Filter theory (see this also), specified that we could only attend to one input at a time (serial processing of stimuli as opposed to parallel processing, whereby you can attend to many stimuli at the same time), and this was originally put forward by Broadbent in 1959. The theory suggested that stimuli can be filtered based upon physical attributes, prior to full processing by the perceptual system. The fact that there is the filter in place, in turn leads to some stimuli not raising attention. Plus, filter theory proposed that attention has a limited capacity due to the fact that you can only process one stimulus at a time. This is where the bottleneck is, as you will be bombarded with lots and lots of stimuli and you can only process one of them at a time.
Neisser (1967) outlined a two-process theory that explained how much the filter affects attention (and hence consciousness). According to Neisser's theory, both the properties of the stimuli as well as semantic factors (language or logic), play a role in attention. Neisser argues for a constructive view of cognition in which perception is shaped by existing knowledge and hence attention is influenced by experience.
So therefore, when Cohen states that:
Much of Neisser’s critique of filtering makes semantic rather than functional distinctions.
he is stating that Neisser is, in most of his theory, purely making logical distinctions within filter theory rather than functional ones.
I interpret this to mean that, Neisser added the function of logic (through knowledge and previous experience) along with language, to the function of the stimulus filter.
References
Broadbent, D. (1958). Perception and Communication. London: Pergamon Press.
Neisser, U. (1967). Cognitive Psychology. New York: Appleton-Century-Croft.