Neogrammarian sound change and lexical diffusion side by side

[Travis B.]

I have heard things saying that the Neogrammarians were Right when it comes to sound change, while I have also heard things indicating that sound change really operates in terms of lexical diffusion, and the Neogrammarians hence were Wrong. However, from observing sound change in the language I am most familiar with, English, I see both Neogrammarian-style sound change and lexical diffusion side by side. In the former case, the NCVS seems to operate Neogrammarian-style; there seems to be no ties to particular lexical items in its operation. In the latter case, flap elision seems to operate by lexical diffusion; it affects some words and not others, and just which words it affects varies by stress and register, in an unequal fashion. So why would one sound change behave in a lock-step, non-lexicalized fashion across the entire lexicon while another would in some (usually high) registers affect only a limited number of words while in other (usually lower) registers affect many (but not necessarily all) words?

[bradrn]

My understanding is that all sound changes spread by lexical diffusion: some eventually spread across the whole lexicon, becoming regular in the Neogrammarian sense, while others remain restricted. As for why some sound changes don’t spread completely, I’m not sure anyone can know that.

[zompist]

My understanding is that all sound changes spread by lexical diffusion: some eventually spread across the whole lexicon, becoming regular in the Neogrammarian sense, while others remain restricted. As for why some sound changes don’t spread completely, I’m not sure anyone can know that.

That's my understanding too. I thought this came up recently? William Labov discusses this quite a bit, and he's the pioneer on studying ongoing sound changes. I'm pretty sure the NCVS does show lexical diffusion-- I remember him saying that nasal environments were the first affected.

Naturally, the 19th century Neogrammarians only looked at past sound changes, which were complete enough to seem quite regular.

[Richard W]

Does universal lexical diffusion disprove the validity of the phoneme?

Now, for a merger, lexical diffusion is consistent with phonemes, because it's just the replacement of one string of phonemes by another, and that is consistent with the claim that completion of the merger happens very rapidly. I can rationalise it as people ceasing to learn the distinction.

[bradrn]

Does universal lexical diffusion disprove the validity of the phoneme?

Why should it?

[Skookum]

Posting this here because it seems to me that grammatically/morphologically conditioned sound changes can be considered a subset of lexically diffused sound changes, i.e., ones that never spread beyond a particular morphological context. Now, explanations of this type seem to have fallen out of favour in historical linguistics, with people resorting to things like analogy as an explanation.

However, this explanation doesn't seem to work when there is no conditioning environment and the sound change targets morphemes of a particular lexical class. For example, in Secwepemctsín (Shuswap) language, fricatives in determiners become voiced, which can be seen by comparing it to the closely related Thompson language: *ɬ(ə) *x(ə) > Sh lə ɣə, Th ɬ x(ə). This sound change occurs no where else in the language, and these voiced forms do not alternate with voiceless ones, so analogy can't be an explanation here. Does anyone know any other examples of sound changes like this that affect only a particular lexical class?

[Emily]

the english stress shift affecting noun/verb pairs (PROject vs proJECT and so on) always seemed like a pretty clear example to me, though i don't remember which direction it went

[zompist]

It seems suspicious that in English, word-initial /ð/ is limited to function words, leaving /θ/ for everything else.

[Skookum]

It seems suspicious that in English, word-initial /ð/ is limited to function words, leaving /θ/ for everything else.

I was going to mention English as a shockingly similar parallel, but my googling surprisingly didn't turn up anything on the historical process involved. It does seem that the English process is more broad, affecting function words generally (e.g. the adverb then), while in Secwepemctsín it only occurs in the determiner system.

[Emily]

my understanding on that example is that it ultimately just comes down to voicing. OE fricatives were allophonically voiced rather than phonemically, and at the beginnings of words they were voiceless. notice that we have very few native ModE words that start with /v/ ("vat" and "vixen" are the only ones that come to mind, both of which are generally regarded as dialect variations), and probably none that start with /z/. my impression is that with the "th" function words, they tended to be unstressed, and pronounced as part of a larger unit with surrounding words, leading them to be voiced intervocalically, even though it was technically across word boundaries. so the sound change was a regular, neogrammarian-friendly conditioned change. (in danish/norwegian/swedish and in frisian, basically the same conditions seem to have led to a contrast between /t/ and /d/ in descendants of PG /θ/, although in frisian the word "thing" is also affected: ding vs NG ting) i can't find or think of a similar neogrammarian-compliant (call it "brugmann-complete" lol) explanation for the EME verb stress shift though

[bradrn]

so the sound change was a regular, neogrammarian-friendly conditioned change.

Not quite: words like ‘the’ and ‘that’ have ended up with a voiced /ð/ even at the beginning of a sentence. That’s easily explained as analogy, but that means it can’t be completely Neogrammarian.

[zompist]

my impression is that with the "th" function words, they tended to be unstressed, and pronounced as part of a larger unit with surrounding words, leading them to be voiced intervocalically, even though it was technically across word boundaries. so the sound change was a regular, neogrammarian-friendly conditioned change.

That may be the case. Certainly voicing in OE is very low-yield. Still, there's a strong ð/θ contrast medially and finally.

The problem with this explanation, though, is that most of these words appear at the beginning of a phrase or sentence, which is not really where we'd expect sandhi/liaison effects.

[bradrn]

Incidentally, Wikipedia agrees with the sandhi explanation (insofar as it can be considered anything approaching a reliable source).

[Ketsuban]

There have been sound changes in English which remove /ð/, e.g. after /r/ (burden, murder). Words which have retained it non-initially do tend to be function-y rather than content-y (either, further, other, whether) except for weather, which went in the opposite direction (OE weder).

Final /ð/ tends to form verbs, some of which have a noun counterpart with /þ/ (breathe, sheathe, teethe, wreathe, bequeath) and some of which don't (seethe, writhe).

[Travis B.]

There have been sound changes in English which remove /ð/, e.g. after /r/ (burden, murder). Words which have retained it non-initially do tend to be function-y rather than content-y (either, further, other, whether) except for weather, which went in the opposite direction (OE weder).

Final /ð/ tends to form verbs, some of which have a noun counterpart with /þ/ (breathe, sheathe, teethe, wreathe, bequeath) and some of which don't (seethe, writhe).

And don't forget words such as father and mother, which had /d/ > /ð/ before /ər/.

[Richard W]

Does universal lexical diffusion disprove the validity of the phoneme?

Why should it?

Because a conditioned change that wasn't a merger would require a phoneme to split into the unshifted and shifted sounds.

[zompist]

Does universal lexical diffusion disprove the validity of the phoneme?

Why should it?

Because a conditioned change that wasn't a merger would require a phoneme to split into the unshifted and shifted sounds.

If it's conditioned, you're simply describing the formation of an allophone, which doesn't in any way conflict with the phoneme.

A more serious problem is if a change is, as Labov has reported, favored in a small number of words, without a condition which specifies what those words are.

But I'd say trying to use lexical diffusion to attack the regularity of sound changes is like using the motion of electrons to attack the notion of heat. If you look at things very close up, they look chaotic, but that very chaos creates the larger patterns you see if you step back. A regular sound change is what you get when the lexical diffusion process has completed.*

*IIRC it's normal for a few words to end up unchanged. When we talk about regularity of sound change, it's important to know what we're contrasting it with. Against the earlier etymologists' tendency where "the consonants count little and the vowels not at all": it's a very important advance. Against later scholars' examination of the history of individual words: it's not a tool to wipe away inconvenient data.

[Emily]

my impression is that with the "th" function words, they tended to be unstressed, and pronounced as part of a larger unit with surrounding words, leading them to be voiced intervocalically, even though it was technically across word boundaries. so the sound change was a regular, neogrammarian-friendly conditioned change.

That may be the case. Certainly voicing in OE is very low-yield. Still, there's a strong ð/θ contrast medially and finally.

The problem with this explanation, though, is that most of these words appear at the beginning of a phrase or sentence, which is not really where we'd expect sandhi/liaison effects.

hardly scientific but i just looked through an OE translation of gen 3:1-19 and counted words beginning with þ. there were 7 found at the beginnings of sentences (even being generous in considering what a sentence is) and 103 in non-sentence-initial position. expanding it to include phrases and subclauses would reduce the disparity a bit, but even those tend to have several non-initial þ-words: For þām þe þū þis dydest; þe iċ þē bebēad þæt þū ne ǣte; etc. there's enough non-initial presence that a stress-based sound change still seems like the best explanation to me

[Richard W]

Why should it?

Because a conditioned change that wasn't a merger would require a phoneme to split into the unshifted and shifted sounds.

If it's conditioned, you're simply describing the formation of an allophone, which doesn't in any way conflict with the phoneme.

A more serious problem is if a change is, as Labov has reported, favored in a small number of words, without a condition which specifies what those words are.

But if the conditioned change only diffuses through the lexicon, then in the same environment, some words have the shift, and others don't. That looks like a phoneme split to me. It's not a matter of allophony, or at least, not conditioned allophony, and it's not free variation either.

[zompist]

But if the conditioned change only diffuses through the lexicon, then in the same environment, some words have the shift, and others don't.

Then it's not conditioned.

That looks like a phoneme split to me. It's not a matter of allophony, or at least, not conditioned allophony, and it's not free variation either.

It strikes me as leaping to conclusions to call it a "phoneme split". Arguably phonemes are things people actually perceive— that's why we have things like rhyme, or minimal pairs. At this stage in an ongoing sound change, people usually don't perceive the changes, and indeed react with surprise if their own words are played back to them.

But I'd emphasize that the Neogrammarians had no theory of ongoing sound changes, because they never noticed any. There is not a competing theory on how a sound change develops, unless you consider it a theory that "on Wednesday /a/ was [a] and on Thursday it was [æ]."

Plus, an ongoing sound change is, well, ongoing. It would be useless to look around on Wednesday and divide words into /a/ and /æ/: the division won't apply to any entire speech community, and it won't apply to any one speaker over time. We don't know the underlying mechanism, beyond the not very helpful observation that some people initiate a change and other people find that cool and imitate it.