I would like to thank the members of the Institute for Advanced Technology in the Humanities for their aid and comfort in working on this project. My special thanks to Thornton Staples, for customizing the database programs, to Susan Gants, for managing the diagrams, and to John Unsworth, for help with the markup.[back]

1. Some brief comments about differences between the Swahili system and those of other Bantu languages. In Swahili no nouns belong to the `locative classes', 16-18 in the traditional Bantu numbering system; these have therefore been excluded from Table 1. Classes 12 and 13, diminutive singular and plural respectively in some Bantu languages, have disappeared in Southern Swahili dialects (including Standard Swahili). Their function of deriving diminutives has been taken over by Classes 7/8 (of which more anon). Class 12 but not Class 13 is reconstructed by Nurse and Hinnebusch (1993) for Proto Northeast Coast Bantu (PNEC)-- apparently by that period Class 8 had already replaced 13 as the plural of 12 (N&H 1993:338-9). Swahili also has no classes exclusively used to derive augmentatives. This function is performed by Classes 5/6, and the function apparently dates back to PNEC (N&H 1993:342). Swahili Class 11, with the prefix u-, results from a merger of PNEC classes 11 and 14 (with prefixes reconstructed as *lu and *Wu respectively, N&H 338). Possible semantic motivations for this merger will be discussed below (Section 4).[back]

2. Alternatively, singular/plural prefix pairs are often used to identify noun classes. Thus the pairing m-/wa- (Classes 1/2) is distinguished from m-/mi- (Classes 3/4), even though the singular prefixes are homonymous. This is the expedient used in dictionaries, which list nouns in their singular form and give the plural in parentheses. The disadvantage of this method is that some nouns have no singular or no plural. If no plural form is listed in the dictionary entry, the class of the noun is often ambiguous. For critical discussion of criteria for defining the classes, see Zawawi (1979) and Adewole (1986).[back]

3. For arguments against this view with respect to Swahili, see Contini-Morava (forthcoming).[back]

4. Williamson (1989:32) makes essentially the same claim for even more remote levels of time depth: "it is therefore logical to conclude that while the noun classification system arose on a semantic basis in pre-Niger-Congo, it had already become a grammaticalized, essentially formal system in proto-Niger-Congo".[back]

5. Due to constraints on this medium, tone markings for Common Bantu forms have been omitted. I represent the "super-high" /i/ and /u/ of Common Bantu and Proto-Sabaki with upper-case I and U.[back]

6. I originally intended to use the more recent Kamusi ya Kiswahili Sanifu (Taasisi 1981), but decided against this because the Kamusi does not identify loanwords.[back]

7. Breakdown of the database by class:

Class 1        334       Class 7        761
Class 3        924       Class 8        3
Class 4        21        Class 9        1404
Class 5        755       Class 10       34
Class 6        94        Class 11/14    471

There are also 253 nouns whose class is unclear from the dictionary entry, e.g. kindi `squirrel' (Class 7 or 9?), and 141 roots listed in more than one class with the same meaning, e.g. bongo (5)/ubongo (11) `brain'; kichochoro (8)/mchochoro (3)/uchochoro (11) `narrow alley'. (Because of these the above totals do not add up to exactly 4650-- this is the total number of records in the database.) The most common alternation is between Classes 5 and 9, but almost all other pairings also occur. The only ones not attested so far are 1/11 and 7/9.

The database does not systematically include nouns derived by fully productive derivational processes, because their meanings are predictable. These include: agentive nouns in Class 1 derived from verbs plus suffixed -ji, e.g. mchezaji `player' (from -cheza `play'); deverbal nouns in Class 3 indicating act/manner/process, e.g. mkopo `act/process of borrowing' (from -kopa `borrow'; Class 7 diminutives, e.g. kilango `small door' (dim. of mlango, `door'); Class 5 augmentatives, e.g. lango `large door' (aug. of mlango, `door'); Class 11 abstract nouns derived from verb or adjective stems, e.g. ubaya `evil [noun]' (from -baya `bad [adj.]'). Exceptions were made in the case of derived nouns that have developed specialized, hence unpredictable, meanings, for example kikono `stump of arm; anything resembling a small hand, e.g. projecting prow of a vessel', literally `small hand' (dim. of mkono `hand').[back]

8. I should point out that number is being treated as a semantic, rather than morphological category here. As in other Bantu languages, the noun classes are generally paired for singular and plural. One might therefore assume that it is redundant to specify the number of each noun, since that can be deduced from its class affiliation. Number is treated as a separate category from noun class affiliation because a number of nouns denote entities that are not amenable to enumeration, such as masses, collectivities, or abstractions. Such nouns may occur either in one of the `singular' classes, e.g. ugali `porridge', Class 11, or in one of the `plural' classes, e.g. maji `water', Class 6. Since the goal of the project is to analyze the semantic structure of the noun classes, I preferred not to assume in advance that `singular' and `plural' were part of the meanings of the class prefixes..[back]

9. For further information about this corpus and associated software, contact Arvi Hurskainen at the University of Helsinki, e-mail Hurskainen has developed a morphological parser for Swahili (Hurskainen 1992) based on different principles from the one developed by Schadeberg and Elias at the University of Leiden, the Netherlands (Schadeberg and Elias 1989). As yet no syntactic parser exists for Swahili.[back]

10. The few exceptions are all nouns denoting animate beings: mdudu, the generic term for `insect', mnyama, the generic term for `animal', and a few names of animals either derived from verbs or from Class 1 nouns, e.g. mpasuasanda `nightjar' [literally, `shroud-tearer', a bird of ill omen].[back]

11. I have not given examples of the most productive categories of nouns in Class 3: the names of plants and trees, of which there are 358 in the database, or the deverbal nouns indicating process/method etc., an open-ended set. The coverage of the remaining nouns is fairly comprehensive: only 53 out of 854, or approximately 6%, do not obviously fit into one or another of the categories on the chart. Examples of these include: mjango `empty, ineffectual visit, without profitable outcome' [related to -janga `grumble'?]; msapata `kind of dance' [from Portuguese sapatear, a dance]; muhuri `seal, signet, crest' [has initial m- in Arabic]; mgombwe `bull's mouth shell'; mchuzi `gravy, soup, sauce'. In some cases the noun may be derived from a verb that is now obsolete; in others there may be cultural associations that are not apparent from the dictionary definition; yet others are loanwords that may have been put in this class because of their initial m-, rather than because of their meaning. [back]

12. The lists of examples include forms reconstructed for Proto-Sabaki (PSA) or Proto-Swahili by Nurse and Hinnebusch 1993, if available. In their notation W means a voiced labial approximant (exact feature specification unclear, cf. N + H pp. 89ff). Where a word is listed as a loanword in SSED I indicate whether or not the source word had an initial syllable matching the prefix of the class to which it was assigned. [back]

13. As mentioned in Section 2, Denny and Creider (1986:221) propose `extended solid figure' as the basic meaning for Class 3 in Proto-Bantu. They do not regard this as a metaphorical extension based on the typical shape of plants; on the contrary, they suggest that it was the shape meaning that motivated the inclusion of plants in this class, rather than vice versa (ibid.:223-4). They may have missed some of the complexities in the internal structure of Class 3 because of their small data set (the total sample for all classes was the approximately 300 nouns reconstructed for Proto-Bantu by Guthrie 1971, vol. 2).[back]

14. This category in Swahili shows interesting parallels to the Ewe morpheme vi, discussed in Heine et al. 1991:84.[back]

15. Khamis (1984) describes this as a `metaphorical' use of ki- `for excess', although `ironic' might be a more appropriate term. I would like to thank Abdulaziz Lodhi (p.c.) for bringing this paper to my attention.
The use of ki- in an inversion of its more literal meaning still seems to be productive in Swahili, as evidenced by the following forms from Ohly (1987): kifurushi `large buttocks'(dim. of furushi `bundle', Class 5), kizigo `large buttocks' (dim. of mzigo `load', Class 3), kibunda `a lot of money' (dim. of bunda `bale', Class 5). [back]

16. Of the 656 Class 5 nouns in my database, only 72, or 10.9%, begin with j-. However for almost all polysyllabic stems the initial j- must be analyzed as part of the noun stem, since it is not replaced by a plural prefix when the noun is pluralized. Thus compare jani 'leaf', pl. majani with jicho 'eye', pl. macho. [back]

17. The prefix for Class 9, originally a nasal consonant in Proto- Sabaki, has been retained only before voiced stops, vowel-initial, or monosyllabic stems, e.g. nguo 'clothing', nyoka 'snake', mbwa 'dog' (cf. N&H 199). Furthermore, since nouns of Class 9 do not change their prefix in the plural, the status of even this vestigial prefix is marginal: it is recognizable only when a Class 9 stem is made diminutive or augmentative (e.g. ny-oka 'snake', j- oka 'large snake'). Thus zero is the most common realization of the Class 9 prefix as well as that of Class 5.[back]

18. There is some controversy about whether "nonce borrowings" should be distinguished from other types of borrowings or from code-switches (see for example Myers-Scotton (1993:181ff.). For present purposes, this issue is irrelevant. My point is that Class 9 acts as the "default" for foreign words that are not frequent enough to have acquired a regular association with a particular noun class, regardless whether one calls these "nonce borrowings" or "code-switches". I will discuss exceptions to this generalization below.[back]

19. I gratefully acknowledge the help of Arvi Hurskainen of the University of Helsinki for extracting the data for me.[back]

20. It will be recalled that extendedness in one dimension is also part of the semantic network of Class 3, by association with plants, especially trees. There is a difference between long things in Class 3 and those in Class 11, however: those in Class 11 tend to be flexible, like fibers, whereas those in Class 3 tend to be stiff, like trees. [back]