Why does stop VOT duration vary depending on place of articulation?

Question

From the (albeit citation needed) section of the Wikipedia article on aspiration:

Spanish /p t k/, for example, have voice onset times (VOTs) of about 5, 10, and 30 milliseconds, whereas English /p t k/ have VOTs of about 60, 70, and 80 ms. Korean has been measured at 20, 25, and 50 ms for /p t k/ and 90, 95, and 125 for /pʰ tʰ kʰ/.

This is also confirmed from my anecdotal explorations in the topic.

The question I have is what causes the different stop consonants to have different VOTs. I couldn't find any good linguistic descriptions through some preliminary googling.

The two hypotheses I had were both based on the (my own) idea that voicing begins once air pressure subsides from high stop-like levels to some trigger point.

The first hypothesis was that air from a far back pressure release has to travel farther and through more obstructed corridors to reach the open air. Whereas air from a /p/ release can immediately and quickly reach the open, air from a /k/ release has to go all the way through the mouth to get outside. The latter trek is slow and results in a slower decrease of air pressure resulting in a later trigger for voicing.

The second hypothesis was that somehow, the backer points of articulation resulted in higher pressure buildups. This would be plausible if stop bursts required some given mass of extra air to release, and this mass, distributed over the oral cavity in /p/ would be spread over a larger volume and consequently have a lower pressure than in /k/. Thereafter, even if air pressure subsided at the same rate, the higher original pressure in /k/ would take longer to hit the trigger point.

Are there any good readings on the subject or anybody who can provide a more confident explanation?

Answer 1

From what I understood of the model proposed by Hanson & Stevens 2000, variation in stop VOT by place of articulation is a consequence of the cross-sectional area formed by articulators. There is a plateau in area before vowel onset during which frication occurs. This plateau duration is proportional to anterior-posterior length of the contact region (longer for velars, shorter for labials), and as a result VOT varies by place. Here is the link to the article [pdf].

Answer 2

Stevens (1999) (as well as his article with Helen Hanson) discusses the relationship between cross-sectional area of the constriction and the size of the active articulator. However, this is provided as an explanation for why more retracted consonants have longer BURSTS, not longer VOT. (VOT duration is a characteristic of the glottis, not the oral articulators.) The idea here is that since the tongue dorsum is a bigger and slower muscle, it moves more slowly away from the point of constriction than the tongue apex or the lips.

If one includes the burst duration in calculating "VOT", then sure, one will tend to have longer VOT duration for more posterior stops. However, if burst duration is distinguished from VOT, the pattern you observe still holds. With relation to voicing, the aerodynamic constraint holds. Yet, you are curious about a general pattern independent of voicing. So, the explanation must lie elsewhere.

Maddieson (1997) provides one explanation that I have found compelling in my own work:

"There is an abduction-adduction cycle of the vocal cords for voiceless stops which is longer in duration than the closure and has a constant time course, anchored to the onset of closure (p. 621). In other words, the duration of the vocal fold opening is considered to be fixed, and when the closure duration is relatively longer, the following VOT becomes relatively shorter (and vice versa)." (from Cho & Ladefoged, 1999)

This particular explanation accounts for the differences in VOT and for observed differences in trading relations in stop closure. In many languages, velar stop closure is shorter than alveolar stop closure. Stevens & Hajek (2004) found preaspiration in Sienese Italian geminates to arise from closure lenition. When there was less closure, preaspiration arose. However, the duration of voicelessness in the geminate remained more or less constant. Their finding seems to agree with Maddieson's proposal.

Answer 3

I would agree with your second hypothesis. Since P_1*V_1 = P_2*V_2 for an ideal gas, and an obstruction farther along the vocal tract creates a larger volume than another, there will be a corresponding pressure decrease.

An interesting related phenomenon is that voiced alveolar stops tend to be more retroflex than their voiceless counterparts (which have longer VOT). Ohala (1983) explains that, counterintuitively, retroflex stops have a larger oral cavity behind the constriction; thus, the gas law above predicts that they will have a shorter VOT.

Ohala, J.J. (1983). The origin of sound patterns in vocal tract constraints. In P. F. MacNeilage (Ed.), The Production of Speech (pp.189-216). New York: Springer