Classifying, categorizing, demarcating - is that what linguistics is all about? Ill-tempered ruminations on two paradoxes about how linguists spend their time.

1/ First paradox:
On the one hand, in language nothing is a hundred percent anything (did Talmy Givòn say that?). Instead of clear-cut categories, we have continua all over the place: from typical nouns to typical verbs (via nominalizations, gerunds, participles and the like),

2/ from typical lexical verbs to typical auxiliary verbs (via catenatives, modals, semi modals etc), from obligatory complements to optional adjuncts (via optional complements and obligatory adjuncts), from compositional noun phrases to typical compounds, from x to y, you name it

3/ On the other hand, linguists spend an estimated 90% of their time categorizing, labelling and (first and foremost) quibbling over how to come up with clear-cut categories and how to label them. Hmm

4/ Second paradox, based on the observation that demarcation issues have two types of sources: linguistic ones and metalinguistic ones.

5/ The linguistic ones lie in the nature of language. It is not (only) that bloody native speakers don’t stick to the rules described by linguists. Gradience is simply an inherent property of language. Discussing linguistic sources of demarcation issues is interesting.

6/ The metalinguistic ones lie in the nature of our theories and in the way we do linguistics. Metalinguistic demarcation issues are less interesting, mainly because they are often home-made, theory-internal, moot or don’t do justice to the linguistic data in the first place.

7/ But discussing metalinguistic demarcation issues accounts for an estimated 90% of the 90% of the quibbling time. Hmm
So should we stop bothering? Of course not.

8/ We need metalinguistic categories to describe language and formulate predictions. If item x can be treated as a member of category a, and we (think we) know what defines category a, then we can predict how x will behave in terms of usage options, constraints etc.

9/ The way out? Maybe we can shift the focus from quibbling over metalinguistic and meta-metalinguistic categories (especially if they are only required to keep some theory nicely consistent) to more nuanced, data-driven descriptions and explanatins of linguistic ones.