Recently I picked up the two earliest pieces of canon on the Vilani language and gave them a more thorough going-through than I have in many, many years. There's some interesting things there... at least to me.
The first item is by John Harshman, birthing Vilani in the “Random Notes” column of JTAS #17 (1983). It's interesting to look back and see the critter in its earliest formulation. One thing that leapet out at me was the existence of sandhi rules (i.e., regular changes that phonemes undergo when coming into contact with neighboring phonemes), which had to be manually applied during and after the randomized generation process:
As Harshman says that GDW was already then using a computer to automate the process, I wonder how he applied this rule – did the program have a step to prevent such sequences? Did he go through the output and fix it manually? If so, how did he fix it: by deleting the second syllable, by eliding one of the vowels, or....?
In any case, this points up the fact that the robo-Vilani words like “aaukha” that crowd later publications were not supposed to happen. Without applying this rule, ~20% of bisyllables would have V+V ligatures, vastly more than Harshman apparently intended.
Three points jump out at me:
1) I've rarely, if ever, seen evidence of this being applied in post-GDW publications where lots of Vilani words appeared. I can't say I've been good about it myself, either.
2) To be fair, it's not specified what is meant by a “similar” consonant. Point of articulation, at least, is about all can be gleaned from the example he gives.
3) Oh, and in that example, we get a phoneme (/t/) that doesn't appear in the list of phonemes he just gave us!! There is no /t/ in any published system for generating Vilani words, but here we can see that it was creeping in from the very outset!
This is another rule that I think has been largely ignored by later sources – including the successor system of MT (below). Which is a pity, as I think it could have been implemented relatively easily in fan-made programs. (Can we take it as a given that 99.9% of all Vilani words appearing in print after the 1990s were cranked out by computers?) It's a major tool in making Vilani look more “Assyrian-ish”, I think – getting more gemminnatted consonants.
There's some other issues with multisyllabic words in this article. To make myself a bit clearer, I'm going to rewrite Harshman's rules using “F” to represent “final consonant” -- so that the four syllable types are V, CV, VF, and CVF. Why does it matter? Because C /= F; F excludes several phonemes that occur in C, and moreover the frequencies of those phonemes they do share are different.
So, using Harshman's system where you strings syllables together, randomly occurring in the ratio 1:4:3:2 V:CV:VF:CVF, you're gonna get sequences like VF+V (say, /ar + a/ > ara). Then, when you encounter the word ara, do you syllabify that as /a + ra/ or /ar + a/? Is the /r/ a C or an F?
This bugs me, because without providing for syllable boundary placement forming minimal pairs (i.e., the Vilani recognize /ar.a/ and /a.ra/ as phonologically distinct from each other), then the frequency distribution of consonants in a large group of randomly-generated words drifts away from the system used to generate it. That's why I appreciate the MT system (below), where C and F are clearly separated, and you can really start dealing with codas as codas.
The second of the two articles is the 'Alien Languages' chapter in 1986's Megatraveller Referee's Companion. Harshman is listed as one of the book's four authors, perhaps he was behind the new alien word generation system, too? The Vilani content is basically contained in a single-page chart, presenting the now-familiar stack of 6x6 tables to dice up a word (of 1-6 syllables, naturally).
The frequency distribution of phonemes within the categories of C, V, and F are the same as those of 1983. What is different is that the issue of syllable ligatures is resolved, and tidily too.
The user is instructed to randomly generate a syllable structure on the “Basic” table if it is the first syllable in the word (“or if otherwise instructed” -- there are no such instructions in the system as written). All subsequent syllables in a given word have their structure rolled on the “Alternate” table – which has only CV and CVF patterns. In other words, they've removed the possibility of V+V and F+V sequences ever occurring, even at the first stage of generation.
So it's a much simpler system to use and understand: one, there's no need to apply mid-/post-generation sandhi rules by hand in order to get 'correct' sequences; two, it lets you look at the given frequency distribution of C, V, and F in the generation system and it will be much closer to 'actual' (finished) output). (Which in turn means, if you want to fiddle with the tables, it's easier to end up with the results you were imagining.)
You gotta recognize, though, that it also means that those who skipped the JTAS 1983 article and instead adapted the “6x6x6x6... tables” system for computers, and who used those programs to churn out huge lists of Vilani words for use in fanzines and official publications, screwed up. Starting with T4, I haven't seen a Traveller product using Vilani words that isn't full of stuff like “iiaaum” and “ushal” and whatnot. Whichever of these two systems post-GDW/MT writers have tried to implement, they've failed.
Let me stress, admittedly, I'm not claiming to have checked every Traveller publication for dodgy Vilani names. But I have sampled every product line I know about, and yes, I did look at their use of 'naming conlangs'. I haven't seen a single one, post-DGP, that didn't have broken Vilani, whether due to their own new robogeneration or by adopting /incorporating previous examples.
Basically, around thirty years ago the OTU was given a fairly clear and simple algorithm in JTAS and a really, really clear and simple system in MT, and has consistently been unable to apply it correctly – and the resulting reams of errors are now canonical.
This concludes my 1st Occasional Hans Ranke Madsen Memorial Rant.
The first item is by John Harshman, birthing Vilani in the “Random Notes” column of JTAS #17 (1983). It's interesting to look back and see the critter in its earliest formulation. One thing that leapet out at me was the existence of sandhi rules (i.e., regular changes that phonemes undergo when coming into contact with neighboring phonemes), which had to be manually applied during and after the randomized generation process:
As Harshman says that GDW was already then using a computer to automate the process, I wonder how he applied this rule – did the program have a step to prevent such sequences? Did he go through the output and fix it manually? If so, how did he fix it: by deleting the second syllable, by eliding one of the vowels, or....?
In any case, this points up the fact that the robo-Vilani words like “aaukha” that crowd later publications were not supposed to happen. Without applying this rule, ~20% of bisyllables would have V+V ligatures, vastly more than Harshman apparently intended.
Three points jump out at me:
1) I've rarely, if ever, seen evidence of this being applied in post-GDW publications where lots of Vilani words appeared. I can't say I've been good about it myself, either.
2) To be fair, it's not specified what is meant by a “similar” consonant. Point of articulation, at least, is about all can be gleaned from the example he gives.
3) Oh, and in that example, we get a phoneme (/t/) that doesn't appear in the list of phonemes he just gave us!! There is no /t/ in any published system for generating Vilani words, but here we can see that it was creeping in from the very outset!
This is another rule that I think has been largely ignored by later sources – including the successor system of MT (below). Which is a pity, as I think it could have been implemented relatively easily in fan-made programs. (Can we take it as a given that 99.9% of all Vilani words appearing in print after the 1990s were cranked out by computers?) It's a major tool in making Vilani look more “Assyrian-ish”, I think – getting more gemminnatted consonants.
There's some other issues with multisyllabic words in this article. To make myself a bit clearer, I'm going to rewrite Harshman's rules using “F” to represent “final consonant” -- so that the four syllable types are V, CV, VF, and CVF. Why does it matter? Because C /= F; F excludes several phonemes that occur in C, and moreover the frequencies of those phonemes they do share are different.
So, using Harshman's system where you strings syllables together, randomly occurring in the ratio 1:4:3:2 V:CV:VF:CVF, you're gonna get sequences like VF+V (say, /ar + a/ > ara). Then, when you encounter the word ara, do you syllabify that as /a + ra/ or /ar + a/? Is the /r/ a C or an F?
This bugs me, because without providing for syllable boundary placement forming minimal pairs (i.e., the Vilani recognize /ar.a/ and /a.ra/ as phonologically distinct from each other), then the frequency distribution of consonants in a large group of randomly-generated words drifts away from the system used to generate it. That's why I appreciate the MT system (below), where C and F are clearly separated, and you can really start dealing with codas as codas.
The second of the two articles is the 'Alien Languages' chapter in 1986's Megatraveller Referee's Companion. Harshman is listed as one of the book's four authors, perhaps he was behind the new alien word generation system, too? The Vilani content is basically contained in a single-page chart, presenting the now-familiar stack of 6x6 tables to dice up a word (of 1-6 syllables, naturally).
The frequency distribution of phonemes within the categories of C, V, and F are the same as those of 1983. What is different is that the issue of syllable ligatures is resolved, and tidily too.
The user is instructed to randomly generate a syllable structure on the “Basic” table if it is the first syllable in the word (“or if otherwise instructed” -- there are no such instructions in the system as written). All subsequent syllables in a given word have their structure rolled on the “Alternate” table – which has only CV and CVF patterns. In other words, they've removed the possibility of V+V and F+V sequences ever occurring, even at the first stage of generation.
So it's a much simpler system to use and understand: one, there's no need to apply mid-/post-generation sandhi rules by hand in order to get 'correct' sequences; two, it lets you look at the given frequency distribution of C, V, and F in the generation system and it will be much closer to 'actual' (finished) output). (Which in turn means, if you want to fiddle with the tables, it's easier to end up with the results you were imagining.)
You gotta recognize, though, that it also means that those who skipped the JTAS 1983 article and instead adapted the “6x6x6x6... tables” system for computers, and who used those programs to churn out huge lists of Vilani words for use in fanzines and official publications, screwed up. Starting with T4, I haven't seen a Traveller product using Vilani words that isn't full of stuff like “iiaaum” and “ushal” and whatnot. Whichever of these two systems post-GDW/MT writers have tried to implement, they've failed.
Let me stress, admittedly, I'm not claiming to have checked every Traveller publication for dodgy Vilani names. But I have sampled every product line I know about, and yes, I did look at their use of 'naming conlangs'. I haven't seen a single one, post-DGP, that didn't have broken Vilani, whether due to their own new robogeneration or by adopting /incorporating previous examples.
Basically, around thirty years ago the OTU was given a fairly clear and simple algorithm in JTAS and a really, really clear and simple system in MT, and has consistently been unable to apply it correctly – and the resulting reams of errors are now canonical.
This concludes my 1st Occasional Hans Ranke Madsen Memorial Rant.
