WALS and empiricism

[Tom Givon]

Fair enough, Martin. But with some caveats: Some of us have NEVER 
conceded "theory" to the Chomskians. So I would suggest that the work of 
many of us has been theoretical from the word go, in the best sense of 
the word--seeking explanations from wider and wider domains. This is 
articulated as clear as a bell in my "On Understanding Grammar" (1979), 
and was an implicit practice since the very start of the anti-Chomskian 
rebellion (dating in my life to Haj Ross & George Lakoff's 1967 paper 
"Is deep structure necessary?"). So it is high time we stop conceding to 
Chomsky more than his necessary dues (which are extensive but not absolute).

Second: Greenberg's work in typology has been theoretical from the very 
start, in at least three senses I am aware of: (a) the work on 
markedness of categories; (b) The work on the diachronic foundations of 
typology; (c) His later forays into diachrony and evolution. It is of 
course true that he didn't explicitly try to explain his famous 
"Greenbergian correlations". But it is also true that he encouraged 
those of us who did try to explain them, and at least in person accepted 
our theoretical explanations as a matter of course. Alas, he was not as 
explicit as Hermann Paul on these issue (indeed, he never cited him, to 
my knowledge).

Lastly: bigvs. small sample. How big is big enough? The advantage of a 
large sample could only be demonstrated if you can show that you have 
discovered a body of important facts that were ignored by those of us 
who "impressionistically" started their theoretical work on smaller 
samples without waiting for a 100% world-wide data-base (and didn;'t 
apologizing for it). In other words, as in all good empirical science, 
you can falsify our hypotheses by collecting data from a larger sample 
and showing facts in them that are incompatible with those hypotheses. 
So go ahead, we're not perfect. But, quoting from memory from a 1980 
interview with George Watson: "...we didn't just want to find the right 
solution. We wanted to find it with the minimum amount of data...".

Cheers,  TG

========




Martin Haspelmath wrote:
> Thanks, Tom, for these interesting comments! But they demand a response:
>> I think Martin, perhaps inadvertently, articulated the concern that 
>> some of us have felt about the WALS project from its very 
>> inception--it's relentless a-theoretical perspective. To me, this 
>> project has chosen to follow the old empiricist misonception  (vis 
>> Bloomfield,  Carnap, etc.) that facts are, somehow, 
>> theory-independent, and that one can do a theory-free typology. 
> That data and theory are interdependent is a truism. But the problem 
> with the term "theory" in linguistics is that it has been virtually 
> monopolized by the generative view of the world. The Greenbergian 
> functionalist theory underlying WALS is of a very different sort.
>
> What matters to me (and to empirical typology in general) is that it 
> makes no sense to postpone typology till we somehow find "the right 
> theory", which provides all the categories that languages might have. 
> This is basically the generative approach, and for this reason 
> generative linguistics has not led to any major insights into 
> cross-linguistic regularities. Boas, Sapir and Bloomfield were right 
> that all languages have their own categories (a position that was more 
> recently articulated by people such as Lazard, Dryer, LaPolla, Croft, 
> Bickel), so the typological concepts have to be different from 
> linguistic categories.
>> This is done by two implicit moves: First, by defining grammatical 
>> phenomena purely structurally, rather than grouping them by the 
>> *grammaticalized functional domains* that underlie them. And second, 
>> by leaving *diachrony* out of the equation. 
> I find these remarks puzzling. Many of the WALS chapters have been 
> conceived of in terms of functional domains, and of course many of the 
> generalizations are ultimately due to diachronic factors. I think most 
> WALS authors are fully aware of this. Still, to make systematic 
> cross-linguistic databases, we need consistently applicable 
> definitions of the types (which one might call "purely structural 
> definitions").
>
> The main difference between WALS and Givonian work is the scale: In 
> WALS, each chapter looks at 200 languages or more (the average number 
> of languages per map is 400). We feel that this is necessary, because 
> the earlier practice of taking a few languages and jumping to 
> generalizations, while suggestive and interesting, does not provide a 
> firm basis about what is truly general.
>> To my mind, the geographical distribution of grammatical phenomena is 
>> neigh meaningless without considering the diachrony of the particular 
>> languages (or families) in the region. 
> I don't know anyone who would disagree with this statement. The 
> problem is that while grammaticalization gives us the beginning of a 
> theory of morphosyntactic patterns, we don't even have the beginning 
> of a diachronic theory of the large-scale areal patterns. That these 
> are so common for all areas of language structure is a fascinating, 
> though currently quite enigmatic observation.
>> It is of course true that a project could choose to be less 
>> ambitious, and simply give us "pure facts", perhaps in anticipation 
>> that theory-oriented people would later on use those facts to build 
>> their theories. But I have to agree with Hanson (and, for that 
>> matter, Chomsky, perish the thought...) that in science facts are 
>> never theory-neutral, and that to propose to do a science of "pure 
>> facts", even as a preliminary exercise to  subsequent 
>> theory-building, is the height of self delusion.
>>
>> Cheers,  TG
> I don't think that the Greenbergian work from the 1960s is rightly 
> characterized as "the height of self-delusion", and WALS represents a 
> continuation of that tradition. While practically all of the famous 
> Chomskyan parameters of the 1980s have dissipated and disappeared from 
> the scene (see my 2008 paper in the Biberauer volume), the great 
> majority of Greenberg's universals from 1963 have survived.
>
> We are still struggling to understand these patterns, but nobody is 
> deluding themselves. World-wide linguistics is much like geology: You 
> first need to do a lot of on-site fieldwork to get a good sense of 
> what the mountain range is like, before you can begin to construct 
> your ambitious (catastrophist, gradualist, etc.) explanatory stories.
>
> Best,
> Martin
>
>
>
>
>> Martin Haspelmath wrote:
>>> Dear Esa,
>>>
>>> Thanks a lot for writing this detailed commentary on the World Atlas 
>>> of Language Structures (WALS). This is the most detailed review that 
>>> has been written, and we are very grateful for it. Many of the 
>>> individual points of criticism are well-taken, and the WALS authors 
>>> should take them into account in future editions. (We're planning 
>>> future online editions of WALS, see the free online version at 
>>> http://wals.info.)
>>>
>>> Just one comment, concerning one of your major points:
>>>
>>> You write (p. 1): "The reader of WALS is encouraged ... to seek 
>>> *correlations* between the results of different chapters, and this 
>>> clearly presupposes a high degree of compatibility between the views 
>>> of different authors."
>>>
>>> Well, I would say: To find true correlations, the chapters must be 
>>> sufficiently correct, but they don't necessarily have to be very 
>>> compatible, certainly not in terminology. Suppose you want to link 
>>> case-marking and plural marking, and ask whether affixal 
>>> case-marking (as opposed to adpositional marking) correlates with 
>>> affixal plural marking (as opposed to pluralization by number 
>>> words). Then even if the two chapters use different definitions of 
>>> "affixal", you might still get a true correlation. But it will of 
>>> course be a correlation between affixal(1) case-marking and 
>>> affixal(2) pluralization, not between "affixal (tout court) 
>>> case-marking and pluralization".
>>>
>>> My view is that typological definitions are inherently 
>>> linguist-specific, and as such the typological concepts of different 
>>> linguists are bound to be different (unless a Chomsky-like figure 
>>> comes along and imposes widespread "agreement by authority"). So 
>>> care has to be taken in interpreting WALS correlations, of course. 
>>> But this is not a flaw in the design of the project.
>>>
>>> Typology cannot be based on some kind of "definitive" set of 
>>> grammatical concepts, because there is no such list (or if there is, 
>>> i.e. if UG exists after all, we're so far away from knowing what it 
>>> is that it's irrelevant for practical purposes). Each language has 
>>> its own categories, so typologists necessarily have to make up their 
>>> comparative concepts that give them the most interesting results.
>>>
>>> (For more on this, see my paper "Comparative concepts and 
>>> descriptive categories in cross-linguistic studies", on my website 
>>> under "Papers and handouts".)
>>>
>>> Martin Haspelmath
>>>
>>> Esa Itkonen wrote:
>>>> Dear Funknetters: By all accounts, World Atlas of Language 
>>>> Structures (= WALS) is a monumental achievement. Still, two 
>>>> intrepid Finnish linguists (= myself & Anneli Pajunen) have 
>>>> ventured to write a 30-page commentary on it, available on the 
>>>> homepage below. Enjoy!
>>>>
>>>> Esa Itkonen
>>>>
>>>>
>>>> Homepage: http://users.utu.fi/eitkonen
>>>>   
>>>
>>
>>
>