The problem of describing palatalization in Japanese centers on a tricky issue I've wodered a lot about: What do postalveolar consonants technically have in common with high front vowels? Informally, you can say they're both "palatal", but how does that translate into distinctive features? Here's one way to illustrate this phenomenon using Articulator Theory style trees.
[+cons -son] [+son] | | Oral Oral | / \ Cor Cor Dors | | | [+ant] [-ant] [+high -back]
See how I had to represent the high front vowel? I stuck a [-ant(erior)] in there because, otherwise, where would the [+ant] consonant get its new value? [-back] is clearly related, but I didn't want to end up with a consonant that's both [+ant] and [-back]. That doesn't sound particularly palatal. Using this representation of the segments, here's how the relinking would go.
[+cons -son] [+son] | | Oral Oral | / \ Cor___ Cor Dors = \ | | [+ant] [-ant] [+high -back]
Another way to handle this sort of palatalization might be to ignore the [ant] feature altogether. The sounds that wind up getting palatalized before /i/ aren't just the [+ant] ones after all; velars do it too. What if you had the following setup?
[+cons] [+son] | | Oral Oral | Dors | [+high -back]
The situation I'm describing here is any consonant (including sonorants) followed by /i/.
[+cons] [+son] | | Oral____ Oral \ | Dors | [+high -back]
Delinking should happen for consonants that are already dorsal, but really nowhere else. This keeps true palatals (palatalized dorsals) separate from postalveolars (palatalized coronals). As I said, I'm ignoring [ant] for this analysis, and I'm actually ignoring [distr(ibuted)] too. Anything with a specification for both Coronal and Dorsal would be inherently [+distr].
I've been working on another way to handle phenomena like this, something I'll call "segment stacking". The idea is that segments (that is, bundles of features) don't really occur in linear order at all; they just include positioning features to make sure they're pronounced in the right order. This way of looking at things might come in handy when dealing with vowel harmony, mispronunciations like my daughter's `drifferent` and the popular `orangutang`, and slips of the tongue like `seech spound` that involve transposing onsets.
Here are some examples of what I mean by "stacking". (The values on the left side of the "=" could go in any order.)
- a + p = pa
- S + t = tS
- t + a + S = tSa
- s + k = ks
- s[init] + k = sk
- s[init] + k + a[init] = sak
- a[init] + s + k = aks
- a[init] + s + k[fin] = ask
See how this sort of thing could explain why the pronunciation of `ask` can somehow get reduced to `axe`? Losing the [fin] feature from the representation makes much more sense than switching the order of two segments. It's not really reduction at all in the latter case.
What I'm saying is that several segments can having the same position in the representation of a given syllable. They would all get started at roughly the same time, but some segments take longer to pronounce than others (vowels > continuants > stops), so what we hear is distinct pieces in a definite order. This helps to explain why the preferred pattern for words across languages is CVCV and why affricates (stop + fricative) seem to be generally easier than combinations of a fricative plus a stop. It also makes English onset aspiration and coda glottalization easier to describe. (Rather than using special features, just throw the existing segments /h/ and /?/ on top of the affected stops.)
Before you start thinking I've gone off on a big tangent, let me get back to Japanese palatalization. Say the segment /i/ has the features [+high] and [-back], which rely on the Dorsal articulator. /s/ uses the Coronal articulator. If /s/ and /i/ have exactly the same position in the underlying representation of the syllable /si/, it could be that the Dorsal features of /i/ don't have to somehow rub off on /s/ to produce [S]; they're just being pronounced at the same time as the Coronal feature(s). The result is that /si/ in Japanese becomes [Si].
Two more syllables to consider are /ti/ and /tu/. Somehow, these get turned into [tSi] and [tsu]. Again, /i/ seems to be giving its [-back] feature to the onset, but what about the [+back] feature of /u/? Rather than call [ts] [+back], we should probably change our terms a little. [tSi] and [tsu] could be called [+pal] and [-pal] respectively.
That's a pretty straightforward description of the palatalization, but where do those fricatives come from? /i/ and /u/ are the only vowels that trigger this, and /t/ and /d/ seem to be the only affected onsets. So on the surface, it looks like [+cor] combined with [+high] necessarily means simple stops aren't allowed (but nasals, flaps, and affrictes are fine). It seems like [+high] is pretty meaningless with stops since they're inherently high enough to make contact already. This feature might be conceptualized as close, but not touching (which is one of the reasons why I like the IPA terms "close" and "open" better than "high" and "low").
I'm visualizing this process as something like segments whose features turn off and on one at a time, resulting in some overlap. The fricatives in [tSi] and [tsu] would be a product of that overlap.
t (S) i [+cor ] [+pal ] [+high ] t (s) u [+cor ] [+high ]
Any ideas on how to formalize this a little better? Notice that I might as well be using privative features here. I like the idea of simplifying the UR by only remembering the [+] features (and assuming [-]). It lets you deal with features more like segments, morphemes, phrases and such. They're either in the UR or not. You remember them or you don't.