Paleo-European languages

[Nortaneous]

Celtic *longā 'boat, vessel' (but 'cinerary urn' in Cisalpine Gaulish lokan /longan/) is a loanword related to Caucasian *lĕqˀV̆ 'a kind of vessel'. Latin lanx 'dish' also comes from this etymology, presumably through Etruscan.

The Celtic word means 'ship'; the Caucasian word typically means a holder for food or liquid.

Like a gravy boat?

[Talskubilos]

Yes, I agree that this is the most solid of your correspondences. But metathesis is common enough that, while suspicious, I don’t see this coincidence as enough evidence for a genealogical connection.

Mmm. I don't think this is exactly "genealogical" in the traditional sense, but it's still a connection.

[Znex]

Celtic *longā 'boat, vessel' (but 'cinerary urn' in Cisalpine Gaulish lokan /longan/) is a loanword related to Caucasian *lĕqˀV̆ 'a kind of vessel'. Latin lanx 'dish' also comes from this etymology, presumably through Etruscan.

The Celtic word means 'ship'; the Caucasian word typically means a holder for food or liquid.

Like a gravy boat?

Quite a few "boat" words have similar usage, eg. Celtic *ɸlexstros is used both to mean "sea vessel" and "cup vessel".

As for the supposed loanword, *longā can very simply be linked to other IE languages; Latin for instance has navis longa as a frequent collocation, as does Germanic with langskip. These alternatives are considerably more attractive than a supposed Caucasian origin, which suggests a far more complex journey (obvious questions to my mind: How was it received into Celtic? Why does it only remain in Celtic? Where does the nasal and vowel come from? How/why is this etymology more attractive than the traditional etymology, apart from its novelty?).

[Talskubilos]

As for the supposed loanword, *longā can very simply be linked to other IE languages; Latin for instance has navis longa as a frequent collocation, as does Germanic with langskip. These alternatives are considerably more attractive than a supposed Caucasian origin, which suggests a far more complex journey (obvious questions to my mind: How was it received into Celtic? Why does it only remain in Celtic? Where does the nasal and vowel come from? How/why is this etymology more attractive than the traditional etymology, apart from its novelty?).

How about Latin lanx 'dish'?

[Znex]

How about Latin lanx 'dish'?

I don't know about the Latin word, but the stem is rather different from the Celtic to make me think the two are related (*long- vs. *lank-), and there are more plausible links elsewhere, eg. to Greek λέκος lékos and λεκάνη lekánē (both of which mean dish or basin too). Greek is errhinophobic, so the correspondence lank - lek is plausible.

Now the Greek word on the other hand does indeed become more similar to the Caucasian, but whether the nasal in the Latin stem can be explained or not, I'd still hesitate to add the Celtic word as a cognate here.

[Talskubilos]

How about Latin lanx 'dish'?

I don't know about the Latin word, but the stem is rather different from the Celtic to make me think the two are related (*long- vs. *lank-),

But if lanx were a loanword from Etruscan, which has no o nor voiced stops, they would match nicely.

and there are more plausible links elsewhere, eg. to Greek λέκος lékos and λεκάνη lekánē (both of which mean dish or basin too). Greek is errhinophobic, so the correspondence lank - lek is plausible. Now the Greek word on the other hand does indeed become more similar to the Caucasian, but whether the nasal in the Latin stem can be explained or not, I'd still hesitate to add the Celtic word as a cognate here.

I didn't know the Greek word, so thank you for the information. Being a Pre-Greek substrate loanword would explain its similarity to the Caucasian protoform, including the voiceless stop k corresponding to glottalic *qˀ in Caucasian.

[Travis B.]

How about Latin lanx 'dish'?

I don't know about the Latin word, but the stem is rather different from the Celtic to make me think the two are related (*long- vs. *lank-),

But if lanx were a loanword from Etruscan, which has no o nor voiced stops, they would match nicely.

and there are more plausible links elsewhere, eg. to Greek λέκος lékos and λεκάνη lekánē (both of which mean dish or basin too). Greek is errhinophobic, so the correspondence lank - lek is plausible. Now the Greek word on the other hand does indeed become more similar to the Caucasian, but whether the nasal in the Latin stem can be explained or not, I'd still hesitate to add the Celtic word as a cognate here.

I didn't know the Greek word, so thank you for the information. Being a Pre-Greek substrate loanword would explain its similarity to the Caucasian protoform, including the voiceless stop k corresponding to glottalic *qˀ in Caucasian.

The question I have here though is were this to be borrowed in this form into Greek from some pre-Greek substate, how did Latin and Celtic end up with a nasal, unless one were to show that this was lent with a nasal into Greek, and then lent from Greek into Latin and Celtic, and only afterwards the nasal was lost in Greek (but that implies that one should be able to see nasal loss as it occurred in Greek, as contact between Greek and Latin occurred in historic time) and the nasal was also either lost so completely in Caucasian that it does not appear in any speculative reconstructions or was lost in some other language before this word was borrowed into some Caucasian language.

[Talskubilos]

The question I have here though is were this to be borrowed in this form into Greek from some pre-Greek substate, how did Latin and Celtic end up with a nasal, unless one were to show that this was lent with a nasal into Greek, and then lent from Greek into Latin and Celtic, and only afterwards the nasal was lost in Greek (but that implies that one should be able to see nasal loss as it occurred in Greek, as contact between Greek and Latin occurred in historic time) and the nasal was also either lost so completely in Caucasian that it does not appear in any speculative reconstructions or was lost in some other language before this word was borrowed into some Caucasian language.

As Celtic and Latin (presumably by intermediation of Etruscan) have *long- and Greek has *lek-, it's quite clear they must be independent borrowings.

[Zju]

The proposed correspondence *Hnēttsˀwǝ̄ (the updated protoform in the NCED) ~ IE *ned-o- leaves us with *ttsˀ ~ IE *d (apparently, there're no traces of the labial glide /w/), which makes sense in the framework of the glottalic theory. If this is correct, the metathesized variant *ttsˀwǝ̄nHē would correspond to IE *don-. However, regional IE *yoini- (presumably a substrate loanword) would derive from the same etymology but with different sound correspondences: the initial affricate gave /y-/ (as in other words) and apparently the labial glide gave /o/.

Yeah... I'm not buying it. Seems much more likely to be just a chance resemblance. At least one or two dozen more instances of the same sound correspondances are necessary to be worth considered as anything more than that.

I disagree. Chance resemblances are quite a different thing.

Apparently, there're no more examples of *ttsˀ ~ IE *d but I could find one with *s instead: Caucasian *ttsˀăqˀV 'strength, power' ~ IE *seģh- 'to hold'.

Neither phonetics nor semantics match here, not even parts of speech match. This is not even a resemblance.

Not as sure as the fomer, but quite reasonable IMHO. Also remember that /e/ is one of the IE Ablaut vowels, so it doesn't count for external comparison.

Chance resemblances are quite this very thing - phonetics and semantics only vaguely similar to each other, with barely any more example examples of the same correspondance. We'd expect one or two dozen chance resemblances between any two languages. Wildcard sounds make matters even worse, as they allow for more chance resemblances.

...
I've seen the results of computational studies of reticulation on Indo-European, and remarkably enough, 'round' and 'mountain' don't lead to a deduction of hybridisation between English and French, and I see no sign of it in the reported outputs. (These meanings don't hold onto their words well.) Computational studies come up with English being distinct from the rest of West Germanic, and reinterpret the Anglo-Frisian group as a hybridisation event between English and Dutch yielding Frisian. What shows up in phylogenetic networks is the Scandinavianness of English within West Germanic.
...
How does creolisation, as opposed to pidginisation, affect genealogy?

Using biology software to make conclusions on linguistic phenomena? I'm not sold to the whole argument then. Biological and linguistic clades are superficially similar, but differ in the details. Languages don't have DNA.

Creoles appear as the language of a generation that grows up listening to a pidgin - a haphazard combination of two languages, with the speakers of neither knowing the other. Creoles have their own proper grammar and lexicon, which differ from both of the parent languages. Creolisation is one-off, 'instant' process and creoles are generally considered to have two parent languages.

OK, so how do you know that your correspondences are not spurious, i.e. are regular and not just chance resemblances? The usual way of doing this is by finding enough cognates that we know they must be regular, which is why I am placing such an emphasis on quantity.

Well, I think the case of the 'reed' words is solid enough. On the one hand, we've got a Caucasian protoform with two metathesized variants, and on the other, two IE protoforms with consistent sound correspondences.

I don't think so.

In that case, I am forced to consider this relationship ‘very unproven’, if not completely non-existent. Please inform us when you have enough regular correspondences to show that these aren’t just chance resemblances.

Quality is always preferrable to quantity. I'm sure we don't need a complete skeleton to identify an extinct hominid species, for example.

Not necessarily. Specifically, at least 10-12 examples of sound correspondances between seemingly unrelated languages are called for before we even begin entertaining the idea that there's something regular about them. Anything less than that and they may as well be chance resemblances.

[WeepingElf]

Using biology software to make conclusions on linguistic phenomena? I'm not sold to the whole argument then. Biological and linguistic clades are superficially similar, but differ in the details. Languages don't have DNA.

Amen. And the mechanisms of change in languages are very different from DNA mutations. Words do not "mutate" randomly, they are affected by regular sound changes. Also, innovations spread laterally, at least among related languages, in a way that is totally alien to DNA. These differences make the application of algorithms that compute clades from DNA mutations to languages meaningless.

[Talskubilos]

Chance resemblances are quite this very thing - phonetics and semantics only vaguely similar to each other, with barely any more example examples of the same correspondance.

I strongly disagree.

We'd expect one or two dozen chance resemblances between any two languages.

Except we aren't comparing languages but reconstructed protoforms, i.e. protolanguages. Of course, this doesn't imply the +2000 items reconstructed for PIE actually belonged to a real language.

Not necessarily. Specifically, at least 10-12 examples of sound correspondances between seemingly unrelated languages are called for before we even begin entertaining the idea that there's something regular about them. Anything less than that and they may as well be chance resemblances.

Unfortunately, we're studying the remainader of unattested (i.e. with no written records) languages, so data available is more scarce. This means we can't use the same methology than with e.g. Latin or Sanskrit.

[WeepingElf]

Chance resemblances are quite this very thing - phonetics and semantics only vaguely similar to each other, with barely any more example examples of the same correspondance.

I strongly disagree.

We know; you have said so multiple times. But you haven't yet given a valid reason for your disagreement.

We'd expect one or two dozen chance resemblances between any two languages.

Except we aren't comparing languages but reconstructed protoforms, i.e. protolanguages. Of course, this doesn't imply the +2000 items reconstructed for PIE actually belonged to a real language.

Protolanguages aren't languages?!? You are making a fool of yourself. Of course, protolanguages are less well known than living languages (where we can go and ask speakers) or languages known from large text corpora like Greek or Latin. But they are languages, though incompletely known ones. At least, they are models of languages of the past. Not perfectly accurate, but reasonably close if the evidence is good and the methods are sound. Indeed, PIE is known better than some scarcely attested historical languages such as Hattic or Etruscan!

Not necessarily. Specifically, at least 10-12 examples of sound correspondances between seemingly unrelated languages are called for before we even begin entertaining the idea that there's something regular about them. Anything less than that and they may as well be chance resemblances.

Unfortunately, we're studying the remainader of unattested (i.e. with no written records) languages, so data available is more scarce. This means we can't use the same methology than with e.g. Latin or Sanskrit.

So because we have fewer data, we should use a less rigorous method? That's nonsense. Of course, you will get more positives that way - but they are false positives, making the whole endeavour worthless. This is the kind of fallacious scholarship Samuel Johnson once characterized as "milking the bull" because "the cow [...] does not give enough milk".

[KathTheDragon]

I hereby reiterate that you will have a far better time if you stop engaging Octaviano.

[Talskubilos]

We know; you have said so multiple times. But you haven't yet given a valid reason for your disagreement.

I quoted an example where both sound correspondences and semantics are good, so Zju's point was a strawman.

Protolanguages aren't languages?!? You are making a fool of yourself. Of course, protolanguages are less well known than living languages (where we can go and ask speakers) or languages known from large text corpora like Greek or Latin. But they are languages, though incompletely known ones. At least, they are models of languages of the past. Not perfectly accurate, but reasonably close if the evidence is good and the methods are sound. Indeed, PIE is known better than some scarcely attested historical languages such as Hattic or Etruscan!

I disagree. The reconstructed PIE, with more +2000 lexical items, could never be anything like a real language!

So because we have fewer data, we should use a less rigorous method? That's nonsense.

I never said that.

Of course, you will get more positives that way - but they are false positives, making the whole endeavour worthless.

As I mentioned before, quality doesn't depend on quantity, and the likelihood of false positives (a better term than "chance resemblances" in this context) actually raises when more data is available.

This is the kind of fallacious scholarship Samuel Johnson once characterized as "milking the bull" because "the cow [...] does not give enough milk".

If you try to milk the bull, you won't get milk but a quite different thing.

[bradrn]

Of course, you will get more positives that way - but they are false positives, making the whole endeavour worthless.

As I mentioned before, quality doesn't depend on quantity, and the likelihood of false positives (a better term than "chance resemblances" in this context) actually raises when more data is available.

OK, this is an interesting argument, so let’s look at it in a bit more detail. The two main variables involved here are sample size (quantity) and methodology (which I presume is what you mean by ‘quality’). Firstly, I do agree with you that the false positive rate rises with sample size — it is a trivial conclusion that the number of chance resemblances increases with the number of words you compare. But this is neglecting the impact of methodology. It is obvious that, no matter how large or small your sample size is, a sloppy methodology will greatly increase the number of false positives you see — and indeed, an analysis suggests that this factor is as large as that of sample size, if not far larger. In this sense, I do agree with you that ‘Quality is always preferrable to quantity’; that is, an analysis of many words with rock-solid methodology is preferable to an analysis of few words with bad methodology. (This, incidentally, is why Greenberg’s proposals never were accepted — he analysed extremely many words with extremely bad methodology, thus failing on both counts.)

So the question becomes: what, exactly, constitutes a ‘bad methodology’? As I see it, there are three things which can cause this: allowing phonetic leeway, by being sloppy with which sounds correspond to each other; allowing semantic leeway, by being sloppy with what meanings correspond to each other; and allowing irregularity, by forgetting that sounds must correspond to each other regularly. As it happens, your methodology is in fact pretty good, compared to people like Greenberg: despite what others are saying here, I do think you have reasonable phonetic and semantic correspondences. However, I am sceptical of your sound correspondences, or lack thereof. And, as a most important component of good methodology, I would say that this sloppiness is enough for me to reject your proposals — at least, until you find some sound correspondences.

I hereby reiterate that you will have a far better time if you stop engaging Octaviano.

[Richard W]

OK, this is an interesting argument, so let’s look at it in a bit more detail. The two main variables involved here are sample size (quantity) and methodology (which I presume is what you mean by ‘quality’).

Is sampling happening here, or do we just have examples? When one uses lists such as Swadesh lists, one is using meanings that tend to keep their principle words. In origin, this was a coincidence. The original idea was to use meanings for which every language would have a word; in part I think it was meant to work with vocabulary lists, for full-blown dictionaries are a luxury item. When working with such lists, the better the quality.

I'm not sure what the 'sampling' process for Tavi's examples is. I suspect it is more a matter of going through an entire documented vocabulary and looking for cognates and picking out what look like good matches - probably many years' work.

[bradrn]

OK, this is an interesting argument, so let’s look at it in a bit more detail. The two main variables involved here are sample size (quantity) and methodology (which I presume is what you mean by ‘quality’).

Is sampling happening here, or do we just have examples? When one uses lists such as Swadesh lists, one is using meanings that tend to keep their principle words. In origin, this was a coincidence. The original idea was to use meanings for which every language would have a word; in part I think it was meant to work with vocabulary lists, for full-blown dictionaries are a luxury item. When working with such lists, the better the quality.

I'm not sure what the 'sampling' process for Tavi's examples is. I suspect it is more a matter of going through an entire documented vocabulary and looking for cognates and picking out what look like good matches - probably many years' work.

By ‘sample size’, I simply meant the number of words which were inspected to find cognates. For instance, ASJP uses a sample size of 40, the Swadesh lists give 100–200, and proper comparative techniques usually use a dictionary (with many thousands of words).

[Richard W]

By ‘sample size’, I simply meant the number of words which were inspected to find cognates. For instance, ASJP uses a sample size of 40, the Swadesh lists give 100–200, and proper comparative techniques usually use a dictionary (with many thousands of words).

And generally, the shorter the list, the better the quality of each match. Huge dictionaries are generally very bad for the average quality of the comparisons. Loanwords, of course, are especially difficult, as the semantics are likely to be significantly changed in the course of borrowing.

[bradrn]

By ‘sample size’, I simply meant the number of words which were inspected to find cognates. For instance, ASJP uses a sample size of 40, the Swadesh lists give 100–200, and proper comparative techniques usually use a dictionary (with many thousands of words).

And generally, the shorter the list, the better the quality of each match. Huge dictionaries are generally very bad for the average quality of the comparisons. Loanwords, of course, are especially difficult, as the semantics are likely to be significantly changed in the course of borrowing.

I did already mention this:

… I do agree with you that the false positive rate rises with sample size — it is a trivial conclusion that the number of chance resemblances increases with the number of words you compare.

(Though it should be noted that this advantage of shorter lists disappears when you make your lists too short: the shorter the list is, the harder it is to apply proper methodologies, since you can no longer confirm sound correspondences nor account for semantic shifts.)

[Richard W]

I did already mention this:

… I do agree with you that the false positive rate rises with sample size — it is a trivial conclusion that the number of chance resemblances increases with the number of words you compare.

It is not a trivial consequence that the proportion increases. As you lengthen the lists, the proportion of words that do correspond historically goes down. Conversely, if you sampled words randomly from a fixed vocabulary, the proportion of false matches would not increase as you increased sample sizes. However, if you enlarge the vocabulary included from the other language, the proportion of false matches will increase.

(Though it should be noted that this advantage of shorter lists disappears when you make your lists too short: the shorter the list is, the harder it is to apply proper methodologies, since you can no longer confirm sound correspondences nor account for semantic shifts.)

The Swadesh lists and the like have very little use when comparing unrelated languages and looking for loanwords.