Is Sumerican a Uralic Language?

[Åge Kruger]

The problem is that 5000 is an arbitrary number. For all we know the limit is 50,000 years.

I don't think it's all that arbitrary. Latin is well attested in written sources, and yet we can't reconstruct it perfectly or purely from it's descendants. We can get close to Vulgar Latin, but there is a certain amount of error in our reconstruction. If we then take several reconstructions, and compare them, then we're just multiplying the error. We can use this to make a good estimate of the margin of error over time and comparisons.

And it's likely the error grows exponentially, right?

That would be assuming that all evidence of relationship is equally likely to be lost or muddled, which is not true. *m is more likely to stay put than *o is, a noun class system is more likely to stay put than a set of locative cases is, etc. A relationship 10K years old is going to be harder to recover than one 5K years old, but not squared-as-hard.

No, it would be assuming that any 'parent' reconstruction would have a margin of error equal to some multiplicative function of the error margins of the child reconstructions.

But this is something we can actually test, and get real numbers for, so go ahead and prove us wrong.

[Tropylium⁺]

The problem is that 5000 is an arbitrary number. For all we know the limit is 50,000 years.

I don't think it's all that arbitrary. Latin is well attested in written sources, and yet we can't reconstruct it perfectly or purely from it's descendants. We can get close to Vulgar Latin, but there is a certain amount of error in our reconstruction. If we then take several reconstructions, and compare them, then we're just multiplying the error. We can use this to make a good estimate of the margin of error over time and comparisons.

And it's likely the error grows exponentially, right?

That would be assuming that all evidence of relationship is equally likely to be lost or muddled, which is not true. *m is more likely to stay put than *o is, a noun class system is more likely to stay put than a set of locative cases is, etc. A relationship 10K years old is going to be harder to recover than one 5K years old, but not squared-as-hard.

No, it would be assuming that any 'parent' reconstruction would have a margin of error equal to some multiplicative function of the error margins of the child reconstructions.

Oh wait, you're talking about the error in the reconstruction, not in whether we've found the correct cladogram…

But this is something we can actually test, and get real numbers for, so go ahead and prove us wrong.

We can? How many attested second-order protolangs there are?

Either way, I'm not proposing some actual number here, so you first, if you please.

[Åge Kruger]

But this is something we can actually test, and get real numbers for, so go ahead and prove us wrong.

We can? How many attested second-order protolangs there are?

None that I know of, but that wouldn't stop is from making a pretty realistic model - we're conlangers, people! We have the technology!

Either way, I'm not proposing some actual number here, so you first, if you please.

I'm game. I don't know how long it will take, but I'm game.

[TomHChappell]

Reconstruct proto-Nostratic, maybe? A start has been made.

Or, just take any two language-families for which the proto-langs have been reconstructed, and reconstruct their common ancestor. (Might this might be easier, though not necessarily more accurate, if you use more than two?)

For instance you could reconstruct proto-Vennemanian, the ancestor of the Celtic and Semitic languages.

If one or some of the protolangs don't have to be reconstructed -- e.g. one or more is/are already known -- this should mean there's even more opportunity.

Perhaps you should pick families that occur in New Guinea and reconstruct proto-Papuan; or maybe reconstruct proto-non-Pama-Nyungan (that hasn't already been done, has it?); or maybe proto-Caucasian.

Or proto-Australo-Pontic, the common ancestor of proto-Australian and the proto-Euxine language (the protolang of all the Black Sea languages).

[Miekko]

But this is something we can actually test, and get real numbers for, so go ahead and prove us wrong.

We can? How many attested second-order protolangs there are?

None that I know of, but that wouldn't stop is from making a pretty realistic model - we're conlangers, people! We have the technology!

Either way, I'm not proposing some actual number here, so you first, if you please.

I'm game. I don't know how long it will take, but I'm game.

One thing we could do here is:
- reconstruct Vulgar Latin from modern Romance languages
- reconstruct Latin from intermediate reconstructions
Compare the result with the real thing.

Using conlangs (or even some kind of mathematical model) would be easier, I guess, though.

[Tropylium⁺]

One thing we could do here is:
- reconstruct Vulgar Latin from modern Romance languages
- reconstruct Latin from intermediate reconstructions
Compare the result with the real thing.

Hasn't at least one of those been done?

Anyway, I thought the discussion was about determining relationships rather than determining the protolanguage. I could well imagine that errors in the latter do grow exponentially, but the situation is different in the latter, because we aren't interested in the absolute strength of the hypothesis (after all, it's well-recognized all reconstructions are approximations) but its strength relativ to other relationship hypotheses. This doesn't even decrease monotonically! For example, take Lithuanian:
— We're pretty sure Latvian is its closest relative.
— We're also somewhat sure that the Slavic languages are the next-closest relatives.
— We're not sure at all what's the next closest… Indo-Iranian? Greco-Aryan? Albanian? Germanic?
— However, we are again pretty sure that all of those are closer related to Lithuanian than, say, Turkic.
And even while I have little idea if Lithuanian (or IE in general) is closer related to Turkic than to Dravidian or Sino-Tibetan, I'm also reasonably sure it is closer related to all of those than it is to !Xóõ.

I did not use the comparativ method it to reach that last conclusion tho, but as far as that goes, the point is that the cutoff point where we can no long use its results to say much about the protolanguage is not the cutoff point where we can not longer say much about the cladogram. Suppose we're only 80% sure there is a good reconstruction of Proto-Penutian; if we're simultaneously 80% sure there isn't a good reconstruction of Proto-Algic-Utian, then that means we're 96% sure that Algic is not a branch of Penutian. Similarly, even a pretty crappy reconstruction of Proto-Nostratic will have implications in the absense of any kind of hints of, say, a Proto-Indo-Austronesian relationship.

It is actually a required assumption here that the strength of a reconstruction does decrease monotonically by age. Otherwise, the 16% odds for both the Proto-Penutian and Proto-Algic-Utian reconstructions being on the right track couldn't be counted as a case where we have evidence against the inclusion of Algic in Penutian. (Well, unless the reconstructions would be sufficiently close for this to become a case similar to the one of IE subgroups.)

Now if we want to put numbers on how fast exactly a reconstructions's usefulness tends to zero, we'll first want an idea how to quantify the error in a reconstruction. Do words that were there but we didn't manage to reconstruct count as errors? (Perhaps, but that might require weighing by frequency.) Does reconstructing *sudɪ rather than *tsudi count as less of an error than reconstructing *hudə? (I think it should — the first is only off by two features/sound changes, the latter by about four.) What about the semantics??

[TomHChappell]

.... the cutoff point where we can no long use its results to say much about the protolanguage is not the cutoff point where we can not longer say much about the cladogram.

QFT.

....
the strength of a reconstruction does decrease monotonically by age.
....

But there's no guarantee that the rate of decrease is comparable cross-linguistically nor that it's predictable given the age.
If those were both true, "glottochronology" would be something that could eventually be given a reliable basis.
But it's now commonly accepted that there can't be any such scientific "glottochronology".

------------

But, what if what's commonly accepted, happens to be wrong?

[Herra Ratatoskr]

One thing we could do here is:
- reconstruct Vulgar Latin from modern Romance languages
- reconstruct Latin from intermediate reconstructions
Compare the result with the real thing.

I also am vaguely thinking that this has been done. What I think would be interesting would be to try reconstructing language families based on only modern languages, i.e. construct Proto-Germanic only using modern Germanic languages, nothing older. Then try reconstructing vulgar Latin based on modern Romance languages, Proto-Slavic based only on modern Slavic languages, etc, and then try to reconstruct Indo-European based solely on those reconstructions. It would be a hell of a lot of work, and might not even be possible, but it might give us an insight into how hard reconstructing second order proto-langs might be. I think the hardest part would be to not "cheat", I would imagine, and allow faulty reconstructions to arise when the evidence might suggest it, or to let our knowledge help us in figuring out difficulties or recognizing cognates. For instance, lacking evidence from Gothic (since it's not modern), I would think it would be important to reconstruct the masculine nominative singular ending as "-r", instead of "-z", even though we "know" that the "-z" is what the proto-form actually had.

[Miekko]

One thing we could do here is:
- reconstruct Vulgar Latin from modern Romance languages
- reconstruct Latin from intermediate reconstructions
Compare the result with the real thing.

Hasn't at least one of those been done?

Yes, Latin has been reconstructed and it was fairly ok. But now the point is: make a number of intermediate reconstructions, then reconstruct latin from it, then compare those results with the result of the previous attempt. Check which one's worse.

[Tropylium⁺]

What I think would be interesting would be to try reconstructing language families based on only modern languages, i.e. construct Proto-Germanic only using modern Germanic languages, nothing older.

I do happen to have a project on those lines cooking. The fact that standard literary English/German/etc. draw from several dialects probably doesn't help, nor do the truckloads of Low German loans in Scandinavian. (Or, that to make things even harder than they have to be, I'm going to be ignoring Icelandic.)

the strength of a reconstruction does decrease monotonically by age.

But there's no guarantee that the rate of decrease is comparable cross-linguistically nor that it's predictable given the age.
If those were both true, "glottochronology" would be something that could eventually be given a reliable basis.

Hmm. Good point. That means that having a good reconstruction of Proto-X-Y and some hints for Proto-X-Z might also be explained by both X and Y being very conservativ and Z being wildly innovativ, in which case Z might be after all actually closer related to X than X is to Y.

But this seems to be a possible problem with reconstructions of *any* strength, not just "long-range" reconstructions. Besides I don't think we should assume conservativity or innovativity to be an invariant feature of a language branch; ie. that we should assume that over long stretches of time, all languages develop at the same average rate.

[Bedelato]

I don't really understand all the flak that the long-range comparers are getting.

Largely because of shoddy work by Ruhlen, Greenberg, and Starostin. It's tainted the field. Also there's the bizarre idea floating around that you can't demonstrate family relationships beyong 8000 years (sometimes this number is higher or lower). Oddly, this number is derived from lexicostatistics, a la Swadesh, which has long been discredited. The meme remains powerful though.

Out of those three, Greenberg's the only one I've heard much about. If the problems with him are limited to that "mass comparison" thing, I don't see what the deal is; mass comparison is just the cognate-collecting phase of the comparative method, and once you've collected the data you can analyze it as normal, looking for sound correspondences and all that. Or is there some other reason why people hate him so much?

Yes, lexicostatistics, glottochronology, etc. as methods for reconstruction were proven dead wrong, but the basic idea – that the lexicon gradually changes over time – is valid. Like I said before, errors multiply with time, which creates a limit beyond which we can't know anything. This limit might be more than 8,000 years, but it does exist.

Other than the shortage of material to work with, I see nothing inherently flawed with the idea of long-range comparison. In theory, you should be able to apply the comparative method indefinitely and end up at the beginning.

The data isn't really all that scarce though. Part of the problem is that there is no standard by which we can say that a relationship has been demonstrated. It's easy to dismiss something as being unconvincing when the goalposts can be so easily moved.

What happened to finding regular sound correspondences? By "shortage of material" I meant that the further back in time you go, the less abundant the data is.

But, real life isn't theory. Random, sporadic mutations, which the comparative method can't deal with, happen all the time, and we can only recover so much of those. The rest are gone forever. Over time these build up and obscure relationships even further.

Sometimes words are lost in all daughter languages which also creates a limit.

Yes.

Because of this little bit of non-ideality, the comparative method has the chronological limits that it does, and going beyond 5000 years or so is a stretch. That's not to say we shouldn't try, though. Just don't get your hopes up.

The problem is that 5000 is an arbitrary number. For all we know the limit is 50,000 years.

Take a look at the first part of this post.

[zompist]

Suppose we're only 80% sure there is a good reconstruction of Proto-Penutian; if we're simultaneously 80% sure there isn't a good reconstruction of Proto-Algic-Utian, then that means we're 96% sure that Algic is not a branch of Penutian. Similarly, even a pretty crappy reconstruction of Proto-Nostratic will have implications in the absense of any kind of hints of, say, a Proto-Indo-Austronesian relationship.

If you are applying two independent probabilities, you need to multiply them. So the hypothesis that depends on two 80% probabilities is only 64% likely.

[zompist]

Out of those three, Greenberg's the only one I've heard much about. If the problems with him are limited to that "mass comparison" thing, I don't see what the deal is; mass comparison is just the cognate-collecting phase of the comparative method, and once you've collected the data you can analyze it as normal, looking for sound correspondences and all that. Or is there some other reason why people hate him so much?

Many reasons, but the chief one is that Greenberg, Ruhlen, and their defenders don't take the above view— that mass comparison is a useful preliminary to the comparative method— but assert that it is enough in itself to establish "accurate" (Ruhlen's term) relationships. Ruhlen 1987 even calls the insistence on using the comparative method a "defect" of Amerindianists.

[Richard W]

List me three languages or language groups that have a sound change affecting original word-initial *m in any way? Or original intervocalic *m, for that matter.

While I actually agree with you, how about:

1) Insular Celtic's mutations and intervocalic lenition, e.g Welsh Rhufain for Rome.

2) The absence of proto-Basque /m/ - it seems unlikely that no ancestral language ever had it.

3) In the Qiaoshang 'dialect' of Gelao, Proto-Gelao initial *m has become /mb/, and the corresponding voiceless (or whatever) labial nasal has become a prenasalised velar or post-velar.

[Richard W]

Or, just take any two language-families for which the proto-langs have been reconstructed, and reconstruct their common ancestor. (Might this might be easier, though not necessarily more accurate, if you use more than two?)

If you take four or more languages, matches in three languages are unlikely to be chance. Of course, they could be other corruption - common borrowing or a correlation between sound and meaning,

[Count Iblis]

Out of those three, Greenberg's the only one I've heard much about. If the problems with him are limited to that "mass comparison" thing, I don't see what the deal is; mass comparison is just the cognate-collecting phase of the comparative method, and once you've collected the data you can analyze it as normal, looking for sound correspondences and all that. Or is there some other reason why people hate him so much?

He's actually done some good work with language typology, so it's not like he was a crank or something. I think Greenberg himself said that mass comparison was just supposed to be a preliminary step to the comparative method. The problem is that Greenberg and Ruhlen never went beyond that preliminary step before declaring languages were related. They never attempted to establish sound correspondences, weed out borrowings and chance resemblances, or eliminate sound symbolic words.

Yes, lexicostatistics, glottochronology, etc. as methods for reconstruction were proven dead wrong, but the basic idea – that the lexicon gradually changes over time – is valid. Like I said before, errors multiply with time, which creates a limit beyond which we can't know anything. This limit might be more than 8,000 years, but it does exist.

I agree that there must be some limit it's just that no one has been able to establish what it is. Instead random numbers are tossed about but said with authority.

What happened to finding regular sound correspondences? By "shortage of material" I meant that the further back in time you go, the less abundant the data is.

The problem is how do we know when a set of correspondences have been demonstrated? There is no standard, although I've heard 5 examples as a rule of thumb. Several months ago I started on a reconstruction of Proto-Finno-Bantu, the most recent common ancestor of modern Finnish and Proto-Bantu. Despite the fact that they probably aren't even related I was able to come up with plenty of correspondences. The methodology was pretty simple actually. Let's say I had a word with Finnish p and Proto-Bantu *k. In that case I reconstruct *kw. The reverse correspondence might exist in other words, so I reconstructed a labial-velar *kp. For Finnish p Bantu *d I reconstructed *dw. So if language A has X consonants and language B has Y consonants then Proto-AB is going to have X*Y consonants. That number could even be trimmed down by assuming that consonant clusters get reduced to single phonemes and by ignoring words that have a good semantic match but a bad phonetic match.That would be a bit like saying that English dog and German hund have an excellent semantic match, but they're not cognate which allows us to avoid having to come up with a proto-phoneme for English d and German h.

This is a cool methodology for coming up with crazy conlangs like Proto-Finno-Bantu, but it also shows that any pair of languages can be proven to be related with regular sound correspondences (of course it quickly becomes unwieldy with more than 2 languages since you could have proto-langs with thousands of consonants).

[Herra Ratatoskr]

What I think would be interesting would be to try reconstructing language families based on only modern languages, i.e. construct Proto-Germanic only using modern Germanic languages, nothing older.

I do happen to have a project on those lines cooking. The fact that standard literary English/German/etc. draw from several dialects probably doesn't help, nor do the truckloads of Low German loans in Scandinavian. (Or, that to make things even harder than they have to be, I'm going to be ignoring Icelandic.)

You sick, masochistic bastard! Tell me you're at least going to be using Faroese. Or at least some of the conservative mainland Scandinavian dialects. Please?

And I take it this will all be done phonetically, so no help from spelling?

[Bristel]

Someone should do a proto-American project with me.

I want to figure out if I can reconstruct proto-Haida-Salishan-Penutian-Dene... lol

[Tropylium⁺]

What I think would be interesting would be to try reconstructing language families based on only modern languages, i.e. construct Proto-Germanic only using modern Germanic languages, nothing older.

I do happen to have a project on those lines cooking. The fact that standard literary English/German/etc. draw from several dialects probably doesn't help, nor do the truckloads of Low German loans in Scandinavian. (Or, that to make things even harder than they have to be, I'm going to be ignoring Icelandic.)

You sick, masochistic bastard! Tell me you're at least going to be using Faroese. Or at least some of the conservative mainland Scandinavian dialects. Please?

Still debating that internally. Perhaps, if I can find a good phonetical source. But the basic point still is to mostly work with those Germanic languages that have undergone major loss of 2nd-syllable distinctions (which is most of them).

And I take it this will all be done phonetically, so no help from spelling?

Certainly. Even descrambling the English trap-bath-split has been bit of a pain so far.

[zompist]

List me three languages or language groups that have a sound change affecting original word-initial *m in any way? Or original intervocalic *m, for that matter.

How about seventeen?

Greek *hemei > heīs
Cushitic *lamo > Somali lába-di
Bantu *-mó > Duala -wó, Fang fokh, Zulu nye...
Austronesian *lima > Barakai len, Marau riŋ
Mixe-Zoquean cf. Sayula máktašp, Texixtepec bacsná
Uto-Aztecan *simay- > Koso cewite, Kawaiisu suu-yu,Yaqui sénu...
Old Chinese *moh > wù, *m-lat > shé
Latin comite > Sp/Pt conde, domina > It. donna
Sanskrit kamala > NIAr. kaβal
Proto-Germanic *heman > English heaven, *emna > even

Many of these are from my numbers list, and I could probably add many more if I had more protolanguages. (There are many families where initial m predominates, and there are exceptions, but I don't know which is original.)

But actually I'm inclined to agree that m is more stable. Austronesian m in *lima is incredibly persistent.

Nonetheless, I think my original question hasn't been answered (and perhaps there is no answer). If m is more stable, why doesn't it accumulate in frequency over time? But so far as I know, there are no known tendencies for certain sounds to be more prevalent today than (say) 4000 years ago.

[jmcd]

Nonetheless, I think my original question hasn't been answered (and perhaps there is no answer). If m is more stable, why doesn't it accumulate in frequency over time? But so far as I know, there are no known tendencies for certain sounds to be more prevalent today than (say) 4000 years ago.

Probably because sound changes going to m are about as rare/common as those going from m.

[Bedelato]

The problem is how do we know when a set of correspondences have been demonstrated? There is no standard, although I've heard 5 examples as a rule of thumb. Several months ago I started on a reconstruction of Proto-Finno-Bantu, the most recent common ancestor of modern Finnish and Proto-Bantu. Despite the fact that they probably aren't even related I was able to come up with plenty of correspondences. The methodology was pretty simple actually. Let's say I had a word with Finnish p and Proto-Bantu *k. In that case I reconstruct *kw. The reverse correspondence might exist in other words, so I reconstructed a labial-velar *kp. For Finnish p Bantu *d I reconstructed *dw. So if language A has X consonants and language B has Y consonants then Proto-AB is going to have X*Y consonants. That number could even be trimmed down by assuming that consonant clusters get reduced to single phonemes and by ignoring words that have a good semantic match but a bad phonetic match.That would be a bit like saying that English dog and German hund have an excellent semantic match, but they're not cognate which allows us to avoid having to come up with a proto-phoneme for English d and German h.

This is a cool methodology for coming up with crazy conlangs like Proto-Finno-Bantu, but it also shows that any pair of languages can be proven to be related with regular sound correspondences (of course it quickly becomes unwieldy with more than 2 languages since you could have proto-langs with thousands of consonants).

Then you have to check how plausible the relationship actually is, based on things like how many "cognates" you have (if there's only two or three phono-semantic matches and the rest are clearly unrelated, then there's probably no relationship) and what your correspondences are (if a phoneme in A has a match with every phoneme in B, then there's probably no relationship, but if there's less than five unique correspondence sets, then there's a pretty good chance of a common ancestor. And of course if /aː/ corresponds to /dʒ/, then you have a problem.) Heuristics play a really big part in reconstruction.

[Travis B.]

Nonetheless, I think my original question hasn't been answered (and perhaps there is no answer). If m is more stable, why doesn't it accumulate in frequency over time? But so far as I know, there are no known tendencies for certain sounds to be more prevalent today than (say) 4000 years ago.

Probably because sound changes going to m are about as rare/common as those going from m.

The primary sorts of sound changes to /m/ I can think of are consonant cluster assimilations, primarily involving /n/ and an obstruent.

[merijn]

About Greenberg and the African phyla:
Even when people don't accept Niger-Congo, they usually do accept that the languages most related to Bantu are spoken in Nigeria and are (some) Benue-Congo languages. In other words, where they disagree with Greenberg, they also disagree with his predecessors (the idea that all West-African languages were related to each other was the consensus before Greenberg) and they accept his innovation (that Bantu is most closely related to some Nigerian languages. Before Greenberg people believed that either Bantu languages were not related to other African languages at all or that it was a sister-group to all West-African languages, and West-African languages were more related to each other than any was to Bantu).
I also disagree that the only reliable way to show that languages are related to each other is by regular sound changes. Sometimes it is obvious that languages are related to each other even before somebody tried to find out sound changes. I don't think anybody doubted that the Germanic languages were related to each other before the advent of the idea of the regular sound change. Similarly, I imagine that it took quite a while to discover all the sound changes that make up Indo-European, but most linguists excepted it from the get-go based on some of the obvious shared lexicon. Going back to Africa, to me it is clear that the Arabic prefix-conjugation and its counterparts in other Afro-Asiatic languages come from one source, and that to me is a clear evidence that they are related, even though it is a long way from regular sound changes and a Proto-Afro-Asiatic languages. This is to a lesser extent also true for the noun-class systems of Niger-Congo (with the possible exception of the noun-class systems of the Atlantic languages). Of course, it could all be coincidence, or they could have borrowed the noun-class system/prefix conjugation, but that they are related is by far the easiest and most likely explanation.

[Yng]

Please, it's 'accept', not 'except'.