Infants and some mammals can apparently perceive VOT continuum categorically, which I guess is evidence for natural auditory sensitivities, and can explain, at least in part, the mechanisms through which phonemes are perceived categorically. But, even if I have an inkling of the impact, one thing I don't really understand after reading these and similar articles is what is meant by "perceiving the VOT continuum". What does it mean to perceive a continuum?
It has to do with two things: the experimental task, and the nature the things being perceived. All acoustic measurements are continuous, i.e. capable of having any number of values. However, the mind does not handle continuity well, and linguistic units are all categories (a small set of discrete values), but are based on physical continua. Tone in a language with downstep is a great example of the categorical vs. continuous, since any number of pitches can occur in an utterance and be treated as H vs. L, depending on phonological context (for instance, in H!H!H!H, the last H may be lower than L in LHHH).
The task is typically to present subjects with stimuli continuously varying along some dimension and tell them to identify a stimulus as a particular word (bad vs.pad, for example). If "perception" of linguistic stimuli were continuous, you would expect speakers to have slowly increasing rates of "bad" over "pad" as you reduce VOT, and instead you get rather rapid rates of "bad" identification.
People are quite capable of "perceiving" such differences when treated as physical signals, though it may take work to get subjects to not treat the task as linguistic-identification. Bilingual subjects can be good, because they have higher degrees of awareness of differences in VOT. The problem is that "perceive" is used in two ways: one can "perceive" pitch and duration continously, or categorially. Likewise though not a obviously, one can perceive that stimuli A, B, C all have different VOT values, or you can treat them as falling into two categories. The second kind of "perception" might be better termed "analysis"; however, I think it is actually correct to call all of these automatic processes as being "perception", just as you can perceive an apple, or a combination of somewhat round, shiny, and red: you don't "perceive" just the low-level stimuli, and there is no conscious analysis involved in integrating facts into an apple or specific phoneme.
Categorical means yes vs no or one of small set of things like red/green/blue. Phonemic features of consonants are usually like this, place of articulation, stop vs fricative, voiced vs unvoiced.
Continuous means on a smooth scale like frequency or a large approximate set of things (like time in milliseconds).
Perception works through sensor devices, special nerve cells like rods and cones in the eye or hair cells in the cochlea.
The action of these sensors is limited to firing once past a threshold and then taking some (very short time) to reset so that it can fire again.
Also there are many cells of the same kind so that they may not fire all at the same time but if exposed to an extended source of one particular kind of energy can seem to, as a set, fire continuously.
A single firing is a sign of a category, the number of firings (by multiple cells, or by a single cell over time) approximates continuity.
Perception of timing is not done in these primary sense cells, but further down the neural pathway. However the same principles apply to these dependent cells.
So despite the inherently categorical (binary) nature of a single cell, a bunch combined together or analyzed over time produce a continuous measure.
Voice Onset Time, in terms of production, is the time from the start of some articulation and the start of vibration of vocal cords at some generic frequency. Timing is a continuous parameter, there's no threshold.
But consonant phonemes in human language are perceived categorically, meaning there is some VOT value that acts as a threshold, that is there is no slow statistical increase in probability of a positive voicing feature, rather a sudden jump from always considered yes to always considered no. When the VOT is less than this threshold then the voicing feature is almost always considered on, when greater then off, and only in the smallest neighborhood of the threshold is it questionably on or off.