Since the mid-20th century, structural linguists have often made use of two types of abstract devices that were not part of the earlier arsenal (which did of course include rules and paradigms): zero elements (or empty positions), and transformations (or derivations, or operations).
Zeroes have become popular since Jakobson’s famous (1939) paper “Signe zéro” (though he did not invent them), and transformations have become popular through Chomsky’s (1957) book “Syntactic structures” (though the concept and the term were used earlier by Zellig Harris). Of course, various kinds of derivational relationships that can be thought of as transformations were used in earlier grammars, in particular the derivation of nouns from verbs.
What zeroes and transformations share is that they have encountered resistance: Many linguists reject one or both of these devices, or treat them with great suspicion. Transformations, in particular, have become markers of communities of linguists – since the 1970s, many linguists have characterized their work as “non-transformational”, “non-derivational” or “constraint-based”. The resistance against zeroes is perhaps not quite as strong, but to some extent it is the same kinds of linguists that resist both transformations and zeroes.
A striking example of a negative attitude towards zeroes is Aronoff’s (1976: 71) outright condemnation of the idea of “zero-derivation” in English (as in butterN / butterV):
“… the concept of a formless phonological substance like this is abhorrent, even ridiculous, when we realize that for every word-formation rule which has no associated phonological operation (and there are several in English (cf. Marchand 1969: 359-389), we must posit a separate such entity, with a resulting proliferation of zeros, one for every rule: Ø1, Ø2, Ø3, …”
But what is the problem with “proliferation”, either of zeroes or of transformations? We have surely seen the latter as well – movement transformations all over the place. Many people find analyses with many movements “excessive”, but by what criteria?
I would like to suggest the following here: Zeroes and transformations are often useful to express language-particular generalizations. They are good for p-linguistic analyses. But whether they should have any status in general linguistics is a different matter.
Take a simple example of determiners in English, where there are reasons to say that there is a zero element: If we say that the singular and plural forms of the core determiners are the/the, this/these, that/those, a/Ø, then this is straightforward and gets the facts right with a minimum of complications. It is true that we could say more, e.g. that the definite article the is actually number-neutral (rather than showing homonymous singular and plural forms), or that there is no determiner when a plural noun is indefinite. These additional statements may be more satisfactory in some sense, but they complicate the story. So from the perspective of simplifying language-particular descriptions, there are good reasons here to adopt the zero analysis – and there are many similar cases elsewhere. It’s the same with transformations: If we say, for example, that the finite verb moves to the second position in German clauses that lack a complementizer, this allows a simplification of the syntactic statements.
But what does this mean for general linguistics – for our understanding of Human Language in general? Are there any obvious consequences that go beyond the simplicity of language-particular description?
Many linguists seem to assume that this is the case, because they do not distinguish properly between p-analyses and questions of g-linguistics. For example, Cavirani & van Oostendorp (2017: §1.1) first note that zero/empty elements allow more elegant language-particular theories/descriptions:
“What are possible arguments in favor of empty positions in the first place? Such arguments typically come from the construction of a theory, which can be set up more elegantly if we assume that a certain position is there even if it receives no phonetic interpretation. Typically the “theory-internal” nature of such positions is also the main point of criticism for the detractors of abstract positions, because their inaudibility makes them less “real” to their minds; such scholars would prefer a more complicated description of the data if it does not take recourse to such emptiness.” (Cavirani & van Oostendorp 2017: §1.1)
But later in this passage they move on to mention “detractors of abstract positions”, who criticize the “theory-internal” nature of zeroes. Note here how they have shifted imperceptibly from a language-particular theory (or analysis) to a general theory: After all, a critique of “theory-internal” considerations makes sense only if the theory is a general one. P-theories necessarily include (theory-)internal abstractions, and nobody denies them. Everyone sets up categories, and these are just as abstract (and “inaudible”) as zeroes. So these detractors are probably detractors of a general theory that involves certain general claims about zeroes. This is not any different from critizing a theory for making other kinds of general claims (e.g. that all syllables have CV structure underlyingly, or that all movement operations are to the left, or that all languages have nouns, verbs and adjectives).
In fact, there is one theory within mainstream generative grammar that is widely rejected even by generativists, namely “Cartography”, which posits extremely widespread zeroes. Cavirani & van Oostendorp (2017: §1.1) cite Cinque (1999), who basically claims that the entire array of many dozens of functional projections is present in every sentence in every language, just not always pronounced:
“[the possibility for] the entire array of functional projections to be present in every sentence […] is the least costly assumption, once we recognize that each head comes with a marked and a default value. This conclusion, if correct, opens up a new view of clausal structure – one that is further removed from what we see, but no less interesting, for that” (Cinque 1999: 127)
So this “extreme” view is controversial even within generative grammar – but why? As Cinque notes, this claim is “interesting”, and he even thinks that it is “less costly” than other assumptions.
I think that we can understand, and resolve, these controversies if we distinguish clearly between p-abstractness and g-abstractness. The former is universally accepted, but for the latter, we do not have more than speculative proposals. Everyone posits abstract notions (categories, constructions, features, relations) to describe language-particular systems, and everyone recognizes that the choices are often determined by elegance and other subjective factors. So there is no real debate about p-abstractness. It is generally necessary, and the details are a matter of taste. (P-analyses with zero can always be rephrased if one wishes.)
G-abstractness is an entirely different matter. Are zeroes constrained in a particular way (e.g. by an Empty Category Principle?) Are transformations/operations constrained in a particular way, e.g. by having particular levels in level-ordered phonology and morphology? These are the sorts of questions that have kept general linguists busy for decades. There are many proposed answers to these questions, but there is no convergence among these answers, as far as I can tell. Linguists tend to work with the kinds of answers that they grew up with, and even though the debate on some of them is now perhaps less heated than in the past, this is not because we know more about the general picture than we knew forty years ago.
So it is unclear to me to what extent it is useful to debate the general properties of zeroes or transformations. I learn much more from studies that test claims about language universals, and these need to be stated in terms of comparative concepts, not in terms of universal building blocks. And while “zero” is often useful for description/analysis, it is not something that can be identified across languages. In comparative grammar, we do not need “zero”, and for this reason, I have recently argued that the key term grammatical marker should be used only for elements that are pronounced (i.e. morphs). Likewise, transformations are not something that can be identified across languages, and they play no important role in works such as the World atlas of language structures.
There used to be a time when it was widely thought that languages differed less in their “deep structures” than in their “surface structures”, and that by analyzing them in terms of transformations, we would be able to somehow identify the common core of languages. Chomsky (1965: 118) put it like this:
“Since the study of deep structure has not been the concern [of Greenbergian typology], it has not attempted to show a corresponding diversity of underlying structures, and, in fact, the evidence that has been accumulated in modern study of language does not appear to suggest anything of this sort… Insofar as attention is restricted to surface structures, the most that can be expected is the discovery of statistical tendencies, such as those presented by Greenberg (1963).”
Maybe there are still some linguists who think that we need to study languages “in depth” before we can make valid generalizations, but almost all large-scale studies of the world’s grammars have used the comparative-concept approach (striving for measurement uniformity), not the universal-building-block approach. So for such comparative grammatical studies of Human Language, zeroes and transformations can be mostly left aside. They are obviously valuable for p-analyses, but their role in g-linguistics is still unclear.
Aronoff, Mark. 1976. Word formation in generative grammar. Cambridge, MA: MIT Press.
Chomsky, Noam A. 1957. Syntactic structures. ’s-Gravenhage: Mouton.
Chomsky, Noam A. 1965. Aspects of the theory of syntax. Cambridge, MA: MIT Press.
Cinque, Guglielmo. 1999. Adverbs and functional heads: A cross-linguistic approach. New York: Oxford University Press.
Greenberg, Joseph H. 1963. Some universals of grammar with particular reference to the order of meaningful elements. In Joseph H. Greenberg (ed.), Universals of language, 73–113. Cambridge, MA: MIT Press.
Jakobson, Roman. 1939. Signe zéro. In Mélanges linguistiques offerts à Charles Bally, 143–152. Genève: Georg & Cie.
Marchand, Hans. 1969. The categories and types of present-day English word-formation: A synchronic-diachronic approach. Munich: C.H. Beck.
Zero elements and transformations can be used to model a language, a group of languages, or potentially all languages. A model is different from reality, though. A model is good in so far as it is useful. The problem arises when linguists cannot acknowledge this and stick to a model religiously, to the point that languages start to be modeled on the model and not vice versa. This happens often times, unfortunately.
I forgot to say that I am skeptical about the possibility to find the same zero-elements- and tranformations-based model applicable to all/most languages, because those mechanisms are usually applied to categories which vary crosslinguistically. First of all, one should try to find crosslinguistic categories.
But what exactly is “a model” of a language? Is it different from “a p-theory”, or “a description”, or “an analysis”? I would think that all these mean the same, and I find it confusing that we have different terms that mean the same thing. – The next question concerns language comparison: Can one come up with a single “meta-model” (or “meta-theory”) that provides a template that can be used to model/describe/analyze any language? It seems that this is what many people are trying to do, but I think this is wrong.
I am a strong supporter of the p-g-distinction. But, Martin, I wonder (and I am not on top of this debate, since I would not consider myself a typologist, and haven’t read the relevant studies discussing this): wouldn’t it be interesting to have a typology of zeroes in the languages of the world, that would say, for instance, that the comparative concept of indefinite articles is zeroed more often than the comparative concept of definite articles? Languages are systems, systems have certain characteristics both in p- and g-linguistics perspectives, and if you have a marker for some distinction, the absence of the marker is also meaningful from the system’s perspective. So even if it is impossible practically, wouldn’t it be interesting to see how zeroing of markers is used in the languages of the world? But maybe that’s just me missing some integral part of the debate…
Yes, Mattis – and in fact, one could say that my project on asymmetric coding is precisely doing this: studying where languages have zeroes and where they have overt coding. HOWEVER: The criteria which are used for establishing descriptive zeroes and typological zeroes are totally different. For example, I say in a forthcoming paper (https://ling.auf.net/lingbuzz/004531) that Chinese has zero plural marking (and thus the singular-plural distinction is treated symmetrically). From a typological perspective, this makes sense (Chinese thus contrasts both with Greek, which as symmetric overt marking, and English, which has asymmetric marking), but for a description of Chinese, this would not make sense at all.
I am not sure I understand the difference between ‘language-particular’ and ‘general’ theories, because I do not understand what ‘language-particular’ means.
As a matter of fact, several of the observations in Cavirani and Van Oostendorp (2017) are based on cross-dialectal comparison of dialects within a language. The argument actually in those cases derives from the microtypology.
As a reminder (or for those readers of your blog who have not read C&vO17), here is a simple case. There are dialects of Dutch in which the first person singular of verbs is the (only) exception to final devoicing which is otherwise completely general. In these dialects, one says ‘geloof’ for the noun ‘belief’ and ‘geloov’ for the 1st person singular for the verb ‘believe’. It can be shown that such dialects occur in different separate regions, but they are always of the edge of some region in which the first person singular is (still) ‘gelovə’ (so with a schwa ending) in the big sea of Dutch dialects which have ‘gelof’. It is actually that particular fact which we think can be understood by invoking the notion of emptiness: the vowel gets (historically) deleted so that it is no longer pronounced, but it can leave behind for a while an empty vowel, protecting the preceding consonant from voicing..
Regardless of whether or not you like the argument, it seems to me that there is no sensible way that one can say that it is ‘language-particular’, because it compares three different languages, albeit languages without an army and a navy.
The whole point of disinguishing p-phenomena and g-phenomena seems to be that ‘language’ is a count noun. That is something which I obviously also accept in everyday parlance (you speak German, I speak Dutch) but which I find very difficult to make sense of from a scientific of scholarly point of view. I would think that language in the end is more like a mass noun, and we all have a bit of it. But if that is the case, it does not make sense to have a ‘language-particular’ analysis of anything, and if we determine that null elements are essential for some part of language, it is proof that it is therefore present in language.
Thanks for this very deep comment! Yes, indeed, we have both individual p-languages (such as Dutch and German), and the string sense that there is such a thing as Human Language that all languages are somehow part of. And in the case of dialects, the distinction gets blurred: Are the various dialects of Dutch different systems (like Dutch vs. German), or are they somehow variants of a single meta-system? After all, there is no speech form that does not have some internal variation. One might therefore also say that there is a single “Dutch-German language” (which even has a conventional name: Continental West Germanic), and that the small differences between the Dutch and German versions of this language are simply variants of a larger system. And moving still higher, we can perhaps say that there is just one single language system of which all languages are just variants.
Now while I agree that this is a logical possibility, the point of insisting on the p- vs- g-linguistics distinction is the claim that without it, one loses a huge amount of insight. It’s not so much that it’s ontologically incoherent (although it is of course very unintuitive) – it’s more that this view clouds many important generalization. If there were no such thing as p-systems, for example, one wouldn’t be able to describe German adjective inflection (a notoriously language-specific phenomenon that is hard to learn and cross-linguistically weird, but still internally systematic). And there is a host of phenomena of this sort. In fact, I think that most papers on grammar are driven precisely by the desire to capture language-particular generalizations (and this has been so for decades, so I say that we are all structuralists: https://dlc.hypotheses.org/2356).
Where exactly to draw the boundary between different systems, and where to say that we have a single system with variants, is not so important. We may also have intermediate solutions (a single verb system, but different noun systems, for example). David Gil has written very insightfully about these issues in 2016: https://www.degruyter.com/view/journals/lity/20/2/article-p439.xml).
There’s a constant tension between comparability and uniqueness. Joos keeps getting quoted with the claim that there are no limits to variation, and Chomsky gets quoted with the claim that there is fundamentally just one language. Both are right, in some sense, but in practical terms, we will not make progress by adopting an extreme approach: either by ignoring the similarities and treating each language as a world of its own without comparing it to other languages (à la Joos); or by ignoring language-internal generalizations and considering only the way in which each variant instantiates Universal Human Language (à la Chomsky). Progress comes from considering both p- and g-aspects.
I obviously agree with my coauthor, and I won’t comment further on the points he already addressed. I’d just like to refer the interested reader(s) to the fact that, indeed, we look at a few varieties (Dutch, but also Italian), as well as to the work by Kayne, who investigated silent elements in even more varieties.
As for zeros in the g-side of the language business, I think that if we can show that they belong to the abstract toolkit of a speaker of a given p-language, we can safely argue that they belong to humans, no matter what they happen to speak. As far as I am concerned, the chance it gives us to investigate human cognitive properties is what makes linguistics interesting. From this perspective, as also claimed by Marc, I don’t see such a big difference between p and g.
Sure, “we can safely argue that zeroes belong to humans, no matter what they happen to speak” – but I’m not so sure we can “safely argue” that they help us investigate human cognition in some direct sense. Linguists keep arguing back and forth, but there is little convergence among their proposals, so it seems that we don’t really know what our “linguistic cognition” consists of. We can investigate the conventions that we obey, and these are p-conventions. Each such convention in some way “belongs to humans”, but we need systematic comparison to find out which of the many aspects of the things that belong to humans are general.