Computational Analysis of French Reborrowing Process for English Loanwords.

Analyzing semantic change of loanwords over time between different languages has been a longstanding sociolinguistic problem. However, few efforts have been put to computational approaches to analyzing the semantics of loanwords. In this paper, we present a new computational method for detecting and tracking the semantic change of loanwords between two languages, specifically for the reborrowing process of loanwords. We trained our model on pre-trained historical bilingual English-French word embeddings aligned with MUSE and proposed two quantitative measures of detecting loanwords reborrowing. The first measure analyzes cosine similarity of one word from two languages in the same year, and it is sensitive to cultural shifts and lexicon reborrowing. The second measure analyzes Pearson correlation from the tendency of the cosine similarity and thus predicts the pattern of these reborrowing loanwords. We show that our model can detect reborrowing loanwords that have been discovered in literature.