Revealing Long-term Language Change with Subword-incorporated Word Embedding Models

AbstractWe propose an augmented word embedding model that better incorporates subword information with additional parameters that characterize the semantic weights of characters in composing words. Our model can reveal some interesting patterns of long-term change in Chinese language, which provides novel evidence and methodology that enriches existing theories in evolutionary linguistics. The resulting word vectors also has decent performance in NLP-related tasks.


Return to previous page