Im chattin :-) u wanna NLP it: Analyzing Reduction in Chat

Abstract

In this paper, we study the conditions in which words are reduced in chat text, in particular whether these conditions resemble those found for reduction in spoken text. We extract instances of five types of reduction from the 2Mw NPS Chat corpus, and apply regression analysis to measure the level of influence for 9 potentially predicting features. Although we do find significant effects, there appears to be no relation to the findings for speech. Furthermore, only the feature representing individual preferences is shown to have a substantial explanatory value. We conclude that for a reliable prediction on a chatter’s choice to use reduction, we need to look further than our limited set of features and apply more intricate analyses, both of which necessitate access to more extensive and more accessible chat data.