Confused by syntax: Some notes on Koeneman & Zeijlstra (2017)

(See also a reply to this critical review by the authors: “Syntax and didactics“)

A new authoritative textbook on Chomskyan syntax

Papers in the framework of current mainstream generative grammar (MGG) are often difficult, or even impenetrable, to read, even when the reader is well-versed in syntax and in other models of generative syntax. They are mostly written for the community of practitioners, who naturally do not see a need to motivate their choices. I was thus happy to see a new textbook (“Introducing syntax”, Koeneman & Zeijlstra 2017), published by an authoritative publisher, and approved by Noam Chomsky himself („A lucid, accessible, and deeply informed introduction… A very welcome contribution“), that explains the approach for beginners, trying to motivate most aspects of the theory. It has helped me understand a number of things that weren’t clear to me before.

The book is very well written (I wish I had the authors’ obvious talent for engaging beginning readers), so the purpose of this lengthy post is not to criticize the book, but to examine the nature of the claims, the evidence presented, and the style of argumentation. Although I am not an expert in MGG, I feel that I know enough about it to make generalizing statements when they apply much more broadly than this book.

Why would one comment in such detail on a textbook? The main purpose of this post is to help those students who are interested in a wider picture to get a bit confused. The picture that is presented in this book is very neat, and alternative approaches are hardly mentioned – only the short foreword briefly says that the book is actually an introduction to minimalist syntax. There is no explanation why it is not accurately titled Introducing minimalist syntax (of English). I sometimes hear that students may get confused if they are presented with several different frameworks at the same time (as in S. Müller’s 2016 textbook). But syntax is difficult, and one could argue that they should get confused and struggle to develop their own thinking about syntactic patterns of human languages.

The nature of the “theory”

Koeneman & Zeijlstra (K&Z) do not provide much background discussion for the “theory” that they build up over ten chapters. In the introduction, they merely say that they are not content with finding the rules of English syntax, but also want to understand the “nature” of these rules (p. 3), or the “principles” that underlie them (p. 3). In practice, this means formulating the rules in a maximally general way for English, but surprisingly, K&Z assert that “the principles underlying these rules are the same for all languages” (p. 3). They never come back to this claim later in the book, and none of their general statements is justified by evidence from languages other than English (with some marginal exceptions, e.g. some Japanese examples in ch. 9). So somehow they must think that studying English alone will give us substantial insight into the universal principles of syntactic rules (and even helps us “to understand the human brain itself”, p. 3). However, this can be so only to the extent that the rules of English are not learned, but follow directly from “properties of the human brain”. There have been many arguments, especially from psycholinguistics and from learnability considerations (“poverty of the stimulus”), that some rule principles are not learned but present at birth. But such considerations play no role in this book.

It seems to me that this is characteristic of MGG in general: When proposing rule principles that are meant to apply to all languages, generative linguists typically appeal neither to broadly cross-linguistic evidence nor to evidence from psycholinguistics or learnability. Instead, they consider individual languages and propose highly general rules (or rule principles) that subsume a number of lower-level rules. Thus, the description of the language is more abstract (and thus perhaps more “elegant”) than in other treatments, but it is not immediately clear whether the principle would work for all other languages as well. In fact, testing the claim that the principles are universal is extremely difficult, and as a result, nobody has ever proposed a list of principles that might underlie the rules of all languages. MGG has been successful in raising many new questions and suggesting a number of intriguing cross-linguistic claims (e.g. D’Alessandro 2017), but it is clear that it has not even come close to its stated goal of finding universal principles.

Of course, some generative linguists have tried to justify their universal principles by considering evidence from many languages, e.g. Cinque (1999) and Baker (2015). But in the end, it seems that arguments from English have always been more important in the development of MGG ideas, so the approach taken by K&Z makes good sense from the perspective of introducing students to the community enterprise. From a broader perspective, and for those readers who are interested in a theory of the language faculty that can say something about any language, there remains a big open question.

The style of argumentation

In the first few chapters of the book, K&Z try to present alternative analytical possibilities, and argue for one of the alternatives. This sort of “syntactic argumentation” is sometimes seen as important to teach to students, so that they learn not only facts, but also how to evaluate competing hypotheses. However, the most important alternative considered early in the book (the only one that gets a section heading, §2.3.2) is a complete strawman: They contrast the “Merge” hypothesis with a hypothetical “Glue” hypothesis, where the latter means that there is no hierarchical structure (and “Merge” does not say more than that there is hierarchical structure). But the existence of hierarchical structure is so evidently true that the issue has never come up in discussions among syntacticians. Something that is less evidently true is universal binary branching of constituent trees (cf. Culicover & Jackendoff 2005), but K&Z do not discuss non-binary branching at all (they briefly mention the possibility on p. 182, but dismiss it because they are “committed to” binary branching).

In later chapters, the discourse style changes definitively from two-sided argumentation to persuasion. The authors imagine readers who first balk at excessive abstractness, but then, step by step, abstract analyses are made palatable to them (“[abstract Case] may be a bit hard to swallow, but…”; “if you’re not yet convinced, here is a third fact…”). But what is the purpose of this additional abstractness? Do we need it to understand which English sentences are well-formed and which ones are not? Not really. Instead, the abstract analyses are justified by the uniformity of mechanisms:

“The hallmark of this analysis is that it uses the same kind of structural configuration that we have already seen many times before.” (p. 145)

“As a consequence, we arrive at our maximally simple theory” (p. 167)

“it would be weird if one grammatical module [i.e. syntax] were able to further manipulate its building blocks, but another module [i.e. morphology] weren’t.” (p. 206)

But what is the purpose of this uniformity? The readers are apparently expected to find it self-evident that this is desirable, even at the cost of enormous complexity that is generated elsewhere in this way. There is some reference to “elegance” or “beauty”, as if these were not in the eye of the beholder.

It is really only at the very end, in the Afterword, that K&Z make reference to the idea that their framework is restrictive:

“Syntax, we conclude, is not a machine with 200 distinct operations but offers a highly constrained way of building phrases and sentences.” (p. 269)

This resembles standard statements in generative grammar that the theory must be restricted in order to solve the poverty of the stimulus problem, but the point is not elaborated further. Clearly, their system contains only a small number of mechanisms, and given enough features and movements, there is no doubt that all structures of English CAN be described with this highly complex machinery. But what is the advantage of doing so? In the principles-and-parameters framework (Chomsky 1981; Baker 2001), the idea was that some logically possible language types cannot be described by the system, and these language types are predicted not to occur. This idea, which generated many interesting predictions (most of which were later disconfirmed), is completely absent in this book, and it seems that it is no longer prominent in MGG as a whole.

It is very likely, of course, that a rather different system of highly general mechanisms could be constructed. At every point, there are many conceivable alternatives, some of which have been pursued by linguists working in other frameworks. How can we tell which of the frameworks is better? If cross-linguistic restrictiveness is no longer a goal of the enterprise, I no longer understand in what sense it is an empirically constrained scientific endeavor.

K&Z refer to alternative generative approaches only once, in the “further reading” section of ch. 6, where they mention the HPSG approach to question-word fronting (“wh-movement”). They briefly comment: “The question here is how to empirically distinguish proposals that at heart are quite similar” (p. 163), leaving the reader puzzled. In HPSG approaches, restrictiveness has never been a prominent consideration, so if their MGG approach is seen as “quite similar”, why should one spend so much energy justifying an alternative approach in great detail? And is there an “empirical” difference between K&Z and any other description of English? It seems that the differences between HPSG and MGG are primarily “conceptual”, but if it is unclear what these conceptual differences are, I have a serious problem.

Confirmation bias

Psychologists have long known that people interpret new information in a way to make it consistent with their earlier beliefs, and confirmation bias is known to be a serious problem in science. Before we can be sure that a hypothesis is correct, we need to look not only for confirming evidence, but also for counterevidence. But in linguistics, especially in MGG, one frequently observes that only confirming evidence is considered. This is very striking in K&Z, where we find rhetorical moves such as the following:

“The obvious goal, therefore, is to see whether accusative case assignment can be reformulated as an agreement relation” (p. 121)

“we will try to investigate whether [binding constraints] can be described by the same agreement mechanism” (p. 124)

“Let us attempt a hypothesis, and see what it gives us…” (p. 145)

“How far can we get in trying to understand Wh-movement as being feature-driven?” (p. 173)

These moves invariably introduce some further arguments in favour of the proposed idea, and the conclusion is that the idea has been “strengthened” or “corroborated”. A bit later, we typically read that the idea was “established”, and the authors go on to build more far-reaching suggestions on it.

But what about possible counterevidence? Perhaps one cannot expect a textbook to confuse its readers by presenting such evidence, too (and K&Z do include some problematic phenomena in their Afterword). But the real problem is that it is generally quite unclear what would count as counterevidence to a syntactic “theory”. The general style of the subdiscipline is argumentation and persuasion, not making predictions that could lead one to look for confirming evidence or counterevidence. Ideally, of course, the truth should eventually be sifted out through a community effort: Those insights that stand the test of time, after many scholars have examined the evidence and considered multiple approaches, are likely to be true. But if syntax is an enterprise of a narrow subcommunity to which access from outside is difficult, couldn’t the power of persuasion get in the way of the necessary sifting?

Interim conclusion of a confused outsider

Before discussing some of the individual chapters below, let me summarize what confuses me about K&Z Introducing syntax, and more generally about the field of MGG:

– the book (and the field) talks about syntax (or syntactic theory), as if it were in any way established that the highly specific approach of MGG has come any closer to the truth than other approaches

– arguments from elegant description of English grammar are taken as evidence for a “theory” that makes a contribution to “understanding the human brain itself”

– despite the claimed cognitive perspective, syntactic patterns are approached in a highly abstract way, as if we knew nothing about memory, language processing, language acquisition, animal language, or syntax-like patterns elsewhere in cognition.

Of course, MGG is institutionally influential at the moment (e.g. at Cambridge University), which also explains that Cambridge University Press has published a book about “syntax” that presents an extremely narrow perspective on syntax. And if young scholars interested in syntax want to get a good job, they should know something about this approach. But if young scholars also want to make a significant contribution to science, is this really the way to go?

I will now make some more detailed comments on the first few chapters.

Notes on Chapter 1: Categories and features

Chapter 1 introduces categories and features, starting with the “insight” that “syntax cares not about individual words, but rather about kinds of words” (p. 8), which is of course the very definition of syntax. The text makes it appear that it is “interesting and striking” that every word belongs to a category, but this is so by definition, not because we discovered anything. There are of course words that behave in a unique way (e.g. neither, or let alone, cf. Fillmore et al. 1988; Culicover 1999), but we still assign them to a category, because rules must be category-based – otherwise they would not be called rules. (And languages not only have rules but also idiosyncratic words, i.e. they have not only syntax.)

The next section about determining category assignment is also confusing because it treats the substitution test first as a “generalization”, but then as a “diagnostic”, as if the categories were already there and the substitution test were just a good way of finding out about them – whereas in fact categories are defined by the substitution test. There is a brief introduction of the notion of a “hypothesis” here (wrongly characterized as an “assumption” intended for testing, p.11), but that “words denoting actions are verbs” is of course not a testable hypothesis – we know that action-denoting nouns (war, dance) are nouns because of the substitution test, which is the usual defining criterion for nounhood. Some of the traditional categories cannot be defined by a substitution test (e.g. adverb, characterized here vaguely as “modifying the meaning of the verb”, p. 15), but they seem to be introduced here primarily for terminological completeness.

Then the perspective is shifted from “belonging to a category” to “carrying a feature” (p. 18), which might just be a notational difference. But it turns out that “feature” actually means “value” of an attribute-value pair (and is thus a value, not a feature in Corbett’s 2012 sense). So saying that car has the “feature” [N] means that its attribute [±N] is set to the value [+N] – in other words, there is no “syntactic category” as an attribute that could take on multiple values. This can be seen in the proposal that pronouns carry both an [N] and a [D] feature (p. 23), i.e. they can belong to two traditional categories simultaneously. But if words don’t actually belong to categories, then it’s odd to say that some of the features are “subfeatures” (p. 20), because there is no taxonomic relation between the features (unlike traditional categories, which could be thought of as organized in a taxonomic hierarchy). It is also odd to distinguish between “categorial features” and “non-categorial features” (p. 18-19), because this distinction does not seem to play a role in K&Z’s system.

In §1.3.1, K&Z then go on to argue that English has a zero indefinite article in plural and mass nouns. This is of course a familiar idea, but they claim that the substitution test “leaves us no other choice” (p. 21). This is plainly wrong – of course one can describe English without zero indefinite articles. Compare the two minimal quintuples in (1) and (2), where the last element is ungrammatical in the given context.

(1) I like {a car / the cars / cars / wine / *car}.

(2) I hope {he knows her / she likes him / he smokes / she blogs / *she likes}.

In order to explain the ill-formedness of *I like car, we must indeed say that car does not have the same features as a car, the cars, cars, and so on, because otherwise it would not fail the substitution test. One straightforward way to do this would be to say that car is not a complete nominal expression (does not carry the feature [NOMEX]), just as the reason why she likes fails the substitution test in (2) is that it is not a complete clause (does not carry the feature [CLAUSE]). K&Z do not mention this possibility, because here they simply do no consider the possibility that multi-word expressions have separate features. Now of course, it would be simpler if the features of multi-word expressions derived in a simple way from the features of their components, but the contrast between she likes him, she blogs and *she likes shows that more is sometimes needed (e.g. a notion of valency requirements, or object obligatoriness) – and similarly, more is needed for *I like car (some notion of article obligatoriness with singular nouns). How one states these additional conditions is an interesting question, and adding zeroes is often convenient (it would work in (2) as well, and indeed it has occasionally been proposed that all verbs have objects, just not always overt objects). But zeroes are never forced upon us, they are only one of many descriptive choices.

In the last part of §1.3.1, K&Z then go on to argue that personal pronouns and proper names have both an [N] and a [D] feature, because they can substitute for nominals (I like her, I like Lee). But this presupposes that nominals like the car have the two features – in other words, it presupposes the possibility that they failed to consider two pages earlier, namely that multi-word expressions have features. They also consider the possibility of a zero determiner for these cases (I like Ød her), but they reject it by appealing to a vague notion of “plausibility” (“if pronouns can never be combined with any expressed article, it is not very plausible to assume that they can be combined with an unexpressed one”). Finally, in their discussion of personal pronouns and proper names, they only appeal to substitutabilty, not to non-substitutability. The substitution test on p. 14 is correctly formulated in such a way that full feature identity entails substitutability, but it does not work in the opposite way: Substitutability of course does not entail full feature identity. Personal pronouns and proper names can replace other nominal expressions (maybe because they share with them the feature [NOMEX]), and the fact that they do not cooccur with articles (*the Lee, *the her) might simply be attributed to the fact that they do NOT carry the feature [N]. (In fact, earlier on p. 14, K&Z themselves define [N] with respect to occurring after an article, and later on p. 39, they say that a name such as Lee is an NP.)

Thus, we see that even in the very first chapter, which many syntax teachers will treat as simple preparatory material to be dealt with quickly, there are many questions that are not really settled (and that cannot ever be settled, it seems, because there are multiple equally good possibilities). It is a virtue of K&Z’s argumentative style that this is brought out clearly on careful reading.

Notes on Chapter 2: “Merge” and constituency

The second chapter introduces constituent structure in terms of “Merge” and argues against a strawman alternative that does not use hierarchical structures and that nobody has ever advocated (confusingly, the “further reading” section says that it has been advocated, which is wrong). The Merge mechanism is described as follows:

(1) Merge: Combine two constituents and create a new constituent that carries the feature of one of those constituents as its feature (p. 35).

This creates binary endocentric constructions, and apparently the claim is that all constructions are of this sort, i.e. binary and endocentric (“one word in a complex constituent is always responsible for the behaviour of the entire constituent”, p. 34). The possibility of ternary constructions is never discussed in the book (even though there are good reasons not to exclude them, as discussed by Culicover & Jackendoff (2005: §3.1.3), and a large literature on morphological templates), and it is of course not easy to see that all constructions should be endocentric: Both clauses (consisting of nominals and verb phrases) and coordinated phrases (consisting of two phrases of the same category and a coordinator) have long been considered exocentric in generative grammar. Making them endocentric (and thus conform to the idea of uniform Merge) requires quite a few extra assumptions that make things much more complicated.

Although K&Z aspire to be “extremely precise and concrete” (p. 35), the formulation in (1) is not precise because it is unclear which feature is meant – after all, ch. 1 argued that syntactic elements may carry several features, including several categorial features. But their Merge mechanism seems to require that one feature is somehow privileged.

From §2.2.2 onward, more familiar (1980s-style) notation is introduced, but why this is needed and how it relates to the newer terminology in terms of Merge did not become clear to me. So here we learn about “heads”, “phrases”, and “maximal phrases” (which get a P at the end of their label: NP, AP, AdvP, etc.), but none of these concepts is explained fully. Heads are first characterized as in (2).

(2) Head: Whenever one combines two words, the word that determines the feature of the new constituent is referred to as the head of that constituent (p. 36).

But later, larger constituents are also said to have heads (e.g. very delicious sausages from Italy, p. 42), and even though intuitively, such larger nominals can be understood as expansions of nouns, it is unclear how the head notion fits into the formal system that is built up here. (In §2.2.3, there is a revised statement of Merge: „If one combines two constituents, then one of them must either be the head (as in delicious + sausagesH), or contain the head (as in expensive + [delicious + sausagesH])“ (p. 39). But it is of course possible to combine two constituents that each contain a head (e.g. [very deliciousH] + [sausagesH from Italy]), so this is not precise. It seems that the notion of „head“ does not help us understand what can happen in the formal system that is built here.)

Likewise, I did not understand how “phrase” is defined. First we read that “a phrase is a constituent with a head” (p. 37), but a page later it is said that “single words can be phrases too” (which seems to contradict the earlier statement). The first motivation for distinguishing between nouns and noun phrases is simply that nouns are also “words” (a notion that is never discussed in the book but simply presupposed, despite the huge problems with it; see, e.g., Haspelmath 2011). But earlier, when introducing the Merge mechanism, it was claimed that noun phrases like delicious sausages have the same distribution as nouns, so there is again a contradiction. (Of course, if one includes compounds, then such noun phrases do not have exactly the same distribution and fail the substitution test: one can have sausage factories, but not *[delicious sausage] factories.)

The notion of “maximal phrase” is salient to older readers because of familiar labels like NP, VP, PP, but a novice may well ask why one needs such a notion. Its definition as “highest node of the same categorial type” (p. 38) is again imprecise, because not only sausages from Italy is an NP, but also Italy itself, even though it is not the highest nominal of type [N] in this phrase. Even more puzzlingly, the notation N’ is introduced for intermediate phrases (delicious sausages in the phrase expensive delicious sausages), even though there is apparently no distributional difference between the two. The notion of maximal phrase is used in a generalization about the “non-head”, namely that it is always a maximal phrase (p. 40), but the text does not say which kinds of possible structures are excluded by this generalization.

Another notational decision that will be puzzling to novice readers is the notation “AdvP” for a degree adverb like very (p. 39). While the AdvP notation for often can be easily justified (because of expansions like very often), the justification for very is limited to the desire “to be consistent” (which could of course be used to add a “P” to all kinds of other forms: demonstratives like this could be said to be DemP, infinitive markers like to could be said to be InfP, etc.). It seems that the main reason why this notation is introduced is to make the generalization about non-heads being maximal phrases true, but this is of course circular and thus does not tell us anything about English (or about syntactic patterns in general).

Two constituency tests, fronting and substitution by a pronoun, are introduced in §2.3, and it is pointed out that they are only sufficient conditions for constituency, not necessary conditions. This means, of course, that we do not have evidence for constituency in ternary constructions like (They) gave1 the money2 to their friends3, where no combination of two constituents can be moved or replaced by a pronoun. However, the authors also say here that by using only a single test, one runs the risk of circular reasoning, and that one needs “independent evidence”. They show that sometimes independent evidence is available, but what if it is not available? For example, the fronting test does not confirm the idea that there is a subconstituent delicious sausages in the nominal expression expensive delicious sausages (fronting is not possible: *Delicious sausages I bought expensive, p. 48). It remains unclear why circular reasoning should not be a problem here.

One-substitution is of course a popular textbook example of a substitution test for constituency, but it should be noted that it has been shown that this test fails to give unique results, because one can be used to substitute either a noun and its prenominal modifier, or a noun and its postnominal modifier:

(3)      a.         You like the green apples from Italy?

  1. … and I like the red ones [= apples from Italy]
  2. … and I like ones from France [= green apples]

The explanation of this must have to do with meaning: green apples from Italy can be construed semantically in two ways, as ‘(green apples (from Italy))’ or as ‘(green (apples from Italy))’, and the pronoun one can target a semantically coherent constituent. Could this mean that the substitution test tests for semantic constituents and may not be relevant for syntactic constituency?

Notes on Chapter 3: Thematic roles

The phenomena discussed in chapter 3 on semantic roles are more English-specific than the very basic phenomena discussed in the first two chapters, but the discussion curiously continues in the same mode, as if all the rules mentioned here were applicable to all languages. It is true that the thematic-role scale for subject and object assignment (agent > recipient > patient/theme > goal) has often been claimed to be universal, but it is clear that no single scale can account for all languages (cf. Newmeyer 2002). Moreover, the very notions of subject and object (used in this chapter, but not explained) are of course highly variable across languages.

I found this chapter less confusing than the first two, but a key question was not answered for me: Is the chapter about semantic restrictions or about syntactic restrictions? In §1.2.2, the authors take pains to distinguish between syntactic ill-formedness and semantic deviance, but this distinction is not taken up here, even though it is clear that *John loves Mary Bill is semantically deviant, and thus would not not need to be ruled out by syntax. Still, §3.2.2 presents the “Theta Criterion” (why “criterion”? why not “principle”, or “hypothesis”) as a cornerstone of the syntactic theory (following Chomsky 1981, but with no argumentation). It is only toward the end of that section that one gets a sense of the syntactic nature of some of the regularities: In English, some verbs have optional objects (Anna is eating (a sandwich)), while others have obligatory objects (John announced *(his resignation)). For about 60 years, such verb-specific differences have been known under the heading of valency (Tesnière 1959), and the present book would have benefitted from this notion. It is the verb-specific behaviour that is relevant to syntax, not the need for arguments and roles to match (a semantic requirement that goes without saying). In view of the importance that chapter 1 attached to the substitution test and to features, it is odd that this aspect is not mentioned at all. But of course valency differences like the contrast between believe in and count on (p. 61) are instances of substitution test failures which point to the need of introducing further (subcategorizing) features. Instead, the authors here talk about “Merge overgenerating” and the need to constrain it, but this is of course precisely the purpose of syntactic features.

The last section of the chapter explains the difference between raising (Mary seems to win the race) and control (Mary hopes to win the race) and argues that one should posit an empty subject in the latter kind of construction (Mary hopes [PRO to win the race]). This is said to be forced by the Theta Criterion, but the authors consider a possible alternative: Allowing a nominal (Mary) to bear two thematic roles, and abandoning the Theta Criterion. The argument against this is that it would “open a Pandora’s box” as one would not be able to explain some basic facts, such as the impossibility of *John loves (where John bears both roles of love, i.e. ‘John loves himself’). But later, they admit that positing a zero subject (PRO) is also a kind of Pandora’s box that needs to be kept in check (e.g. by restricting its occurrence to nonfinite clauses, p. 75). But then, couldn’t one equally say that nominals may bear two semantic roles only if one of them is assigned by a nonfinite verb? So again, positing a zero element is only one of many different descriptive devices, and it just happens that the Chomskyan tradition has been favouring the PRO option. For some reason, this same tradition has not taken much interest in the fact that a close relative of *John loves (‘John loves himself’) is actually possible: John washes (‘John washes himself’). If the solution in terms of PRO were particularly insightful, one might expect that the restrictions on it would also shed light on the restrictions on zero reflexives with transitive verbs (John washes vs. *John loves), but the notion of nonfiniteness seems completely irrelevant to the latter (likewise, the idea that PRO lacks case, proposed in §4.3.2, does not seem to help with zero reflexives). Instead, I would suggest that we consider the explanatory principle of predictability: Wash is a grooming verb, i.e. the kind of verb where the object tends to be coreferential with the subject (cf. Haspelmath 2008), and hope is a desiderative verb where the subordinate subject tends to be coreferential with the matrix subject (Haspelmath 2013). Thus, both are quite predictable and hence more dispensable than other arguments.

Notes on Chapter 4: Case theory

The phenomena discussed in ch. 4 are even more idiosyncratic to English, and yet the lead the authors to suggest some completely novel concepts that have no counterparts in other approaches to syntax. But curiously, the fact that we are talking about English moves more and more into the background – the word “English” is only mentioned briefly on p. 84-85. Readers who miss these passages might get the impression that these are very common (or even universal) aspects of human language, whereas they are English-specific and in fact cross-linguistically unusual.

The chapter starts with case distinctions in personal pronouns (He visited her/*Him visited she) and notes that in English, verbs and prepositions generally require that pronouns following them occur in the Accusative case (I capitalize the labels of the cases because case forms are language-specific , cf. Haspelmath 2010: §5). When a pronoun precedes a verb, it is in the Nominative case, though this is not possible with infinitival verbs in English.

On the analogy of thematic-role assignment, K&Z start talking about “assigning case” (following Chomsky 1981, of course) and make the assumption (p. 88) that cases are always assigned by syntactic heads. For verbs and prepositions (which sometimes also assign idiosyncratic prepositions, e.g. believe in, because of), this makes good sense (though the earlier term “govern” might have been adopted instead of coining the new term “assign”). But the main proposal here is that Nominative case is actually assigned by a feature ([Finite]) of a head (“Fin”) whose exact nature seems elusive. At first, one gets the impression that this is a category for auxiliary verbs in English, i.e. those verbs that can occur before not (will not, must not, may not, has not, is not), but as the chapter progresses, one gets a sense that Fin is not a category, but rather a fillable position in the tree. Conventional syntactic trees have nodes that correspond to kinds of linguistic forms, but in the approach that K&Z introduce, tree nodes seem to be places rather than categories: On p. 95, we read that “Fin is occupied by the feature [Finite]”, that “the feature [Finite] (is) in the head of FinP”, and that the verb stem love and the 3rd singular ending -s (in John love-s sausages) are “generated in two distinct positions” (on p. 96, we read that elements may “reside in Fin”, again a spatial metaphor). There is no explanation for this shift in perspective, and there is also no explanation for the claim that Fin is a head. Semantically, will read and may read are similar to often reads and probably reads, where the pre-verbal elements were earlier said to be non-heads (p. 33). Better criteria for head vs. non-head status may be syntactic properties, but which ones?

Heads are supposed to have the same distribution as the whole construction, and this criterion would seem to allow both elements to have head status (They may take it can be replaced both by They take it and by They may, though the latter sounds elliptical). Another question is what evidence there is that the auxiliary phrase is a constituent: If it is an infinitival phrase, it can perhaps be fronted (?To go home tomorrow, I hope), but hardly replaced by a pronoun-like element (I hope so has a different meaning). But if it is not an infinitival phrase, it cannot be fronted (?*Leave soon he may), and it cannot be replaced by a pronoun-like element (*He may it, *He may so). Thus, what we learned about combinations of words and constituents in earlier chapters does not seem to help us here. K&Z have chosen to keep the same machinery (following MGG practice), but it is far from obvious that this is a good choice. Many other possibilities must exist, but they are not mentioned.

Then on p. 93, we suddenly (without explanation) see the negative element not likewise treated as a head, even though here there is good evidence that it is a non-head: It can be expanded like other adverbs (not at all, comparable to often enough), and it can be replaced by the prepositional construction by no means.

While the idea that Finiteness contexts (or perhaps more precisely non-Infinitival contexts) require Nominative case in English is not too far-fetched, the claim that there is a special Fin position also seems to fail to account for simple clauses like John loves sausages, where -s indicates Finiteness but is in the wrong position. However, K&Z are not deterred by this and simply refer the reader to a later chapter.

Perhaps the overall motivation for the introduction of abstract elements like Fin which are then claimed to be heads is to make the syntactic relationships more salient in the diagrams, and the large number of trees in the book do probably make it easier to read. But it seems much harder to argue that the abstract elements in the tree make the formalism somehow more elegant or explanatory. The same facts of English grammar could be described in a much simpler way if one used a different and richer set of concepts.

Clearly, one of the goals of the book is to show to the reader that one can do a lot of different things in English with a very small set of concepts, and it certainly succeeds at this. But is it a goal in itself to minimize the number of grammatical concepts? If so, why? Does it make language acquisition easier (or does it make it possible, in the face of the poverty of the stimulus)? Language learners (and linguists) are faced with the challenge of learning/describing the lexicon as well, and nobody doubts that the lexicon of each language contains thousands of entries. Why couldn’t grammatical systems contain many dozens or even several hundreds of categories and rules?


I had originally intended to write similar notes on the other chapters as well, but to be honest, I ran out of steam. There is a lot more that could be said, but of course it’s not nice to keep criticizing one’s colleagues, so I will stop here. (And few readers will be interested in still more details about a book that is after all not more than a textbook.)

And as I said at the beginning, the purpose is not to be negative about this really admirable book, but to ask hard questions of the MGG approach, and to challenge syntacticians who feel that they have learned a lot from it to focus on those aspects that are indeed lasting insights.

Is there anything in this book that most syntacticians would agree with? Can we use parts of it to build a bigger theory that works for all languages?

Or is this a high-risk enterprise, in the sense that while it COULD be that the approach is on the right track, it could also be that it’s completely wrong? Noam Chomsky has often said that the specifics of what he is pursuing could well be wrong, and he seems to be happy with merely inspiring people to continue thinking about the issues.

But if we want to build a cumulative science of language where later generations can build on the insights of earlier generations, then maybe we need something rather different. Readers who are confused will have to think about syntax on their own. The field is wide open.


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

Baker, Mark C. 2015. Case. Cambridge: Cambridge University Press.

Chomsky, Noam A. 1981. Lectures on government and binding. Dordrecht: Foris.

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

Corbett, Greville G. 2012. Features. Cambridge: Cambridge University Press.

Culicover, Peter. 1999. Syntactic nuts: Hard cases, syntactic theory, and language acquisition. Oxford: Oxford University Press.

Culicover, Peter W. & Ray S. Jackendoff. 2005. Simpler Syntax. Oxford: Oxford University Press.

D’Alessandro, Roberta. 2017. The achievements of Generative Syntax: A time chart.

Fillmore, Charles J., Paul Kay & Mary Catherine O’Connor. 1988. Regularity and idiomaticity in grammatical constructions: The case of let alone. Language 64(3). 501–538.

Haspelmath, Martin. 2008. A frequentist explanation of some universals of reflexive marking. Linguistic Discovery 6(1). 40–63. doi:10.1349/PS1.1537-0852.A.331.

Haspelmath, Martin. 2010. Comparative concepts and descriptive categories in crosslinguistic studies. Language 86(3). 663–687.

Haspelmath, Martin. 2013. On the cross-linguistic distribution of same-subject and different-subject ‘want’ complements: Economic vs. iconic motivation. SKY Journal of Linguistics 26. 41–69.

Koeneman, Olaf & Hedde Zeijlstra. 2017. Introducing syntax. Cambridge: Cambridge University Press.
Müller, Stefan. 2016. Grammatical theory: From transformational grammar to constraint-based approaches. (Textbooks in Language Sciences 1). Berlin: Language Science Press.

Newmeyer, Frederick J. 2002. Optimality and functionality: A critique of functionally-based optimality-theoretic syntax. Natural Language & Linguistic Theory 20(1). 43–80.

Tesnière, Lucien. 1959. Eléments de syntaxe structurale. Paris: Klincksieck.







Cite this blog post
Martin Haspelmath (2018, May 25). Confused by syntax: Some notes on Koeneman & Zeijlstra (2017). Diversity Linguistics Comment. Retrieved April 22, 2024, from

2 thoughts on “Confused by syntax: Some notes on Koeneman & Zeijlstra (2017)

  1. One way to justify a universal principle (or collection of them=part of a framework/theory) is to show that it/they predict structures that are so rare in the PLD that they can’t plausibly be learned from it. 100 million words might be a plausible estimate for that (20mw/y*10), but we don’t know what rate a construction has to occur at in to get learned, and what other conditions are relevant

    • Yes, stimulus poverty arguments for innate knowledge are convincing (in principle), but syntacticians almost never appeal to them. And some very rare patterns are cross-linguistically variable. An example that I recently came across is the use of “even” with verb-initial conditionals: “Even had she told me about it earlier, I would not have been happy”. This seems to be fine in English, but the German counterpart is completely impossible – so it can hardly be attributed to a universal principle. Apparently, we can learn this (*Selbst hätte sie mir es gesagt), even though such verb-initial constructions are not common (and quite formal), and selbst+conditional combinations are not common either.

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.