BriefGPT.xyz
Oct, 2022
IsoVec:控制词嵌入空间的相对同构性
IsoVec: Controlling the Relative Isomorphism of Word Embedding Spaces
HTML
PDF
Kelly Marchisio, Neha Verma, Kevin Duh, Philipp Koehn
TL;DR
本文提出一种通过在Skip-gram损失函数中加入全局同构度量来提高单语词向量空间同构性的方法以改进词向量之间的跨语言映射,从而实现对通用数据条件下的双语词典诱导、领域不匹配和训练算法不匹配的提高。
Abstract
The ability to extract high-quality
translation dictionaries
from monolingual word embedding spaces depends critically on the geometric similarity of the spaces -- their degree of "
isomorphism
." We address the ro
→