Paul Egré (CNRS, Paris, http://paulegre.free.fr/) will give a talk on:
“Vagueness and Moving Thresholds” (joint work with Steven Verheyen)
Thursday 25th, 14:00-16:00
Room 508 (Philosophy building).
All are welcome. If you are not from King’s, simply sign in and follow the signs to the philosophy building and then room 508 (you may need to ask your way around).
According to Barker (2002), vague expressions can be used to update the context in relation to two kinds of uncertainty: uncertainty about the standard of comparison operative in the conversational context (what does “tall” mean around here?), and uncertainty about the stimulus described (“is John tall?”). Lassiter (2011) and Lassiter and Goodman (2015) have proposed a probabilistic account of vagueness focusing mostly on the first kind of uncertainty. Other probabilistic accounts of vagueness on the other hand (including Verheyen et al. 2010, Egré 2017) focus mostly on the second kind of uncertainty, namely on the idea that categorization decisions are made relative to a fixed threshold, but with various probabilities. In Egré (2017), probabilities are introduced in relation to the idea that categorization is a process of noisy magnitude estimation. In the models used by Verheyen et al. (2010), probabilities are supposed to be a function of the item’s similarity to a given anchoring value, but can basically be rationalized in the same way. The models introduced independently by Verheyen and by Egré are meant to support the idea that vagueness is fundamentally a semantic phenomenon, that is, speakers can vary faultlessly in positioning their thresholds for categorization differently along the same latent dimensions. Arguably, however, the idea of a fixed threshold within each speaker may appear to cohere better with an epistemicist view of the meaning of vague terms, that is with the idea that vague expressions fundamentally denote crisp properties. In this paper we extend the threshold model proposed in Egré (2017) to combine it with the idea that thresholds themselves could be probabilistic. The model incorporates the two levels of uncertainty distinguished by Barker (2002) and by Lassiter (2011), but prima facie, this time it is free of any remaining epistemicist commitment. We discuss two difficulties for that approach: the first is to show that those two levels are effectively needed to model vague judgements. The second is how to interpret such moving thresholds within an individual for the same category. Several options can be considered: the fluctuation could be due to the same individual changing criteria of application over time; or to the individual being uncertain about specific contextual properties; or to the individual sampling over plausible collective values but being uncertain about any single value. We discuss those various assumptions and outline ways to tease apart those hypotheses.