[Lingtyp] homeostatic property clusters

Martin Haspelmath martin_haspelmath at eva.mpg.de
Thu Feb 11 10:23:35 UTC 2021

Many thanks, Adam, for providing this additional perspective, trying to 
defend (aspects of) generative grammar!

Yes, I had seen your 2018 dissertation on Tuparí, and I quoted from it 
in my blogpost on the role of generative grammar in language description 

You are absolutely right (and I think these points are not 
controversial) that
– "generative grammar" is internally diverse and includes HPSG, LFG and 
– many generative grammarians do careful, detailed, and honest work on 
non-European languages
–  work that has come out of the generative tradition has greatly 
contributed to our understanding in the past

In fact, it seems to me that there has been a drastic reversal since the 
1980s: During that period, linguists working on non-European languages 
felt that generative grammar was not serving them well and often turned 
to "typology". But now, people working on English, German or other 
bigger languages are more likely to work with corpora and to identify as 
"construction grammarians" or "usage-based linguists".

By contrast, "C-generative grammar" (the tradition of Chomsky 1981/1995) 
has become very typological, as seen in work by influential linguists 
such as Polinsky, Baker, Bobaljik, Legate, and Preminger. (LFG has been 
"typological" since the 1990s, but this is less so with HPSG and the 
others.) This is why I have been paying a lot of attention to generative 
grammar over the last decade and have been trying to understand it.

And I've come to think that Croft's (2009: 157) characterization of the 
practical problems with C-generative grammar was insightful:

"the other serious problem with the generative method: ... the arbitrary 
selection of a subset of distributional contexts (constructions) in a 
language in order to identify a theoretical grammatical 

The "structurally defined positions" (in vP) for argument types can be 
based on all kinds of diverse diagnostics, and the same is true for "the 
process of fine-tuning the CP/the left periphery". While "fine-tuning" 
does indeed suggest "decomposition", the expectation is always that the 
ultimate building blocks are the same in all languages, and there is no 
real place for emic descriptions (and as Jürgen Bohnemeyer put it: it 
should be "blatantly obvious" that the etic-emic distinction must be 
maintained, while at the same time emic descriptions should be inspired 
by "etic decomposition/fine-tuning").

But I'm not saying that the "building block uniformity" approach of 
generative grammar MUST be wrong – ultimately, we could find a set of 
innate building blocks (maybe smallish and abstract, as in "successive 
cyclic movement" proposals; or maybe largish and concrete, as in Cinque 
1999). But like Croft (2009), I don't think there's any prospect of 
finding these building blocks with the "one language at a time" method. 
So for the time being, we need to adopt "measurement uniformity" instead.


Am 10.02.21 um 23:28 schrieb Adam Singerman:
> Dear all,
> Sorry in advance for the long message. I'm not sure how many
> subscribers to this list are like me in being broadly sympathetic to
> both the typological approach and the generative approach, without
> ever feeling fully satisfied by either. But I'd just like to respond
> to some of Martin's comments, which I don't feel do complete justice
> to how many generative grammarians view their approach. (Martin, maybe
> this message of mine can serve as a response to your 2018 blogpost
> critiquing the syntax chapter of my dissertation, which I meant to
> reply to but never did!)
> Let me say, at the outset, that in my view generative grammar isn't
> just Minimalism and its immediate ancestors but also includes a bunch
> of alternative/competing formalisms that have been developed over the
> past few decades, ranging from Lexical Functional Grammar and
> Head-Driven Phrase Structure Grammar to Combinatory Categorial Grammar
> and Tree Adjoining Grammar.
> Re: labels like vP and CP, I think the former at this point is better
> defined than the latter. vP should be thought of as where the thematic
> relationships between the predicate and its arguments are established.
> Here "arguments" don't translate directly into the traditional notions
> of subject and object, which don't survive as intact notions (as far
> as I can tell) for either functional-typologists or generative
> grammarians, but rather onto structurally defined positions: internal
> argument as the sister of the verbal root, external arguments
> introduced slightly higher up in the structure, etc. A fair amount of
> generative work has gone toward figuring out how well these
> *structurally defined positions* map onto more traditional grammatical
> categories, without assuming a clear correspondence between them. And
> CP was originally proposed in the 80s, if I am not mistaken, as an
> endocentric solution to the puzzle of the category S, which was
> intuitively rather than structurally defined and which was exocentric.
> Subsequent research has complicated the idea of CP since many
> different proposals have been offered about how fine-grained the
> structure of these "Complementizer Phrases" could become, but I don't
> see how that process of fine-tuning the CP (or the "fine left
> periphery," as it's sometimes called, following Rizzi) is in principle
> all that different from what typologists do when they realize that
> some kind of general correlation that used to be seen as unitary needs
> to be further decomposed.
> As someone whose research focuses on a non-Indo-European language
> (Tuparí, iso code tpr) with grammatical categories rather different
> than the ones I'm used to from more familiar languages, I am *very
> sympathetic* to the view that generative grammar as a field of inquiry
> needs to pay much more attention to what it's assuming in terms of
> crosslinguistic comparability. This is why I think the biggest
> successes to come out of generative grammar (which, again, I take to
> include not only Minimalism and GB but also LFG, HPSG, GPSG, CCG, TAG)
> have to do with figuring out issues related to locality: it's not hard
> to find evidence that something has been displaced, but how can this
> intuitive idea of displacement be defined and constrained? In his
> e-mail, Martin referred to the "extremely complex movement operations"
> of generative syntax, but I don't think this is fair. In formalized
> Minimalism the movement operations are actually NOT complex. Rather,
> they're well-defined and apply locally. What can look superficially
> complex at times is that movement is an iterable operation. This is
> the key idea behind analyses of what's been termed "successive
> cyclicity," where it looks as if something that shows up in the matrix
> clause (often a content interrogative word or phrase) is interpreted
> thematically in a lower clause. This is also the case for what's been
> called raising. So I would dispute the characterization of these
> movement operations as complex. Of course, movement can be invoked
> inappropriately / without sufficient empirical support, in which case
> we would say that the analysis is weak; but I don't see how this is
> different, in principle, from phonological rules that are invoked
> without sufficient support.
> There are also ways to formalize "movement" in ways that don't
> actually invoke movement, or at least, don't invoke it to the same
> extent as mainstream Minimalist analyses do — this is where other
> formalisms, like Tree Adjoining Grammar and Combinatory Categorial
> Grammar, really shine in my opinion. The most interesting findings
> from the generative tradition, broadly construed, have to do with how
> to best categorize surface displacement in a computationally tractable
> way. Formalisms that succeed on this front (like TAG and CCG) don't
> say anything about a lot of grammatical issues that fieldworkers like
> myself face when dealing with raw data — but they're not designed to
> do that, anymore than the grammaticalization literature is designed to
> deal with successive cyclicity. I wish more extensive cross-linguistic
> work were done in TAG and CCG because I see a lot of underexplored
> potential there, but the nature of our field is that different
> formalisms get developed by different groups of researchers who tend
> to focus only on particular languages, blah blah blah...
> Finally, just as I don't think it would be appropriate to lump all
> historical linguists together or all functional-typologists together,
> the generative world is internally very diverse and full of
> disagreement. In my view much of the least interesting (and,
> empirically, least defensible) stuff to come out of the generative
> tradition is crummy precisely because it doesn't bother to contend
> with basic typological variation. But there are plenty of generative
> grammarians who do careful, detailed work on very non-SAE languages
> and are quite dedicated to being honest to their data while also
> trying to stay internally consistent in their theorizing. In the end,
> I don't think Martin's dichotomy — between trying to understand
> generative grammar and trying to understand languages — is
> sustainable, because the best generative theorizing is based on a rich
> empirical foundation, and detailed descriptions of individual
> languages need to make reference to structural phenomena that we
> understand (to the degree that we can say we understand them) thanks
> at least in part to work that has come out of the generative
> tradition.
> All the best,
> Adam
> _______________________________________________
> Lingtyp mailing list
> Lingtyp at listserv.linguistlist.org
> http://listserv.linguistlist.org/mailman/listinfo/lingtyp

Martin Haspelmath
Max Planck Institute for Evolutionary Anthropology
Deutscher Platz 6
D-04103 Leipzig

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/lingtyp/attachments/20210211/7510b32d/attachment.htm>

More information about the Lingtyp mailing list