Preference, attitudes and sentiment are part of stable mental representations shared across homogeneous groups of people. So far, evidence for the patterned variation between groups of people (e.g. defined by gender, language, etc) remains marginal in work on sentiment ratings (e.g. Warriner. Kuperman, & Brysbaert, 2013). Similarly, these factors are often assumed to be subjective and peripheral to theories of meaning (cf Murphy, 2004). This study will evaluate to what degree these factors contribute to the meaning across a variety of well-studied semantic domains and vary systematically across groups. In contrast to previous work on lexical sentiment, we assume that the aspect of sentiment is encoded at the category level (e.g. “comfort” for the category of clothing, “ferocity” in the context of animals, “safety” in the context of vehicles, “prestige” in the context of professions, “nutritional value” in the context of food), and thus provides a sensitive context to test assumed differences). The goal of this project is to investigate a new method to measure preferences for concepts in categorical context and test whether such preferences are reliable and stable across individuals. For example, we will ask participants to judge their preference in the context of a specific category (e.g. ANIMALS: indicate preference for cat, dog, sheep). These preference ratings will be compared with previous studies where concepts are judged out-of-context to test reliability and systematic differences for this new preference measure. The ratings will also be compared with semantic measures from previous work such as goodness-of-exemplar measures and pairwise similarity to test the hypothesis that preference contribute to meaning. Finally, systematic differences between groups (here in terms of gender) will be explored to identify the prevalence (i.e. how common) and the nature (what semantic domains) vary systematically between groups.
Research Questions / Hypotheses
Are preference rankings stable within categories and can variability be explained in terms of individual differences related to age, culture or personality?
109 participants. 98 participants who qualified on the basis of data quality and demographics.
The materials consisted of 10 semantic categories that vary in concreteness (e.g. FRUIT, ACADEMIC DISCIPLINE) each containing 25 exemplars (e.g., lemon, apple in the case of FRUIT). For each category participants judged their preference in a category context by positioning them on a zoomable scale. This was followed by a forced-choice task in which participants were asked to pick the most preferred pair drawn from their previous judgments to assess reliability, together with a verbal description of what aspect was considered in their preference judgment (e.g. lemon vs fig: I prefer lemons because they're tastier).
Our results show that subjective quantities such as preferences are highly reliable and vary consistently between groups of speakers with different gender/language characteristics. While preference judgments correlate with existing judgements for valence, these correlations tend to vary depending on the category and are moderate on average. We expect the context will provide a more accurate measurement of sentiment in terms of preference, which should result in fine-grained distinctions between groups often missed in context-free judgments of sentiment.
This study has the potential to change current theories of concepts and semantics as it might show that preferences are part of a concepts' meaning that is shared among groups of people and can be attributed to preferences for categorical aspects (i.e. dimensions like price/aesthetic appeal for artifacts or ferocity for animals) rather than global lexico-semantic factors such as valence or arousal. Moreover, we also believe this could advance such theories by taking into account systematic individual variation, which might be underestimated in context-free judgements.