Tonal interactions and autosegmental analysis

[Tengado]

A while ago I did a lot of research on tone interactions, mostly because I stumbled across it and it seemed really interesting and then I decided to use it in a conlang so I needed to understand it well. Recently a few tonal conlangs have appeared, and someone suggested we need a “Tone” thread in the L&L museum, so I thought I’d get the ball rolling and summarise what I learnt and list links to other resources.

Basics of Tone

There are basically two kinds of tone systems - pitch accents, and proper tonal languages. As in a stress accent system, in a pitch accent system one syllable per word is the stressed syllable. The difference is the accent is realised as a change in pitch, not stress. For example, Japanese syllables can have either high or low tone. In most words, the first syllable is low, and all the following syllables are high, until the stressed syllable - after the stressed syllable the pitch falls again. If you know which syllable is stressed, then you know the whole pitch pattern of the word. Each word has one and only one fall in pitch [some are unstressed and so remain high pitch, and if the stress is on the first syllable, then the first syllable is obviously high].

In tonal languages, each syllable has a tone - the tone is as much an important part of the syllable as the vowel [in Chinese, the tone carries just as much information as the vowel - saying a toneless syllable is just as hard to understand as saying the syllable with the correct tone but no vowel]. There are basically two types of tone that a syllable can have - a level tone, where the whole syllable is at the same pitch [Low, High, Medium,...] and contour tones, where the pitch changes [Rising, Falling, Falling-Rising].

For ease, H M L F R will be used instead of high, mid, low, falling, rising. If there are different falling or rising tones, they can be distinguished by HM [falling from heigh to mid ], LH etc.

Tone Systems in Languages

Different languages vary a lot in how many different tones they have

From A Model of Tone Systems. Jean-Marie Hombert. [Department of Linguistics. University of California, Los Angeles]

About 30% of the world’s languages are tonal - “This figure would go up to about 50% if we would consider a sample of languages in which each language family would be represented by a umber of languages proportional to the actual number it contains. This discrepancy between the two figures comes from the fact that most of the languages spoken in areas of great linguistic diversity [e.g. Chinese and southeast Asian languages, Papuan New Guinea languages and Northwestern Bantu languages] are tonal.”

30% of tonal languages have 2 tones
30% have 3 tones
15% have 4 tones
10% have 5 tones
10% have 6 or more


Of 2 tone systems:
~90% are H, L
3 tone :
~ 65% are H, M , L
~ 20% are H, L, F
4 tones:
~ 40% are H, L, F, R
~ 30% are H, M, L, F
5 tones:
~ 60% are H, M, L, F, R
6 tones:
~ 60% are 2 level tones [usually L, M], 2 Falling tones [HM, ML] 2 rising tones [LM, MH
7, 8 tones:
~no clear patterns found
[end of material taken from that link]

Tonal Universals [“They’re more .. guidelines than what you’d call actual rules”]

There are also a few universals of tone systems.

One is that no language distinguishes more than 5 tone levels.
This means that there are a maximum of 5 level tones possible in any language, and all contour tones must start and stop on one of those 5 levels. These are often numbered 1 being the lowest, 5 the highest. Tones can thus be written as 55 - a high level tone [the first tone in Mandarin Chinese], 35 a rising tone from the mid-level to the high level [The rising tone in Mandarin] etc.

Some more universals, taken from SIL Electronic Working Papers, SILEWP 2007-007:

a. A larger number of tone levels occupy a larger pitch range than a smaller number
(~20 Hz for two tones, 50 Hz for four tones)
b. Systems in which high tones are marked are more frequent than systems in which
low tones are marked.
c. If a language has contour tones, it also has level tones.
d. A language with complex contours also has simple contours.
e. Rules raising tones are more common than rules lowering them.
f. Perseverative rules are more common than anticipatory ones.
g. Tonal polarity is more common than polarity with other features.
h. Lower vowels tend to have lower tone.
i. Low-toned vowels tend to be longer than high-toned ones.
j. Vowels with rising tone tend to be longer than vowels with falling tone.

With the exception of the (a), these universals are general tendencies rather than true universals. (a) means that if A system has two tones, then the difference in pitch between the highest and lowest will be different than the difference in pitch between the highest and lowest tone of a five tone system. So the H and L in different languages won’t always sound the same.

How do tones behave? Interactions

This is the interesting bit.

If you’re planning a tonal conlang, you need a way to represent the interactions of tones in order to plan phonotactics. There are lots of these, I would imagine, but the only one I know, and which appears to be rather good, is autosegmental theory.

Autosegmental theory treats the tones as being a separate feature that is attahced to the syllables, but that can move around independantly. Little diagrams like this are very useful:

Code: Select all

H  H   H  L H
|  |   |  | |
ma ma ror ire

[The tie lines show which syllable the tone is attached to.]
This is the Kikuyu word “they looked at them.”
The structure of the Kikuyu word is quite simple

Code: Select all

Subject       (Object)       Root              Tense
to ‘we’        mo ‘him’      ror ‘look at’ 
ma ‘they’      ma ‘them’     tom ‘send’         ire

Kikuyu has two tones, low and high - the tones of these morphemes are

Code: Select all

to ma mo ma ror tom ire 
L  H  L  H   L   H   H 

However if words of the above pattern are looked at, it turns out that not every theoretically possible sequence of these tones actually occurs:

* The subject “to” [we] always has L tone
* The subject “ma” [they] always has H tone
* The morpheme immediately after the subject [ie object or root] always has the same tone as the subject - regardless of what its own tone.
* The last syllable of the tense suffix “ire” is always high, but the first syllable varies: if the root is “ror” [look at], the first syllable of “ire” is L, if the root is “tom” [send] then the first syllable is H.

Why such apparently random restrictions?

This can be anaysed quite neatly using the little tie line diagrams mentioned above.

Consider the word “ma mo tom ire” “they sent him.” Its tones are:

Code: Select all

H  L   H   H
|  |   |   |
ma mo tom ire

Tones must be attached to a vowel in the suface realisation - that means, when the word comes to be pronounced, all the tones are attached to soemthing, they can’t just float about nowhere. Usually, they would be linked to their own vowel, but in Kikuyu [and many languages] there’s another criteria: tones want to be as close to the right edge of the word as possible [ie within the other constraints of the languages phonology]. This effect is very common [something to do with tones taking longer to articulate than the other features, so pronunciation of the tone tends to lag behind the rest of the syllable, so they drift to the end of the word].

The tones really all want to be at the right edge - on the final syllable. This is impossible in Kikuyu - this would create a HLH contour on the final syllable, which is against the languages phonotactics: level tones only. In Kikuyu, there is a condition [the well-formedness condition]: each tone can be attached to only one vowel, and vice versa [this isn’t true in all languages]. Two high tones can merge togetehr though.

So the maximum possible right drift of tones is

Code: Select all

H   L  H  H
 \  \   \ |
ma mo tom ire

All the tones are as far right as they can go without creating contour tones.

The tone pattern of the word has changed from
HLHH to
0HLH

This has left the first syllable [the subject] toneless. But remember Kikuyu’s well formedness condition: each vowel must be attached to one and only one tone; and each tone must be attached to one and only one vowel. This requires that
1. No tone can be linked to more than one vowel
2. No vowel can be toneless.

One of these must be violated - in Kikuyu, the second takes priority, and so the word becomes

Code: Select all

 H  L  H    H
 | \  \  \  |
ma mo tom ire

the initial high tone gets attached to two vowels, rather than leave a toneless vowel. In otehr languages the priority is dfferent, and the syllable would remain toneless.

Thus the word has changed from
HLHH to
HHLH.

All the other observations of the tone distribution above can be explained in the same way: by the tones trying to drift to the right as far as they can.

More complicated systems - contours
This same analysis can be used for contour tones too. Contour tones are analysed as sequences of H and L tones on one vowel:

Code: Select all

 H  L  F  R
 H  L  HL LH
 |  |  \| |/
 ma ma ma ma

An example from Margi [an African language, not a woman]. Margi has three tones, H, L, R. The definite suffix is “-ari,” and has H tone. If the suffix is added to a noun ending in a high vowel, then the high vowel becomes a glide:
imi+ari = imyari
ku+ari = kwari

* If the noun ends in a consonant, there is no tone change - unlike Kikuyu, no tones drift.
* If the noun ends in a high vowel with H tone, there is also no tone change
* if the noun ends in a high vowel with L tone, “ari” changes to rising tone

Why?
Consider ti + ari = tyari. The inital tones are

Code: Select all

L  H H
|  | |
ti a ri

but then the /i/ changes into a glide, tyari, the word goes from three syllables to two

Code: Select all

 L H  H
  \|  |
 tya ri

The glide cannot carry a tone anymore. The tone wants to be attached to a vowel [well-formedness condition] and so it attaches to the next vowel, the a, causing a rising tone.

These principles can be used to decide on tone sandhi in a conlang. Decide on the restrictions on tone structure - only one tone per vowel, no toneless vowels, etc. One vowel can only take one tone, so contour tones must be on diphthongs or long vowels only. Whatever seems logical. Then you can apply the rightward drift to get tone interactions in the final words.

This way you can devise systems where the morphemes only have level tones, but the finished words have more complicated contur tones [due to drifitng like in the Margi example].

More interesting bits - Floating Tones and Wordless Tones

Because of this drfiting tendency of tones, some languages have tones that float off on to the next word, and even tones that do not belong to any word. They are a morpheme in their own right:

In San Miguel El Grande Mixtec of Mexico, some nouns
have a suffixal High tone which is underlyingly unassociated. This tone shows its
effects on the following word, if one is present.
• The same pattern occurs with Lango of Uganda.
• In Kalam Kohistani of Pakistan, some words cause a Low tone to occur on the
following word
• In Usarufa of Papua New Guinea, tones set afloat by vowel
deletion in noun phrases manifest themselves on the following word.

An example of a suffixal high tone: In Konni [another African language that sounds like it’s a woman]. In possessive/genitive constructions - such as “my hat, the cat’s tail” the word order is the same as English “possessor + possessed.” When the possessor is a third person noun or pronoun, the possessed object always has a H tone on it’s first syllable.

This can be analysed as their being a H tone suffixed to the end of the possessive form of the third person noun:

Code: Select all

L H        L H H
| |        | | |
u          bua
his/her   child’s    

This floating High tone wants to be attached to a vowel [well-formedness condition again], and so it attaches itself to the right - to the first syllable of the following word, the possessed noun.

[Tengado]

More complex systems
One analysis of more complex systems, such as Chinese, uses two features: tone and register. Each of the realised surface tones is composed of two parts, Register [Upper or Lower] and Tone [High or Low]:

Code: Select all

Register Tone  Realisation
Upper    High  Extra-high
         Low  High
Lower    High  Mid
         Low   Low

The analysis is the same, except that Tone and Regsiter can drift and reattach independantly of each other.

For example, In Ewe [a language, not a sheep], mid vowels become extra-high when surrounded by high vowels. That sounds quite bizarre - why don’t they just become high? Why become even higher than the surrounding vowels?

Consider it in terms of Register and Tone. H, L mean Tone; Up, Lo mean register. The High-Mid-High sequence is then Up+L - Lo+H - Up+L. But the two features are independant, so we can draw a diagram like the earlier ones for the Register alone: Up-Lo-Up

Code: Select all

Up Lo Up
|  |  |
a  a  a

The rule is assimilation: the Lower register simplifies to being the same as the surrounding vowels.

Code: Select all

Up Up Up
|  |  |
a  a  a

So the sequence of vowels is now Up+L - Up+H - Up+L: High - Extra high - High.

Neat, eh? A logical and systematic way to get some unexpected effects. Just have a rule that says Register will assimilate, but Tone won’t change. Or vice-versa.

Another example, from Chaozhou Chinese [one of my friends speaks Chaozhou. It sounds like Cantonese gone wrong ]. Chaozhou has three level tones, H, M, L. Strange things seem to happen in the fust syllables of compound nouns:
With the word “hue,” goods
hue-lung, cargo ship
HM-H
hue-ts’ng, warehouse
ML-L
- hue appears with two tones: high-falling and low falling.
With the word “hue,” fire
hue-ba, torch
MH-HM
hue-tsi, rocket
LM-LM
- hue appears as high-rising and low-rising.

Seems strange, unless you look at it this way:
* Chaozhou only has three height distinctions, so

Code: Select all

Register Tone  Realisation
Upper    High  High
         Low   Mid
Lower    High  Mid
         Low   Low

Mid-level tones can either be Up+L or Lo+H.
* Register spreads from the second noun to the first noun.
* The tones on “goods” are HL, on “fire” LH

If the register spreads back, that combines with the Tone to give the surface forms listed.

Tone can also Spread separately of Register [Zhenzhai dialect of Chinese].

This model doesn’t explain why sometimes Tone and Register spread together. There are more complex models to explain those situations [which I do not understand], but for situations where they are independant, this model is fine.

For a conlanger’s perspective I would suggest that this is more than enough - the independant spreading of Tone and Register can give interesting effects in the surface forms.

Theoretically I guess you could make any rule you like: Tone spreads in one way, Register spreads the other. One of them assimilates, one spreads.

That was a very long post! I hope it makes sense. If anything needs clarification let me know. and I’ll do my best.

[Noriega]

Very good summary! Thanks for taking the time.

Because of this drfiting tendency of tones, some languages have tones that float off on to the next word, and even tones that do not belong to any word. They are a morpheme in their own right:

In San Miguel El Grande Mixtec of Mexico, some nouns have a suffixal High tone which is underlyingly unassociated. This tone shows its effects on the following word, if one is present.

I was wondering about that, if there are no words following, then the utterance ends, and some tones have nowhere to float. Doesn’t that violate the “well-formedness condition”?

Something I’d love to see would be a multimedia presentation of an utterance in, let’s say, Yoruba. It would play the speech, show it in IPA or some writing, and trace the F0 fluctuations graphically. As soon as you actually hear the tones, I feel that it becomes much easier to understand.

[Noriega]

I did a quick analysis of a sentence in Yoruba (sentence #80 here http://archive.phonetics.ucla.edu/Langu ... 72_01.html)

The UCLA page lists it as [mo fɛ́ lɔ ɽí ɔmɔ na].

The sentence can be heard as an mp3 (112 kbps mono, 20 kb) here: http://83.233.156.71/linguistics/yoruba.mp3

Here’s a Praat analysis, with the blue line showing the pitch of the utterance. You can see two important tops, marked as H. These are high tones on the syllables "fe" and "ri".

[sangi39]

Well P'ar dinstinguishes between 3 level tones, 5, 3 and 1 which then interact in long vowels and diphthongs to bring about another 9 contour tones; 45, 35, 13, 53, 21, 31, 55, 33 and 11. There's also a neutral tone which may only occur on the first vowel of a word and takes on the tone of the previous syllable, so after 45 a single vowel neutral tone syllable will have tone 5, etc.

I've been thinking about tone sandhi and all I've come up with is that after a syllable ending in tone 1, a syllable beginning in tone 5 will have this lowered to tone 4, and after a syllable ending in tone 5 a syllable beginning in tone 1 will rise to tone 2. Long level tones 55 and 11 will thus become 44 and 22 to distinguish them from 45 and 21. Neutral tones will therefore take either 1, 2, 3, 4 or 5 depending on how the previous syllable was affected:

45+11+53+N5+13
=
45+22+53+35+23

[Tengado]

n San Miguel El Grande Mixtec of Mexico, some nouns have a suffixal High tone which is underlyingly unassociated. This tone shows its effects on the following word, if one is present.

I was wondering about that, if there are no words following, then the utterance ends, and some tones have nowhere to float. Doesn’t that violate the “well-formedness condition”?[/quote]
I dpn't know. In the exmples I've seen, he floating tones occur only in words that always have a following word [eg the possessives]. But also, it should be remembered that
a) there is not only one "well formedness condition" different lanagues have thir own version of what is well formed, and as in Kikuyu the example above, often one condition must be violated - which one takes precedence varies from language to language. Maybe the well-formedness condition can be violated and the floating can just be lost in some languages. Maybe it must be attached and so reverses itself back onto the word before in otehr languages. I guess both are possible.

Well P'ar dinstinguishes between 3 level tones, 5, 3 and 1 which then interact in long vowels and diphthongs to bring about another 9 contour tones; 45, 35, 13, 53, 21, 31, 55, 33 and 11. There's also a neutral tone which may only occur on the first vowel of a word and takes on the tone of the previous syllable, so after 45 a single vowel neutral tone syllable will have tone 5, etc.

Have you analyisde this with autosegmental analysis? If the tone drifts from the last syllable of one word onto the toneless first sylable of the next, does that change the tone on the last syllable of the previous word If not, why not?

Also, you say the 1, 3, 5 level tones interact to make the other tones. How do they interact to make a 21 tone? Where did the 2 come from?

Why can the neutral tone only occur on the first syllable? That sounds pefectly in accordance with the rightward drifting tendency, but then to keep the phonotactics in harmony with that principal the final syllable of the word should also undergo changes.

I've been thinking about tone sandhi and all I've come up with is that after a syllable ending in tone 1, a syllable beginning in tone 5 will have this lowered to tone 4, and after a syllable ending in tone 5 a syllable beginning in tone 1 will rise to tone 2. Long level tones 55 and 11 will thus become 44 and 22 to distinguish them from 45 and 21. Neutral tones will therefore take either 1, 2, 3, 4 or 5 depending on how the previous syllable was affected:

I would recommend analysing this in the way I described above - You have a L, M, H system, where all the contour tones [except that random 21] begin or end on one of those three]. If I were you [and I'm not, so you don't have to do this] I'd analyse your tones in terms of the Register-Tone system I mentioned above, and then devise rules about register or tone spreading to account for the surface tone changes. The mid level tones could be Upper+Low or Lower+High, so apparently similar words could do different things dependning on which of the these two mid-tones it is.

[sangi39]

Well P'ar dinstinguishes between 3 level tones, 5, 3 and 1 which then interact in long vowels and diphthongs to bring about another 9 contour tones; 45, 35, 13, 53, 21, 31, 55, 33 and 11. There's also a neutral tone which may only occur on the first vowel of a word and takes on the tone of the previous syllable, so after 45 a single vowel neutral tone syllable will have tone 5, etc.

Have you analyisde this with autosegmental analysis? If the tone drifts from the last syllable of one word onto the toneless first sylable of the next, does that change the tone on the last syllable of the previous word If not, why not?

Also, you say the 1, 3, 5 level tones interact to make the other tones. How do they interact to make a 21 tone? Where did the 2 come from?

Why can the neutral tone only occur on the first syllable? That sounds pefectly in accordance with the rightward drifting tendency, but then to keep the phonotactics in harmony with that principal the final syllable of the word should also undergo changes.

I've been thinking about tone sandhi and all I've come up with is that after a syllable ending in tone 1, a syllable beginning in tone 5 will have this lowered to tone 4, and after a syllable ending in tone 5 a syllable beginning in tone 1 will rise to tone 2. Long level tones 55 and 11 will thus become 44 and 22 to distinguish them from 45 and 21. Neutral tones will therefore take either 1, 2, 3, 4 or 5 depending on how the previous syllable was affected:

I would recommend analysing this in the way I described above - You have a L, M, H system, where all the contour tones [except that random 21] begin or end on one of those three]. If I were you [and I'm not, so you don't have to do this] I'd analyse your tones in terms of the Register-Tone system I mentioned above, and then devise rules about register or tone spreading to account for the surface tone changes. The mid level tones could be Upper+Low or Lower+High, so apparently similar words could do different things dependning on which of the these two mid-tones it is.

Tbh, if you're going to ask about 21 you should probably be asking about 45 as well. It was this that made me think that maybe I should change the tones to 4, 3 and 2 but then the low tone would only be affected by the high tone it was 45 and the high tone would only be affected by the low tone if it was 21. I guess 45 and 21 in a 1-3-5 system just made more sense to me.

The neutral tone can only occur on the first vowel because the second one determines whether it is level, rising or falling. If the neutral tone is followed by a 1 tone vowel it will be either low or falling, a 3 tone vowel will cause it to be level, rising or falling and a 5 tone vowel will cause it to be high or rising. A second neutral vowel would essentially just cause the vowel to be level, which is not something I was aiming for.

As far as I've gotten so far, the tone is only affected in one direction, from start to finish, so a changed tone will not affect the one before it.

[Wycoval]

Thanks for typing all that up, Tengado. That's bound to be very helpful to those who are interested in doing tonal conlangs or who are learning tonal natlangs.

I'll pitch in with a link to a very good reference source for tone and autosegmental analysis.

http://www.vanoostendorp.nl/fonologie.php

[Qwynegold]

h. Lower vowels tend to have lower tone.

What? I was told by ZBB:ers that there is no correspondence between tone and vowel quality, when I said that I thought that sounds more higher pitched than the other vowels.

i. Low-toned vowels tend to be longer than high-toned ones.
j. Vowels with rising tone tend to be longer than vowels with falling tone.

Ah, I have a conlangs where vowels with rising tone are long and falling tone half-long, which now is supported by rule I. But then I had also stated that the rising tone can also be high level and falling tone low level, depending on the speaker, which goes againt rule J. Oh uh, but those are only guidelines and not universals so I guess it doesn't matter.

[Tengado]

Tbh, if you're going to ask about 21 you should probably be asking about 45 as well.

Well, yes. I just didn't spot the 4.

It was this that made me think that maybe I should change the tones to 4, 3 and 2 but then the low tone would only be affected by the high tone it was 45 and the high tone would only be affected by the low tone if it was 21. I guess 45 and 21 in a 1-3-5 system just made more sense to me.

The way you said the 1 3 5 tones interact to produce the other tones. How exactly do they interact to produce 45 and 21?

The neutral tone can only occur on the first vowel because the second one determines whether it is level, rising or falling.

...then you say ...

As far as I've gotten so far, the tone is only affected in one direction, from start to finish, so a changed tone will not affect the one before it.

Don't these contradict each other? The second vowel's tone affects the first one, this is effect from finish to start.
My question is, why is the first syllable toneless in the first place? Why does it have no tone? Where did its tone go? Why does the tone of the second syllable determine the tone of the first's?

Neither of these things are unrealistic - in fact, both happen in the Kikuyu examples. But in Kikuyu they are the result of two systemaitc processes: the tendency of tones to move to the right, and the fact that every vowel must have a tone. The first causes ALL the tones to move right, leaving the first syllable toneless, and the second causes the second syllable's new tone to attach itself to the first syllable as well.

In your case, it seems pretty random that the first syllable is toneless.

If the neutral tone is followed by a 1 tone vowel it will be either low or falling, a 3 tone vowel will cause it to be level, rising or falling and a 5 tone vowel will cause it to be high or rising. A second neutral vowel would essentially just cause the vowel to be level, which is not something I was aiming for.

Do you have explanations for why these changes happen? You list several possibilities for each neutral + level combination - under what situations do the different possibilities happen?

I'm not sure your system sounds too realistic.

[Tengado]

Thanks for the link Wycoval!

h. Lower vowels tend to have lower tone.

What? I was told by ZBB:ers that there is no correspondence between tone and vowel quality, when I said that I thought that sounds more higher pitched than the other vowels.

You were told wrong then lol

i. Low-toned vowels tend to be longer than high-toned ones.
j. Vowels with rising tone tend to be longer than vowels with falling tone.

Ah, I have a conlangs where vowels with rising tone are long and falling tone half-long, which now is supported by rule I.

Rule J, you mean, not I.

But then I had also stated that the rising tone can also be high level and falling tone low level, depending on the speaker, which goes againt rule J.

I think you not the letter wrong - what does Rule J [Rising tone vowels tend to be longer than falling tone vowels] have to do with this? If the rising tone was longer than the falling one, the length distinction could remain after they become level tones. Or the length could change too [unless the length is distinctive]. Rule J is an observation that it physically takes longer to pronounce a rising tone than a falling tone.

In Cantonese, the High falling tone 53 merged with the high level tone 55 in most people's speech, and the low falling tone 21 has changed to 11 in most people's speech.

Oh uh, but those are only guidelines and not universals so I guess it doesn't matter.

They're observations of what usually happens - rising tones tend to be longer because they physically take time to say, whereas it's possible to say a very short and choppy falling tone.

[sangi39]

The way you said the 1 3 5 tones interact to produce the other tones. How exactly do they interact to produce 45 and 21?

I guess using the word "interact" probably wasn't the best word to use. Contour tones only occur on long vowels and diphtongs rather than the tone of the two vowels interact with each other to form a contoured tone, it is just represented in the writing system I devised for it as so, so I might have been confused by my own work.

My question is, why is the first syllable toneless in the first place? Why does it have no tone? Where did its tone go? Why does the tone of the second syllable determine the tone of the first's?

Neither of these things are unrealistic - in fact, both happen in the Kikuyu examples. But in Kikuyu they are the result of two systemaitc processes: the tendency of tones to move to the right, and the fact that every vowel must have a tone. The first causes ALL the tones to move right, leaving the first syllable toneless, and the second causes the second syllable's new tone to attach itself to the first syllable as well.

In your case, it seems pretty random that the first syllable is toneless.

It seems here there was a little confusion. Each word has only one sylable which can be made of either a short vowel, a long vowel or a diphthong. Only the long vowels and diphthongs can have contoured vowels. A neutral tone can be level, rising or falling with the original tone being determined by the tone of the monosyllabic word before. Aslo, the neutral tone is more prevalent in clitics and postpositions rather than nouns, verbs and adjectives so I guess it could have arisen by means of tone sandhi in an earlier stage of the language that caused the tone of the monosyllabic word to be lost and instead based on the previous tone instead while the level, rising or falling nature of the tone remained.

[Tengado]

So you actually have a 5 level tone system, but the only level tones that appear are 55, 33, 11, yes? That's ok. Mandarin has a 5 level system and only one level tone - 55. The others are 35, 213, 51.

It seems here there was a little confusion. Each word has only one sylable which can be made of either a short vowel, a long vowel or a diphthong. Only the long vowels and diphthongs can have contoured vowels. A neutral tone can be level, rising or falling with the original tone being determined by the tone of the monosyllabic word before

What do you mean by a neutral tone? Neutral tone usually means no tone.
Here you say it can have a tone. Do you mean a word with no tone of its own?

Also, the neutral tone is more prevalent in clitics and postpositions rather than nouns, verbs and adjectives so I guess it could have arisen by means of tone sandhi in an earlier stage of the language that caused the tone of the monosyllabic word to be lost and instead based on the previous tone instead while the level, rising or falling nature of the tone remained.

So the shape of the tone [level, rising, and falling] stays the same, but the starting point of it must be fixed by the previous word? Strange. There's something similar called downdrift or updrift - the actual pitch of a series of H and L tones isn't constant - If a sentence has HLH, the second H will be at a lower pitch than the first. In a long sentence, the last H might even be lower than the first L, but is still perceived as a H. Still, your system seems a little odd.

[sangi39]

So you actually have a 5 level tone system, but the only level tones that appear are 55, 33, 11, yes? That's ok. Mandarin has a 5 level system and only one level tone - 55. The others are 35, 213, 51.

That makes it sound a lot less messy :p

What do you mean by a neutral tone? Neutral tone usually means no tone.
Here you say it can have a tone. Do you mean a word with no tone of its own?

I guess "dependent tone" might be a better way of describing it.

So the shape of the tone [level, rising, and falling] stays the same, but the starting point of it must be fixed by the previous word? Strange. There's something similar called downdrift or updrift - the actual pitch of a series of H and L tones isn't constant - If a sentence has HLH, the second H will be at a lower pitch than the first. In a long sentence, the last H might even be lower than the first L, but is still perceived as a H. Still, your system seems a little odd.

Only the shape of the "dependent tone" has no fixed start point. Other contour tones, the high, low and mid rising and high, low and mid falling contour tones have a more or less fixed start point. So I guess there are two basic kinds of tone, "non-dependent tones" and "dependent tones".

[Tengado]

As for your tone change rules, I would heartily recommend trying the Register-Tone method. You could do it this way
5 Up+H
4 Up+M
3 Up+L, Lo+H
2 Lo+M
1 Lo+L

and then have rules about spread and assimilation of register or tone. If you make sensibe sounding rules, you should end up with naturalistic and plausible sound changes for any situation.

One example you gave was

after a syllable ending in tone 1, a syllable beginning in tone 5 will have this lowered to tone 4

Here the sequence is [Lo+L] + [Up+H].
The rule could be spread of register to the right, giving you a change of 5 3 after 1.
Or the rule could be rightward spread of tone,
giving 5 4 after 1.

5 4 was the rule you wanted, so that means tone is spreading, not register. So check that all the other tone interactions are similar rightward spreading of the tone feature. This way you’ll get consistent natural tone sandhi.

Only the shape of the "dependent tone" has no fixed start point. Other contour tones, the high, low and mid rising and high, low and mid falling contour tones have a more or less fixed start point. So I guess there are two basic kinds of tone, "non-dependent tones" and "dependent tones".

I'm not sure that having a completely dependent tone with only shape and no pitch level whatsoever is plausible.
How would you explain that? Think about the kind of analysis I did above - what feature is attached to the vowels?
If it's a diphthong with a rising tone, for example, what feature is attached to the two vowels that is higher on the second vowel than the first, but has no actual pitch value?

[sangi39]

As for your tone change rules, I would heartily recommend trying the Register-Tone method. You could do it this way
5 Up+H
4 Up+M
3 Up+L, Lo+H
2 Lo+M
1 Lo+L

and then have rules about spread and assimilation of register or tone. If you make sensibe sounding rules, you should end up with naturalistic and plausible sound changes for any situation.

One example you gave was

after a syllable ending in tone 1, a syllable beginning in tone 5 will have this lowered to tone 4

Here the sequence is [Lo+L] + [Up+H].
The rule could be spread of register to the right, giving you a change of 5 3 after 1.
Or the rule could be rightward spread of tone,
giving 5 4 after 1.

5 4 was the rule you wanted, so that means tone is spreading, not register. So check that all the other tone interactions are similar rightward spreading of the tone feature. This way you’ll get consistent natural tone sandhi.

Only the shape of the "dependent tone" has no fixed start point. Other contour tones, the high, low and mid rising and high, low and mid falling contour tones have a more or less fixed start point. So I guess there are two basic kinds of tone, "non-dependent tones" and "dependent tones".

I'm not sure that having a completely dependent tone with only shape and no pitch level whatsoever is plausible.
How would you explain that? Think about the kind of analysis I did above - what feature is attached to the vowels?
If it's a diphthong with a rising tone, for example, what feature is attached to the two vowels that is higher on the second vowel than the first, but has no actual pitch value?

Hmmm, you seem to be right about the dependent tone. It seems too... out of place considering the other tones and the very nature of the dependent tone itself. I may well get rid of it actually. I think it would lower the syllable count down to about 6,000 but considering Mandarin is working fine on about 1,200 so it's not like it's a huge loss considering I would have reached the point where some words would be compounds.

So it looks like I'll be working with three level tones and six contour tones from now on. Now just to work on the sandhi.

[sangi39]

Would these be plausible:

45+5>45+5, 45+3>45+3, 45+1>45+2
35+5>35+5, 35+3>35+3, 35+1>45+2
13+5>13+5, 13+3>13+3, 13+1>13+3
53+5>53+5, 53+3>53+3, 53+1>53+3
21+5>21+4, 21+3>21+3, 21+1>21+1
31+5>31+4, 31+3>31+3, 31+1>31+1
55+5>55+5, 55+3>55+3, 55+1>55+4
33+5>33+5, 33+3>33+3, 33+1>33+3
11+5>11+4, 11+3>11+3, 11+1>11+1

[Tengado]

How would getting rid of the dependant tone lower the syllable count? The dependant tones would still have been realised as one of the other tones, would they not? So surely your syllable count is still the same. Just that now each syllable has an assigned tone, instead of the not-yet assigned tones of the dependant tones.

If you want something similar to dependant tones, you could employ Kikuyu-esque right-drift of tones. This is a very common feature of tonal languages. The reason is articulatory: you can move your mouth to pronounce the next consonant/vowel much faster than you can change the pitch of your vocal chords, so the tone will tend to change later - on the next syllable.

Instead of having syllables with an inherent tone pattern, you could have rules of rightward spread that alter their tone. Perhaps on all syllables, perhaps only some, and in certain situations. Maybe you only allow pitch rising changes: so if a syllable is followed by a syllable of a lower tone, the lower tone rises.

The tone spread could cause lengthening of vowels - a tone spreads right onto a single vowel. Single vowels can only have one tone, so if the two tones aren't the same the vowel lengthens to accomodate the contour tone.

Look at the Kikuyu and Margi examples, and the floating tone thing. There are lot of possibilities.

[sangi39]

How would getting rid of the dependant tone lower the syllable count? The dependant tones would still have been realised as one of the other tones, would they not? So surely your syllable count is still the same. Just that now each syllable has an assigned tone, instead of the not-yet assigned tones of the dependant tones.

You make a good point. What I meant was now every syllable has an assigned tone so I can't use the depedent tone syllables for words anymore. Really I was aiming only to use the dependent tones in clitics in a kind of "tone harmony" if you will, but I still wanted each actual word to have an assigned tone, not a dependent one.

[Tengado]

Would these be plausible:

45+5>45+5, 45+3>45+3, 45+1>45+2
35+5>35+5, 35+3>35+3, 35+1>45+2
13+5>13+5, 13+3>13+3, 13+1>13+3
53+5>53+5, 53+3>53+3, 53+1>53+3
21+5>21+4, 21+3>21+3, 21+1>21+1
31+5>31+4, 31+3>31+3, 31+1>31+1
55+5>55+5, 55+3>55+3, 55+1>55+4
33+5>33+5, 33+3>33+3, 33+1>33+3
11+5>11+4, 11+3>11+3, 11+1>11+1

Erk! This table is unnecesarily complicated. The 5 in 13+5 and 33+5 should change exactly the same way. What difference does the beginning of the contour tone make if the contour tone doesn't change? so just list them as 3+5 5 etc

To answer your plausibility question USE AUTOSEGMENTAL ANALYSIS! What is the underlying rule that makes these changes happen? Kikuyu has three: all tones try to drift right, only one tone on each vowel, no toneless vowels. If you apply these rules to every situation, you get all the observed tonal sandhi.

One thing I wonder about is why the contour tone never changes. Is this explained by your rule?

What you need to do is to make a set of rules, and then apply them to all the combinations above.

5 Up+H
4 Up+M
3 Up+L, Lo+H
2 Lo+M
1 Lo+L

X1+5 4 can be explained by assimilation of Tone, regardless of register: the tone [L, M, H] of the second changes to be the same as that of the end of teh preceding contour tone. Apply this same rule to all of those combinations, and see if you get the tone changes in your table above. If you do, then it is plausible.

Do these sound changes only apply to single vowels which follow a long vowel/diphthong?

[Tengado]

How would getting rid of the dependant tone lower the syllable count? The dependant tones would still have been realised as one of the other tones, would they not? So surely your syllable count is still the same. Just that now each syllable has an assigned tone, instead of the not-yet assigned tones of the dependant tones.

You make a good point. What I meant was now every syllable has an assigned tone so I can't use the depedent tone syllables for words anymore. Really I was aiming only to use the dependent tones in clitics in a kind of "tone harmony" if you will, but I still wanted each actual word to have an assigned tone, not a dependent one.

A sort of "tone harmony" is fine - it would just be assimilation of Tone/Register from the main word to the clitic. The clitic could maybe have a value for Register, but no value for Tone, and take its Tone value from the preceding word.

So clitics would be Upper or Lower. Upper clitics could then appear as 5, 4, 3, Lower clitics as 3, 2, 1. 4 and 2 are not legal monosyllables though, so you'd need a rule to deal with that. Maybe they raise to 5 and 3 respectively.

[Tengado]

If you had Upper and Lower clitics, the results would be this:

Code: Select all

                Clitic realisation
Word Tone value Upper Lower
X5     H        5          3
X4     M        4          2
X3     L,H      3,5      3,1
X2     M        4          2
X1     L        3          11

If instead the clitics had Tone values but no register, you'd get

Code: Select all

                    Clitic realisation
Word Register value H   M   L
X5     Up           5   4   3         
X4     Up           5   4   3      
X3     Up, Lo       5,3 4,2 3,1     
X2     Lo           3   2   1       
X1     Lo           3   2   1      

The row for X3 has two values because 3 could be Up+L or Lo+H.


You'd then need a rule to get rid of the 4 and 2.

EDIT: I just notcied that in the table you posted above, you have 4 and 2 as possible tones for the clitic. No problem then. Choose register spreading or tone spreading and go for it!