Paul Teller (UC Davis) gives a talk at the MCMP Colloquium (18 April, 2013) titled "Separating Truth from Its Idealization". Abstract: Science never succeeds in providing representations that are both perfectly precise and completely accurate. Instead science constructs models that are always in some ways inexact – imprecise, not perfectly accurate, or both. If this goes for the results of science, how much more should we expect it to hold for human knowledge generally! I explore this expectation for the project of modeling what it is for a statement to be true. The familiar model of characterizing truth in terms of predicating a precisely delimited property of a precisely delimited referent succeeds famously in characterizing semantic structure, but falters with questions about application to the world because we rarely, if ever, succeed in perfectly determinately picking out properties and referents. I sketch an alternative model-building approach that takes advantage of the ubiquitous occurrence of imprecision: For an imprecise statement to be true is for its precise “semantic alter-ego”, though false, to function as a truth.