Against traditional grammar – and for normal science in linguistics

I was recently invited to give a talk for the IGRA doctoral programme during a retreat workshop in beautiful Hohenstein-Ernstthal. I am grateful to the organizers (Gereon Müller and Jochen Trommer) for the talk invitation, and to the participants for a very good discussion after the talk. Here is the gist of what I said.

I apologize for the somewhat negative talk title (and content), but as in some of my earlier work (Haspelmath 2006), I also say “what to replace it with” – namely normal science instead of traditional grammar. Traditions are indispensable in human life, and science is part of human life, but in science we always have to question our traditions. The various traditions of grammatical research are often presented as in this figure:

So in a sense, all of this is “traditional grammar”, but for a scientific approach, just working in a particular tradition cannot be sufficient. Many people work in a particular framework, and this can be seen in publications like the recent Handbook of Morphological Theory (which lists over a dozen frameworks), and also earlier in the 2010 Handbook of Linguistic Analysis (to which I contributed a paper on framework-free theorizing, which basically made the same points).

In grammatical research circles, the following is a typical narrative:  Different “approaches” have different strengths, but one has to choose one of them – each framework gives us a set of assumptions and technical terms, and it also provides comfort and support (as well as a community of friendly reviewers). So one needs a framework to do theoretical grammar research, and one has to make a choice among the existing ones.

But this is not normal science. Normal science does not consist in continuing a tradition, but in finding true facts and true explanations.

Why would one need a tradition to insert oneself into (let alone a community with an intellectual leader) to engage in theoretical research on grammar? It seems to me that there are two prerequisites for normal science in structural linguistics (= the study of language structures, as opposed to phonetics, psycholinguistics and sociolinguistics): Normal-science linguistics is (A) based on pedigree-independent thinking, and (B) pursues clearly stated goals.

Why is the latter an issue? Because linguists have a range of diverse goals, which are often not distinguished very well. First of all, we need to distinguish between p-linguistics (particular linguistics, the study of individual languages or language families) and g-linguistics (general linguistics, the study of Human Language). Linguists often seem to think that studying a particular language is evidently relevant to g-linguistics, but this is not true. P-languages have many truly idiosyncratic properties (in lexicon and grammar) which are due to historical accidents.

So, what are our goals? Within p-linguistics, there are at least three different goals:

  • description of the conventions of particular languages (dictionaries, grammars)
  • description of how particular languages are mentally represented (e.g. mental grammars)
  • historical reconstruction of earlier languages and language genealogies (this is the research focus of my primary research unit, the Department of Linguistic and Cultural Evolution at MPI-SHH Jena)

Likewise, within g-linguistics (the general study of Human Language), we can puruse different goals:

  • general study of the mechanics and acoustics of pronunciation (phonetics), of language processing (psycholinguistics), of language in the brain (neurolinguistics)
  • general study of shared structural properties of the world’s languages (comparative linguistics)
  • characterization of the innate grammatical knowledge (“UG”)

Which of these goals are we pursuing? Many linguists don’t say clearly what they are after. In particular, they often say that they are interested in “theoretical” research, but all of the above is theoretical in the sense that it does not have applied goals. (Most of it is also “descriptive” in that it involves description – of p-languages, of speaker behaviour, of the world’s languages, or of UG.)

I often find it difficult to understand what goals my colleagues are pursuing. The sad truth is that general linguists often talk past each other, unless they belong to pedigree-based communities – and even within such communities, there is no clear sense of progress. Many papers written 30 or 40 years ago are not clearly outdated, except to the extent that they addressed very narrow concerns. But none of the larger issues that were discussed half a century ago have been resolved.

This lack of progress was brought home to me recently when I was reading Lieber’s (2009) morphology textbook. In the section where she talks about different models of morphology (lexical vs. realizational, etc.), she concludes:

“We will end this section by simply saying that the jury is still out on whether item-and-arrangement, item-and-process or realizational models of morphology constitute better models of how morphology is organized in the human mind” (2009: 183)

And in the section where she discusses incorporation in the context of lexicalism vs. anti-lexicalism, she ends up on a similar note:

“… There is a great deal that might be said about the pros and cons of this analysis, although we cannot do so here. I should point out, though, that while some linguists find the evidence for this analysis convincing, others are less convinced and prefer to work within theoretical models that treat noun incorporation as the result of morphological rules, and allow less interaction between morphology and syntax. As with many other theoretical issues in morphology, the jury is still out on the best way to treat the relationship between morphology and syntax.” (Lieber 2009: 186)

Since it is a subjective matter whether we find something convincing or unconvincing, it is not surprising that we cannot resolve our controversies. Another example for the lack of progress in resolving contentious issues could be witnessed at a workshop on NP vs. DP at the recent DGfS conference in Bremen in March 2019. The organizers described the workshop topic as follows:

“Does the DP-projection dominate the NP or does the latter dominate the determiner system? As of yet this simple question has not received a conclusive answer. While nominal phrases figure prominently in the theory of syntax, the DP  vs. NP debate remains unresolved, despite the decades-long success of the DP-hypothesis across theoretical frameworks.”

I attended very good talks by Martin Salzmann, who defended the DP view (despite being unconvinced by most of the earlier arguments in its favour) and by Benjamin Bruening, who reasserted his 2009 claim that the NP analysis does not fare worse and should be preferred. Both talks were very persuasive, but how can we determine who was right? Is there more to g-linguistic science than argumentation and persuasion? Frank van Eynde’s talk provided some history. As an HPSG practitioner, he saw many things differently from MGG, and he pointed out that “Jackendoff (1977) was our last common ancestor”. The following can be taken as a partial genealogy:

But this does not help if we agree that pedigree-dependent thinking will not lead us forward – it merely makes us aware how much of linguists’s practice is based on traditions. So let us get back to the other issue, the lack of clear goals.

What exactly are the goals of the debates around morphological models, lexicalism and DP vs. NP?

– Are they perhaps primarily p-linguistic goals? After all, many of the papers contain arguments deriving from particular descriptions of particular languages – so they seem to be relevant to p-linguistics, and not to g-linguistics. For example, in the NP vs. DP debate, Bruening (2009) discusses English and Passamaquoddy, and Salzmann (2019) discusses English and Serbo-Croatian. If one had g-linguistic goals, shouldn’t one provide evidence that clearly bears on Human Language in general, rather than only on a few individual languages?

– Or are they g-linguistic goals? In the Chomskyan tradition, it has often been taken for granted that the goal of theoretical linguistics is to describe innate grammatical knowledge (“UG”), so I am assuming that the debates around morphological models, lexicalism and NP vs. DP are debates of this kind. But there is something missing – an awareness of the methods by which one can study innate grammatical knowledge. As far as I can see, this can be studied only in one of three ways.

Three methods for studying innate grammatical knowledge

(i) By arguments from the poverty of the stimulus – because structures that cannot be learned from the input must be innate. However, stimulus poverty arguments (which are often discussed in the philosophical literature on the Chomskyan approach) play no role in the NP/DP debate or the lexicalism debate, and in practice, they are basically irrelevant to what the everyday working linguist does.

(ii) By comparative research that finds universals which have no plausible functional explanation (e.g. “No grammatical rule involves counting of elements”). The facts from particular languages cannot by themselves make a contribution to g-linguistics, because they could be historically idiosyncratic. However, there is no comparative research programme in recent generative syntax, and broadly comparative work such as Dryer’s (2014) and Becker’s (2018) on articles is typically not even cited by authors such as Bruening and Salzmann when they discuss  DP/NP. Broadly comparative work is often associated with functional explanation, rather than with innate grammatical knowledge, perhaps because many universals that were found in this way have an obvious functional explanation.

(iii) By hypothesizing innate natural-kind categories (and architectures) of grammar on the basis of particular languages, and hoping that we will hit upon the right categories – because these categories seem to work for other languages as well (e.g. DP seems to work for Russian as well, Pereltsvaig 2007). Indeed, languages are often similar, so a category hypothesized for one language often also works for another language. But how do we know that we have found the right categories? As we saw earlier, this is often unclear, and one’s persuasions tend to align with one’s pedigree.

So is there an alternative to pedigree-dependent g-linguistics? Can linguists practice normal science instead of some sort of “traditional” grammar? I believe that the answer is yes, and I will give a small example (based on my 2008 paper on reflexive pronouns).

Normal science in general comparative linguistics

In many fields of science, comparative research is based on objective measurements, not on categories that are hoped to be universal natural kinds. In linguistics, we can work with objectively defined comparative concepts (Haspelmath 2010). I have elaborated on the distinction between comparative concepts of g-linguistics and descriptive categories of p-linguistics elsewhere, and here I will merely give one example.

In the study of reflexive constructions, we can work with the following (category-like) comparative concepts:

reflexive construction: ‘a grammatical construction that can only be used when two participants of a clause are coreferential and that contains a special form that signals this coreference’

reflexive nominal (or “reflexive pronoun”): ‘a special nominal form that expresses one of the coreferential participants in a reflexive construction’

reflexivizer: ‘a special form that indicates that a construction is a reflexive construction’

length of a form: its number of consonant or vowel segments (where diphthongs and long vowels count as two segments

Once we have comparative concepts such as these, we can compare languages and come up with objectively comparable claims, e.g. (cf. Haspelmath 2008, based on much earlier research):

(1) In all languages, the reflexivizer for extroverted verbs (e.g. ‘see’, ‘kill’, ‘hate’) is at least as long as the reflexivizer for introverted verbs (e.g. ‘wash’, ‘dress’).

(2) In all languages, the 3rd person reflexive pronoun is at least as long as the 3rd person nonreflexive pronoun.

(3) In all languages that have reflexive pronouns beyond the minimal clause domain, the clausemate reflexive pronoun is at least as long as the long-distance reflexive pronoun.

These claims can be easily and objectively tested for the world’s languages, without requiring “in-depth” analysis of particular languages, because the comparative concepts are defined in the same way in each language.This allows the use of the survey method, as elsewhere in social science: Using a questionnaire, we can collect systematic data from the world’s languages.

Argumentation, persuasion and personal opinions are not relevant in this procedure – as is expected in normal science. Of course, such more subjective aspects are part of the overall enterprise. In particular, there is no objective way of choosing the phenomena to be compared, and of choosing the comparative concepts. But once these concepts are in place, the comparative survey itself can be carried out objectively, and claims about universals can be tested as elsewhere in science.

What explains the universals of reflexive constructions?

Explanation is the difficult part – there is no objective method for distinguishing true explanations from false explanations, again as elsewhere in science. My own take is that the systematic length differences are probably due to a highly general principle of efficient coding: More predictable information requires less coding effort, and thus shorter forms. (I have come to call this the efficiency theory of asymmetric coding, as briefly summarized in this earlier post.)

But alternative explanations are certainly possible, including explanations in terms of innate grammatical knowledge (“UG”). However, innate grammatical knowledge has typically been invoked not in order to explain universal tendencies in the world’s languages, but to explain parochial facts of particular languages, e.g.

“Principle A: An anaphor is bound in its governing category” (Chomsky 1981)

This is a generalization about the English forms himself/herself and each other. There is no general definition of “anaphor”, so the term is not a comparative concept that can be applied to other languages. Thus, there is no way to test systematically whether this principle is general across languages and thus a candidate for innate knowledge. All we can do is hope that some category hypothesized for some language will eventually turn out to be correct for all languages.

The normal-science research programme that I sketched above is very demanding, but as an intermediate goal, linguists as a discipline could work on consistent terminology – a kind of “IPA of morphosyntax”. If each term of general linguistics had a clear definition (e.g. the terms sentence and clause), this would make consistent comparison of languages much easier.

And here there may actually be some convergence with linguists who otherwise pursue a different methodological programme. A few years ago, David Pesetsky complained about bad terminology in linguistics (opaque terms like specifier, pseudopassive, EPP), and he ended up with a concrete suggestion:

“I think at some point in the histories of fields like chemistry and physics, commissions were set up to regularize nomenclature and make it sensible. How about us? This sounds like such a boring activity, but on the other hand, for how many years are we going to be telling people that there’s no pseudo in pseudopassive and that specifiers don’t specify anything…”

Thus, having a nomenclature committee might be a very concrete way for g-linguists to make some real progress for the discipline, and to help grammarians to practice normal science.


11 thoughts on “Against traditional grammar – and for normal science in linguistics

  1. Martin, I wonder if you have any comment on the fact that your concept of “normal science” seems almost the opposite of the one proposed by Thomas Kuhn. Cf.

    “In this essay, ‘normal science’ means research firmly based upon one or more past scientific achievements, achievements that some particular scientific community acknowledges for a time as supplying the foundation for its further practice.” (Kuhn, The Structure of Scientific Revolutions, p. 10)

    • I admit that I haven’t read Kuhn, but I don’t really see a conflict. I see most linguists’ research firmly based in past scientific achievements of linguistics – the advances of the 19th century and of the distrubutional method are an indispensable foundation for modern normal-science linguistics. This is different with generative linguistics – it is not normal science, because it builds on a lot of speculative ideas.

      • The expression “normal science” has come to be associated with Kuhn’s work to the degree that when you google it, you are taken to a Wikipedia article saying: “Normal science, identified and elaborated on by Thomas Samuel Kuhn in The Structure of Scientific Revolutions, is the regular work of scientists theorizing, observing, and experimenting within a settled paradigm or explanatory framework.” So my implicit point was that if you want to propagate your views on how linguistics should be done using that term, it would be good if you made clear how your understanding of it relates to the Kuhnian concept. And I do think there is room for confusion. For Kuhn, or at least as he is usually understood, normal science takes place within a paradigm. I am not sure how Kuhn would characterize the more eclectic kinds of linguistics that I think both you and I subscribe to, but something like Chomskyan linguistics most certainly fits his view of a paradigm. So another thing that should be clarified is how your notion of “pedigree-dependent thinking” differs from Kuhn’s notion of a paradigm.

        • I don’t really see a Chomskyan “paradigm” – instead, what I see is several communities of linguists who are held together by a shared set of assumptions and technical terminology (going back to Chomsky’s work between 1957 and 1981, and occasionally to later work of his), but I don’t see any “explanatory framework”, let alone “past achievements”. In Baker’s work, one sometimes sees (or saw in the past) a research programme, but in most contemporrary Chomskyan work, even that is hard to discern.

  2. Although I agree with most things in the post, I think it’s important to point out that there is a vicious circle between the lack of progress and the lineage thinking. On the one hand, the lineage thinking hinders the progress by making the communication difficult or impossible. On the other hand, in the absence of visible achievements that any person (and the general public) could appreciate, one needs to belong to a certain clan in order to get acknowledged, published and cited. While the ‘normal’ science was involved in promoting the Enlightenment and its values (such as common sense, utility for the humanity, objectivism and universalism), linguistic theory is stuck in the middle ages with clans, feudal lords, dynasties, scholastic disputes, masters and apprentices. We see a game of thrones instead of a battle of falsifiable theories. Since the clans thrive on their differences, which justify their existence, and on obscurantism, which is vital for hiding their scientific irrelevance, I don’t think that clear goals and transparent procedures, let alone common nomenclatures, which presuppose common objects of investigation, can be established any time soon. We are dealing with intellectual machines whose primary goal is to reproduce themselves, rather than true facts and explanations.

  3. I would say that the problem with the ‘normal science’ approach is that it makes it seem natural to turn to a structural description as a mode of ‘doing good linguistics’. But I think that this is misleading as to the nature of language. The current challenge of the radical potential coming from the constructionism (covering Fillmore, Croft, Goldberg) on the one hand and cognitive semantics (Lakoff, Talmy, Langacker) on the other makes a purely structural description (even imbued with functionalism) less appealing as the foundation of a newly scientificised linguistics.

    It’s curious that it’s physics and chemistry that gets mentioned in this context with their fully mathematised personas and not biology or geography. In both of those, precise definitions are much more provisional and iterative. Even foundational terms such as species or gene are much more fluid and less well defined than it might seem (I recommend Keller’s ‘Century of the Gene’ for an account of how discrepancies on how gene is defined among various labs was actually beneficial for the development of genetics).

    That’s not to say that I strongly disagree with any of your proposals but they don’t particularly make me excited to do linguistics. I found Dixon’s ‘Basic Linguistic Theory’ an exhilarating read but it was not because I felt that programme would lead to more consistency but because it was a radically new proposal (despite its claims to the contrary) for a theoretical basis for a comparative linguistic programme. Which is also why I like your body of work.

    But I doubt that the road ahead is in better definitions. I’m not opposed to them just skeptical that they will lead to much. The road ahead is in better data and better theory. I think that between corpus linguistics, frame semantics and construction grammar we can get both. I proposed the analogy of ‘dictionary and grammar being to language what standing on one foot is to running’ (http://metaphorhacker.net/2014/11/what-language-looks-like-dictionary-and-grammar-are-to-language-what-standing-on-one-foot-is-to-running). I think linguistics needs to embrace the dynamism of language and a human property rather than as a fixed effect (to borrow Clark’s phrase). Fillmore and Kay’s early writing on construction grammar was a first step but things seemed to have settled into the bad old ways of static structural description.

    Data and theory need each other in a dialectic fashion. You need data to create a theory but you needed some proto-theory to see the data in the first place. And then you need your theory to collect more data and that data then further shapes your theory which in turns let you see the data in different ways. The difference between biology and linguistics is that our proto-theories of the biological world correspond much better to the dynamic structures which can be theorised based on systematic data collection and its modeling. Which is why folk taxonomies of the biological world are much closer to those of botany or zoology than folk taxonomies of language are to linguistic ones. (They are much more elaborate – to start with – at least at the level available to human perception.)

    My proposal is to take seriously human ability to reflect (hypostaticaly) on the way they speak (cf. Talmy’s defense of introspection) because this is at the start of any process that bootstraps a theory of language. We then need to be mindful of the way this awareness interacts with the subconscious automaticity in which the patterns of regularity we call structures seem to be used. In the same way that Fillmore and Kay asked any theory of grammar to account for the exceptions (or even take them as a starting point), I’d want to ask any theory of language to take bilingualism and code mixing as its starting point (inspired by Elaine Chaika) and take seriously the variability of acquisition of the ability to automatically use those structures.

    None of this is precludes or denies the utility of your great work. But it is what I think would lead to linguistics being a better science (at least, in the sense of Wissenschaft or ‘natural philosphy’ rather than in the sense implied by the physics envy which often characterizes these efforts).

    • My proposal may not be “exciting”, and of course, as individuals we need to be personally motivated. But I fear that for a whole discipline, “excitement” is not a good goal. We have seen a lot of exciting new proposals over the last few decades, but do we know whether they are true? Biology (including genetics) has made actual discoveries (as can be seen from technological applications), but in linguistics, it’s hard to distinguish proposals from discoveries. And this will remain so as long as we cannot even agree on what some of our basic terms mean…

  4. What you are proposing will reveal patterns in the data (which, at the very best, would be only somehow meaningful or interesting if you manage to compile incredibly large samples). But even when disregarding this point the serious question is: What will any of what you propose reveal about the way in which the human language ability works? By identifying these patterns you will learn nothing about the mechanism that gave rise to them and the potential solutions to such a problem then are infinite: that is, you cannot reverse-infer the mechanism from a pattern in the data, best you can do is some comparisons of different algorithms which, again, will have to be constructed on the basis of some kind of theory (like it or not).

    All in all, what you’re proposing is nothing more than theory-free “data-driven” research, as practiced in the machine learning and AI communities. In effect, what you’re proposing is, unlike to what you claim in the title, actually in direct opposition to normal science and would at best qualify as “abnormal science”, as practiced in the machine learning and AI community. Because, as you may have noticed, these fields have not produced any understanding (in the sense of “normal science” where understanding is mechanistic explanation) whatsoever, they’re interested in practical application which is why they are not “normal science” by any reasonable standard. That’s fine, of course, because they don’t aim at being “normal science”, they want to produce something practically useful. But why would you think that it should be the goal of linguistics to find out that the reflexivizer for extroverted verbs is at least as long as the reflexivizer for introverted verbs using this method of “abnormal science”? That may be the goal of a data scientist who is training an AI and needs to identify criteria for extracting patterns that allow the algorithm to do its job, but in the end you will have learned and, more importantly, understood nothing at all. And there is also no reason to suppose that the patterns you have identified relate to the actual algorithm that generated them in any way. The best you will be able to do is to say that “these are the patterns that we (currently) observe across languages”, then you may add a diachronic perspective and, in the end you end up with an awful lot of data (some of which cannot be analysed because it is historic and does no longer exist) … and derive more patterns, or slightly different ones, and the result will change depending on what your input to the computation. Even if you had a complete dataset covering all human languages synchronically and diachronically, automated analysis would tell you nothing about the mechanism that generated this data … you’ll have to construct hypothesis that go beyond what you observe and thereby explains as many data points as possible with the simplest mechanism one may postulate … as is normal science since.

    Hence, I fail to see how your approach is “normal science” and how it would generate any scientific understanding, in the sense of normal science, whatsoever.

    • Thanks for this detailed comment! But I think it shows that we need to be clear about our goals, and you seem to have in mind a question concerning the biological underpinnings of human languages (“the way the human language ability works”). This is indeed not something that structural linguistics (with whatever approach) can say very much about. Maybe psycholinguistic and neurolinguistic methods are better suited to answering this question. The question that I am concerned with primarily is why human languages have the properties that they have. There are still other goals (e.g. elegant descriptions of particular languages, solving Plato’s Problem, etc.), but among structural linguists, the question why human languages have the properties they have seems to be quite prominent.

      • Hi Martin
        I really love this essay but I am afraid I agree with puzzled linguist. I think linguists will never discover *why* languages have the properties they have, because the answer lies in studying the propeties of brain and behaviour (psycholinguistics and neurolinguistics). I think linguistics methodologies can only answer the question *what* properties do languages have and, perhaps more crucially, what properties do languages share/not share. The “what” questions are crucial questions but very different to “why” questions, which can only be answered by studying the organ that created language – the brain.

        • Thanks, Caro – but I understood “Puzzled Linguist” a bit differently, as implying that only the study of the biological mechanism underlying language is an interesting research question. Now you seem to be conceding that “what” questions may be interesting, while at the same time claiming that “why” questions cannot be answered without brain/behaviour methods. But again, I would say that there are distinct goals: All of these questions and methods are legitimate and valid, but not all methods are suitable for all goals. Studying English speakers (or experimental subjects) will not help us understand the properties of “agreement” (or of “DPs”) in general – it will only help us with p-linguistics of English. To make progress in learning about Human Language (either “what” or “why”), we need to add a comparative perspective, especially when we don’t work with experimental methods.

          As for “why” questions, I’m not so sure that one needs brain/behaviour methods in all cases, because the brain is not the only limiting factor. Of course, languages cannot have properties that the brain does not support, but the brain can probably process languages that the society would not support. For example, universal Zipfian efficiency effects (frequent words being short and rare words long) seem to be due to convergent cultural evolution, not to the properties of the brain directly.

Leave a Reply to A puzzled linguist Cancel 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.