The LSCD Benchmark addresses the heterogeneity in modeling options and task definitions for lexical semantic change detection (LSCD), which makes it difficult to evaluate models under comparable conditions and reproduce results.
The benchmark exploits the modularity of the LSCD task, which can be broken down into three subtasks: 1) measuring semantic proximity between word usages (Word-in-Context), 2) clustering word usages based on semantic proximity (Word Sense Induction), and 3) estimating semantic change labels from the obtained clusterings.
The benchmark integrates a variety of LSCD datasets across 5 languages and diverse historical epochs, allowing for evaluation of WiC, WSI, and full LSCD pipelines. It provides transparent implementation and standardized evaluation procedures, enabling reproducible results and facilitating the development and optimization of LSCD models by allowing free combination of different model components.
The authors hope the LSCD Benchmark can serve as a starting point for researchers to improve LSCD models, by stimulating transfer between the fields of WiC, WSI and LSCD through the shared evaluation setup.
To Another Language
from source content
arxiv.org
Principais Insights Extraídos De
by Dominik Schl... às arxiv.org 04-02-2024
https://arxiv.org/pdf/2404.00176.pdfPerguntas Mais Profundas