Building a robust sentiment lexicon with (almost) no resource

TL;DR

This paper introduces a resource-efficient method for building sentiment lexicons across languages by transferring words via aligned word embeddings, eliminating the need for extensive machine translation resources.

Contribution

The paper proposes a novel approach using aligned word embeddings and linear transformation to create multilingual sentiment lexicons without relying on in-domain machine translation.

Findings

Comparable performance to machine translation-based methods

Effective across four different languages

Reduces resource requirements for lexicon creation

Abstract

Creating sentiment polarity lexicons is labor intensive. Automatically translating them from resourceful languages requires in-domain machine translation systems, which rely on large quantities of bi-texts. In this paper, we propose to replace machine translation by transferring words from the lexicon through word embeddings aligned across languages with a simple linear transform. The approach leads to no degradation, compared to machine translation, when tested on sentiment polarity classification on tweets from four languages.