#natlangs

[glossopoesis]

False Friends #1 - A Gallic Gaul or a Gael?

This is going to be the first post in a series of words you’d really think are related, but in fact are not

There are lots of Celtic peoples and place names. Gael, Gallia, Gaul, Galicia, Galatia, hey, wait a minute, why all these Ga(?)l names? Those have to be related right? Well, most of these are actually unrelated, coming from at least three different sources

Gael doesn’t actually coming from the Gaelic languages, but originally from Old Welsh. It first appears in Irish as Goídel meaning “an Irishman” (predating the large-scale Gaelic migration to Scotland) which is an apparent loan of the Old Welsh Guoidel appearing as a personal name and a common noun meaning “warrior”, “wild man”, or “raider”, apparently deriving from a word for “woods”, “forest”, or “wilderness”. The Old Welsh comes from a reconstructed Proto-Celtic *wēdelos and so no initial stop would be expected in the Goidelic or Gaulish languages 

Gallia was part of the name of several Roman provinces, covering an area broadly synonymous with Modern France. It is similar to many Latin place names in that it is formed from the name of a people group (the Gallī) with the suffix -ia. Unfortunately not much more can be said although it appears to have been borrowed from Gaulish, perhaps related to a Proto-Celtic root *gal “to be able” (cf Modern Welsh “gallu” of the same meaning) with the people name meaning “those who are able”

Gallia is usually referred to as Gaul and many people assume that Gaul is just the French reflex, but this isn’t actually the case; the French reflex of Gallia would be Jaille (which is attested in some placenames such as “La Jaille-Yvon”)

Gaul has actually had a very long journey, from Old Frankish via French, and the same root as Modern English Wales. This root is reconstructed as *walhaz “foreigner”, “Celt”, “Roman” in Proto-Germanic, apparently a result of a very early borrowing the Celtic tribal name Volcae, itself usually believed to be from a root *wolkos “hawk” present in personal names like Gaulish Catuvolcus) . The age of this borrowing is shown by the fact that the *k has undergone Grimm’s Law. The g in Gaul is not unexpected, but actually the regular reflex of Old Frankish *w (cf guerre “war” from Old Frankish werra, cognate with the Modern English war)

Galicia & Galatia are less clear, the former appears in Latin as Gallaecī; the latter appears in Greek as Γᾰλᾰτῐ́ᾱ (Galatíā). Both are apparently borrowings from the respective Celtic languages and so assumed to be from the same root as Gallia (Proto-Celtic *gal “to be able”)

There’s also the word Celt, from Greek Κελτός (Keltós) which might also be from Proto-Celtic *gal, making it almost identical to the form giving rise to Galatia. Why the difference in the initial consonant though? Well, Greek had a three-way phonation distinction in its stops (between voiced, tenuis, & aspirated), whilst the Celtic languages had merged the Proto-Indo-European breathy-voiced stops into the voiced stops so had a two-way phonation distinction (between voiced and voiceless stops); if the Celtic voiced stops were only lightly voiced (or potentially tenuis, with the “voiceless” stops aspirated), it’s natural that the Greeks may have perceived it to have been intermediate between their γ & κ sounds and so spelt it inconsistently. This may have also reflected differences between the pronunciations in the various Celtic varieties

So Gael and Gaul are entirely unrelated from the other words, although the many other words (Gallia, Galatia, Galicia, & Celt) probably come from a single root, originally “to be able”

[glossopoesis]

Informational Linguistics

I’ve mentioned this a few times now, both on this blog, and on my non-linguistics blog before I made this so I thought I should go into a bit more detail about what I mean by it.

Essentially, I mean the application of information theory to problems in linguistics; that is approaching such problems from an information theoretic perspective. For this, the most important concept we need to introduce is that of the self-information Is of a segment s:

That is, the information in bits (i.e. how many yes/no questions it would take a sensible person to correctly guess the segment) is equal to the power one needs to raise 2 to, in order to get the (proportional) frequency of that segment. The - sign just ensures that the information is positive. 

This graph shoes the self-information on the vertical axis, against the proportional frequency on the horizontal axis. This can be thought of intuitively as quantifying the notion that infrequent segments are more specific than frequent ones. I.e. "crimson” is used less frequently than “red” but conveys a more precise colour and so contains more information. 

We can also calculate the entropy H of a system of segments which is just the frequency-weighted average of the self-informations of all segments

So far, this is just information theory with nothing linguistic anywhere so let’s look at a more concrete example. Spanish has an orthography that is unambiguous going from writing to pronunciation which is nice, so based on a translation of the first five books of the bible I found online (not an especially representative text so obviously the specifics may not be entirely accurate to spoken Spanish) the phoneme frequencies (as a percentage), self-information (in bits) and the entropy of these phonemes (based on the standard peninsular Spanish I was taught in school) is:

So 8.58% of all phonemes are /s/ but only 0.18% are /ɲ/ (<ñ>) so an /ɲ/ conveys more 5 extra bits of information

Of course, this is just the information conveyed by a phoneme in isolation without additional structure. In actual language, if we know the preceding segments, we can guess the next segment much better than the odds above would suggest. This is because of redundancy

So far, we’ve talked about how much information each segment conveys, but the information conveyed by each segment is not necessarily independent and, in fact, is extremely redundant and most of the information conveyed by any given segment is also carried by other segments. Some of this redundancy occurs due to phonotactics (i.e. in Hawai’ian, if the most recent phoneme is a consonant, we can be certain the next phoneme will be a vowel) and other redundancy comes from grammar, especially agreement (e.g. if I’ve heard “los hermanos son alt...” so far, I can be almost certain this will continue “...os”)

We can then calculate the bit rate Rb (i.e. the rate at which information in bits is transferred) from the entropy H (in bits per segment), the redundancy r (actually getting a value for this is not at all trivial), and the speed of speech c (in segments per unit time):

When we control for speaker-fluency, background-noise, formality etc, this quantity is independent of the language (presumably the fastest reliably producible and interpretable by the human tongue and ear respectively). I.e. on average, it takes the same amount of time to say the same thing regardless of language

Combined with the fact there are some physiological and mental limits on these quantities, this can put some restrictions on the sort of linguistic changes that are possible and, potentially, argue for certain changes being preferred

It’s also worth noting that this analysis has to be applied in two directions, once by the speaker, and once by the listener. In the first case, the relevant segments are phonological as they describe how the speaker produces the utterance, but in the second case, the relevant segments are the phonetic segments the ear hears. Of course, as people can usually only reliably tell the difference between sounds they can distinguish in their own speech, it seems likely the brain converts this phonetic information into phonological information before actually analysing it; this provides an extra opportunity to lose information, but means the brain only needs to store a single model of each segment rather than both a phonological and phonetic one suggesting that the slight reduction in bit rate this would require is less than the effect having two models for each segment would have on the rate one can produce speech

I think that covers most of the necessary concepts. Hopefully I’ll discuss this more in future

[glossopoesis]

Yes/No Questions

We don’t think much about yes/no questions but they’re surprisingly complicated. A question can be phrased in a positive or negative way (e.g. “is he fast” vs “is he not slow”) and it is not obvious how responses to these two phrasings should relate to each other, should “yes” indicate agreement with the statement in the question or assertion of an unnegated verb in the answer? Some languages solve this by adding extra words to disambiguate.

Obviously, responses indicating negated and unnegated verbs have to be different meaning there are the following possible systems of responses to yes/no system

the 4-form system (4)

the first 3-form system (3a)

the second 3-form system (3b)

the first 2-form system (2a)

the second 2-form system (2b)

Romanian used the 4 system with the responses "da”/"nu”/"ba da”/“ba nu”. There are some claims that Middle English used the 4 system, but this has always been inconsistent at best. The rules in Middle English is usually stated as having the responses “yea”/“nay”/“yes”/“no”, although there is such widespread confusion of “yea”/“yes” and “nay”/“no” to the extent that it may be more accurate to say that Middle English had two 2a systems in parallel, one “yea”/”nay”, and on “yes”/”no” with this “rule” invented to try and introduce a distinction between these two sets of responses where none actually existed.

German, French, the Scandinavian languages, and Hungarian all follow the 3a system with the responses “ja”/”nein”/”doch”, “oui”/”non”/”si”, “ja”/”nej”/”jo” (or similar), and “igen”/”nem”/”de” respectively

To my knowledge, no language has a 3b system (presumably because asserting a negative is less frequently needed than a positive one)

Spanish follows a 2a system with the responses “sí”/”no”

Japanese follows a 2b system with responses “hai”/”īe”

English meanwhile is in a mess of situation where no-one’s entirely sure whether it follows a 2a or 2b system (although people tend to lean towards 2a) and so negative questions are notoriously ambiguous.

Additionally, many languages (notably the Chinese languages, the Goidelic languages, & Finnish) lack specific response words and instead rely on echoing a cut down version of the question either with or without negation. E.g. in the example above, it behaves like a 2a system with responses translating to “is”/”isn’t” but to a question “does he run?”/”doesn’t he run?” it is again like a 2a system but with responses translating to “runs”/”doesn’t run”. 

Where does that put us from a conlanging point of view? Echoing is almost never an unnaturalistic option, the 3a system seems to be an areal feature (having spread from the Germanic languages in our world), 2a & 2b both seems naturalistic in most instances (just be careful to know which system you’re following, and not to accidentally import the ambiguity of English), the 4 system seems to be unusual but is clearly not implausible. 3b on the other hand does seem to be unnaturalistic and should be avoided in situations where naturalism is desirable

[glossopoesis]

A New Qart-Ḥadašt

Qart-Ḥadašt “new city”, better known by its Latinate name Carthage, was a city on the site of Modern Tunis and the centre of a large trading republic (with trade routes extending North, to Britain and the North Sea, and further South along the African coast further than anyone in the Mediterranean would travel for almost two millennia and, depending on how seriously you take their accounts, potentially as far as Cameroon and the Gulf of Guinea) that, in and around the 3rd century BC vied with the burgeoning Roman Republic for dominance in the Western Mediterranean.

Carthage was a Phoenician colony and its language largely continued the late standard forms of Phoenician found in the Eastern Mediterranean, particularly around Tyre in modern-day Lebanon where most of its settlers originated from. The Phoenicians spoke a Canaanite Semitic language closely related to Biblical Hebrew, and there are simple correspondences in the consonants between the two languages.

Like the other Semitic languages of its day, the Phoenician language (and its later Carthaginian form, Punic) was written with an abjad not originally marking vowels at all. Later on, some consonants came to be used to mark long vowels at the end of words (and word-internally for loanwords and foreign names), a situation preserved in Modern Hebrew (although scripture is written with a much more complete system of vowel marking using diacritics and developed in the second half of the first millennium AD). In this abjad, the city is called 𐤒𐤓𐤕𐤇𐤃𐤔𐤕‎ “qrtḥdšt” or, for convenience using the Hebrew abjad (which has a 1:1 correspondence with the Phoenician abjad), קרתחדשת.

This makes it difficult to deduce the quality and location of vowels, particularly word-internally, except through careful examination of transcriptions in other languages (particularly Greek and Latin) and cognates in surviving Semitic languages (particularly Hebrew, and Aramaic, which are fairly closely related to it),  So, let us now examine the name of the city itself.

It is borrowed into Greek as Καρχηδών “Karkhēdṓn” /karkʰeːdɔ́ːn/ or /karxeːdɔ́ːn/, into Latin as Carthāgō /kartʰaːgoː/, and into Etruscan as Carθaza /kartʰaʦa/. The cognates are קרת חדשׁה‎ qeret khadashah /keret xadaʃa/ in Hebrew, ܩܪܝܬܐ ܚܕܛܐ q’ritha’ khadta’ /q(ə)riθɑ(ʔ) ħædtaʔ/ in Syriac Aramaic, and قَرْيَة حَدِيثَة‎ qarya khadītha /qarja ħadiːθa/ in Arabic.

Where does that leave us then? The borrowed forms all suggest the first element as “qart” with a short /a/. The Cognates reflect a root q-r-(y) (presumably “to reside” or similar), with the usual feminine -(a)t ending (or that suffix’s reflex) and a vowel between the q and r of the root; this would suggest a Punic form qart, qaro(t), qaryo(t), or possible qarit, obviously the first of these is the one compatible with the borrowings, so it is the voaclisation we must reconstruct for qrt.

Moving onto the second element, which reflects an adjective from the root ḥ-d-θ (in Phoenician as in Hebrew and apparently Canaanite in general, θ > š, hence the difference in forms between the various cognates), with the usual feminine ending by agreement. Hebrew places an a-vowel between each consonant, Arabic replaces the middle of those vowels with an i-vowel, and Syriac Aramaic simply lacks that vowel.

Meanwhile, the borrowings are trickier to interpret, but appear to reflect a vowel immediately following the ḥ. This leaves the following possible vocalisations: ḥV1dV2št, or ḥVdšo(t), or ḥV1dV2šo(t). The former of these does not appear in any of the cognates, the second in Syriac Aramaic, and the third in both Hebrew & Arabic. The borrowings all include just two vowels in the second element which, whilst only weak evidence (as additional syllables could have been recognised as inflectional endings and not included in the transcription), suggests one of the first two vocalisations is the proper one. Together, these two pieces of evidence suggest ḥVdšo(t) as the proper vocalisation, but only weakly.

Let us now consider the medial consonant (cluster) in the transcriptions which, depending on the vocalisation ought to reflect /d/ or <dš> (potentially /dʒ/ or /tʃ/ by voicing assimilation). Greek has /d/, Latin /g/, and Etruscan /ʦ/ (potentially voiced to /dz/ intervocalically). /d/ being transcribed as /g/ or /ʦ/ would be unusual, but the transcription of /dʒ/ or /tʃ/ as /d/, /g/, and /ʦ/ or /dz/ is entirely plausible. Again, this suggests that ḥVdšo(t) is the correct Punic vocalisation.

This leaves open the quality of the first vowel in the second element. In all cognates this is an a-vowel of some sort. Phoenician has two a-vowels, /a/ and /a:/. An a-vowel is also reflected in both the Latin and Etruscan forms and, whilst Etruscan does not mark vowel length, the Latin shows a long vowel /a:/. Greek shows a long e-vowel though which is somewhat odd. Usually in Greek transcriptions of Phoenician, this reflects original stressed short /i/ (which presumably had been lengthened and lowered to /e:/, merging it with original /ay/), but it could also plausibly reflect a Phoenician ā-vowel /a:/ as Attic Greek (the standard form of Ancient Greek, centred on Athens) disallows /a:/ except after /e/, /i/, or /r/ raising it to /e:/ in all other positions. All of this evidence agrees that, despite the relative rarity of /a:/ in Phoenician (original /a:/ having been rounded to /o:/ in the Canaanite shift, and then raised to /u:/ in the Phoenician shift), the most plausible vocalisation of 𐤒𐤓𐤕𐤇𐤃𐤔𐤕‎ “qrtḥdšt” is qart ḥādšot.

It follows that the conventional vocalisation of 𐤒𐤓𐤕𐤇𐤃𐤔𐤕‎ “qrtḥdšt” as Qart-Ḥadašt is incongruous with the evidence of both foreign transcriptions and cognates within the Semitic language family, and that the original Phoenician name of Carthage was actually Qart-Ḥādšat, developing to Qart-Ḥādšot and then Qart-Ḥādšo in later Punic.

edit: clarified which stage of the language I’m talking about

[writingwithcolor]

so i'm constructing an alternate fantasy world and the main characters speak english becuz that is the language i speak altho it is called by a different name in the story. many parts of the story take place internationally and many characters are multilingual. i will be doing some conlang, but would it be acceptable to take other real world languages (spanish, japanese ect) called by different names? and if so what would be the best way to do this?

Renaming Existing Languages in Fantasy

Unless you’re Terry Pratchett I wouldn’t recommend just cloning real languages (natlangs) and renaming them.  Somehow I let him get away with just calling bad French “Quirmian” where I would’ve thrown anyone else at a wall.  Please see the 2nd part of the answer to this ask (East Asian-based Fantasy, Appropriation, and Fantasy Languages) for some tips on how you might create the general aesthetic “feel” of a natlang without simply ripping off the language itself.  

I recommend conlanging as opposed to copying natlangs wholesale for various reasons, but one big one is to avoid buying yourself heaps of trouble when it comes to marginalized or endangered languages.  How would you feel if you were an indigenous person whose ancestral language was taken away from you or pressured out of your family until they stopped using it, and then suddenly it shows up in someone’s story but the author’s renamed it and given it a fictional history and culture as if it sprang out of thin air?  I bet you’d feel massively disrespected.  Don’t put someone else in that position.

Using English (or whatever publication language) as the medium of your story is different because you say everyone is speaking “Fantasylandian” or whatever but you just assume there’s a translation convention in effect so readers can understand anything in your story.  It’s trickier when there’s other non-main-story languages involved.

~Mod Nikhil

[technologist-revolution2]

@tipsforlanguagelearning

Saw your recent post on Japanese and was wondering, do you have any recommendations for learning Japanese online? I signed on with Linguti but I’m not sure how useful it will be or accurate it is... and I doubt other resources will hurt. 

Especially curious about the accuracy since it is translating, for example “kare wa nonde imasu” as “he drinks” but this website here suggests “nonde imasu” is in the progressive aspect so I would think “kare wa mizu o nonde imasu” would be more like “he is drinking the water”? And the indicative “kare wa mizu o nomimasu” would be better translated as “he drinks the water”?  

[gringacubanal]

i give up long live natlangs, for they are superior