Hanamura (2005) describes what it is when the element by itself is fuzzy. That is, an element may also be not exactly 10 but “approximately 10” or “10±10%”. This occurs more frequently as it appears at first glance. All measurements aren’t absolute, but rather they are afflicted with measuring tolerance. Strictly speaking, the value indicated by a measuring device must not be taken unconditionally, but must always be provided with some measuring tolerance.
This procedure is obviously in measurement technology. Vividly, a fuzzy number can be seen as a small fuzzy set, an interval, whereby locating the middle number and determining the width based on measuring tolerances.
For example, let’s consider that a thermometer shows a body of 36℃, then the tolerance amounts to ca.±1%. Through this technique, a triangular process of the membership function has proved itself as practically convenient.
One knows the measured value (36℃) and the interval border that appears during the tolerance data. Poor measuring devices with larger tolerances leads to larger intervals, good measuring devices without any tolerance provides a unique discrete value. The vertical line about the measuring value (such as figure 38 below) indicates that a value taking tolerance into consideration is important.
How does one now determine the membership grade of a fuzzy number to a fuzzy set? The most plausible way is to select the maximal membership function value at the intersection of both membership functions. For example, a body temperature of 36.0℃±0.4℃ creates the below curve shape (figure 39).
花村嘉英（2005）「計算文学入門－Thomas Mannのイロニーはファジィ推論といえるのか？」より英訳 translated by Yoshihisa Hanamura