The Great Proto-Indo-European Thread's Sequel

[WeepingElf]

From my own research, I've found out the classical genealogical tree isn't an adequate model for the IE family,

If only there were hundreds years worth of linguistic research - done by dozens of linguists - which establishes and ascertains the tree model for the IE family...

In fact the tree model is pretty bad, in general. François has a nice preprint which goes into considerable detail about why and how. I strongly recommend that Talskubilos read it, in order to understand how historical linguistics actually works.

Yes. I am also into the wave model, which IMHO describes language families such as Indo-European better than the family tree model. Such language families start as expanding dialect continua, in which innovations spread from various centres, resulting in intersecting isoglosses, until the continuum breaks up into different languages. We see such intersecting isoglosses in IE; for instance, Germanic shares some features with Italic and Celtic and others with Balto-Slavic, which in turn shares other features with Indo-Iranian, and Greek completes the circle by sharing some features with Indo-Iranian and others with Italic.

The difficulty of subgrouping in IE is due to the rapid expansion of the family in the 3rd millennium BC. In 3000 BC, it is limited to the Pontic Steppe (already a rather large area - it is about 2,000 km from the Danube to the Ural - so there will have been different dialects); 1000 years later, it extends all the way from the Atlantic seaboard to the Hindukush.

I have two questions:
1. How did they all stay protolanguages for that long? (also, how did they stay distinct enough to be recognizable as different protolanguages?)

Surely, I didn't mean these protolanguages lasted for so long, but the whole process which ultimately lead to the historical IE languages. This would be a series of successive expansions (= waves) which replaced the previous one. An example of this idea will be this diagram (not mine):

I know that chart, it is from the book Economy and Society in Prehistoric Europe by Andrew Sherratt (1997). It has partly been obsoleted by archaeogenetics (the Neolithic farmers of Europe were genetically quite different from the Indo-Europeans, originated in a different region and probably spoke languages unrelated to IE), and it does not corroborate your thinking about the IE family.

And as for keenir's question: "Protolanguage" as used in this discussion is not a language type, it only describes the role of a language as the common ancestor of a family of languages. keenir may have confused this with a different sense the word is used in language origins studies, where it refers to a not yet fully evolved language spoken by early hominins - in times much more remote than what we are talking about here. But I may have misunderstood him, in which case I apologize.

2. how many language families do these protolanguages represent/belong to?

If I understood well, you refer to the number of protolanguages involved and their relationship to the historical IE languages. Although I still think this is a simplification, the late Spanish Indoeuropeist Rodríguez Adrados, who studied IE morpohology, proposed 3: PIE II for Anatolian alone, PIE III A for the Indo-Greek group (Indo-Iranian, Greek, Phyrgian and Armenian), and PIE III B for NW IE European (Celtic, Italic, Germanic and Balto-Slavic) plus Tocharian.



But in my IMHO these would be the more recent expansions, because at lexical level, there're remnants of the languages spoken by Neolithic farmers and Mesolithic hunter-gatherers, which weren't necessarily related to the later ones.

Surely, the languages of the Neolithic farmers and the Mesolithic hunter-gatherers did not disappear without trace, but left substratum loanwords in the IE languages that replaced them, the same way Romance languages contain loanwords from pre-Roman languages of Western Europe such as Gaulish. No Indo-Europeanist worth his stripes denies this! But the largest part of the lexicon and the grammatical framework of the IE languages can be traced back to a single ancestor language - PIE.

[bradrn]

But the largest part of the lexicon and the grammatical framework of the IE languages can be traced back to a single ancestor language - PIE.

I think you missed the bit where he denied the existence of a single common ancestor (or at least sounded like he denied it, as usual he’s never straight with his answers):

If I understood well, you refer to the number of protolanguages involved and their relationship to the historical IE languages. Although I still think this is a simplification, the late Spanish Indoeuropeist Rodríguez Adrados, who studied IE morpohology, proposed 3: PIE II for Anatolian alone, PIE III A for the Indo-Greek group (Indo-Iranian, Greek, Phyrgian and Armenian), and PIE III B for NW IE European (Celtic, Italic, Germanic and Balto-Slavic) plus Tocharian.

[WeepingElf]

But the largest part of the lexicon and the grammatical framework of the IE languages can be traced back to a single ancestor language - PIE.

I think you missed the bit where he denied the existence of a single common ancestor (or at least sounded like he denied it, as usual he’s never straight with his answers):

If I understood well, you refer to the number of protolanguages involved and their relationship to the historical IE languages. Although I still think this is a simplification, the late Spanish Indoeuropeist Rodríguez Adrados, who studied IE morpohology, proposed 3: PIE II for Anatolian alone, PIE III A for the Indo-Greek group (Indo-Iranian, Greek, Phyrgian and Armenian), and PIE III B for NW IE European (Celtic, Italic, Germanic and Balto-Slavic) plus Tocharian.

But as far as I understood Adrados, he has "IE IIIA" and "IE IIIB" descend, together with Anatolian, from "IE II" (and that from a hypothetical "IE I", which he characterized as "pre-flectional"), so this is not much different from the standard model, and all IE languages descend from a single common ancestor language, which is what Talksubilos denies. The only thing that is controversial in Adrados's model (apart from his characterization of "IE I" as "pre-flexional", of course) is that he construes the "NW IE" and "Greco-Aryan" clusters as branches of IE, while there is much uncertainty whether these typological clusters ("Greco-Aryan" keeps the tripartite verb aspect system intact in the earliest attested languages, while in "NW IE" it is gone in the oldest attested stages already, and some IEists claim it was never in place there) correspond to actual branches in a family tree.

The problem with this is that we are comparing different time stages here: the oldest known "Greco-Aryan" languages are earlier than the oldest known "NW IE" languages, and by the time the earliest known "NW IE" languages were set down in writing, the tripartite aspect system was already collapsing in most "Greco-Aryan" languages. Also, the tripartite verb aspect system appears to have left quite easily recognizable traces in the "NW IE" languages, and accounts for the changes in those languages quite well, and the only branch of IE where it seems unlikely that it ever had the tripartite verb aspect system is Anatolian.

[Talskubilos]

The difficulty of subgrouping in IE is due to the rapid expansion of the family in the 3rd millennium BC. In 3000 BC, it is limited to the Pontic Steppe (already a rather large area - it is about 2,000 km from the Danube to the Ural - so there will have been different dialects); 1000 years later, it extends all the way from the Atlantic seaboard to the Hindukush.

This is more less the Kurgan hypothesis (i.e the expansion of people from the Pontic Steppes) according to archaelogical.

Surely, the languages of the Neolithic farmers and the Mesolithic hunter-gatherers did not disappear without trace, but left substratum loanwords in the IE languages that replaced them, the same way Romance languages contain loanwords from pre-Roman languages of Western Europe such as Gaulish.

As well as other non-attested languages, including IE ones. For example, Gaulish itself has some loanwords from a Baltoid (i.e. similar to Baltic) language, which would be part of the substrate called "Sorothaptic" by the Catalan linguist Joan Coromines.

No Indo-Europeanist worth his stripes denies this! But the largest part of the lexicon and the grammatical framework of the IE languages can be traced back to a single ancestor language - PIE.

It's rather on the contrary, as Indo-Europeanists have underestimated the role of pre-Kurganic (I prefer to use this term instead of PIE to avoid confusion) languages. Surely grammar (i.e. morphology) is more recent, but not all the +2000 lexical items reconstructed for PIE have the same chronology.

[Talskubilos]

I think you missed the bit where he denied the existence of a single common ancestor (or at least sounded like he denied it, as usual he’s never straight with his answers):

If the classical genealogical tree model is inadequate for the IE family, it follows there's no single common ancestor for IE languages.

[bradrn]

I think you missed the bit where he denied the existence of a single common ancestor (or at least sounded like he denied it, as usual he’s never straight with his answers):

If the classical genealogical tree model is inadequate for the IE family, it follows there's no single common ancestor for IE languages.

Ah-ha — finally, we get to the root of the confusion!

Your logic here is understandable, but incorrect. In fact it is entirely possible for a group of languages to evolve from a common ancestor, while at the same time having a complex history which cannot be well represented as a tree. The paper I linked goes into considerable detail about how this happens.

The model can be summarised as follows: a language, spoken by a single community, gradually expands, creating a set of dialects. As it expands, the dialects stay in contact, exchanging sound changes and loanwords with each other, as well as with surrounding unrelated languages.

Eventually, these dialects diverge into separate languages. The resulting language family does not fit well in a tree model: the languages are connected both to each other and to surrounding families in overlapping ways. However, these languages have still descended from a single common ancestor, and this is reflected in the modern-day languages by at least the core of their shared vocabulary and grammar.

If you want to see how this works from the perspective of a single language, let’s take English. We know that English has an enormous amount of loanwords: over half its vocabulary comes from French or Latin. Even within Germanic, its subgrouping has some vagueness — it is closest to Frisian and Dutch, but has been deeply influenced on all levels by Old Norse. But, nonetheless, we can trace its basic grammar and lexicon directly back to Proto-Germanic. (And Proto-Germanic is marginally attested in its own right, so we know that this common ancestor existed.)

It is the same with Indo-European, on a larger scale. Yes, there have been expansion and diffusion processes, with massive amounts of loaning. But that does not change the fact that we can reconstruct a single Proto-Indo-European language from which the basic grammar and lexicon has descended.

[WeepingElf]

I think you missed the bit where he denied the existence of a single common ancestor (or at least sounded like he denied it, as usual he’s never straight with his answers):

If the classical genealogical tree model is inadequate for the IE family, it follows there's no single common ancestor for IE languages.

Ah-ha — finally, we get to the root of the confusion!

Your logic here is understandable, but incorrect. In fact it is entirely possible for a group of languages to evolve from a common ancestor, while at the same time having a complex history which cannot be well represented as a tree. The paper I linked goes into considerable detail about how this happens.

The model can be summarised as follows: a language, spoken by a single community, gradually expands, creating a set of dialects. As it expands, the dialects stay in contact, exchanging sound changes and loanwords with each other, as well as with surrounding unrelated languages.

Eventually, these dialects diverge into separate languages. The resulting language family does not fit well in a tree model: the languages are connected both to each other and to surrounding families in overlapping ways. However, these languages have still descended from a single common ancestor, and this is reflected in the modern-day languages by at least the core of their shared vocabulary and grammar.

If you want to see how this works from the perspective of a single language, let’s take English. We know that English has an enormous amount of loanwords: over half its vocabulary comes from French or Latin. Even within Germanic, its subgrouping has some vagueness — it is closest to Frisian and Dutch, but has been deeply influenced on all levels by Old Norse. But, nonetheless, we can trace its basic grammar and lexicon directly back to Proto-Germanic. (And Proto-Germanic is marginally attested in its own right, so we know that this common ancestor existed.)

It is the same with Indo-European, on a larger scale. Yes, there have been expansion and diffusion processes, with massive amounts of loaning. But that does not change the fact that we can reconstruct a single Proto-Indo-European language from which the basic grammar and lexicon has descended.

Exactly.

[Talskubilos]

It is the same with Indo-European, on a larger scale. Yes, there have been expansion and diffusion processes, with massive amounts of loaning. But that does not change the fact that we can reconstruct a single Proto-Indo-European language from which the basic grammar and lexicon has descended.

OK. My point is the PIE reconstructed by Indo-Europeanists isn't just a "common ancestor" but a kind of Frankenstein from several sources.

[WeepingElf]

It is the same with Indo-European, on a larger scale. Yes, there have been expansion and diffusion processes, with massive amounts of loaning. But that does not change the fact that we can reconstruct a single Proto-Indo-European language from which the basic grammar and lexicon has descended.

OK. My point is the PIE reconstructed by Indo-Europeanists isn't just a "common ancestor" but a kind of Frankenstein from several sources.

Well, our disagreement is perhaps merely in degree then. Nobody denies that IE languages contain words from different sources. Also, the standard reconstruction of PIE probably mixes features from different dialects and time stages. I have been conjecturing for some time that the phonology of "Handbook PIE" may reflect an earlier stage than the morphology: while the phonology accounts for Anatolian easily, the morphology does less so. This may mean that the phonology belongs to Pre-Anatolian PIE and the morphology for Post-Anatolian PIE. Alas, we simply don't know.

[keenir]

I have two questions:
1. How did they all stay protolanguages for that long? (also, how did they stay distinct enough to be recognizable as different protolanguages?)
2. how many language families do these protolanguages represent/belong to?

thank you

(1) is a very strange question to ask. A protolanguage is simply a reconstructed ancestor language: PIE and all its ancestors ad infinitum have always been protolanguages, because they are ancestors of the modern IE family (if not others too). Although, in practice, it seems that the term is reserved for the latest common ancestor of a linguistic group, specifically

yeah, I think thats what was going through my head {the final statement there} when he said it. thank you for the clarification and explanation.

And as for keenir's question: "Protolanguage" as used in this discussion is not a language type, it only describes the role of a language as the common ancestor of a family of languages. keenir may have confused this with a different sense the word is used in language origins studies, where it refers to a not yet fully evolved language spoken by early hominins - in times much more remote than what we are talking about here. But I may have misunderstood him, in which case I apologize.

No, I had confused "any and all ancestor languages which led to PIE and its descendant branches" {which everyone else was using}, with "PIE, which is ancestor to the IE family" {which i had presumed}

[WeepingElf]

OK, so I apologize for the misunderstanding.

[keenir]

OK, so I apologize for the misunderstanding.

it was an easy misunderstanding on/for all sides, so no worries.

[dɮ the phoneme]

The tree model is perfectly adequate. Species, too, go through a period of relatively free internal gene flow (cf. how humans of different genetic backgrounds can interbreed), but if two populations are separated for long enough they may speciate, i.e. genetically diverge to such a degree that horizontal gene flow becomes impossible. The tree model is a model: that is to say, it is by nature a simplification of reality, but it is a simplification which, if used within its domain of applicability, accurately describes some feature of reality and informs us about its behavior. Even though horizontal gene flow exists in biology, the tree model is still a good model of biological evolution because horizontal gene flow is in most cases temporally restricted, and on large timescales the map of all living organisms does in fact look like a tree. To assert the validity of the tree model is not to deny the reality of horizontal gene flow! Likewise, we know that horizontal "feature flow" exists in linguists, in the form of loanwords and areal influences, things which are especially pronounced among closely related lects in geographic proximity. None the less, the fact is that on large scales the web of influences and relationships between different language varieties over time comes out looking like a tree. This is likewise a fact about reality, a fact which the tree model describes.

What everyone in this thread seems to be opposing is overapplication of the tree model, application on small timescales which fall outside of its domain of applicability. But this does not mean the tree model is a bad model—far from it, it is the most successful model of language relatedness yet devised and I believe it will remain so. It simply means that, as with all models, one should not apply the tree model outside its domain of applicability. Duh!

[bradrn]

None the less, the fact is that on large scales the web of influences and relationships between different language varieties over time comes out looking like a tree. This is likewise a fact about reality, a fact which the tree model describes.

No, this is simply incorrect. On large scales it doesn’t look like a tree at all (except in special cases). The overlapping subgroups don’t magically go away once you give varieties sufficient time: they simply stay overlapping subgroups, albeit old ones.

(For more, I will direct you too to the paper I linked earlier, which has plenty of examples of families for which trees are insufficient.)

[dɮ the phoneme]

None the less, the fact is that on large scales the web of influences and relationships between different language varieties over time comes out looking like a tree. This is likewise a fact about reality, a fact which the tree model describes.

No, this is simply incorrect. On large scales it doesn’t look like a tree at all (except in special cases). The overlapping subgroups don’t magically go away once you give varieties sufficient time: they simply stay overlapping subgroups, albeit old ones.

(For more, I will direct you too to the paper I linked earlier, which has plenty of examples of families for which trees are insufficient.)

I haven't finished the paper, so maybe I should hold off on commenting, but I think the core place it goes wrong is here:

One could propose that the two models are complementary, in the sense that trees
might be well-designed for representing the genealogical relations among separate
LANGUAGES; whereas waves are only concerned with the relations between DIALECTS
within the boundaries of each language. The two models would then both be useful, but
at different scales of observation. We think that this view is mistaken, for one important
reason: namely, that many language (sub)families — as we will see below — have in fact
arisen from the diversification of earlier dialect continua. To the extent that local
innovations are faithfully transmitted across generations, the resulting languages preserve
the traces of earlier entangled isoglosses. If trees fail to represent genealogical relations
between dialects, then they must also fail to capture the relations between the languages
that are descended from them.

I obviously don't buy into some hard "language/dialect" distinction, but I think the view they're dismissing here is essentially correct. The wave model works well for high-mutual-intelligibility high-contact scenarios, the tree model starts making more sense when mutual intelligibility becomes low or contact between two speaker communities becomes infrequent—consider, for instance, how English and Hindi have participated in entirely different sets of sound changes in the last few centuries. Just because different branches of a tree can retain entangled isoglosses from the "wave stage" doesn't mean that language change at this later date is still best described via a wave model! Far from it, at a certain stage language varieties often diverge enough, as they have in IE, to start participating in wholly separate sound changes, and it's clear that this produces a tree. Of course there are occasional exceptions (cf. r > ʁ in Western Europe), but we already have a model for these exceptions and it's call "areal influence". I suppose you could rename the latter to "wave model" if you felt like it.

In particular, this line from the above quote does not seem to me a valid inference, and I would go even farther and say that it is a fundamentally misguided way to reason about scientific models:

If trees fail to represent genealogical relations between dialects, then they must also fail to capture the relations between the languages
that are descended from them.

In biology, too, genetic material is often distributed unevenly throughout a population, with what might be called "overlapping isoglosses" (I don't know the proper biological term). For instance, the genes that result in monolids are IIRC relatively common among Irish people, as they are in East Asians, and this is hypothesized to be the result of horizontal gene flow between human populations via the Eurasian steppe and not the result of there existing some hypothetical recent common ancestor between East Asians and Irish people to the exclusion of everyone else (don't quote me on this particular example please, I am not a geneticist and it is half remembered; the essential point is valid though). Overlapping isoglosses! However, if a founder population from Ireland and a founder population from Japan were to move to separate planets for a few million years and speciate, the result you got would be best modeled as a tree, even if both new species retained those monolids (perhaps differentiating them from other human descendant species that didn't).

Horizontal gene flow and founder effects (which are collectively what the above example is describing) do not invalidate the tree model in biology and they do not invalidate it in linguistics. It is a model, it does not and cannot describe every detail. It doesn't need to. The large scale picture of language change does look like a tree, which is why the tree model has been so successful!

Fundamentally I think we all agree on the facts here, we all agree that horizontal feature flow exists and we all agree that English and Hindi (for instance) are currently participating in wholly separate sound changes despite being related. The upshot of these facts—in particular high horizontal feature flow in the high mutual intelligibility, high contact regime—may be that reconstructing internal subgroupings at a certain level of granularity is not just practically but conceptually impossible. Well, then, so be it I suppose. We don't get to choose how reality behaves, we merely get to study it.

[Ketsuban]

I haven't finished the paper, so maybe I should hold off on commenting

Part of the point the paper makes is that the wave model reduces to the tree model given a set of constraints, so all arguments for the utility of the tree model are actually arguments for the utility of the wave model; it handles all the cases the tree model handles while also handling all the situations where there is clear relatedness but no clear treelike pattern.

[Richard W]

In biology, too, genetic material is often distributed unevenly throughout a population, with what might be called "overlapping isoglosses" (I don't know the proper biological term).

The biological concept is "incomplete lineage sorting". There's also introgression, which is more akin to loans between closely related languages.

[bradrn]

I obviously don't buy into some hard "language/dialect" distinction, but I think the view they're dismissing here is essentially correct. The wave model works well for high-mutual-intelligibility high-contact scenarios, the tree model starts making more sense when mutual intelligibility becomes low or contact between two speaker communities becomes infrequent—consider, for instance, how English and Hindi have participated in entirely different sets of sound changes in the last few centuries. Just because different branches of a tree can retain entangled isoglosses from the "wave stage" doesn't mean that language change at this later date is still best described via a wave model! Far from it, at a certain stage language varieties often diverge enough, as they have in IE, to start participating in wholly separate sound changes, and it's clear that this produces a tree. Of course there are occasional exceptions (cf. r > ʁ in Western Europe), but we already have a model for these exceptions and it's call "areal influence". I suppose you could rename the latter to "wave model" if you felt like it.

I think you’re misunderstanding what the ‘wave model’ actually means. It doesn’t mean that every branch is strongly related to every other: that would be obviously absurd. Instead, it means that many neighbouring branches can and do share sound changes, in ways that the tree model does not predict happening.

Let’s look at Hindi. Of course, English and Hindi have shared no sound changes in the past, say, 2000 years. But Hindi has shared sound changes with its neighbours, the other Indic and Iranian languages. Indeed, it shares enough sound changes with the other Indic language that we can group them as a single group, and then enough with Iranian languages that we can be justified in grouping them further as ‘Indo-Iranian’. Further up the cladogram, Indo-Iranian shares some sound changes with Greek, Armenian and possibly others — but not enough to draw any exclusive subgrouping, which is the requirement of the tree model.

If you read to the end of the paper they address all of this. Their key innovation is proposing a metric which allows assessing the coherence of any proposed subgroup in the wave model. If the subgroups happen to be non-overlapping, it reduces to the tree model (as Ketsuban correctly noted). But, in general, they will overlap, and the tree model will be insufficient to describe the actual history.

[dɮ the phoneme]

I haven't finished the paper, so maybe I should hold off on commenting

Part of the point the paper makes is that the wave model reduces to the tree model given a set of constraints, so all arguments for the utility of the tree model are actually arguments for the utility of the wave model; it handles all the cases the tree model handles while also handling all the situations where there is clear relatedness but no clear treelike pattern.

Fair enough, this more or less answers my objection (although I still think "language change looks tree-like on large scales" is undeniably true).

In biology, too, genetic material is often distributed unevenly throughout a population, with what might be called "overlapping isoglosses" (I don't know the proper biological term).

The biological concept is "incomplete lineage sorting". There's also introgression, which is more akin to loans between closely related languages.

Thanks!

[bradrn]

(although I still think "language change looks tree-like on large scales" is undeniably true).

On this point, you just need to look at the difficulties with subgrouping in families like PIE. Some people say, for instance, that Greek forms a distinctive subgroup with Indo-Aryan; others have argued for a relationship with Armenian; still others suggest similarities with Albanian. Which of these is correct? The simplest conclusion is that all of these overlapping groupings reflect different parts of the history, with varying degrees of ‘subgroupiness’ (as François calls it) depending on how much they share.