An experience in automatically building lexicons for affective computing in multiple target languages

Abstract

Affective Computing in text attempts to identify the emotional charge reflected in it, trying to analyse the moods transmitted while writing. There are several techniques and approaches to perform Affective Computing in texts, but lexicons are their common point. However, it is difficult to find solutions for specific languages different from English. Thus, this article presents an experience in automatically generating lexicons to perform Affective Computing following a multiple- target languages approach. The experience starts with some initial seeds of words in English that define the emotions we want to identify. It then expands them as much as possible with related words in a bootstrapping process and finally obtains a lexicon by processing the context sentences from parallel translated text where the terms have been used. We have checked the resulting lexicons by conducting an exploratory analysis of the affective fingerprint on a parallel corpus with books translated from and to different languages. The obtained results look promising, showing really similar affective fingerprints in different language translations for the same books.