[Lingtyp] homeostatic property clusters

Mon Feb 8 11:01:24 UTC 2021

Dear all,

A brief reminder: since the current discussion is mainly concerned with matters of cross-linguistic categorization in phonology (originally, at least), it may be useful to point out that we had a similar debate about the relationship between language-particular and cross-linguistic (‘comparative’) categories in the area of morpho-syntax in 2015-2016, which resulted in a special issue of Linguistic Typology (2016, vol. 16-2) as well as round-table discussions at conferences.

For example, in the special LT issue some argued that
- all categories are the products of human cognition (cf. Locke 1689/1825: 322), i.e. there is no fundamental difference between ’natural’ and other categories (also outside linguistics);
- the way we set up categories (‘carve languages’), and which features or characteristics in our analysis should count as ‘relevant’ or ‘necessary’, depends on one’s theory, goals, method, data, and cultural factors. See also Round’s comments on Febr. 6: “…what appears to be a genealogical/areal fact about languages may actual be, to some extent, a fact about linguists.”

If this is correct, disagreement is an inherent aspect of any attempt at categorization.

But this doesn’t mean it is impossible “to do good typology”, as suggested by the word ‘prospects’ in the title of my contribution in LT 2016-2 (333-363) ‘Crosslinguistic categories in morphosyntactic typology: Problems and prospects’. Minimally required are (i) a proper sampling method (representative, unbiased etc.) and (ii) a transparent way to define category membership (strictly distinguishing between formal, semantic and functional features). This way cross-linguistic categories can support scientific induction (see also Spike 2020).

Best, Jan Rijkhoff

School of Communication and Culture, Aarhus University
Jens Chr. Skous Vej 2, Building 1485-621
DK-8000 Aarhus C, DENMARK
URL: http://pure.au.dk/portal/en/linjr@cc.au.dk

________________________________________
From: Lingtyp <lingtyp-bounces at listserv.linguistlist.org> on behalf of Martin Haspelmath <martin_haspelmath at eva.mpg.de>
Sent: Saturday, February 6, 2021 3:31 PM
To: lingtyp at listserv.linguistlist.org
Subject: [Lingtyp] homeostatic property clusters

Erich Round mentions Spike's (2020) paper and suggests that Spike showed that one need not distinguish descriptive categories and comparative concepts – and here I would like to bring up the notion of "homeostatic property concepts" that Dahl (2016) brought into the discussion.

Erich Round wrote:

  *   You’re right I want to “have it both ways”, to treat languages as systems and compare them. This is a familiar red line that has separated your views from other typologists for a while now. I appreciate that you had an argument couched in philosophical terms that seemed to lead to your conclusion. Formulating such arguments is hard and valuable work, but in my view Spike (2020) demonstrates that the argument fails, predominantly because it’s based on premises that turn out to be false. This is fine; typology benefits from such debates: you raised some interesting problems; Spike engaged with them and showed them to be apparent, not real.

Spike's discussion is mostly at an abstract philosophical level that many linguists will find hard to understand (see some reactions from me here<https://dlc.hypotheses.org/2410> and here<https://dlc.hypotheses.org/1963>), but he cites the concrete example of Östen Dahl's work on tense-aspect categories. What was innovative about is was that it was based on a parallel questionnaire and other parallel texts, and Dahl found "gram clusters" such as "perfect grams", "imperfective grams", "habitual grams".

According to Dahl (2016: 435), we can see these as similar to Boyd's (1999) homeostatic property clusters (HPC):

"According to HPC theory, a natural kind is a group of entities with stable similarities, where there may however be no properties shared by all and only the members of the group. The only condition is that the similarities are stable enough to make better than chance predictions and that there are maintained by “homeostatic causal mechanisms”. In the case of biological species, these mechanisms are inheritance of shared genetic material and environmental pressures."

So this is presented by Spike (2020) as an alternative to the tripartition between descriptive (p-)categories, comparative (g-)concepts, and innate natural-kind categories.

However:

(i) Dahl and Spike do not really suggest that such "clusters" can serve as language-particular descriptive categories (the English Perfect still needs to be distinguished from the Spanish Perfect, because they don't have exactly the same conditions of use)

(ii) HPC theory does not help us understand how generative grammar operates (the main reason I introduced the notion of a "natural-kinds programme" was that I wanted to explain why generative linguists are doing what they are doing; e.g. here: https://dlc.hypotheses.org/1012)

(iii) Even though the Dahlian tense-aspect clusters are of course extremely interesting typological generalizations, we do not understand their "homeostatic causal mechanisms" well.

(iv) Clearly, in order to arrive at Dahlian clusters, one needs comparative concepts of the token-based type, e.g. questionnaire translations, or parallel texts. There is no counterpart to this in biological HPCs – the "similarities" are not defined in the same (semi-arbitrary) way as in linguistics.

(v) Spike argues that “The utility of some kind does not require clear-cut, exceptionless definitions, but rather a track record of being used in successful inferences... Agronomists can tell you what to plant, geologists have a good idea of where to look for oil...”

But while agronomists and geologists have had successes which are evident from usesful applications, the same can hardly be said about theoretical linguistics. So we don't have an independent way of assessing how successful or concepts are.

So while Spike (2020) made some interesting contributions (just like Round & Corbett 2020, on which see https://dlc.hypotheses.org/2415), there's no reason to think that there is a problem with the usual way of dealing with uniqueness<https://benjamins.com/catalog/alal.20032.has> – but on the other hand, I also wish Erich a lot of success with his attempts at having his cookies and eat them too :-) Maybe it will eventually turn out that both (or all three) approaches are right, but for different domains.

Best,
Martin

References

Boyd, Richard. 1999. Homeostasis, species, and higher taxa. In Wilson, R. (ed.), Species: New interdisciplinary essays. Cambridge MA: MIT Pres.
Dahl, Östen. 2016. Thoughts on language-specific and crosslinguistic entities. Linguistic Typology 20(2). 427–437. (doi:10.1515/lingty-2016-0016<https://doi.org/10.1515/lingty-2016-0016>)
Round, Erich R. & Corbett, Greville G. 2020. Comparability and measurement in typological science: The bright future for linguistics. Linguistic Typology. De Gruyter Mouton 24(3). 489–525. (doi:10.1515/lingty-2020-2060<https://doi.org/10.1515/lingty-2020-2060>)
Spike, Matthew. 2020. Fifty shades of grue: Indeterminate categories and induction in and out of the language sciences. Linguistic Typology. De Gruyter Mouton 24(3). 465–488. (doi:10.1515/lingty-2020-2061<https://doi.org/10.1515/lingty-2020-2061>)

--
Martin Haspelmath
Max Planck Institute for Evolutionary Anthropology
Deutscher Platz 6
D-04103 Leipzig
https://www.shh.mpg.de/employees/42385/25522