Politicize and Depoliticize: A Study of Semantic Shifts on People's Daily Fifty Years' Corpus via Distributed Word Representation Space

摘要

The semantic meanings of words are always changing with time. In this paper, we focus on semantic shifts, a certain type of semantic change, which indicates the meaning changes of a bunch of words that are influenced by social trends in specific time period. By training distributed word representation spaces for segmented time periods in diachronic corpus and mapping them into a universal semantic space, the semantic shifts of a certain cluster of words can be reflected as an offset vector in the universal space. Further study shows that this semantic shift vector can be used as standard pattern to trace the words which have the similar semantic shift between other time periods.