A few months ago, I was invited to give an online presentation to the Center for Linguistic Sciences of Beijing Normal University (my host was Chia-Jung Pan). This blogpost summarizes the main points of the talk (based on the talk handout; there is also a video of the talk, with questions, on Bilibili). The title of the talk was: “Structural, evolutionary and biocognitive explanations are mutually compatible”.
The purpose of the talk was to emphasize that there is no deep reason for ideological divisions in linguistics: we do not need to have “commitments” or “tenets” (e.g. Lakoff 1991; Goldberg 2003); we do not need to decide “what linguistics is about” (e.g. Hornstein 2019); we do not need to “subscribe to” a framework and defend it; and we do not need to perpetuate diverging terminologies. All these things are a reality, but they are not necessary. Different methodological choices (“approaches”) are more compatible with each other than many people think. For the purposes of this talk, I distinguished three types of explanations:
Structural explanations: Language systems are more orderly than one may think at first glance, and we can often reduce apparent variety to deeper regularities.
Evolutionary explanations: In diachronic change, speakers often select variants that increase the fitness or utility of their language system – language systems are (to a significant extent) the product of evolutionary adaptation.
Biocognitive explanations: Language systems are constrained by being acquired by humans with their particular biological properties. Not every logically possible language is attested, so innate biocognitive constraints can explain observed limits on worldwide diversity.
We may thus want to distinguish (i) structural linguistics, (ii) functional-adaptive linguistics, and (iii) biolinguistics, but only if these are not thought of as “competing ideologies” (or “schools of thought”) – the approaches should be complementary. Competition makes sense between competing hypotheses (= competing claims), and perhaps between competing methods (though methods are often complementary, too), but not between competing ideologies (let alone competing “camps”). Next I give concrete examples of different kinds of explanations, and then I say why they are not mutually incompatible.
(I) Structural explanations
Structural explanations reduce apparent variety to deeper regularities. Examples come from vowel systems, word order in German, and differential object marking in Sakha.
This order is quite rigid in German, and it can be described by setting up an abstract template:
[prefield – P1 – middle field – nonfinite verb – P2]
With this abstract template, the number of rules that we need can be simplified: arguments or adverbials can occur in the prefield or the middle field; subordinators occur in P1 and preclude a prefield; and the finite verb (singt, hat) occurs in P1 unless this field is filled by a subordinator. Otherwise the finite verb occurs in P2. (This “field template” goes back to structuralist work on German in the 1930s.)
(C) Sakha differential object marking
The patient object in Sakha (a Turkic language) is accusative-marked, but only when it is definite (Baker 2015), as seen in (5). This corresponds to a word order difference: a definite (and accusative-marked) object precedes an adverb, as seen in (6).
Baker (2015: 126) says that this is because the domain for accusative case assignment is the TP, and when the object is not moved out of the VP in (6a), it is not accessible for case. This may be seen as a “more modern” way of describing syntax than the German field template, but its essence is that it is a kind of structural explanation that generalizes over certain facts by proposing an abstract pattern.
(II) Evolutionary explanations
Evolutionary explanations say that language systems are (to a significant extent) the product of evolutionary adaptation. Let us look at vowel systems and differential object marking again.
(A) vowel systems: Symmetric triangular vowel systems can be explained as making optimal use of the vowel space (dispersion theory: Liljencrants & Lindblom 1972; Gordon 2016). Moreover, vowel systems are constantly recreated through diachronic change, as is illustrated by changes from Latin to French:
u > y mūrus [myʀ] mur ‘wall’
a > ɛ carus [ʃɛʀ] cher ‘dear’
e > a tectum [twa] toit ‘roof’
o > u collum [ku] cou ‘neck’
e > i lēctum [li] lit ‘read’
These changes tend to be of a kind that yields a well-dispersed vowel space (cf. Martinet 1955: Économie des changement phonétiques) – in other words, the changes are adaptive, even though they cannot be said to be directed.
(B) differential object marking: Object marking of definite or animate objects can be explained as making optimal use of role-marking, because those types of nominals that are least likely to be objects are marked (Bossong 1991). The changes are of a kind that yields such systems, e.g.
Latin ad ‘to’ > Spanish a
Latin per ‘though’ > Romanian pe
Russian -a (genitive) > -a (accusative)
German -en (stem marker) > -en (accusative)
Chinese bǎ 把 ‘take‘ > bǎ(acc. preposition)
Again, we can say that the changes are adaptive. “Evolutionary explanations” have also been called “functional explanations” – but I prefer “evolutionary” or “adaptive”, because they operate at the level of language change (like biological evolution). There is no claim of synchronic language-particular “functionality” – synchronically, languages often show nonfunctional or dysfunctional features. The functional-adaptive factors explain general trends, not necessarily language-particular features.
(III) Biocognitive explanations
Innate biocognitive constraints can explain observed limits on worldwide diversity. Specifically: languages are constrained by formal universals (architectures of grammar and types of rules) and substantive universals (specific features and categories) (Chomsky 1965) (it has also been said that they are constrained by an innate blueprint https://dlc.hypotheses.org/2462 for grammar). This can again be illustrated by the three kinds of phenomena.
(A) vowel systems: They are possibly constrained by the universal features of Chomsky & Halle (1968), which all phonological systems are said to be made up of:
“That there must be a rich system of a priori properties – of essential linguistic universals – is fairly obvious… general linguistic theory might propose, as substantive universals, that the lexical items of any language are assigned to fixed categories such as noun, verb and adjective, and that phonetic transcriptions must make use of a particular, fixed set of phonetic features… We will be concerned with the theory of “universal phonetics,” that part of general lingusitics that specifies the class of possible phonetic representations” (Chomsky & Halle 1968: 4)
(B) differential object marking: This is possibly constrained by a substantive set of universally possible case assignment rules, as in Baker’s (2015) theory of “dependent case”. (The rule formulations in (7) are very much simplified; see my discussion in Haspelmath 2018 https://zenodo.org/record/3464933).
(7) a. High case in TP (clause) is ergative.
b. Low case in TP (clause) is accusative.
c. High case in VP is dative.
d. High case in NP is genitive (there is no low case in NP).
e. Unmarked case is nominative-absolutive.
(C) German word order: Again, word order is possibly constrained by universal functional hierarchies (CP – IP – VP) and constraints on movement (cf. Haider 2010). Then we can say that the German prefield is really the specifier of CP, and that the finite verb is in the C position (“complementizer”). The latter is an odd claim (because C is a category, not an abstract “position”), but this analysis has been widely proposed. The hope is thus that the general “CP – IP – VP” system is innate and thus explains German word order.
Why these three types of explanations are not incompatible
After illustrating these three types of explanations, using three types of phenomena, I will now say why I do not regard them as mutually incompatible.
(i) Structural and biocognitive: In generative grammar, both types of explanation are pursued, and the idea is that the structural building blocks are part of the innate grammar blueprint (“Universal Grammar”). So they are obviously compatible. But they are logically independent of each other, and many generative linguists do not want to commit themselves to rich innate grammatical knowledge (especially since Chomsky 2005, who reversed his earlier position and no longer claims that the structural building blocks are innate (cf. Fitch 2016; and this earlier blogpost).
(ii) Structural and evolutionary: In general, structural description is compatible with evolutionary explanation – and indeed, the two are obviously complementary. We need description before we can move on to evolutionary explanation. Consider vowel systems again: A vowel system like /i e a o u/ can be reduced to three binary features: [±high], [±low], [±back], as I noted above. One can propose an elegant description in terms of binary features, and at the same time advance an evolutionary explanation. The same applies to differential object marking (4B above), but I do not know any evolutionary explanation of German word order – this is one of the many random patterns that languages exhibit.
A structural explanation answers the question: Why do speakers talk the way they do? (because they have internalized a certain language system). An evolutionary explanation answers the question: Why are grammatical systems the way they are? (because certain systems are optimal, and adaptive change favours optimal systems). These are why questions at different, compatible levels.
In a sense, both questions are explanatory/theoretical: The first leads us to propose particular theories (p-theories), and the second leads us to general theories (g-theories) (see Haspelmath 2021b). Each grammatical description is a theory of the language (“A grammar of the language L is essentially a theory of L.” (Chomsky 1957: 49). Thus, it makes no sense to contrast theoretical and descriptive linguistics – descriptive linguistics (which describes language structures) is usually a theoretical enterprise (unless it has applied goals) (it also doesn’t make sense to contrast “theory” vs. “typology”).
(iii) Evolutionary and biocognitive: Finally, I claim that functional-evolutionary and biocognitive explanations are compatible. Many authors have framed the approaches taken by different linguists in terms of “functionalism” vs. “formalism” (e.g. Newmeyer 1998; Thomas 2020) and indeed, we observe almost disjoint communities of scholars, with different conferences, journals, and so on. However, I would like to claim that these groups of scholars are divided merely by habits: by different notations and scientific styles, as well as different “hunches”.
The approaches taken by them would be conceptually incompatible if they were committed to radical positions, e.g. if “functionalists” insisted that all general aspects of language structures can be explained by functional-adaptive forces, and that the difference between humans and other species has nothing to do with domain-specific capacities. Or if the “generativists” presupposed that all general aspects of language structures can be explained by innate grammatical knowledge, and that functional-adaptive forces play no role in language structures.
But these are caricatures, and in reality, most functionalists do accept that we are quite far away from explaining everything in functional-adaptive terms (and many of them also use complex formalisms). In addition, all generativists do accept that some structural properties of languages are functionally motivated (and many of them even think that very little is innate; Chomsky 2005). Thus, in each domain, it is an empirical question what the best explanation is – innate knowledge or functional-adaptive forces.
In fact, we find very similar complementarities in biology and linguistics. In biology, some properties of organisms are explained by the makeup of DNA (its discovery was a major breakthrough) – BUT this has not made evolutionary-adaptive explanations superfluous: Organisms are constrained both by DNA and by evolutionary adaptation.
In linguistics, some properties of languags are explained by innate knowledge (UG) (its discovery would be a major breakthrough) – BUT this has not made evolutionary-adaptive explanations superfluous. Languages are constrained both by UG and by functional adaptation.
Competing claims: Differential object marking again
I said earlier that there should be no competing ideologies or camps (because we are all scientists), but there can of course be competing claims (= competing hypotheses).
For example, there are (at least) two competing possibilities for explaining differential object marking patterns, like those in (8)-(10).
A functional-adaptive hypothesis says that object marking occurs when it is most needed, namely with definite and/or animate nominals (which are less likely to be on object position) (this explanation was formulated clearly by Bossong 1991, but he built on earlier work). A biocognitive explanation (Baker’s “dependent case” theory mentioned above) says that accusative case is assigned when the object NP is “close enough” (in a specific way) to the subject NP. Thus, in Sakha and Chinese, the object moves out of the VP and hence gets “closer” to the subject NP, so that it can be assigned case.
The functional-adaptive explanation actually has a much wider scope and allows us to understand the universal finding in (10) (see Haspelmath 2021a).
(10) Differential object marking universal: If a language has an asymmetric split in object marking depending on some prominence scale, then the special marker occurs on the referentially prominent P-argument.
This applies to six different prominence scales (animate > inanimate, definite > indefinite, 1st/2nd > 3rd), and it is based on the general observation that in language use, referentially prominent arguments tend to be agents, while nonprominent arguments tend to be patients.
The biocognitive explanation (Baker 2015) also has wider scope because it allows us to understand why definite nominals show differential accusative case when they occur in a different position from indefinite nominals (cf. Sakha in (6) above, and Chinese in (9)). However, there are several ways in which this theory is more limited: (i) Baker (2015) does not extend his explanation to Chinese – he wants to limit it to “case marking”, not to all kinds of markers that flag nominals (for reasons that did not become quite clear to me). (ii) When the differential marking is conditioned by animacy (as in Spanish), Baker’s biocognitive explanation does not seem to make any prediction. (iii) Even when differential object marking is conditioned by definiteness, it may not be associated with a clear positional difference, as in Hebrew:
I conclude that the evolutionary-functional explanation has much better empirical coverage than the biocognitive-generative explanation, and moreover, it appeals to highly general explanatory factors: efficiency of coding (“marking occurs where it is most needed”) is merely a special case of efficiency of action (I elaborate on this in my 2021 paper on coding asymmetries and predictability).
In general, functional-adaptive explanations are often superior, so why do linguists keep pursuing biocognitive explanations of the generative type? I have long thought about this and would like to suggest an answer: Because they conflate structural and biocognitive linguistics. The problem is not that for ideological reasons, they reject functional-adaptive explanations – the problem is that they do not properly distinguish structural explanations (p-linguistics) and one type of general explanations (g-linguistics). I elaborate on this in the next section.
The conflation of structural and biocognitive linguistics
To recapitulate in a simplified way, I said above that (i) structural explanations explain speaker behaviour (“language description”); (ii) evolutionary explanations explain general trends (“why languages are the way they are”); and (iii) biocognitive explanations explain the differences between species (“why chimpanzees don’t talk”).
Now we see that many linguists conflate structural and biocognitive explanations, assuming that structural explanations must make use of universal building blocks. For example, Aissen (2003: 439) said in her famous paper on differential object marking:
“Optimality Theory (OT) provides a way, I believe, to reconcile the underlying impulse of generative grammar to model syntax in a precise and rigorous fashion with a conception of differential object marking (DOM) which is based on prominence scales. The purpose of the present paper is to develop an approach to this phenomenon within OT which is formal and at the same time expresses the functional-typological understanding of DOM.”
Now what does Aissen mean here by (i) “modeling syntax” (ii) in a “precise and rigorous fashion” and (iii) with a “formal approach”? Aissen seem to mean a “formal framework” that is the same for all languages – the sort of notation that is often taught in syntax textbooks. But it can be the same for all languages only if the framework is thought to be innate – and if the framework is innate, then it also provides an explanation for some of the limits on languages (this is the Principles & Paramaters framework, e.g. Roberts 1996; Baker 2001).
We see this very clearly in Aissen’s Optimality Theory notation:
All the constraints and many of the constraint rankings are thought to be innate – this makes it possible to both describe all languages in the same framework, and to offer this framework as an explanation.
However, even though this reflects widespread practice, there are very good reasons why one cannot (or should not) conflate structural frameworks and claims about innateness in this way. First, it is biologically implausible that rich descriptive frameworks should be innate (Chomsky 2005; Berwick & Chomsky 2016). Second, Aissen (2003) wants to “capture” the insights of economy explanations, but in her conflating approach, there is no link between economy/efficiency and her constraint rankings (I discussed this briefly in §4.5 of my 2008 paper). Third, while different languages often show intriguing similarities, there are often many differences in detail (e.g. “definiteness” is subtly different in Sakha, Chinese and Hebrew) – these need to be described anyway, and the coarse-grained general categories are not sufficient for this. Fourth, functional-evolutionary explanations often have a disparate range of effects – for example, efficiency of argument coding affects objects, subjects, recipients, patients, and a range of different prominence scales (Haspelmath 2021a); these cannot possibly be captured by a single innate framework.
We all agree that linguists must describe (or “model”) the structures of languages in a “precise and rigorous fashion”. In this sense, we are all structuralists (see this blogpost). And unless we concentrate on applied goals, we are all theoretical linguists (Haspelmath 2021b).
And we all agree that formal methods of different kinds are often useful for linguistics. But due to a complicated sociological process, the term has become associated with Chomskyan generative linguistics (see also this blogpost). And we all agree that (i) some aspects of language structures are due to efficiency (e.g. abbreviations such as “MPI” for “Max Planck Institute”), and (ii) human minds and chimpanzee minds differ in ways that gives humans language.
So we really all agree that we need (i) structural explanations (formal models of syntax); (ii) evolutionary explanations (adaptive accounts of efficiency effects), and (iii) biocognitive explanations (explanations of species differences); and that these three must be mutually compatible. So is there an ideological difference between a functionalist and a generativist? No, the difference primarily consists in scholarly traditions, and in “hunches”. The functionalist’s hunch is that the precise nature of the formal description is probably not very important for understanding grammar. The generativist’s hunch is that considerations of functional efficiency are probably not very important for understanding grammar.
For differential object marking (DOM), I have argued that functional efficiency is very important, because it explains many cross-linguistic tendencies (Haspelmath 2021a). But: I do not have any particular “commitments” or “tenets” (Lakoff 1991); I do not “subscribe to” any particular framework (Haspelmath 2010); and I do not say that “linguistics is about languages” (Comrie 1978) rather than universal cognition (Hornstein 2019). Linguistics is about both of these.
Given that scholars are humans, and humans live in traditions, we will probably continue to work in such traditions. But we should not confuse our traditions and hunches with competing ideologies.
Aissen, Judith. 2003. Differential object marking: Iconicity vs. economy. Natural Language & Linguistic Theory 21(3). 435–483.
Baker, Mark C. 2001. The atoms of language. New York: Basic Books.
Baker, Mark C. 2015. Case: Its principles and parameters. Cambridge: Cambridge University Press.
Berwick, Robert C. & Chomsky, Noam. 2016. Why only us: Language and evolution. Cambridge, MA: MIT Press.
Bossong, Georg. 1991. Differential object marking in Romance and beyond. In Kibbee, Douglas & Wanner, Dieter (eds.), New analyses in Romance linguistics, 143–170. Amsterdam: Benjamins. (https://www.rose.uzh.ch/dam/jcr:ffffffff-c23e-37d9-0000-00006e1a9200/Bossong_80.pdf)
Chomsky, Noam A. 1957. Syntactic structures. ’s-Gravenhage: Mouton.
Chomsky, Noam A. 1965. Aspects of the theory of syntax. Cambridge, MA: MIT Press.
Chomsky, Noam A. 2005. Three factors in language design. Linguistic Inquiry 36(1). 1–22. (doi:10.1162/0024389052993655)
Chomsky, Noam A. & Halle, Morris. 1968. The sound pattern of English. New York: Harper & Row.
Comrie, Bernard. 1978. Linguistics is about languages. Studies in the Linguistic Sciences 8(2). 221–236.
Fitch, W. Tecumseh. 2016. Reinventing linguistics – again (Review article of “What kind of creatures are we?”, by Noam Chomsky). Inference: International Review of Science 2(3). (https://inference-review.com/article/reinventing-linguistics-again)
Goldberg, Adele E. 2003. Constructions: A new theoretical approach to language. Trends in Cognitive Sciences 7(5). 219–224.
Gordon, Matthew Kelly. 2016. Phonological typology. Oxford: Oxford University Press.
Haider, Hubert. 2010. The syntax of German. Cambridge: Cambridge University Press.
Haspelmath, Martin. 2010. Framework-free grammatical theory. In Heine, Bernd & Narrog, Heiko (eds.), The Oxford Handbook of Linguistic Analysis, 341–365. Oxford: Oxford University Press.
Haspelmath, Martin. 2021a. Role-reference associations and the explanation of argument coding splits. Linguistics 59(1). 123–174. (doi:10.1515/ling-2020-0252)
Haspelmath, Martin. 2021b. General linguistics must be based on universals (or nonconventional aspects of language). Theoretical Linguistics (to appear). (https://ling.auf.net/lingbuzz/005158)
Hornstein, Norbert. 2019. The stupendous success of the Minimalist Program. In Kertész, András & Moravcsik, Edith & Rákosi, Csilla (eds.), Current approaches to syntax: A comparative handbook. Berlin: De Gruyter.
Lakoff, George. 1991. Cognitive versus generative linguistics: How commitments influence results. Language and Communication 11(1/2). 53–62.
Liljencrants, Johan & Lindblom, Björn. 1972. Numerical simulation of vowel quality systems: The role of perceptual contrast. Language 48(4). 839–862.
Newmeyer, Frederick J. 1998. Language form and language function. Cambridge, MA: MIT Press.
Roberts, Ian. 1996. Comparative syntax. London: Edward Arnold.
Thomas, Margaret. 2020. Formalism and functionalism in linguistics: The engineer and the collector. New York: Routledge.