Thread by @thai101

Thai has 5 tones. It's quite reasonable to think every tone should be possible on every (mono)syllable. But sound change reveals exactly why we basically never see this!

(A comically long thread in my drafts for ages, because I was procrastinating revising the manuscript.) 1/

As it turns out, the distribution of tones in a given tonal language of (South)East Asia has been directly constrained by past segmental sound change, especially syllable shape. 2/

Wait, you might say, syllable shape? Isn't that area of the world famously full of monosyllables and relatively few consonant clusters? How can syllable shape be so important? 3/

Well a lot of that area is like that *now*, yeah. That was part of the trade-off! Over the course of a few millennia the whole linguistic area underwent a massive shift from *segmental* cues upholding phonemic contrasts to *suprasegmental* cues doing the heavy lifting instead. 4/

(This monosyllable~tones connection is still a bit chicken and egg, to the honest, but a pathway between them is clear.) 5/

You may know the basics of tonogenesis by now: phonemic contrasts made on consonants "transphonologize" onto new phonetic cues. The loss of onset voicing is maybe the most famous case:

pa > pá
ba > pà

(What used to be signaled by voicing became signaled by pitch etc) 6/

The study of which lexical tones can occur with which consonants is often called (surprise!) "consonant-tone interaction"

So sometimes I cheekily call what I describe below as "consonant-tone non-interaction" 7/

For all the tones in a language to be possible on all syllables, tone has to be phonologically atomic in some way, something independent of segments/syllables and thus combinable with any of them.

This is key: tone in this region is fundamentally *not* independent of segments. It's is what I've called "desegmental", meaning comparative work lets us trace tonal contrasts back to segmental contrasts. But why does that make modern tones not independently combinable? 8/

This is where syllable shape comes in, and the concept of "checked" and "smooth" syllables. Linguists have used a *lot* of different names for these two, but the concept is simple: Checked = obstruent (usually stop) coda! Smooth = sonorant or no coda! 9/

Virtually every tonal language in East and Southeast Asia went through a stage of 4 tones (which isn't a coincidence at all; story for another thread). But those 4 tones are really 3+1 tones. Meaning 3 tones on smooth syllables, with checked syllables developing separately. 10/

Exactly what gave rise to the 3 tones on smooth syllables varies, and is sometimes mysterious. The usual suspects are either coda consonants (which were lost), or things like glottal constriction, creak, breathiness (which in some languages survive as a secondary cue). 11/

So that gets us to 3+1 tones. It's often assumed checked syllables just had their own tone. Eventually they did. But there's also reason to think checked syllables remained toneless for a long intermediate period. Why? Their syllable shape. 12/

Most famously, tonogenesis results in the *loss* of the former segmental contrast. Like how the loss of onset voicing led to tones. But unlike the 3 free syllable tones, in most languages checked syllables kept their stop codas (until recently, anyway). 13/

After other tones had lost the environments that conditioned their origins, checked syllable tones remained checked. Tone would have been a redundant cue. This redundancy is a big reason to think checked syllables remained atonal longer than smooth syllables. 14/

Then along came the East Asian Voicing Shift, actually rather late in the tonogenesis game. The loss of onset voicing led to a *split* across all existing categories: a doubling. So the 3+1 became 6+2, depending on whether the initial consonant used to be voiced or voiceless. 15/

Btw the East Asian Voicing Shift is the biggest sound change ever described, I believe, and it affects a mind boggling number of languages/families, well beyond just tonal tones. See my recent LSA poster with @RyanGehrmann for more on that: 16/

BUT WHAT ABOUT TONES AND SYLLABLE SHAPE, you scream at your screen (or screen scream, if you will). Patience lol. 17/

Languages/families vary in what happened after the East Asian Voicing Shift. Let's look at Tai languages. We call the proto-tones A, B, C (smooth) and D (checked). They *kept splitting* based on onset. So that 3+1 which had become 6+2 looks like 12+4. But wait, there's more! 18/

In many Tai languages, D (checked) also split based on vowel length. So now 12+8 = 20 cells. Hoo boy. And no, nobody thinks any language split ALL of these, just that each one mattered in some language. The ceiling of concurrent tones in Tai languages is ~7 tones. 19/

Each cell in a tone box represents a set of words that pattern together tonally in modern languages. A similar concept to J.C. Wells' 24 lexical sets for English vowels. Tai has 20 tone lexical sets (some claim more), which boil down to typically 4-7 modern tones. /20

We match up these historical lexical sets up to modern tones in a tone box (first image), using a diagnostic wordlist. Different splits and mergers means languages carve up the tonal space differently (second image). And other language families have tone boxes, too. /21

At last! We arrive back at consonant-tone "non-interaction". Here are Bangkok Thai's 5 tones (from Abramson 1962). The previous tweet shows that all 5 appear on smooth syllables, but only 3 appear on checked syllables (summarized in the second image). /22

A big question of consonant-tone interaction for Thai (the most studied Tai language by orders of magnitude) has been: why do we only see 3 of the 5 tones on checked syllables? People have posited underlying glottal features, covert low tones, duration, etc. /23

I argue this: tonogenesis is inseparable from syllable shape, and modern tones too. There has never been any stage of synchronic language where tones have been fully independent from segments. So not seeing all tones on all syllables is explained and expected. /fin!