Traditionally, they are not generally called "relational nouns", but this is probably simply down to the fact that it is simply not the current tradition. (Traditions in Swahili teaching terminology are calling the applicative voice the "prepositional", calling non-initial prefixes "infixes" although Swahili lacks true infixes, and calling the middle-voice/mediopassive suffix the "stative extension" in spite of it predominantly having dynamic uses.)
Most of the initial elements of Swahili "compound prepositions" are quite transparently derived from nouns. Ndani and nje both possess the typical prenasalising prefix of class 9/10 and then take class 9 agreement on -a. Then there are cases like miongoni mwa, where the first element has the locative suffix -ni, which triggers agreement with the locative classes 16, 17 and 18 and we indeed find that the mwa includes the class 18 locative agreement.
Other examples that are quite transparently nominal, as the base noun from which they are derived is still in common use:
- usoni pa "in the face of", "on the surface of", "in front of"
- ukingoni mwa "on the edge of", "on the brink of"
- mvunguni mwa "under".
Of course, in many languages that use relational nouns, such as Turkish and Māori, the noun itself is marked with a preposition or a case that indicates that it is being used as an adjunct, for example the kei marking roto in the Māori example and the -de marking iç- in the Turkish example below.
kei roto ahau i te whare
PRES.LOC inside.space 1S REL.LOC DEF.SG house
"I'm inside the house".
ev -2n iç -4(n) -d2 -(y)4m
*Roto ahau i te whare and **evin içiyim are ungrammatical with the intended sense, however that is roughly equivalent to the structure we find in Swahili.
(When the relational location noun in Māori is in the predicate, marked by tensed prepositions, the subject usually stands before the remaining noun phrase. The morpheme iç- is bound in Turkish, always occurring with a possessive suffix. The numbers I have used in the morphemic breakdown indicate different vowel harmony classes: 2 = e/a; 4 = i/ı/ü/u.)
One argument to say that the Swahili equivalent are adverbs, not nouns, is that they are not themselves marked as locative through a case or a preposition. For example, "inside" is just ndani ya, not *katika ndani ya. However, this touches on a more general difficult in Swahili to test what is a noun and what is an adverb; unmarked nouns in Swahili can also be used adverbially.
This is most noticeable for locative nouns. The locative ending, -ni changes the class of the noun to one of the locative classes, C16, C17 or C18, but nouns in these classes can be subjects, objects or adjuncts with no additional marking associated with their use as adjuncts. I'll show it with hapa "here", which is a demonstrative in class 16 (essentially meaning "this (precise) place") rather than a noun, but it illustrates the issue nicely as it can be used as subject, object or as an unmarked adjunct.
"It's nice here."
"This place is nice.
"I like it here."
"I like this place."
"We will sleep here."
"We will sleep in this place."
This is not restricted to locative class demonstratives and nouns (mahali and those ending in -ni) and also applies to place names and, in certain cases, other nouns without any locative marking and which keep their non-locative class.
Another example is the classes ki- (C7) and vi- (C8) which are frequently used as adverbs and adjectives, both with and without additional marking with -a depending on the word.
Then there are cases of voice changing operations applied to verbs, where noun phrases that one would expect to be excluded are still present, unmarked, as though modifying the verb in a way that provides scope or context to the meaning. Sorry that I have expressed that in a rather wishy-washy way and cannot think of any examples of that. I'll come back to that if there is interest, and if I have the points to comment by then.
As an example of this class of word, the word juu is regarded by dictionaries as both an adverb meaning "up", "up high" and a class 9 noun meaning "top" in Swahili. It comes from Proto-Bantu *ìgʊ̀dʊ̀, meaning "sky" or "top" (and, incidentally, is cognate to the word "Zulu") and, in Swahili, it has changed from C5 (where it would be *juu la) to C9 (giving juu ya), possibly (my own speculation) by analogy to other "relational nouns" such as ndani and nje. Because the consonant /ɟ/ at the beginning of juu can be prenasalised, I would expect it to have the form *njuu if it had switched to C9 in the stage of Swahili's development before the prenasalising prefix stopped being productively applied to members of C9/C10, so I suspect it changed class relatively recently, although, again, this is just my speculation.
In any case, it clearly fits within the nominal system as it can be used as subject, object. Its additional ability to be used unmarked as an adjunct lends to its interpretation as an adverb, but does not set it apart from other nouns which can do likewise in Swahili. Drawing a hard line between nouns and adverbs in Swahili is tricky.