Sandhi phenomena and "linking morphemes"

"Linking morphemes" (traditionally called "relational prefixes") are a hotly debated topic in South American linguistics. While some linguists deny their existence as segmentable morphemes (Fortune 1964, Salanova 2009), some find them to be important pieces of evidence for a purported genetic relationship between three major South American families—Tupí, Karib, and Macro-Jê (Rodrigues 2000). A more plausible analysis, in my opinion, lies somewhere in between.

Back in 2007, I sent a query to the HistLing list about the possible relationship between "linking morphemes" (that is, elements that "join together" two constituents in a compound or phrase) and sandhi phenomena. I was then, and have remained so, particularly interested in possible cases where "linking morphemes" may have emerged through the morphologization of once-predictable phonological alternations, either through the reanalysis of stem-initial consonant alternations, or through the reanalysis of alternating forms of a person marker (such as *i-).

The similiarities between the "linking morphemes" in all three families are pointed out by Aryon Rodrigues (see data sample below, from Rodrigues 2000:102). In languages of the three families under consideration here (which are typically SOV), the "linking morpheme" occurs whenever a (noun, verb, and postpositional) root of the relevant morphological class is preceded by its absolutive argument (a possessor, for nouns; an object, for transitive verbs and postpositions; etc.). Roots belonging to this class will have at least two different stem-forms: one, with the "linking morpheme", the other, with a default, third-person marker (although, in some languages, a few stems can also occur "bare", prefixless).

Proto-Tupí-Guaraní *r- ~ *s-
Panará (Jê family) j- ~ s-
Hixkaryána (Karib) j- ~ 0-

My opinion falls somewhere between both extremes: although there are cases in which "linking morphemes" are obviously inherited, they cannot be necessarily seen as an example of "shared aberrancy" when comparing different families, since their distribution seems to point to an origin that may ultimately have been phonologically motivated. That is, given the right (phonological and syntactic) environments, "relational prefixes" could have developed independently in different families.

My own Macro-Jê comparative studies strongly suggest that the alternations involving the so-called relational prefixes (see examples below) can indeed be traced back to Proto-Macro-Jê. For instance, the linking prefix in the Parkatêjê example, j-, is clearly a cognate with the Karajá linking prefix d-, whereas the Parkatêjê third-person marker h- is clearly a cognate with the Karajá third-person marker ɗ- (as fully corroborated by the ongoing lexical comparison).

(1) Karajá (Karajá family, Macro-Jê stock)

(a) N d-e 'N's wing'
(b) ɗ-e '(its) wing'

(2) Parkatêjê (Jê family, Macro-Jê stock)

(a) N j-arkwa 'N's mouth'
(b) h-arkwa '(its/his/her) mouth'

Not surprisingly, similar alternations are also found in the Jabutí family, whose inclusion in the Macro-Jê stock was only made possible (on solid grounds) by recent documentation efforts (cf. Djeoromitxí hako ~ -rako 'mouth', a cognate of Parkatêjê h-arkwa ~ j-arkwa above; Ribeiro & van der Voort 2005).

Recent advances in the comparative studies of the other families, however, seem to suggest independent origins for the "relational prefixes" in some cases. For Karib, a possibility is that the "linking morpheme" *j- is, after all, a cognate of the third-person marker *i- (a result of glide formation in constructions such as "John his-house"). A similar origin cannot be discarded for (Pre-)Proto-Macro-Jê either. [Maybe the fact that a geographically distant (and genetically unrelated) language family, Algic, presents a similar phenomenon — the "intercalated -t-" discussed by Greenberg (1987:47) in support of his "Amerind" — makes the "shared aberrancy" status of "linking prefixes" even more questionable.]

Morphology vs. phonology? Morphology plus phonology?

In all the language families discussed here, linking morphemes occur in environments which may favor sandhi phenomena of some sort (for instance, all the stems are vowel-initial and tend to form a stress unit with the preceding co-constituent). In addition, the alternations generally involve a "hard" consonant and its "softer" counterpart; etc. Therefore, I would very much appreciate any examples that may contribute to a better understanding of the genesis of linking morphemes (not only in the aforementioned languages), including the following possible scenarios:

☞ Insertion of new phonological material (for instance, cases similar to r-insertion in English, etc.).
☞ Modification of existing phonological material (such as lenitition/fortition, etc.).

Counting stems

Out of Lemle's (1971) 221 reconstructed Proto-Tupí-Guaraní forms (a fairly random sample of the lexicon), around 20 percent (41) are body-part terms. Of these, nearly half (19) belong to the class of stems that take the "relational prefix" r-. Were r- the stem's initial consonant, that would be a statistical aberrancy; other consoants wouldn't come even close (8 stems are p-initial, 4 are k-initial, etc.).

Even though only 112 stems were reconstructed by Davis (1966) for Proto-Jê, the numbers are proportionally similar to the Tupí-Guaraní ones. The reconstructed lexicon includes 25 body-part terms, of which 8 are j-initial (j-being, in my analysis, the linking prefix), 4 are m-initial, 5 are k-initial, 2 are p-initial, 2 are t-initial, 2 are s-initial, 1 is ng-initial, and 1 is n-initial. [While all 8 j-initial stems are certainly reconstructible for Proto-Jê, several among the remaining stems were reconstructed by Davis without enough attestation.]

One crucial argument that tends to be overlooked by those who deny affixal status to such "linking alternations" is statistical in nature: if, as Salanova (for Jê), Fortune (for Karajá), etc. suggest, the "relational prefix" is merely the initial consonant of the stem, why would there be so many d-initial stems in Karajá, j-initial stems in Jê etc., much more than any other consonant? Maybe the answer is that, no matter how one analyzes them synchronically, such alternations trace back to one single affix attached to vowel-initial stems, as in the scenario suggested above for Karib.

Towards a conclusion

A conclusion is that, no matter how one analyzes particular instances of such alternations, their value as evidence for genetic relationship has to be seen with caution. It may be that, at the end, "relational prefixes" will be demoted from their status as "shared aberrancy" to a rather trivial case of "pan-Americanism" (the prefixation of *i-).


Fortune, David. 1964. Karajá Grammar. Unpublished manuscript. Rio de Janeiro, Arquivos Lingüísticos do Museu Nacional.

Greenberg, Joseph. 1987. Language in the Americas. Stanford: Stanford University Press.

Ribeiro, Eduardo & Hein van der Voort. 2005. The inclusion of the Jabuti language family in the Macro-Jê stock. Paper presented at the "Simpósio Internacional sobre Lingüística Histórica na América do Sul", Belém: UFPA & Museu Goeldi.

Rodrigues, Aryon. 2000. 'Gê-Pano-Carib' x Jê-Tupí-Karib': sobre relaciones lingüísticas prehistóricas en Sudamérica. In Miranda, Luis (editor), Actas: I Congresso de Lenguas Indígenas de Sudamérica, tomo I. Lima: Universidad Ricardo Palma.

Salanova, Andrés Pablo. 2009. Não existem prefixos relacionais nas línguas Jê. In Braggio, Silvia L. B. & Sinval M. de Sousa Filho, Línguas e Culturas Macro-Jê, p. 259-271. Goiânia: Editora Vieira.

(Página criada em 1253110143|%e %b %Y, %H:%M %Z|agohover. Última edição em 1283373267|%e %b %Y, %H:%M %Z|agohover; comentários e informações adicionais são sempre bem-vindos.)

Este site é melhor visualizado com senso crítico. © 2009-2011 Eduardo R. Ribeiro