A tantalizingly brief note of 1931 in the Gold Coast Review describes an ethnic group called the Mpre, found only in the village of Butie in central Ghana (8° 52' N, 1° 15' W) near the confluence of the White and Black Voltas, apart from a few emigrants in Debre. According to the author's description, the Mpre people, once more widespread, were reduced to a single village in the course of comparatively recent wars with the Asante. Noting that their language was “different to that of the surrounding tribes”, he lists 106 words of Mpre. This short vocabulary appears to be the only existing record of the language, which is believed to be extinct. The gap is all the more unfortunate because Mpre turns out to be of some taxonomic significance. It is not closely related to any of its neighbors, and Heine and Nurse (2000) treat it as unclassified. A friend of mine's paper dealing partly with this will be appearing sometime soonish, but I won't spoil the surprise...
You might think, given all this, that it was impossible to retrieve any information on its grammar. However, you would be wrong! Fellow language geeks may find it an interesting exercise to try their hand at extracting grammar information from the wordlist, which Blench gives a copy of, before reading on...
The wordlist strongly suggests a noun class prefix system still at least partially productive. The highly lopsided initial letter statistics would alone suggest this: 31 entries begin with e-, 21 with a-, and 12 with n-, together accounting for the majority of the wordlist. This speculation is confirmed by distributional analysis for the e- which appears in the numbers 1-5, but disappears in 11-13 and 20-30; it is presumably to be identified with the Ga prefix é- observable in the same numbers. (The change of ekpe “one” to mpe in “11” is noteworthy, if it is not a typo.) Likewise, comparison of kelafa “100” with lefanyo “200” reveals a prefix ke- - with precise analogues in Ch./Kr. kʌ́-, Na. gʌ́-, and Go. ká- in the same numbers. Of the 21 entries with a- (corresponding to 19, or possibly 18, distinct words), five are glossed as plural in English, while another four are glossed as collective nouns; no entries not beginning with a- are glossed as plural. I therefore conclude that a- is a marker of plurality - suggesting that ado (the formative element in “20”, “30”, ...) is the plural of edu “ten”. This jibes nicely with other languages of the area: a plural prefix a- is found in Gonja, Twi, Lejana, Akpafu, and Avatime, for example.
Identifiable compounds include zingilzi-nogha “bush cow” (cf. zingelza “bush”, nogha “cow”), sunko kawuseggi “earth owner or tindana” (cf. sunko “earth”), nkemnzui “son” (cf. nzui “child”), lefanyo “200” (cf. enyo “2”, kelafa “100”), eputo nasi “foot” (cf. eputo “leg”) ; all suggest a word order type Modifier-Modified. “Lion” (jikpajikpakoseggi) must surely be a compound, in which I would identify the final koseggi with kawuseggi “owner (?)” above? Also, ataza “finger” and atazai “toe” are clearly related, but it is unclear whether one is a compound form or whether both are simply different transcriptions of the same word.
One short sentence is given - agbem aba “it rains” (cf. agbem “God”). Assuming that this is of the form SV, this could be taken to suggest verb agreement in gender (or at least number) with the subject; however, this is by no means certain.
Statistical etymology: A Votic example
16 minutes ago