Multilingual Word Embedding and Linguistic Linked Open Data for Tracing Semantic Change

This article proposes a methodology for combining natural language processing techniques for diachronic analysis and linguistic linked open data models to detect and represent semantic change. The change in meaning over time of words, phrases, or concepts encompasses complex phenomena that cannot be fully explained by distributional methods alone. We argue that by joining corpus-based and lexicographical evidence and modelling the results in an interoperable format can provide more solid ground for drawing conclusions and possibilities of re-use by other applications. We define a basic schema for a resource aggregator and a model called LLODIA (Linguistic Linked Open Data for Diachronic Analysis). To illustrate our approach, we use a multilingual dataset, in French, Latin, Hebrew, Old Lithuanian, and Romanian, and build a sample derived from word embeddings and dictionary resources, encoded by means of the proposed model.