Relationship between stress location and branching direction

[chris_notts]

I've just been flicking through "The Study of Stress and Word Accent", and there's a chapter that makes the claim that stress location and branching order is correlated. The initial data they have seems to suggest a weakish but statistically significant correlation, but Tokizaki argues that a number of Altaic languages are misclassified due to a distinction between initial stress and final tonal or pitch accent phenomena, and by the end of the chapter seems to be suggesting that other limitations of the data might mean the relationship is a strong one.

Anyway, the suggested correlation is that languages with stress close to the left edge tend to show left branching syntax, whereas languages with stress close to the right edge of words tend to show right branching syntax. If I understand correctly, the proposed explanation is:

1. Phrasal stress / most heavily stressed elements in phrases tend to follow the same directional pattern as word stress
2. There's a preference for modifiers to receive phrasal stress over heads (since heads are often more predictable or less contrastive in complex phrases?)
3. Therefore:
approx. initial word stress -> initial phrasal stress -> preference for modifiers first
approx. final word stress -> final phrasal stress -> preference for modifiers last

It also seems to be claimed that the effect is stronger for smaller structures, so maybe compounds and/or N+A most likely to align with stress alignment, then N+G, with clause level syntax least strongly correlated.

Tokizaki claims this hypothesis has been discussed before in other contexts, but I can't remember seeing it before. Have any of you seen this claim before, and if so are you convinced by it?

[hwhatting]

Have any of you seen this claim before, and if so are you convinced by it?

No (but that doesn't mean anything), and hard to say. I wouldn't be able to come up with any good counter-examples, but that also doesn't mean anything.

[zompist]

I've just been flicking through "The Study of Stress and Word Accent", and there's a chapter that makes the claim that stress location and branching order is correlated. The initial data they have seems to suggest a weakish but statistically significant correlation, but Tokizaki argues that a number of Altaic languages are misclassified due to a distinction between initial stress and final tonal or pitch accent phenomena, and by the end of the chapter seems to be suggesting that other limitations of the data might mean the relationship is a strong one.

How many languages is he considering, and how strict is his definition of initial and final stress?

WALS gives stress data for 502 languages, of which just 143 (28%) have stress on the first or last syllable. That strikes me as a pretty low N. If just moving the "Altaic" languages around changes the data, his N is probably even lower, and not too trustworthy.

The problem with this sort of analysis is that languages are not a random sample. All the more so well-studied languages, the ones with good grammars available. Languages affect each other, and not just within families; ordinary statistical techniques can't be used without a pile of salt.

I think you'd also have to be careful with word size. E.g. how do you classify Quechua? It has penultimate stress... but most roots are two syllables, which means the first syllable is stressed. (Words are often longer, but you'd have to do some corpus analysis or something to really know if stress is normally toward the beginning or the end of the word... a grammar can't tell you.)

Edit: I haven't read the paper, maybe it's better than it sounds. It's just that when someone finds a weak correlation, then tweaks the data to make it stronger, alarm bells should sound.

[chris_notts]

How many languages is he considering, and how strict is his definition of initial and final stress?

WALS gives stress data for 502 languages, of which just 143 (28%) have stress on the first or last syllable. That strikes me as a pretty low N. If just moving the "Altaic" languages around changes the data, his N is probably even lower, and not too trustworthy.

I don't have it with me now, but since he defined it more as initial-ish (typically first two syllables) and final-ish (penultimate or ultimate), I think the numbers were better? But even so, I also had my doubts about the way the word correlation was used but the wording later on almost suggested relatively strong causation one way or the other.

I think you'd also have to be careful with word size. E.g. how do you classify Quechua? It has penultimate stress... but most roots are two syllables, which means the first syllable is stressed. (Words are often longer, but you'd have to do some corpus analysis or something to really know if stress is normally toward the beginning or the end of the word... a grammar can't tell you.)

This was definitely another issue. I also read a PDF online by the same guy and I remember Khmer being mentioned as final due to the iambic / sesquisyllabic pattern there, but, since many of the words are short, it probably meets the criteria for both initialish and finalish stress!

[chris_notts]

But even though I had my doubts, I thought it was interesting partly because I can also see how the proposed mechanism is vaguely plausible (preferences for location of sentence or phrasal stress correlating with word level structures), and because my current conlanging project has both initial stress with caveats but is very much head initial, so if it were a real language if would break the proposed rule.

[chris_notts]

Here's a PDF by the same guy (although not the same paper) where the numbers look a bit dubious:

https://core.ac.uk/reader/230317740