Verbal probabilities are phrases used instead of numerical values to describe the likelihood of events, such as certain, maybe, and impossible. We encounter them daily, yet each person interprets them somewhat differently. To achieve more consistent interpretations, many fields have adopted standardized lexicons and their numerical translations. However, standardization can be problematic because people naturally rely on their own lexicons and cannot easily disregard their personal interpretations.
In this thesis, we explore an alternative to standardization by examining how the probability lexicon of one individual can be translated into the lexicon of another. We model verbal probabilities using membership functions and introduce a new elicitation method, chips and bins, which we compare with the commonly used slider-based approach. We also propose a new similarity measure between membership functions that, unlike existing approaches, accounts for the entire distribution of the function. Analysis of data collected through an online survey shows that the chips and bins method generally produces membership functions with narrower breadth and more pronounced peaks, making them particularly useful for translating between lexicons. Contrary to related studies, we also find that translations based on identical phrases outperform translations based on equally ranked phrases.
|
