Linguists treat many technical terms as so well-established that they are not in need of explanation or definition, or that any further discussion is a secondary matter. But even a widely used term such as “word” is not well-understood (Haspelmath 2011), so it is not surprising that more specialized terms such as “noun incorporation” suffer from the same problems. This is clearest if one defines “noun incorporation” in terms of wordhood, as in de Reuse (1994a: 2842), where noun incoproration is defined as
“the morphological construction where a nominal lexical element is added to a verbal lexical element; the resulting construction being a verb and a single word”
This definition (as well as Gerdts’s (1998) very similar definition) is fully in line with Wilhelm von Humboldt’s original use of the term “einverleibend” (= incorporating) for examples such as Classical Aztec ni-naka-kwa [1SG-meat-eat] ‘I eat meat’, where -naka- ‘meat’ seems to be part of the complex verbal word ninakakwa, at least according to the spelling. But how exactly do we know that ni-naka-kwa is a single word, apart from the spelling? Couldn’t we write it ni naka kwa and regard it as a syntactic phrase? To be sure, this syntactic phrase has some peculiarities, but then so do many other syntactic phrases. Clearly, a definition of noun incorporation in terms of wordhood is only as good as one’s definition of word is, or one’s definition of “morphological compound”.
In Rochelle Lieber & Pavol Štekauer’s (2009) introduction to their Handbook of Compounding, they discuss a range of criteria for distinguishing compounds from syntactic phrases, and their summary begins with the following remarks:
“The picture that emerges here may seem a bit dispiriting: what are we to think if there are (almost) no reliable criteria for distinguishing compounds from phrases or from other sorts of derived words? … Nevertheless, the majority of theorists – us among them – seem to believe that it’s worth looking further.” (p. 14)
It’s never wrong to try again, but probably the field of linguistics as a whole should acknowledge that at present it doesn’t know what compounding is (as a general process; there may of course be language-particular compound categories which are quite well-defined). If we want to make general claims about compounding in language, we cannot rely on people knowing what we mean.
Given this, it’s interesting to see a typological definition of noun incorporation that does not rely exclusively on wordhood. Caballero et al. (2008) count a noun as incorporated “if it occurs between parts of the inflected verbal complex”, as in the Aztec example ni-naka-kwa. The authors allow the verbal complex to consist of non-affixed material as well, e.g. the subject agreement marker in post-object position in Car Nicobarese in (1):
(1) tínŋɛ ́n tətmák an céˑms
sent.away Tetmak 3SG.SBJ James
‘James sent Tetmak away.’
Here we have post-incorporation of the object, because the inflectional marker follows the object. This is the normal position of the agreement marker, so noun incorporation is obligatory in the language. But if such cases are regarded as noun incorporation, then the same should be said about German compound tense clauses as in (2):
(2) Ich habe ein Buch gekauft.
I have a book bought
‘I bought a book.’
Conversely, if a language forms compounds but does not insert them “into the verbal complex”, this does not count as incorporation. Thus, de Reuse (1994b) has argued that Lakota has noun incorporation, but unlike in Aztec, the incorporee does not follow the person prefixes:
(3) napé-ma-kìpazo (*ma-nape-kipazo)
‘show me (your) hands’
The problem with Caballero et al.’s definition is thus that it leads to a concept of “noun incorporation” that is quite at odds with common usage. Even though Humboldt may originally have had in mind the Aztec-type situation with an affix occurring before the N+V complex (or after the V+N complex), in view of the “in-corporation” metaphor, linguists have not normally required the presence of inflectional affixes in a particular position. Note also that on this view of incorporation, isolating languages could not have incorporation by definition.
There is a completely different direction in which the term incorporation has been extended over the last 15 years: Since the Aztec-style incorporation prototype tends to be associated with nonreferential and generic semantics, some linguists have used “incorporation” in a semantic sense, also for constructions that have no syntactic peculiarities (e.g. She took the train), or for constructions with articleless (“bare”) noun phrases (e.g. He attended class) (see van Geenhoven 1998, Carlson 2006). This is somewhat like the extension of the term case to mean semantic role: A term for a formal pattern is taken and given a meaning that derives from some of the salient semantic properties of the formal term (see Lehmann 2007 for this general tendency in the development of linguistic terminology).
So what should we do? When a term has acquired too many senses, my general preference is to abandon it (see, e.g., Haspelmath 2006 on “markedness”), and use other terms instead. That this may be workable can perhaps be illustrated from the following generalization:
(4) If a noun is used as an argument but is indefinite, does not refer to a specific referent, has narrow scope, and is number-neutral, then languages tend to express it without article, tend to disallow modifiers and tend to give it little positional freedom with respect to the verb.
This is equivalent to the statement “formally incorporated nouns tend to be semantically incorporated” but makes no use of the term incorporation. It seems that (4) has a much better chance of being testable than the corresponding statement with the undefinable term incorporation.