This is interesting.
There are different notions of text here, depending on point of view.
Certainly a sentence could contain a mixture of scripts.
When I said "run of text" I meant it from the point of view of how font featues are applied to text. Usually it is a small chunk of text, with chunks separated by white space or other delimiters.
In web browsers, etc, if a run of text is, say, from the Armenian Unicode range, the script is judged to be Armenian for the purposes of matching OpenType lookups.
(The information of what *language* is meant by the text must be supplied otherwise, in this case, in a 'lang' attribute).
Now, you could imgaine applying an OpenType lookup to, say, a Malayalam letter immediately followed by an Arabic letter, but this is .. I think... really pointless.
Even in text containing very mixed scripts, at least the words from the different scripts are separated by white space (or other punctuation etc).
So the mechanism of determining the "script" of a "run of text" does typically make good sense.
That all said:
It's not clear to me how this is implemented in ConTeXt. I'll play with it once I have some examples working. My guess is, the author has to explicitly indicate the script each run of text (by specifying the attributes of the font to apply to the run).