Wait, wait. If a phone alternates with another phone under some condition, doesn't that make them allophones of a single phoneme? How can one phoneme undergo transformation to another phoneme?DesEsseintes wrote:[...]Sandhi - as I understand it - is when a phoneme undergoes transformation to a different phoneme altogether due to its environment at a word or morpheme boundary. [...]
Could you not say that /ṣ/ has an allophone [k] before /s/ and /s/ has an allophone [ṣ] after /ṣ/?DesEsseintes wrote:The verb stem dviṣ combined with the 2SG suffix -si becomes dvekṣi (the vowel change is a regular unrelated process), whereby the disallowed cluster ṣs has been resolved by turning it into sth wildly different.
This is actually not an argument, I think, because there are chain shifts that are considered allophony (e.g. /a/>[e],/e/> under some condition) and neutralizations that are considered allophony (e.g. German final devoicing where /d/ > [t] and /t/>[t]).DesEsseintes wrote:All of k ṣ s are full phonemes in their own right, so this is clearly not an example of allophony (at least not in the strictest sense of the word).
DesEsseintes wrote:One of the first things I taught myself to say in Sanskrit when I was a silly teenager was I am the most beautiful. I made a (probably incorrect) superlative of śobhanah → śobhanatamah (most radiant ?) and then added the 1SG form of the verb 'to be' āsmi. Now sandhi operates here, too. The combination ah + ā becomes o and the whole thing becomes śobhanatamosmi. Again o is its own phoneme in Sanskrit. This is not allophony.
Okay, I agree that this is not allophony, but I would say that this is because it is fusion i.e. the number of segments is changed. /o/>[o] existing as a separate phoneme is not a good reason IMO.
[hr][/hr]
Now, maybe this has to do with a slightly different understanding of allophony. In my understanding it does not really rely on native speaker analysis, but instead on complementary distribution and alternation.
If one segment changes into another segment under certain phonological conditions, they are allophones.
If two sounds are in complementary distribution, i.e. they only occur under different, non-overlapping conditions, they are allophones only if they are also sufficiently similar phonetically.
To illustrate why this does not depend on other existing phonemes, let's say we have a language with intervocalic voicing and word final devoicing.
[at] but no *[ad]
[ada] but no *[ata]
and [at]+[a] is pronounced [ada]
and [ada] is clipped to [at] under a certain condition.
also [a]+[da] is [ada] and [a]+[ta] is also [ada]
Notice that they still contrast word initially. [ta] means something different from [da].
This means we have the following phoneme allophone pairs. Note how we have voiced and voiceless phonemes that each map to voiced and voiceless allophones. The existence of one phoneme does thus not prevent other phonemes from having the same allophone. This is what I would call neutralization.
/t/ [d]
/t/ [t]
/d/ [d]
/d/ [t]
I hope this has been somewhat understandable.