Consonant correspondences

These correspondences hold among the consonants.

Vowel correspondences

These correspondences hold among the vowels.

Identical correspondences

For identical sets, the reasonable assumption is that the proto sounds is the same as what we find in all the daughter languages.

Other reconstructed categories

For some nonidentical sets, it's possible to reconstruct a sound distinct from those already proposed without creating any complications.

A few comments are in order. Proto */t/ was chosen because it's much more natural for a stop to become an affricate or fricative (especially before a high front vowel like /i/, palatalization, which is one of the world's most common sound changes). "Majority rules" would favor /č/, but that's a less plausible direction of change.

It's not possible to say that /r/ was inserted after a word-final vowel in Tupinambá, since the word kiʔa contradicts this; so we have to reconstruct */r/ and assume it is deleted in the other languages. (That deletion becomes relevant for some other patterns below.)

Based on this limited set of data, we could either assume that /ʔ/ was inserted between vowels in the first two languages, or that it was deleted (regardless of position? or at least between vowels) in the second two languages. The latter sort of change is a bit more typical, but not strongly so.

Finally, we see that a nasalized vowel in most of the languages corresponds to a vowel plus velar nasal in Tupinambá. Most typical is that the consonant was there originally but in most cases was lost in three of the languages, leaving behind only its nasalization on the vowel. (The same sort of change happened in French and Portuguese.)

Final deletion

We're beginning to see a pattern whereby most of the languages lose word-final consonants, and only Tupinambá retains them. That should serve as a clue that the word-final /b/ in this correspondence was originally there, but deleted in three of the languages.

Labial from velar

The last correspondence could be treated as */p/ based on these very limited examples; or, going with majority rules, we could treat it as */k/ with a conditioned change to /p/ before /w/.

A change of the general type /kw/ to /pw/ is also known from other languages: it happened in Ancient Greek (compare Latin quo "where" with Greek pou, both from the same Indo-European source), and more recently in Romanian (with loss of /w/: Latin aqua, Romanian apa).

Summary of changes

Here are the changes needed for each language.

Further examples would likely show that */t/ palatalized only before /i/, but we don't have proof of that in this data set, and so that condition doesn't have to be included.

Another way to think about the loss of final consonants is to first nasalize vowels before a nasal consonant in all but Tupinambá, and then delete all word-final consonants in those three languages.

Subgrouping

Based on the many shared changes among the languages, we can hypothesize the following subgroups for the family.

Proto-Tupi-Guaraní */kitiŋ, tiŋ, meʔeŋ, kwar, kib, kiʔa, abati/
Tupinambá */k/ > /p/	Proto-G-S-G */t/  >  /č/ before /i/ */vŋ/  >  /ṽ/ */b/, */r/  >  Ø finally
	Guaraní no further changes	Proto-S-G */ʔ/  >  Ø
		Sirionó */č/  >  /š/	Guarayo no further changes

Notice that the /š/ of Sirionó is treated as the outcome of two stages: first palatalization */t/ > /č/, then de-affrication /č/ > /š/. Not only is the the expected sequence, but it is consistent with the other evidence for the subgrouping shown.