Particularly in the context of learning to become literate in an alphabetic script, people talk a lot about the principle that (spoken) words can be taken apart and decomposed into smaller pieces, which relate in some way to abstract units which you might call phonemes or segments. (Note that that relation is not that speech “consists of” phonemes/segments – an (important) separate issue.)
But saying that ‘words are decomposable into’ segments is not quite synonymous with saying that words ‘come apart into’ segments. There is a critical difference in what these two phrases imply. An orange comes apart into segments, because that’s what an orange is like once you unpeel it: the segments and the boundaries between them are immediately obvious – it’s decomposable into segments in a natural way. But an apple doesn’t ‘come apart’ into segments – you have to cut it into pieces yourself. It’s ‘decomposable’ in the sense that it’s certainly possible to divide it, and people obviously frequently do divide apples up into sensible pieces that suit them, say quarters length-ways, but considered in itself it’s an undivided whole.
Spoken words, if I may say so, are much more like apples than they are like oranges. Evidence comes from all directions to explode the idea that speech can consist of discrete, independent, atomic particles – whether from inspection of waveforms and spectrograms, listening to speech in an unfamiliar language, or the recognition that the entire vocal tract is in constant motion throughout an utterance, and that in the motor behaviour of speech producers there are as many different configurations of the vocal tract as there are moments of time in the duration of the utterance.
And David Abercrombie makes the same point:
“‘To segment,’ therefore, has two kinds of meaning. There is ‘artificial’ segmentation: it is possible to cut a bit off anything, however much of a continuum it might be, as with a circle; and there is ‘natural’ segmentation. In the first case, we make boundaries; in the second, we detect boundaries. The fundamental problem of linguistic segmentation is whether there can be a natural segmentation of continuous speech. If not, if speech is a continuum in which no natural segmentation is detectable, then any segmentation of it must be artificial.” (Abercrombie 1991: 28; see also chapter 16 of Ladefoged 2006)
Without a recognition of this property of the sounds of speech, any account of the literacy acquisition process – however popular – is bound to flounder into incoherence, and fail to be even descriptively adequate, never mind providing any realistic kind of explanation.
Abercrombie, D (1991), Fifty Years in Phonetics. Edinburgh: Edinburgh University Press
Ladefoged, P (2005), Vowels and Consonants. Oxford: Blackwell