Japanese te-gerunds

Friday May 13 was the TWIST conference in Leiden. The TWIST conference is a conference that is especially for students of Linguistics and Comparative Linguistics to do a talk about their current research.

This year Furukawa Kaori presented a paper on Japanese vowel epenthesis and its historical development. I'm not sure if I agree with her suggestions, but nevertheless, her attempt to tackle this problem is good.

In Japanese you have a so called gerund construction which takes the suffix -te after the stem. This construction is found in classicla Japanese and is very straight forward. After roots that end in a vowel -te  is simply added. tabe- 'to eat' > tabete. After roots that end in consonants though, you would get consonant clusters not allowed in Japanese. For example: kak- 'to write' > *kakte. To solve this issue, in classical Japanese a i was epenthesised to break up the cluster. So

ap- 'to meet' apite, yob- 'to call' > yobite, kat- 'to win' > katite, kak- 'to write' > kakite, oyog- 'to swim' > oyogite, yom- 'to read' > yomite, hanas- 'to speak' > hanasite, tor- 'to take' >  torite.

Obvious enough right? And for some reason Japanese did not agree with a system so elegant and clear, and made a mess. So let's see what happened.

For a lot of the consonant final roots, Japanese decided to get rid of the epenthetic vowel again, creating consonant clusters which were then made homorganic.

apite > atte, katite > katte, torite > totte

For the m and b roots it should be mentioned that b was probably a prenasalised bilabial stop, as were all voiced stops at some point (igirisu < portuguese inglês), and you can see this quite obviously in their forms.

yobite > yonde, yomite > yonde

The s roots for some reason did not feel like participating in this epenthetic vowel deletion and stayed as is to this day.

Finally we come to something really funky with the velar roots. Japanese underwent a somewhat irregular soundshift of k > ø, which can by no means be depended on, but you definitely see it show up in numerous words. First of all all i-adjectives like siro-i 'white' used to be ki-adjectives siro-ki, which can still be seen in all other forms, like the adverbial form siroku.

But even this adverbial form is sometimes found without k. hayaku means 'early/quickly', and what is the gretting for 'good morning!'? o-hayou gozaimasu < o-haya(k)u gozaimasu 'it is early!'.

Another place where we lost the k is, you guessed it, in the gerund.

kakite > kaite

But mysteriously, although g is not lost at all, we also find.

oyogite > oyoide

Which has ellided the g and moved up the voicing to the next syllable, without ever getting rid of the epenthetic vowel. This development is really difficult to understand.

oyogite > oyogte > oyogde > oyode would have been probable, but with the epenthtic vowel in place, it is really difficult to understand why the voicing would spread to te. Not to mentions that we don't understand why g is elided, it gives the impression that the voicing of g was lost at some point in this form.

oyogite > oyokide > oyoide. But why?

So to get back to Furukawa's article, she supposed two sound laws that more or less dragged the whole system into the mess it is today. I'm not sure if I find her ideas convincing, but they're definitely better than anything I can come up with to solve this.

The first sound law is: tit > tt. This then explains katite > katte.

The second is the one discussed above: k before a high vowel i/u > ø (but this law is extremely irregular).

So now we have katte, Furukawa tries to explain the forms atte and totte as a paradigmatic changes to facilitate the syllable shape HEAVY-LIGHT in gerunds. Since all t roots and all k roots (and probably g roots?) had this shape. Then by this same need to make all gerunds HEAVY-LIGHT yomite and yobite were forced to becoems yonde and yonde.

Can you spot the glaring problem here? Right, hanasite is still unexplained. Furukawa ventured, completely unconvincingly that this was to avoid homophones. Japanese does not seem particularly bothered by homophones created by the paradigm as evidenced by yonde and yonde, and we can find numerous more kap- 'to buy' and kat- give katte and katte (though apparently there's a tonal difference here).

I'm willing to go along and say that the HEAVY-LIGHT structure was considered so favorable that roots created it even without accompanying sound laws to make them this way, but I fail to see any reason why the s stems would not join in.

And actually, why wouldn't the vowel stems join in either? i- 'to exist' > itte instead of the regular ite?

I think it is fascinating how Japanese had an extremely regular and transparent system, and during a historically recorded period, it completely turned itself inside out, to settle on a new system, without having any idea what exactly happened. It is clear that it cannot all be explained by regular sound shifts and that a lot of analogy took place, but what parts of it are regular sound shifts and which parts gave basis for what analogy, is completely obscure.

Comments

[Etherman23]

I know almost nothing about Japanese so this idea is pure speculation. Suppose that Japanese had a schwa vowel. Assume the gerund was actually @te. Since schwa tends to be a weak vowel it could easily be lost. For example, after a root ending in a vowel, besides schwa, the schwa is dropped. After a consonant the schwa is also dropped. However, after another schwa the first schwa dissimilates to i and then the second schwa is dropped.

As for the disappearing velars my rampant speculation is that they were originally velar fricatives and in some situations (which ones? don't ask me)were dropped but in others were hardened to stops.

[PhoeniX]

Appreciate the brainstorming. But the idea does not cover all the problems.

You would want to see hanasite as hanas@+@te, There is no obvious reason why all stems ending in s would have an @, while all stems ending in any other consonant don't have it.

The velars definitely have something to do with fricatives. We find Ryukyuan languages that indeed have h for k, and I believe some Japanese languages show the same thing.

Part of the reason that the rule is irregular is due to dialectal diffusion, this sound rule took place during the times that the centers of power in Japan were still actively shifting, and this has ended up with a sound law which was happening around the kyoto area being transferred only partially to the edo (now tokyo) area. Then later again the influence of Tokyo Japanese made the shift of the lost k irregular in kyoto area too.

This seems convoluted, and I'm not entirely clear on the development here, but I do know it plays some role.

We see more of these weird double-dialect stuff in modern Japanese. For example, Why is the polite negative -masen from -masu while impolite negatives are sasanai from sasu?

Well that's because of the Kansai area that still has form with -hen these days (s shifted to h) influenced the polite speech back in the days. http://en.wikipedia.org/wiki/Kansai_dialect#Negative

While it is still annoying that k is so irregularly lost, we at least sort of understand why it is irregularly lost. But we still do not fully understand why i in -site wasn't lost while it was lost everywhere else.

[Hans]

"While it is still annoying that k is so irregularly lost, we at least sort of understand why it is irregularly lost. But we still do not fully understand why i in -site wasn't lost while it was lost everywhere else."
Can it be that phonotactics at one point simply didn't allow for -st- while -tt- was fine? Let's we assume that at this point the variants with retained /i/ and with syncope of /i/ were variants (compare English where the vowel in -ed is normally silent, but can be realised as /@/ in poetry), but when -st- became disallowed, only -site remained in use, while for cases where the result was a geminate, the variant with /i/ was eliminated:
Stage I: -site -pite -tite
Stage II: -site / -ste, -pite/-pte -tite/-tte
Stage III (after assimilation of stops): -site / -ste, -pite/-tte -tite/-tte
Stage IV: (elimination of inadmissible variants) -site, -tte, -tte
If we assume the rule did not just concern /s/, but other fricatives as well, we can also solve the riddle of /k/ and /g/:
Stage I: -kite -gite
Stage II: -kite/-kte -gite/-gde
Stage III: -hite/-hte -γite/-γde
At Stage II or III the variants -gite/-gde or -γite/-γde influenced each other with the result that /d/ was transferred to the variant containg /i/ - it seems reasonable to assume that the syncopated form was the more frequent one and the rule for creating the "long" variant was to insert /i/ between the consonants of the short form.
Stage IV: (elimination of inadmissible variants)
-hite, -γide
Stage V: Loss of the spirants.
Now, I don't no anything about Japanese historical phonotactics and sound laws, so I don't know if this explanation is at all plausible?

[PhoeniX]

It's a bit convoluted but, I like the way you were able to incorporate the Velar elision into the explanation as well.

We have no evidence that the velars were ever spirants of course. But I guess it's hard to get rid of /g/ at least in a different way. /k/ could also have gone on to be a glottal stop and then disappear.

Yeah, very nice idea.

But of course it's ad hoc, it's a plausible scenario that only explains this phenomenon and has no explanatory power outside of this construction. Because of that, there is absolutely no way to check whether it's true.

But due to the irregularity of this part of the system, I think it'd probably one of the best things we can do sadly.

[Hans]

"But of course it's ad hoc, it's a plausible scenario that only explains this phenomenon and has no explanatory power outside of this construction. Because of that, there is absolutely no way to check whether it's true."
You are of course right. Are there similar developments (syncope of vowels with subsequent assimilation of stops) elsewhere in the development of Classical to Modern Japanese? What happens to /g/ elsewhere? If it was always prenasalised, we would expect "-nde" here as well, wouldn't we? Are there any cases of geminate voiced stops in Japanese? What happened to other consonnant clusters with /k/ as first element?

[minus273]

An idea that just popped out my head: isn't the k- -> ite, g- -> ide formation parallel to the Kansai rule of p- -> ute, b -> ude?

Not clear what that would imply, though.