Sinhala language evolution reveals uneven semantic change
Summary and headline written by AI from the source article. How we work
Researchers developed a new computational method to trace how the meaning of words changed in the Sinhala language of Sri Lanka between the 13th and 20th centuries.
The team aligned word embeddings, numerical representations of word meaning, from different centuries using two techniques, Similarity Matrix Based Alignment and Orthogonal Procrustes. They found Orthogonal Procrustes better maintained consistent relationships between words over time. The research goes beyond simply measuring overall shifts in meaning.
The team introduced a Bidirectional Semantic Impact Pruning approach, using a fine-tuned version of the Llama-3.1-8B language model to pinpoint specific sentences driving semantic change. This allowed them to differentiate between broad, systemic shifts in language and temporary expansions of a word’s multiple meanings. Analysis revealed that semantic drift wasn’t uniform. Instead, a relatively small number of contextual instances significantly influenced how words evolved, suggesting that change isn’t always gradual or consistent.
The method offers a starting point for studying language evolution in Sinhala and other languages with limited digital resources, while also acknowledging the balance between model precision and available data.
