cross-linguistic categorization

[Martin Haspelmath]

Thanks, Edith Moravcsik, for the systematization of the issues, and Tom 
Givón, for a historical perspective. Here are some reactions from me:
> (a)	THE LEGITIMACY OF CATEGORIZATION
> Following Bill C., Martin H. and Matthew Dryer, one might question the legitimacy of crosslinguistic grammatical categories on grounds that the members of any one of these categories differ from each other. I cannot see why this is a problem. ... The key to the idea is that things can be different in some respects but the same in other respect. Thus, as long as there is some likeness among entities, we are justified in assigning them to the same category without incurring a contradiction.
>   
Yes, of course -- such categories are what I call "comparative 
concepts". The crucial point is that they are not identical to the 
descriptive categories that we use to analyze languages. As Dan Everett 
points out, the IPA is very useful as a set of comparative concepts in 
phonetics/phonology, even though "no two phonemes and no two phonetic 
segments are exactly the same cross-linguistically").
     Many linguists (not just generativists) assume (often implicitly) 
that the notions that typologists work with are also the notions that 
descriptive linguists working on a single language should use, and that 
categories in different languages are not just similar, but can be 
equated. This is what I object to. (For phonology, Bob Ladd has pointed 
out that many phonologists have worked with the asumption that something 
like the IPA is a universal alphabet of segmental phonology.)
> (b)	CRITERIA FOR CATEGORIZATION
> In an absolute sense, criteria of classification are arbitrarily (or, in Bill C.’s terminology, opportunistically) chosen. However, there are two ways to justify them. First, if we choose criteria for categorization so that they serve a particular research goal, the choice becomes principled instead of being arbitrary. 
I think the choice of criteria for descriptive categories (for 
language-particular analyses) is much less arbitrary than the choice of 
criteria for comparative concepts, because the goal is simple: To 
describe the language in a consistent and complete way. There are 
different ways of doing this, but the possibilities are fairly limited, 
compared to the possibilities of comparing languages with different 
structures.
> (c)	LANGUAGE-SPECIFIC VERSUS COMPARATIVE CONCEPTS
> It seems to me that Martin H. is right in saying that there are categories that are useful in crosslinguistic comparison but that do not play a role in individual language descriptions; but I don’t think this is necessarily so. For example, in a crosslinguistic study, the concept of argument alignment varying over accusative, ergative, and other types is important but in the grammar of a language that is, say, purely ergatively structured, the concept will not play any role. However, another concept, such as of subject-verb agreement, may be a useful category both in single-language grammars and also in a crosslinguistic typology – even if the details of the construction differ across languages. 
>   
"Agreement" is indeed a highly general concept that at first glance 
seems to be suited both for language-particular description and 
cross-linguistic comparison (but of course not "subject-verb agreement", 
because the meaning "subject" varies strongly across languages). But as 
Corbett (2006) has shown, the kinds of phenomena that linguists subsume 
under "agreement" are fairly diverse, and whatever precise definition 
one chooses, one will only capture part of what usually goes by the term.
     In any event, the main point is that a large number of comparative 
concepts are irrelevant in language description (such as alignment), and 
a large number of descriptive categories are irrelevant in language 
comparison, so the two kinds of entities need to be kept separate in 
principle.

T. Givón writes:
> So now a new coalition of alpha males are splitting off and, in a reprise of 
> Bloomfield's maneuver [of splitting off from Hermann Paul], are narrowing the domain once again. This history is, leastwise to me, profoundly depressing. 
I don't see any narrowing of the domain anywhere. Hermann Paul and the 
neogrammarians were narrow in that they disregarded the true range of 
cross-linguistic variation, largely limiting themselves to the languages 
of the European nation states. Franz Boas's name is missing in Givón's 
historical narrative -- his lesson on the radical differences between 
languages needs to be re-taught again and again, because most linguists 
work on English or some other major language and forget the lesson too 
easily (not to mention the seductive simplicity of innate universal 
grammar). 
      The categorial particularist position advocated by Matthew Dryer, 
Bill Croft, Gilbert Lazard, Sonia Cristofaro and myself is fully 
consonant with Hermann Paul, Franz Boas, and Joseph Greenberg, and we 
were all strongly influenced by T. Givón as well, so I don't see any 
"coalition warfare". Just normal scientific debate, without any 
revolutionary rhetoric.
> So far, I have found "universal" categories such as noun/verb or subject/object, and  the complex theory behind them, to be indispensable in my own descriptive field work.
What is clear is that the Greenbergian typological approach has led to 
much more interesting (and transparent) descriptive grammars -- if one 
doesn't know at all what to expect, one cannot easily distinguish what 
is banal from what is special (cf. Dryer 2006). But as in Boas's time, a 
good grammar describes the language in its own terms, with precisely 
defined language-particular categories, not in terms of intuitive 
pretheoretical concepts of "noun/verb" or "subject/object", or in terms 
of vague prototypes (crucial as these often are for comparison and 
explanatory theory).

Greetings,
Martin Haspelmath