control for possible biases.