Origin of the habitual tense in Swahili

Question

How far back can a tense like hu, which fails to agree with its subject, be reconstructed in Swahili? Alternatively, do we know which construction it developed from?

This question is based on reading the Wikipedia article on Swahili grammar. That's also where the example comes from.

One of the Swahili tenses, the habitual seems unlike the rest. It a) does not agree with the subject of the clause and b) appears not to have a negative counterpart.

In the sentence below, the absence of agreement on the verb hula obscures whether ng'ombe is singular or plural. I think in Swahili it is normal (or perhaps required?) for the verb not to agree with the direct object when it is a full noun phrase.

Ng'ombe           hula    nyasi.
head(s)-of-cattle HAB-eat grass
Cows eat grass

Given that this tense is so different from the others, I'm wondering if we know how it developed or, alternatively, how far back the habitual tense or another tense with similar agreement rules can be reconstructed.

Answer 1

This is touched on in §3.3 of Nurse & Hinnebusch Swahili and Sabaki. The elements on which it is founded are "very old", perhaps in proto-Bantu. As a habitual morpheme, the form is dialectally variable, being u-, hu-, kʰu, nkʰu and niku, but the dominant Standard Swahili form is hu-. The derivation niku → hu does not come from totally regular sound laws (one would expect ku), but it is attested elsewhere that nk→h sometimes.

The construction is the copula plus the infinitive, thus "is to V" or some similar gloss that lacks time referent. The copular proclitic ni- does not bear subject agreement, which is why the descendant form does not agree with the subject. Progressives and habituals in Bantu frequently are built out of a copula plus infinitive construction. The construction seems to be wide-spread enough in the Sabaki subgroup that one can posit that it was present at that level, with the caveat that (as observed in Swahili itself) there is a lot of borrowing between these languages, and also not a huge amount of grammatical documentation published on e.g. the Mijikenda languages.

The non-agreement property of ni- is universal in Bantu, as far as I know.