However, although voicing is generally important in English, the voicing difference between the two fricatives written th, , has a very low functional load: It is difficult to find meaningful distinctions dependent solely on this difference. One of the few examples is thigh vs. thy although they can be distinguished from context alone. Similar is the difference of j [dʒ] vs. zh [ʒ], as in virgin vs. version. The difference between the two ng sounds, , found in singer and finger, is so unimportant that it makes no practical difference if one mixes them up. The functional load is nearly zero.
The term "functional load" goes back to the days of the Prague School; references to it can be found in the work of Vilem Mathesius in 1929. Its most vocal advocate was André Martinet, a historical linguist who claimed it was a factor in the likelihood of a phonological merger. He predicted that perceptually similar pairs of phonemes with low functional load would merge. This has not been proved empirically; indeed, all empirical tests have come out against it.
There is no standard measurement for Functional Load. A popular measure is the number of minimal pairs, but this does not take into account word frequency and is difficult to generalize beyond binary phonemic oppositions. Charles Hockett proposed an information theoretic definition in 1955, which has since been generalized. Now, given a large text corpus, one can compute the Functional Load of any phonological contrast including distinctive features, suprasegmentals, and distinctions between groups of phonemes. For instance, the Functional Load of tone in Putonghua (Mandarin Chinese) is as high as that of vowels; meaning that the loss of information when all tones sound alike in Putonghua is approximately equal to that when all vowels sound alike in the language.