Two methods for comparative grammar: Measurement uniformity and building block uniformity

At this year’s annual meeting of the DGfS in Hamburg (2020), I organized a workshop on the empirical testing of grammatical universals, because I feel that universals are too often taken for granted (here is the handout of my talk). The well-known example of a universal morphology-syntax distinction is just the tip of the iceberg. Weirdly, Bauer (2019: 2) says in his recent book on the foundations of morphology:

“Any hard-and-fast definition of ‘word’, even in English, is likely to be problematic. Like most books on morphology, this book will ignore the problem. It will be assumed that orthographic words represent ‘words’ in the more general sense… This is far from ideal, but the only way of making any progress.” (Bauer 2019: 2)

But what if the alleged syntax-morphology distinction is just based on a spelling-based stereotype (the “word”, which becomes salient to all of us by the age of primary school, or by the time we stdy English if we are Chinese or Thai), rather than on anything that is actually present in the languages? A syntax-morphology distinction (or a “word” notion) is not logically necessary, so shouldn’t we set up tests, rather than simply presuppose its existence (as is very often done in our field?).

In order to move from such “presupposed universals” to empirically responsible supported universals (that we can rely on and build more ideas on), we need to study a wide range of languages and compare them on the basis of uniform concepts for comparison. In my Hamburg talk, I distinguished two types of uniformity: measurement uniformity and building block uniformity (see also §8 of Haspelmath 2019).

(1) Measurement uniformity

In many sciences, the comparison of a diverse range of objects requires uniform yardsticks of comparison. Physical units of measurement (especially the SI units such as metre, gram, and second) are crucial for the most basic dimensions, but many other dimensions of reality are measured by quantitative sciences. In comparative linguistics, when we adopt a quantitative approach, we need uniformly defined comparative concepts for our measurements. In other words, quantification is based on measurement uniformity. This may seem a totally trivial observation, but it is still rather uncommon to find the term “measurement” in comparative linguistics. Bickel (2007) may have been one of the first papers to use the term:

“In order to capture and test distributions, typologists develop variables that measure similarities and differences between languages…    If there are large-scale connections between linguistic structures, or between linguistic structures and geography, they consist in probabilistic (and therefore exception-ridden) correlations between independently measured variables; they are not expected to follow from absolutely defined or “ideal” types.” (Bickel 2007: 241; 245)

There is now also a recent paper by Round & Corbett (2021) called “Comparability and measurement in typological science”  which also highlights the measurement aspect of a comparative linguistics that must ultimately be quantitative.

For measurement uniformity, it is crucial that the concepts for measurement are defined in the same way for all languages, because the comparative concepts are not more than that: units of measurement. They measure aspects of the structures of the languages, but they are not claimed to be parts of the languages. When we compare words across languages (e.g. English drive, German fahren ‘go by surface vehicle’, and Russian exat’ ‘go by horse or wheeled vehicle’), we need to settle on comparison meanings (e.g. ‘go by car’), because the parts of different languages are not coextensive. The three words drive/fahren/exat’ are obviously similar, and we want to measure this similarity, but the languages do not have the same parts. In other words, we need etic concepts for comparison, while the parts of languages are best described in terms of language-particular emic categories. The etic comparative concepts can be standardized (cf. the Concepticon, a standard list of comparison meanings), just like other units of measurement have been standardized.

This approach in terms of measurement uniformity has yielded many important results (e.g. Shopen (ed.) 2007; Dixon 2010-2012), and there are by now quite a few online databases that work with measurement uniformity (e.g. WALS, APiCS, SAILS).

(2) Building block uniformity

But many other linguists prefer a different approach, in terms of building block uniformity. They think that the parts of particular languages that we identify in our analyses belong to a universal inventory, which is part of the innate grammar blueprint (sometimes called “universal grammar”).

Quite rightly, Baker (2001) compared this research programme to the identification of chemical elements in the 19th century. Just as Dmitri Mendeleyev discovered the Periodic Table of Elements by studying diverse chemical compounds, linguists can search for the building blocks of the innate grammar blueprint by studying diverse languages. On this view, languages are inherently comparable once one analyzes them in the right way. And the right analysis is not something that can be observed immediately: It is often the result of a laborious process of “in-depth study” (see also my 2019 paper on in-depth analyses and their problems).

In this view, no special set of comparative concepts is required for measuring differences between languages – because all languages ultimately consist of the same parts. The incommensurability of the structures of diverse languages that was highlighted by linguists such as Franz Boas, Gilbert Lazard, Bill Croft and Nick Evans is thought to be merely illusory. Even though languages look diverse at first blush, a closer look reveals that there is a universal set of phonological features, semantic primes, and syntactic categories (perhaps over a hundred universal functional heads of the syntactic spine, as in Cinque 1999).

Like the chemical elements, the innate building blocks of language systems are thus natural kinds – categories of nature that are independent of human observation. I have therefore called this research programme the natural-kinds programme (Haspelmath 2018).

After Antoine Lavoisier (1743-1794) had originally proposed the idea of chemical elements and compounds, several decades of research by chemists added more and more proposals of elements, and by the 1860s, this work had been completed (resulting in Mendeleyev’s Table). Similarly, linguistics may be in a process of working out the innate building blocks of grammars, and this is certainly what many textbooks of phonology and syntax teach. Clearly, this process is complex and laborious, because

“there are often multiple options and we can’t explore them all; and unexpected discoveries require us to question our conclusions (and retrace our steps)” (2020 interview with David Pesetsky)

But while all linguists who practice the building block approach agree that the process is slow and fraught with difficulties, there is one important presupposition that I want to highlight here: Building block uniformity requires the building blocks to be innate if one posits them on the basis of diverse language systems. In the 20th century, Noam Chomsky and his followers did not question the innateness of “universal grammar” (on the contrary – it was thought to provide the crucial solution for Plato’s Problem), and if a building block is given in advance, either to children or to linguists, then one can hypothesize its presence and accept it on the basis of a diverse range of indications. In fact, it makes sense to posit a “Uniformity Principle”:

“Uniformity Principle
In the absence of compelling evidence to the contrary, assume languages to be uniform, with variety restricted to easily detectable properties of utterances.” (Chomsky 2001: 2)

Since the innate building blocks interact in diverse ways, we may not see them in the same way in all languages, but we can hypothesize them to be there unless we have counterevidence. But there is no counterevidence to the universality of binary branching, X-bar theory, the c-command condition on binding, the universality of the cartographic spine, merge, Agree, and feature checking – anymore than there is counterevidence to the universality of the morphology-syntax distinction. All these hypothesized innate building blocks can be found to be relevant in a host of languages, so maybe they are indeed all innate (who knows?). It may well be that the research programme that assumes universal building blocks will eventually converge on a universal list, analogous to the list of distinctive features.

However, in what I find the most enigmatic development of linguistics of the 21st century, many colleagues seem to want to combine two positions: (i) agnosticism about the innateness of the building blocks, and (ii) the Uniformity Principle as a guideline for their language-particular research. This makes no sense, as far as I can see, because we cannot assume that some hidden uniformity underlies the observable diversity unless there is a causal mechanism for the uniformity. Innateness provides a credible causal mechanism, but if the building blocks of languages are not innate, then there is no reason to assume that all languages are made of the same building blocks.

Compare an analogy from a different domain of cultural diversity: On the basis of European medieval societies, one might hypothesize that all human societies consist of three “estates”: Nobility, Clergy, and Peasants. Ancient Roman society would then be mapped onto this by saying that the patricians are in fact the Nobility, the plebeians are the Peasants, and the clergy would perhaps be seen as an empty position, not filled by Roman society. In modern Western societies, one might say that the owners of capital are the Nobility, the scientists and educators are the Clergy, and the workers (white-collar or blue-collar) are the Peasants. This could be considered an initially successful analysis, and more research would hopefully bring better mappings to light, perhaps by decomposing the universal three estates (e.g. decomposing peasants into burghers and true peasants).

If there were any reason to assume that there is an innate division of societies into three or more “estates”, this would make good sense. But since nobody makes that assumption, such an approach would not be taken seriously by anthropologists. To compare societies, we need comparative concepts that are defined in the same way for all societies, i.e. we need measurement uniformity.

Now for languages, we do have fairly good reasons to assume innateness of many building blocks – Plato’s Problem (as mentioned earlier), and in particular, the undisputed universality of complex grammatical structures in all societies. Clearly, human linguisticality is biologically grounded, and languages are not just cultural inventions like ploughs and bicycles.

So building block uniformity cannot be discounted, and must be taken as a serious competitor of measurement uniformity (even though it is of course quite implausible that there is a large number of innate building blocks; but science has often found the truth to be what was earlier considered implausible). However, building block uniformity is incompatible with Chomsky’s 21st century view that there is no innate grammar blueprint. If only “merge” is innate (as Chomsky seems to have concluded on the basis of Darwin’s Problem; cf. Berwick & Chomsky 2016), then comparison must be based on measurement uniformity. If there is no rich UG, then one cannot appeal to the Uniformity Principle. If the new Chomsky is right, then languages do not share parts, and each language is better described in its own terms rather than in terms of concepts that were developed for some other language.

References

Baker, Mark C. 2001. The atoms of language. New York: Basic Books.

Bauer, Laurie. 2019. Rethinking morphology. Edinburgh: Edinburgh University Press.

Berwick, Robert C. & Chomsky, Noam. 2016. Why only us: Language and evolution. Cambridge, MA: MIT Press.

Bickel, Balthasar. 2007. Typology in the 21st century: Major current developments. Linguistic Typology 11(1). 239–251.

Chomsky, Noam A. 2001. Derivation by phase. In Kenstowicz, Michael (ed.), Ken Hale: A life in language, 1–52. Cambridge, MA: MIT Press.

Cinque, Guglielmo. 1999. Adverbs and functional heads: A cross-linguistic approach. New York: Oxford University Press.

Dixon, R. M. W. 2010. Basic linguistic theory. Vol. 2. Oxford: Oxford University Press.

Haspelmath, Martin. 2018. How comparative concepts and descriptive linguistic categories are different. In Van Olmen, Daniël & Mortelmans, Tanja & Brisard, Frank (eds.), Aspects of linguistic variation: Studies in honor of Johan van der Auwera, 83–113. Berlin: De Gruyter Mouton. (https://zenodo.org/record/3519206)

Haspelmath, Martin. 2019. Ergativity and depth of analysis. Rhema 2019(4). 108–130. (doi:10.31862/2500-2953-2019-4-108-130)

Haspelmath, Martin. 2020. Human linguisticality and the building blocks of languages. Frontiers in Psychology 10(3056). 1–10. (doi:10.3389/fpsyg.2019.03056)

Round, Erich & Corbett, Greville G. 2021. Comparability and measurement in typological science: The bright future for linguistics. to appear. (https://www.academia.edu/40915670/Comparability_and_measurement_in_typological_science_the_bright_future_for_linguistics_Erich_Round_and_Greville_G._Corbett_)

Shopen, Timothy (ed.). 2007. Language typology and syntactic description. 3 vols. Cambridge: Cambridge University Press.


Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.