DUKweb, diachronic word representations from the UK Web Archive corpus

Area of Science:

• Computational linguistics
• Digital humanities
• Natural Language Processing

Background:

• Lexical semantic change analysis is crucial for understanding language evolution and for NLP applications.
• Diachronic word embeddings are the standard resource for this task but are computationally expensive to create.
• Existing resources for diachronic word embeddings are scarce, limiting research accessibility.

Purpose of the Study:

• To present DUKweb, a novel, large-scale resource for the diachronic analysis of contemporary English.
• To provide readily available diachronic word embeddings to the scientific community.
• To demonstrate the utility and quality of DUKweb for lexical semantic change detection.

Main Methods:

• DUKweb was constructed using the JISC UK Web Domain Dataset (1996-2013) from the Internet Archive.
• The resource includes yearly word co-occurrence matrices and two types of word embeddings.
• A case study on word meaning change detection was performed to validate the resource.

Main Results:

• DUKweb offers a comprehensive set of diachronic word embeddings spanning 1996-2013.
• The resource facilitates the study of lexical semantic change in contemporary English.
• The case study confirmed the potential for reusing DUKweb for linguistic analysis.

Conclusions:

• DUKweb addresses the scarcity of accessible diachronic word embedding resources.
• The resource enables new avenues for research in lexical semantic change and NLP.
• DUKweb provides a high-quality, reusable dataset for diachronic language analysis.