The Great Proto-Indo-European Thread

[KathTheDragon]

<snip>

I strongly suggest you actually familiarise yourself with PIE, because it's really not hard to disprove this. For starters, Greek and Indo-Iranian are not the be-all-end-all of distinguishing *D and *Dʰ. Germanic reflects these as *T and *D, for example, and di-aspirate roots therefore show two *D, such as *beudaną < *bʰewdʰ- "be awake, wake up". As for minimal pairs, take the aforementioned and Germanic *bautaną < *bʰewd- "beat, hit" for contrasting *DʰeDʰ- and *DʰeD-, and Germanic *treganą < *dreǵʰ- "sadden, distress" and Germanic *draganą < *dʰreǵʰ- "pull, drag" for contrasting *DʰeDʰ- and *DeDʰ-. Thus your belief can be shown to be contrary to the facts.

[Zju]

<snip>

I strongly suggest you actually familiarise yourself with PIE, because it's really not hard to disprove this. For starters, Greek and Indo-Iranian are not the be-all-end-all of distinguishing *D and *Dʰ. Germanic reflects these as *T and *D, for example, and di-aspirate roots therefore show two *D, such as *beudaną < *bʰewdʰ- "be awake, wake up". As for minimal pairs, take the aforementioned and Germanic *bautaną < *bʰewd- "beat, hit" for contrasting *DʰeDʰ- and *DʰeD-, and Germanic *treganą < *dreǵʰ- "sadden, distress" and Germanic *draganą < *dʰreǵʰ- "pull, drag" for contrasting *DʰeDʰ- and *DeDʰ-. Thus your belief can be shown to be contrary to the facts.

Why is *ew once reflected as *eu and once as *au? Is the latter actually from an o grade *bʰowd-?

[KathTheDragon]

Yes, I didn't bother to indicate the stem formation because only the root is important here.

[Soap]

<snip>

I strongly suggest you actually familiarise yourself with PIE, because it's really not hard to disprove this.

...


Thus your belief can be shown to be contrary to the facts.

Okay...

As for minimal pairs, take the aforementioned and Germanic *bautaną < *bʰewd- "beat, hit" for contrasting *DʰeDʰ- and *DʰeD-, and Germanic *treganą < *dreǵʰ- "sadden, distress" and Germanic *draganą < *dʰreǵʰ- "pull, drag" for contrasting *DʰeDʰ- and *DeDʰ-.

If you read my post, youll see that, of course, I answered your objection in the counterexamples section. In fact I gave a better counterexample than you did: *ghaydos "goat", which is attested in two different branches of IE, and thus may have an origin that goes back to a common ancestor within the IE family, even if not to PIE itself.

The example you gave, treganą , is attested only in Germanic, and thus its PIE form, if there ever was one, is anybody's guess. I could not find a list of PIE roots that lists anything resembling that Germanic stem.

I'd say Germanic is not a good place to look for evidence because a large portion of its vocabulary has no known PIE origin. I assume youre familiar with the Germanic substrate theory?

For starters, Greek and Indo-Iranian are not the be-all-end-all of distinguishing *D and *Dʰ. Germanic reflects these as *T and *D, for example, and di-aspirate roots therefore show two *D, such as *beudaną < *bʰewdʰ- "be awake, wake up".

Again, please read my post and use the Quote function to reply to something that I actually wrote if you want to criticize my theory. What youre replying to now is nothing that I wrote in this post or any earlier one.

[jal]

*brings out the drinks and the lounge chair*


JAL

[hwhatting]

The example you gave, treganą , is attested only in Germanic, and thus its PIE form, if there ever was one, is anybody's guess. I could not find a list of PIE roots that lists anything resembling that Germanic stem.

The root *dreg'^h- is also attested in Lithuanian (Pokorny 226/27, LIV 125)
Maybe I overlooked this: in your theory, what happens to words with an aspirated stop and a laryngeal, like *d^heH₁- "put"?

If, according to your argumentation, we exclude *b^h and look only for CVC roots, because in CeRC roots the second stop could be a root extension, I can find the following roots with plain voiced and aspirated voiced stop in one root: *g'hed- "shit" (Pok. 423, LIV 172); *g'eb^h- (s. below); *g^hed- "grab" (LIV 194, Pok. 437-438).
Minimal pairs: *g'eb^h- eat, chew (Pok. 382, LIV 161) vs. *g^heb^h- "seize, give" (Pok. 407-409, LIV 193) - this shows that *b^h is compatible both with an aspirate and a non-aspirate voiced stop in the root. *ghed- "grab" (s. above) vs. *g^hed^h- "meet, join" (Pok. 423/24, LIV 195).
If I'd include CREC, there would be more examples for M-MA roots and minimal pairs.

Again, nice on paper, but there are problems. For example, what about roots containing a laryngeal and a breathy-voiced stop such as *h₁egʷʰ-? How does this determine which voiced stop becomes aspirated when there are two of them?

That would just be [hegʷ] as per rule 1, presuming the stress is on the root.

Counterexamples (I only quote LIV here as Pokorny doesn't have laryngeals anyway and I only looked at HeC, not HREC or HERC)): *Hed^h "say" (LIV 222), ved. perf.'āha (< *HeHódh-e-) - should be 'āda as per your rule; *h₂eg^h- "be sad / afraid" (LIV 257) gr. ἄχομɑɩ.(should be ἄɣομɑɩ as per your rule).

[Vijay]

This is why Grassman's Law has, in Sanskrit, also been referrred to as aspiration throwback: in any "diaspirate" root, aspiration can appear on either the leading or the trailing consonant: always the trailing consonant when in an open syllable, and always the leading one when the syllable is closed.

Isn't b the leading consonant in badhnāti?

[hwhatting]

This is why Grassman's Law has, in Sanskrit, also been referrred to as aspiration throwback: in any "diaspirate" root, aspiration can appear on either the leading or the trailing consonant: always the trailing consonant when in an open syllable, and always the leading one when the syllable is closed.

Isn't b the leading consonant in badhnāti?

To play devil's advocate here, the syllabification could be ba.dhnā.ti. Does anyone know or can check a source on Sanscrit syllabification?

[KathTheDragon]

I don't have time to argue with someone who clearly doesn't know how even the comparative method works, so I'm just going to ignore you. If you really want me to waste my time on you, go educate yourself properly.

[Vijay]

This is why Grassman's Law has, in Sanskrit, also been referrred to as aspiration throwback: in any "diaspirate" root, aspiration can appear on either the leading or the trailing consonant: always the trailing consonant when in an open syllable, and always the leading one when the syllable is closed.

Isn't b the leading consonant in badhnāti?

To play devil's advocate here, the syllabification could be ba.dhnā.ti.

Then which consonant would be "the leading one"? Because then there are two leading consonants, just in different (adjacent) syllables, right?

EDIT: Wait, then it would be the "trailing consonant" that's the relevant one. So I guess that would be dh? Or would it be n?

[hwhatting]

For Grassmann's law, only the consonants that can be aspirated are relevant, so "n" doesn't count.

[WeepingElf]

I don't have time to argue with someone who clearly doesn't know how even the comparative method works, so I'm just going to ignore you. If you really want me to waste my time on you, go educate yourself properly.

I second that. Soap is more and more going on my nerves, too, with his talking about things he flatly doesn't know about (we have a word for such people in German, Flachpfeife, literally 'flat whistle'). He is beginning to remind me of Octaviano, though in Soap's case it is probably really just lack of knowledge while Octaviano probably knew the relevant facts but refused to accept them. We should simply ignore his posts here (as I already ignore his conlangs, which seem to reflect a deep-rooted obsession with labial and labialized obstruents). If he wants to talk about Proto-Indo-European, he should first attentively read a handbook (B. W. Fortson, Indo-European Language and Culture is IMHO the best - well written, easy to get into and free from speculative nonstandard ideas, and not too expensive). Then, we'll talk.

[Vijay]

not too expensive

Or even available for free online.

[Soap]

The example you gave, treganą , is attested only in Germanic, and thus its PIE form, if there ever was one, is anybody's guess. I could not find a list of PIE roots that lists anything resembling that Germanic stem.

The root *dreg'^h- is also attested in Lithuanian (Pokorny 226/27, LIV 125)

Thank you. But it still seems suspicious, roughly on par with *ghaydos above, as it is attested in only 2 languages, and the Lithuanian form does not match what the supposed PIE root would predict. (where does the /i/ come from?) And also because Lithuanian would not have distinguished between initial /d/ and initial /dʰ/ in the first place, so it may well be either unrelated or a loan from the same source that gave us the Germanicword.

Maybe I overlooked this: in your theory, what happens to words with an aspirated stop and a laryngeal, like *d^heH₁- "put"?

...


Counterexamples (I only quote LIV here as Pokorny doesn't have laryngeals anyway and I only looked at HeC, not HREC or HERC)): *Hed^h "say" (LIV 222), ved. perf.'āha (< *HeHódh-e-) - should be 'āda as per your rule; *h₂eg^h- "be sad / afraid" (LIV 257) gr. ἄχομɑɩ.(should be ἄɣομɑɩ as per your rule).

Those would likely be [|deh|] and such in original PIE. We dont see patterns like that in the attested lajguages because none of the Graeco-Aryan languages preserve laryngeals as fricatives in any attested material. Instead, they evolved new /h/ sounds to fill the gap while, at least in Greek, Grassman's Law continued to operate on that new sound.

If, according to your argumentation, we exclude *b^h and look only for CVC roots, because in CeRC roots the second stop could be a root extension, I can find the following roots with plain voiced and aspirated voiced stop in one root: *g'hed- "shit" (Pok. 423, LIV 172); *g'eb^h- (s. below); *g^hed- "grab" (LIV 194, Pok. 437-438).
Minimal pairs: *g'eb^h- eat, chew (Pok. 382, LIV 161) vs. *g^heb^h- "seize, give" (Pok. 407-409, LIV 193) - this shows that *b^h is compatible both with an aspirate and a non-aspirate voiced stop in the root. *ghed- "grab" (s. above) vs. *g^hed^h- "meet, join" (Pok. 423/24, LIV 195).
If I'd include CREC, there would be more examples for M-MA roots and minimal pairs.

I will look at those another time .... I never said that there were *no* counterexamples, just that theyre outnumbered by traditional diaspirates by a wide margin. But I note that even the examples you provided mostly contain /bʰ/, and are thus not coutnerexamples to my theory. Inner PIE Phonetic realizations of /gebʰ/ vs /gʰebʰ/ would thus likely reflect earlier surface realizations of [|gebh|] vs [|gheb|], with the /b/ > /bʰ/ unconditionally. Again, my theory is that aspiration was a suprasegmental feature, so in aroot with 2 stops, it msut appear on one and only one stop.

This is why Grassman's Law has, in Sanskrit, also been referrred to as aspiration throwback: in any "diaspirate" root, aspiration can appear on either the leading or the trailing consonant: always the trailing consonant when in an open syllable, and always the leading one when the syllable is closed.

Isn't b the leading consonant in badhnāti?

I'm sure there's some explanation for that, perhaps there was once a vowel after the /dʰ/. I dont see any people denying the existence of Grassmann's Law in Sanskrit so I'm sure somebody has the answer.

[Soap]

I don't have time to argue with someone who clearly doesn't know how even the comparative method works, so I'm just going to ignore you. If you really want me to waste my time on you, go educate yourself properly.

I second that. Soap is more and more going on my nerves, too, with his talking about things he flatly doesn't know about (we have a word for such people in German, Flachpfeife, literally 'flat whistle'). He is beginning to remind me of Octaviano, though in Soap's case it is probably really just lack of knowledge while Octaviano probably knew the relevant facts but refused to accept them. We should simply ignore his posts here (as I already ignore his conlangs, which seem to reflect a deep-rooted obsession with labial and labialized obstruents). If he wants to talk about Proto-Indo-European, he should first attentively read a handbook (B. W. Fortson, Indo-European Language and Culture is IMHO the best - well written, easy to get into and free from speculative nonstandard ideas, and not too expensive). Then, we'll talk.

When you proposed, a few pages back in this thread, that PIE had undergone an unconditional change of all ejectives to implosive stops, followed by another unconditional change of all implosive stops to plain voiced stops, I asked you if there had ever been such a change attested in natlangs, and why it wouldnt be simpler to cut out the middle step by altering one other related sound change. I havent heard from you since then until this post where you quote someone who criticized me. If this is the best that an expert linguistic scholar has to offer, Im proud to be ignorant.

[KathTheDragon]

The latter step is well-attested, and apparently implosives and ejectives do interchange, which supports the former step. A direct change from ejectives to voiced stops is possible, but there is greater utility in supposing that a system with implosives directly preceded the traditional reconstruction - and perhaps even that the traditional reconstruction is only valid for a subset of IE - as argued here.

[hwhatting]

Maybe I overlooked this: in your theory, what happens to words with an aspirated stop and a laryngeal, like *d^heH₁- "put"?

Those would likely be [|deh|] and such in original PIE. We dont see patterns like that in the attested lajguages because none of the Graeco-Aryan languages preserve laryngeals as fricatives in any attested material. Instead, they evolved new /h/ sounds to fill the gap while, at least in Greek, Grassman's Law continued to operate on that new sound.

You're losing me here. Those attested languages that show different reflexes for Media (M) and Media Aspirata (MA) clearly show different reflexes for them in roots of the typ CEH, see e.g. *d^heH₁- "put" vs. *deH₃- "give". If your theory of suprasegmental aspiration were right, both *deH- and *d^heH- should show the same outcome.

I never said that there were *no* counterexamples, just that theyre outnumbered by traditional diaspirates by a wide margin.

Are they?
If we exlude *b^h, CREC, and CERC; we have:
One aspirate - 2 cases: *g'^hed- "shit" (Pok. 423, LIV 172); *g^hed- "grab" (LIV 194, Pok. 437-438).
Diaspirates – 3 cases: *d^heg^wh- "burn" (Pok. 240/41, LIV 133), *g^hed^h- "meet, join" (Pok. 423/24, LIV 195), *g^whed^h- "ask, wish" (Pok. 488, LIV 217).
That doesn’t look so overwhelming…

But I note that even the examples you provided mostly contain /bʰ/,

One of three is not “most”.

and are thus not coutnerexamples to my theory. Inner PIE Phonetic realizations of /gebʰ/ vs /gʰebʰ/ would thus likely reflect earlier surface realizations of [|gebh|] vs [|gheb|], with the /b/ > /bʰ/ unconditionally. Again, my theory is that aspiration was a suprasegmental feature, so in aroot with 2 stops, it msut appear on one and only one stop.

So your sound law "/b/ > /bʰ/ unconditionally" works after the stage when aspiration was a suprasegmental feature? How then do you explain the cases of /b/ in PIE? And wouldn’t that mean that in pre-split PIE it wouldn’t have been a suprasegmental feature anymore and that in pre-split PIE, there would have been a phonemic difference between monoaspirate and diaspirate roots?

This is why Grassman's Law has, in Sanskrit, also been referrred to as aspiration throwback: in any "diaspirate" root, aspiration can appear on either the leading or the trailing consonant: always the trailing consonant when in an open syllable, and always the leading one when the syllable is closed.

Isn't b the leading consonant in badhnāti?

I'm sure there's some explanation for that, perhaps there was once a vowel after the /dʰ/. I dont see any people denying the existence of Grassmann's Law in Sanskrit so I'm sure somebody has the answer.

Well, of course, with assuming two underlying aspirates this form doesn't pose a problem at all - of two consecutive but not adjacent aspirates, the first one loses its aspiration, and the position before /n/ is not one where *dh would become a voicelsss stop and lose its aspiration.

In any case, let me summarize my misgivings with your theory:
1) You exclude roots containing /b^h/, and either exclude or at least haven’t checked on structures CERC and CREC, while the existing theory accounts for all of them;
2) The attested outcomes for roots of the type HEC and CEH don’t seem to be those you predict
3) You don’t offer any explanation for counterexamples and minimal pairs showing that there were both monoaspirate and diaspirate roots
4) In my opinion, your theory doesn’t explain anything better than the existing theory.
Considering what seems to be your motivation:

It seems awfully strange to posit that PIE would lose its laryngeals, which must have included at least two fricatives, with /h x xʷ/ being a common proposal, and yet keep phonemic aspiration after its voiced stops, when such aspiration was, at best, equivalent in acoustic strength to the weakest of the laryngeal fricatives.

But the laryngeals were still there when PIE broke up, so your objection only applies to the daughter languages, most of which do not continue the traditionally posited system of phonemic aspiration. Those that do rebuilt it (Greek went for voiceless - voiceless apirates – voiced; Indo-Iranian introduced a four-way distinction including voiceless aspirates) and also created a new phoneme /h/ (probably only after the fall of the laryngeals)
And as you well know, there are other proposed answers: (1) that aspiration wasn’t phonemic in PIE and the Systems with phonemic aspiration in Greek and Indo-Iranian are the result of the reordering of a different system (e.g. the glottalic theory, or some system with lenis / fortis contrasts); (2) that the typological unusualness of the PIE system was exactly the reason why it didn’t survive anywhere.

[Soap]

You're losing me here. Those attested languages that show different reflexes for Media (M) and Media Aspirata (MA) clearly show different reflexes for them in roots of the typ CEH, see e.g. *dheH1- "put" vs. *deH3- "give". If your theory of suprasegmental aspiration were right, both *deH- and *dheH- should show the same outcome.

They should, yes, since they are both underlyingly /deh/. It's not impossible those two roots were once the same word, splitting into two the way "other" and "or" did in English. But Im sure there are examples where the meaning is not similar. I would say that these roots arose only after the allophony had died off, likely after the layngreals had.

So your sound law "/b/ > /bʰ/ unconditionally" works after the stage when aspiration was a suprasegmental feature? How then do you explain the cases of /b/ in PIE? And wouldn’t that mean that in pre-split PIE it wouldn’t have been a suprasegmental feature anymore and that in pre-split PIE, there would have been a phonemic difference between monoaspirate and diaspirate roots?.

Well, the notation is distracting me. In my setup there is no /bʰ/, just /b/. [|bʰ|] would have been an allophone at first, but later apparently became the only allowable realization of what had been /b/. I know it's weird, but I can't see any explanation for a phonology with /bʰ/ and no /b/ being anything less than weird.Thats why I started out a few weeks ago by asking about Chinese, before i decided that I dont believe Chinese ever had /bʰ/ without /b/ either.

So your sound law "/b/ > /bʰ/ unconditionally" works after the stage when aspiration was a suprasegmental feature? How then do you explain the cases of /b/ in PIE? And wouldn’t that mean that in pre-split PIE it wouldn’t have been a suprasegmental feature anymore and that in pre-split PIE, there would have been a phonemic difference between monoaspirate and diaspirate roots?

The standard reconstruction says there was never any phonemic /b/ even a little bit; there was only /bh/. Im working based on that assumption. I believe that aspiration was still allohponic at least into the stage of pre-attested Graeco-Aryan, but that it died once the laynrgeals did, which is why we see only relics of the former setup such as Grassmann's Law.

But the laryngeals were still there when PIE broke up, so your objection only applies to the daughter languages, most of which do not continue the traditionally posited system of phonemic aspiration. Those that do rebuilt it (Greek went for voiceless - voiceless aspirates – voiced; Indo-Iranian introduced a four-way distinction including voiceless aspirates) and also created a new phoneme /h/ (probably only after the fall of the laryngeals)

Im not sure what you mean by this but I want to disengage for the time being. Thank you for being polite and asking pointed questions,m but I think this a little too much for me to handle right now. I have developed minor visual problems and its sometimes hard to read the letters on the screen unless i magnify to an inconvenient size. (phpBB does not seem to have text size zoom, so i have to use browser settings, which makes all images also large.e)

And as you well know, there are other proposed answers: (1) that aspiration wasn’t phonemic in PIE and the Systems with phonemic aspiration in Greek and Indo-Iranian are the result of the reordering of a different system (e.g. the glottalic theory, or some system with lenis / fortis contrasts); (2) that the typological unusualness of the PIE system was exactly the reason why it didn’t survive anywhere.

M<y idea is based on glottalic theory, yes, but without the need for any phonemic glottalics. My stop system is just /p t k kʷ b d g gʷ/. (I dont believe in palatovelars either, as I argued a few pages back, but that's not related to my theory about the aspirates.)

Well, of course, with assuming two underlying aspirates this form doesn't pose a problem at all

On https://en.wikipedia.org/wiki/Grassmann%27s_law they describe how these are just 2 equivalent ways of describing the same phenomenon. whether there are 2 underlying aspirates or none, the surface realization is always one. What I deny is that in PIE there were ever any roots with just one underlying aspirate. the standard theory states that there were never any roots with one aspirate plus a voiceless stop, but seems to allow for roots with one aspirate and one voiced stop. Im just not convinced that any of those roots are actually real ... .they all fit into one of the categories i listed above in another post. Most, especially in Germanic, are probably loans. Others are compounds. /bh/ seems to have behaved differently from the other stops. The dʰeH- type root, otoh, may be just a red herring. If Ive changed my mind on anything it's on that, since it doesnt seem that laryngeals counted as "aspirates", at least not all the time. Perhaps /h1/ did but /h2 h3/ did not? I'll look into it more later.

[hwhatting]

You're losing me here. Those attested languages that show different reflexes for Media (M) and Media Aspirata (MA) clearly show different reflexes for them in roots of the typ CEH, see e.g. *dheH1- "put" vs. *deH3- "give". If your theory of suprasegmental aspiration were right, both *deH- and *dheH- should show the same outcome.

They should, yes, since they are both underlyingly /deh/. It's not impossible those two roots were once the same word, splitting into two the way "other" and "or" did in English. But Im sure there are examples where the meaning is not similar. I would say that these roots arose only after the allophony had died off, likely after the layngreals had.

If allophony only died as late as you say here:

I believe that aspiration was still allohponic at least into the stage of pre-attested Graeco-Aryan

then this simply is not feasible, as both roots are attested in all major branches, including Anatolian.

The standard reconstruction says there was never any phonemic /b/ even a little bit; there was only /bh/.

No. The standard resconstruction says that /b/ was quite rare, but there is still a small number of words / roots where /b/ is commonly reconstructed. /b/ must have been a phoneme in Late PIE; the reconstructions you refer to normally assume that PIE did was /b/-less in an earlier stage and that /b/ arose secondarily (loans, /p/ + /H₃/). That again is a problem for your assumption that aspiration was allophonic until quite late.

Well, of course, with assuming two underlying aspirates this form doesn't pose a problem at all

On https://en.wikipedia.org/wiki/Grassmann%27s_law they describe how these are just 2 equivalent ways of describing the same phenomenon. whether there are 2 underlying aspirates or none, the surface realization is always one.

I know that description, but the point is that the second one (with only one aspiration per root) does work only in Sanscrit and contradicts the evidence of the other IE languages. But that is what we’re discussing here.

Im just not convinced that any of those roots are actually real ... .they all fit into one of the categories i listed above in another post. Most, especially in Germanic, are probably loans. Others are compounds

Well, I listed two more, which you ignored. As I said, if we only take CeC and ignore /bh/, we have two roots with one voiced aspirate against three with two aspirates, so the two roots I listed can’t be simply written off.

/bh/ seems to have behaved differently from the other stops.

Well, it does in your model, but that’s another weakness; in the standard model it behaves like the other voiced aspirates.

Im not sure what you mean by this but I want to disengage for the time being.

No problem, come back when you feel like it .

[Zju]

Why did PG *gulþą 'gold' give OE gold and not *golth? Why did middle Dutch golt, gout 'id' give Dutch goud - how did final voicing happen?

[KathTheDragon]

The usual explanation for words like this and many others is that the Proto-Germanic paradigm contained two variants, *gulþ- and *guld-, due to an accent alternation and Verner's law.

[Salmoneus]

Could also be analogy from inflected or derived forms with the alternant. Wiktionary actually suggests the entire plural in PGmc was in -d.

[EDIT: oh, yes, that's what you said, isn't it? Sorry, misread. In which case I'll offer the other possibility: there may already have been two variant lexical entries in PIE itself; wouldn't be the only couplet.]

But actually in this case a better question might be why we reconstruct it with /T/ at all. Given that it's been allegedly independently replaced by the form in -d in seemingly every Germanic language, and that the derivative adjective in -i:naz has independently been replaced in every language by the form in -d, and that the derivative verb in -ijana has been independently replaced in every language by the form in -d, maybe we should just take the hint and accept that it just had -d in PG? Or does the Norse reflex rule that out?

[Ephraim]

Could also be analogy from inflected or derived forms with the alternant. Wiktionary actually suggests the entire plural in PGmc was in -d.

But actually in this case a better question might be why we reconstruct it with /T/ at all. Given that it's been allegedly independently replaced by the form in -d in seemingly every Germanic language, and that the derivative adjective in -i:naz has independently been replaced in every language by the form in -d, and that the derivative verb in -ijana has been independently replaced in every language by the form in -d, maybe we should just take the hint and accept that it just had -d in PG? Or does the Norse reflex rule that out?

The Norse reflex gull or goll does rule out -d here.

Non-initial PG *d and *þ almost always have the same outcome in North Germanic (i.e. d~ð), but the clusters *nþ/nd and *lþ/nþ are an exception to this:
1. *nþ and *lþ assimilated to *nn and *ll
2. *nd and *ld remained as clusters. At the Old Norse stage at least, he outcome is nd and ld with a stop pronunciation, unlike for example rð from PG *rþ and *rd.

Modern Swedish usually has guld with a -d. The most common Old Swedish form was probably gull (although written <gul>, sometimes goll (written <gol>). A variant form with -d(h) is apparently attested in Old Swedish as well (forms like gol(d)t are apparently attested as well at in Early Modern Swedish). The -d is normally explained as (Low) German influence. The SAOB article (Svenska Akademiens Ordbok), written in 1929 indicates that at that time, gull was still the most common form in speech (although not in writing), but that the pronunciation with ld was gaining ground (due to influence from the written language, presumably).

Danish apparently has guld as well, with a -d in the written language at least.

The outcome of *gulþīnaz is gyllene in Swedish, with no -d, but according to the SAOB the form is actually borrowed from Low German. Icelandic gullinn (without umlaut) must derive from a form with *lþ.
Similarly for Icelandic and Swedish gylla from *gulþijaną. The Swedish form is archaic, replaced by förgylla, influenced by Low German.

The Gothic reflex gulþ also points to *þ. Gothic had final devoicing, but the dative seems to still be gulþa. I'm not sure what the Gothic outcome of *ld under final devoicing would be: *lþ or *lt.

Also, doesn't German Gold actually point to a form with *þ rather than *d? Compare kalt from *kaldaz. From what I understand, the most common outcome of PG *þ is *d while the most common outcome of PG *d is *t. But I'm not sure about this particular environment.

It might actually be the other way around, that we only have to reconstruct *-þ for Proto-Germanic. For the Dutch, Low German and English reflexes:

Why did PG *gulþą 'gold' give OE gold and not *golth? Why did middle Dutch golt, gout 'id' give Dutch goud - how did final voicing happen?

Similarly to High German, I think the regular Dutch and Low German outcome of PG *þ is actually d, which in this case is the same outcome as PG *d.

For Middle Dutch golt, perhaps it's just that final devoicing was indicated in the orthography? Compare these examples from Old Dutch: bigerlika over golt vs in gevon sal imo fan goldi Arabie:
http://www.etymologiebank.nl/trefwoord/goud

It has apparently been claimed that the following innovation is common to the West Germanic languages: Change of /lþ/ to stop /ld/ (except word-finally).
https://en.wikipedia.org/wiki/Germanic_ ... velopments

For this claim, Wikipedia cites Campbell, Alistair (1983-01-01). Old English Grammar. Clarendon Press. p. 169. ISBN 9780198119432

So it's possible that English gold also goes back to the form with *þ (if the shift occured at a time when *þ was not final here).

The usual explanation for words like this and many others is that the Proto-Germanic paradigm contained two variants, *gulþ- and *guld-, due to an accent alternation and Verner's law.

The word might still have had this alternation, though, since the alternation seems to have been somewhat regular for neuter a-stems.

[Salmoneus]

Also, doesn't German Gold actually point to a form with *þ rather than *d? Compare kalt from *kaldaz. From what I understand, the most common outcome of PG *þ is *d while the most common outcome of PG *d is *t. But I'm not sure about this particular environment.

That makes sense. However, with *gelda, the German has likewise ended up with -d instead of -t, so the same could be true here. Wiktionary notes that earlier forms had -t, and blames borrowing, either from Low German (which always had -d), or from unspecified High German dialects in which -lt regularly becomes -ld, so a similary process may have happened here. However, your explanation seems more probable.

It has apparently been claimed that the following innovation is common to the West Germanic languages: Change of /lþ/ to stop /ld/ (except word-finally).
https://en.wikipedia.org/wiki/Germanic_ ... velopments

Huh. Good to know. Would be nice if Wikipedia could standardise their sound changes, so that you didn't need to look at the pages for the phonological history of english, the phonological history of old english, proto-germanic, western germanic, ingvaeonic and apparently now also 'germanic languages' to get everything...

[Ephraim]

It has apparently been claimed that the following innovation is common to the West Germanic languages: Change of /lþ/ to stop /ld/ (except word-finally).
https://en.wikipedia.org/wiki/Germanic_ ... velopments

Huh. Good to know. Would be nice if Wikipedia could standardise their sound changes, so that you didn't need to look at the pages for the phonological history of english, the phonological history of old english, proto-germanic, western germanic, ingvaeonic and apparently now also 'germanic languages' to get everything...

I had to check, and Ringe apparently does not think that this change is common to all West Germanic languages, only to Northern West Germanic. Here, Northern West Germanic means ‘Ingvaeonic’, or the dialects ancestral to Old Frisian, Old English and Old Saxon, although he writes that "whereas the dialects ancestral to Old English and Old Frisian participated fully in most of [the Northern West Germanic] changes, those ancestral to Old Saxon exhibit a more ambiguous development."

Ringe, Don & Taylor, Ann (2014) A Linguistic History of English, Volume 2: The Development of Old English, p. 155 f:

It is often suggested that inherited *lþ became *ld in PWGmc word-internally (cf. e.g. Luick 1914–40: 833–4, Campbell 1962:169), but the facts do not support so broad a generalization. In particular, the d of OHG wildi ‘wild’ and faldan ‘to fold’ (seldom faltan, Seebold 1970: 183) is the regular outcome of PWGmc *þ by a much later, specifically OHG sound change (whereas PWGmc *d would have become OHG t), and in these words it probably cannot have been levelled in from word-final position: wildi has no consonant-final forms, and d is unlikely to have been levelled through the paradigm of faldan starting from the endingless pres. iptv. 2sg. and past indic. 1, 3sg. (though such a change is admittedly not impossible). Moreover, two potential examples might reflect a Verner’s Law alternation *þ ~ *d, namely the words which survive in OE as gold ‘gold’ < *gulþa- ~ *gulda- (see vol. i 4.3.4 (i), p. 270) and feld ‘field’, early OE -felth in place names, ⟵< *felþu- ~ *feldaw- ⟵ *felþu- ~ *fuldaw- (zero grade and *-d- also in the related folde ‘earth’). But it does seem that word-internal *lþ became *ld by regular sound change in NORTHERN WGmc; the following clear examples can be cited:

PGmc *falþaną ‘to roll up, to fold’ (Goth. past faífalþ ‘he rolled (it) up’) > PWGmc *falþan (OHG faldan) > *faldan > OE fealdan (it seems a bit less likely that *d was levelled through the paradigm from the default past and past ptc., though that is not impossible);
PGmc *wilþijaz ‘wild’ (Goth. wilþeis, ON villr) > PWGmc *wilþī (OHG wildi) > *wildī (OS wildi, OF wilde) > OE wilde;
PGmc *balþaz ‘bold, brave’, masc. nom. pl. *balþai (Goth. adv. balþaba, ON ballr) > PWGmc *balþ, *balþē (OHG bald, balde) > *balþ, *baldē ! *bald, *baldē (OS bald, balda) > OE beald, bealde;
PGmc *wulþraz (*-iz?) adj. ‘worth’ (Goth. wulþrs; cf. wulþus ‘glory’) > PWGmc *wulþr > neut. *wuldr ‘glory’ > OE wuldor (cf. wuldortorhtan ‘splendidly bright’, 3 syll. at Beo 1136);
PGmc *gulþīnaz ‘golden’ (Goth. gulþeins, ON gullinn) > PWGmc *gulþīn (OHG guldīn) > *guldīn > OE gylden, OF gelden, OS guldin.

Still, Old High German a general shift θ > d although not in the oldest written language. During the transition from Old Dutch to Middle Dutch, apparently there was a shift θ ð θθ > t d ss, but since f θ s had become voiced v ð z syllable-initally, many instances of PG *þ ended up as d rather than t.
https://en.wikipedia.org/wiki/Middle_Dutch#Phonology

So in other words, the reflex of PG *lþ is often ld in Middle Dutch and High German. Word-finally, Dutch might never have had a voiced sound here, so golt might be the result of a failure to voice in this position rather than final devoicing.

I wonder if there are any examples of <lth> in Old Dutch and Old High German.