Improving on observational blends research: regression modeling in the study of experimentally-elicited blends

Abstract

In this paper, we discuss the results of a blend production experiment and how it relates to previous research that was nearly exclusively based on observational data. Specifically, we study three different findings from published research, namely that (i) the shorter source word contributes more of itself to the blend than the longer source word, (ii) source word2 determines blend stress (more than source word1), and (iii) blending maximizes similarity between source words and blends. Using statistical techniques so far not employed in research on blends, we show that most findings from observational data regarding the three hypotheses studied are supported, but also occasionally tampered down.

Keywords

• experimental blend production
• observational blend collection
• source word lengths
• similarity
• regression modeling