Study of the polysemic term of minerality in wine: Segmentation of consumers based on their textual responses to an open-ended survey

Over the past 20 years, the word ``minerality'' has been increasingly used in the description of wines. However, a precise definition of the concept of minerality appears to be inexistent, and no consensual meaning, even among wine professionals, can be identified. Although this word usage seems to spread out from wine professionals to consumers, research on what consumers assume about minerality is scarce. This paper aims to study the various concepts about minerality held-by consumers by using an open-ended questionnaire. A total of 1697 French-speaking consumers responded to an, online survey and their free answers were analysed using statistical textual methods. The clustering around latent variables (CLV) method was used, taking into account both the lexicon used and the personal characteristics of consumers to classify them. Word associativities were then computed by means of renormalized Markov associativities, generating textual networks associated to each group, as well as to personal characteristics of the consumers. Typically, the most inexperienced consumers confess to have never heard about minerality in wine. Then, young women, also endowed with little wine competences, mainly associate minerality to mineral ions as those found in bottled water. Slightly older consumers embed the concept of minerality into the idea of terroir. Finally, the most experienced consumers refer to sensory perceptions such as gunflint or acidity. Those findings are consistent with a lexical innovation process, diffusing from wine professionals to consumers, referring to the mineral kingdom (as opposed to animal or vegetal), and aiming to stress that the style of their wines has changed-towards more subtlety. Beyond the specific minerality issue investigated in this paper, the methodology (CLV approach used in conjunction with renormalized Markov associativities) demonstrates its ability to generate informative clusters of textual networks, highlighting the cores of prototypical sentences, and apt to investigate the meaning of new concepts. (C) 2016 Elsevier Ltd. All rights reserved.