Halliday & Matthiessen (1999: 39, 40):
The problem of multi-dimensional organisation in phonological systems was solved by the introduction of oppositions that can intersect (such as front/back, open/closed, and rounded/unrounded for vowels); in Trubetzkoy’s (1939) theory, the oppositions were treated as classificatory dimensions and the values as ‘sound properties’. In the late 1940s, however, Roman Jakobson (1949) reinterpreted these properties as components or distinctive features of the phoneme; this was in effect a reification of the earlier sound properties. Phonemes were said to consist of components (just as longer phonological sequences consisted of phonemes) instead of being said to realise terms in phonological systemic oppositions. In other words, a paradigmatic abstraction (sound property in an opposition) was given a syntagmatic status (component of a phoneme). Jakobson’s notion of component or distinctive feature was then taken over into generative phonology. …
Many of the problems with componential analysis arise from misinterpreting the 'components' as if they were constituents of some structure, rather than being paradigms of abstract features; …