What are Georgian harmonic clusters phonetically?

Question

According to Georgian: A Reading Grammar by Howard I. Aronson, Georgian has many "harmonic clusters" consisting of two consonants pronounced with only a single release. (The consonants must be stops or affricates, I guess, but the distinction between stops, affricates, and fricatives in Georgian is something I don't understand very well since it seems that some phonemes can be rendered in multiple ways.)

The examples he gives, which are supposed to be exhaustive, are as follows (using the official romanization: apostrophe to denote ejectives, q to denote a uvular or thereabouts stop or affricate or fricative, and kh/gh to denote an unvoiced/voiced probably post-velar affricate or fricative):

bg, dg, dzg, jg
pk, tk, tsk, chk
p'k', t'k', ts'k', ch'k'

bgh, dgh, dzgh, jgh
pkh, tkh, tskh, chkh
p'q', t'q', ts'q', ch'q'

My question is how these clusters are actually produced (and more generally whether Aronson's description is correct). It seems impossible to literally have a single release. What I suspect is meant is that the tongue starts with two points of contact against the roof of the mouth, but at the forward point it is held against the mouth rather loosely so that when the back stop releases the released air almost immediately causes the front stop to release.

It happens so fast that the releases are perceived as simultaneous and the writing system even puts the front stop first. I think I can produce such an effect when only stops are involved, but I find it hard to apply this theory to the clusters in the fourth and fifth lines above (those involving kh and gh).

Answer 1

There is a PhD dissertation, The Consonant Phonotactics of Georgian (2002), by Marika Butskhrikidze where this issue was tested (see pp. 123-132 [137-146 in the PDF]). Basically, her results are:

1. Most harmonic clusters are in fact produced with two releases.

2. Based on a perception experiment, "harmonic Cc clusters behave as single consonants in terms of hit rate, but as a sequence of two unrelated consonants in terms of detection time".

Answer 2

"Georgian Harmonic Clusters: Phonetic Cues to Phonological Representation" by Ioana Chitoran (1998) does an acoustic analysis and finds two bursts associated with these clusters, as well as a longer duration (whereas doubly articulated consonants were expected to have a duration similar to a non-cluster), suggesting they are separate phonetically. They are pronounced in the order written. Occasionally, word-medially, she finds that the dorsal consonant is lenited.

She mentions Aronson's claim:

With respect to the fifth property [of harmonic consonant clusters], (6e) [simultaneity of closure and release], a number of authors (Zgenti 1965, Aronson 1982, 1991, Deprez 1988) describe the two members of a harmonic cluster as having simultaneous closures and only one release. This description is purely impressionistic.

Later:

In a pulmonic egressive airstream mechanism, a coronal and a velar closure cannot be formed simultaneously. In the case of [pk tk bg dg], any front closure, whether labial or coronal, must be released before the velar closure can be formed. The two closures cannot be simultaneous, with only one release. We are dealing then with a sequence of two stops.