140 EMPIRICAL TOOLS AND EMPIRICISM 



ure accurately the "boiling temperature"— and adopt a thermometer 

 as the instrumental core of denotation. Do we then simply read oflF 

 the temperature shown by a thermometer immersed in the boiling 

 liquid? Not at all! We find that, due to "superheating," a boiling 

 liquid stands at a temperature irreproducibly higher than that of its 

 vapor. We then specify use of a "boiling point apparatus," in which 

 we arrange to measure the vapor temperature, following a procedure 

 designed to minimize superheating of that \^apor. We have now an 

 instrument, an apparatus, and a procedure— but at most only part of 

 the procedure: the actual thermometer readings will require "cor- 

 rection." Using a well-authenticated colligative relation, we make a 

 "stem correction" to allow for the fact that, although the mercury in 

 the thermometer bulb is at the temperature of the saturated vapor, 

 the mercury in the stem is ordinarily at some other, lower, tempera- 

 ture. Another colligative relation warns us that boiling temperature 

 is a function of pressure: we must then correct the thermometer 

 reading to allow for the "barometric pressure." But now, to establish 

 this last, we will require still other corrections. One such correction 

 term is a function of the temperature of the barometer, which af- 

 fects both the density of the mercury and the length of the measur- 

 ing scale; another in\'olves the latitude and elevation of the labora- 

 tory, which aflFects the gravitational acceleration, and hence the 

 formula for the conversion of barometric heights to barometric pres- 

 sures. And so on. Without going any further, we see clearly that 

 establishment of the denotation of "boiling temperature" requires a 

 procedural protocol of substantial complexity. 



How grave are the risks that the scientist will be led seriously 

 astray by the element of human subjectivity irreducibly involved 

 whenever, by making "corrections," he ventures to "tamper with the 

 facts"? The relations used in making corrections have been inde- 

 pendently checked many times o\'er. Moreover, in some crucial cases 

 we have made more elaborate experiments in which the need for 

 many if not most corrections is eliminated. In these cases we have 

 generally arrived at the very same results obtained much more con- 

 veniently in simpler experiments in\'olving multiple corrections. A 

 second point: corrections may enhance predictive reliability. For 

 example, Dulong and Petit's law we now know to be only a crude 

 approximation. We find it a better approximation when we use as 

 specific heats not the values actually measured, ordinarily at con- 



