David Adger "Functional Heads and Interpretation"

Eric Hoekstra

Glot International 1.1, 9-10.

1. Introduction1

The thesis that is before us has been written from a perspective that may be termed "interdisciplinary" or rather "intertheoretical", since it incorporates research from four different theories. In spite of this, all theoretical parties concerned cannot but agree that the end result is very interesting. In what follows, I will discuss the thesis mainly from the point of view of minimalist generative syntax, not from the points of view of discourse representation theory, generalised quantifier theory or head-driven phrase structure grammar.

2. Deriving that DPs are ambiguous in AgrP , but not below

2.1. One of the strong points of this thesis is that it directs attention to a fact which is perhaps not generally known: agreement contributes to semantic interpretation. Whenever an agreement process is optional, the presence of agreement forces a familiar interpretation on the agreeing DP whereas the absence of agreement does not do so.2 This is referred to as the familiarity clause. It turns out that this fact recurs in a variety of languages, under a slightly varying guise. Thus in Dutch, the effect is parasitic not on the optionality of morphological agreement but on the optionality of leftward scrambling. The abstract explanation is the same in each case, as is clear from my paraphrase of the final version of the familiarity clause (p.104): definite DPs and DPs that entertain a movement relation with Agr must be familiar.

2.2. This effect only obtains if movement to AgrP is optional. This is claimed to follow from considerations of economy (p.93-95), which we will examine in some detail. The idea is that for the non-familiar reading the derivation without movement to Agr (and without reconstruction back into the VP at LF) is more economical than the derivation with movement to Agr and reconstruction. In languages in which movement to AgrP is obligatory, reconstruction is the only option for the non-familiar reading.
We can make a square of possibilities, representing vertically the position of the DP, horizontally the readings it may receive:

(1) familiar non-familiar
inside AgrP + -
below AgrP + +

One part of the square has not been accounted for: how can a DP within the VP receive a familiar interpretation? One would expect that the unmoved variant is unambiguously interpreted as non-familiar. An appeal to LF-movement to AgrP in order to derive the familiar reading won't help, because of Procrastinate, the principle favoring covert movement over overt movement. After all, if the familiar reading can be derived by covert movement then this option will be preferred over overt movement, and agreement will not be found. Thus it would follow that agreement cannot be optional, which is contrary to fact.
Hence Adger's proposal implies that Procrastinate be given up. Only if Procrastinate is given up can the familiar reading be derived either by overt movement to AgrP or by covert movement to AgrP, in a language in which agreement is optional. Both derivations will then involve the same movement, and thus will be equally costly. It would not have been undesirable if the consequences for Procrastinate had been pointed out, followed by a critical discussion of the arguments for and against Procrastinate.
It must also be mentioned here that Adger’s account of the semantic contrast relies on a global interpretation of economy. That is, complete derivations must be compared: it is not (or not only) the case that at each step in the derivation economy dictates the next step. This global interpretation of economy is extremely costly from a computational point of view. In connection with this, the question arises: how does the native speaker know what the set of competing derivations consists of? Surely, this set does not consist of all possible derivations of the grammar. In the case at hand, we are comparing derivations with and without agreement. Which principle tells us that we should exactly choose these derivations as being of interest? Surely, it would be ad-hoc to define the set of derivations to be compared in terms of a superficial property like that of morphological agreement being optional. To sum, an appeal to global economy may well raise the wrong kind of questions (cf. Chomsky 1992:48 for discussion from a related perspective).

3. Agreement: features or heads?

3.1. Feature manipulation is taken care of by means of Spec-Head agreement (p.12-14), rather than by means of the mechanisms offered by other frameworks. However, here the thesis shows a certain amount of tension between the varying views on agreement which each framework has to offer. The feature unification outlook is rather innocuous since it is compatible with almost anything. Nevertheless, a lot hinges on whether one chooses to work with relatively unconstrained features, or whether one chooses to treat features a heads, making them subject to all the restrictions to which heads are generally subject. This becomes clear in Adger's analysis of agreement in Celtic.

3.2. Let me present a simplified picture of the facts. In Celtic agreement on heads only shows up with obligatory pro-drop; with full NPs there is no agreement; pronouns can neither occur with uninflected nor with inflected prepositions. This clearly suggests an analysis according to which the pronoun incorporates into the head. Nevertheless, an incorporation analysis is rejected since the phenomenon is also found in coordinated structures (p.30):

(2) 's toigh leum fhin is thu fhein coffaidh
COP liking with-1S EMPH and you EMPH coffee
“Me and you like coffee”

Note that emphatic pronouns are found here. An incorporation account would need to violate the coordinate structure constraint, thus rendering this option unattractive. There are two problems with this argument.
First, conjunctions regularly seem to violate the coordinate structure constraint. For example, a similar example can be constructed on the basis of asymmetric coordination in Dutch (a phenomenon first noticed out by De Vries 1911-1912:179ff):

(3) a. Als-ie te laat thuis komt en hij heeft geen sleutel bij zich, dan ...
if-he-CL too late at home comes and he has no key on him then
b. Als- [[-ie te laat thuis komt] en [hij heeft geen sleutel bij zich]], dan ...
if-he-CL too late at home comes and he has no key on him then

Here cliticisation of the clitic pronoun -ie on COMP violates the coordinate structure constraint if it is analysed as in (3b). Needless to say, the clitic can only occur in the first conjunct. On Adger’s reasoning, (3) would imply that clitic movement should be given up in order to maintain the coordinate structure constraint. This type of example can be multiplied for other movements as well. However, we may now wonder whether the reasoning is not too narrow, and something different is going on.
Second, consider Adger’s own account. His solution is to posit a complex feature structure for Agr given below as (4a), and the parameter in (4b) (my formulation):

(4) a. Agr A-phi-feature F-phi-features
b. Coalesce A- and F-features in Celtic

The features in A are matched against those in F, which is reminiscent of matching DP against Agr in a Spec-Head configuration. As a result of (4b), “the agreeing element and the argument are competing for the same slot in a morphological representation of the functional head Agr” (p.52). Hence agreement and overt DPs are in complementary distribution. This is exactly the same result as would obtain under an incorporation account. Thus, it does not come as a surprise that (4) cannot account for (2) either. Hence, if (2) is a counterexample to incorporation, it is also a counterexample to Adger’s morphological slots analysis.

3.3. All other things being equal, there is a conceptual reason why the incorporation analysis should not be easily given up. Consider first that the alternative Adger proposes, morphological slots in some lexical representation, is a mechanism which basically does the same work as Spec-Head agreement.
Note now that this type of analysis is an addition to the theory of syntax. If incorporation can obviate the addition, then the incorporation option is to be preferred since we know that incorporation is anyhow necessary, and movement must anyhow be assumed.
Finally, as Adger adopts the minimalist framework, it seems to me that he might just as well sell all of his soul to the devil, instead of just a part of it. In this respect, then, marriage of minimalist syntax to a HPSG view of agreement seems unsuccessful: here the two views cannot both be correct. More generally, it seems to me to be desirable to treat features as heads, whenever possible, so that restrictions on heads will apply in full force and a separate theory of features is superfluous.

4. Concluding Remarks

It is remarkable that this thesis has only 136 pages. Surely, a tendency can be perceived over the years towards shorter theses. This may be taken as an indication that linguistics is moving out of the Faculty of Letters and into the Faculty of Sciences, in order words, it is becoming an exact science. After all, short theses of 100-200 pages are the rule in the exact sciences, whereas the humanities still seem to need to justify themselves by bulk.3 I am convinced that most (not all) theses of 400 pages could have been condensed into 200 pages, and gain in clarity, besides. Returning to Adger's thesis, I find that he has done an excellent job by resisting the temptation to merely fill pages. The shortness of this thesis is more than justified by the interest its ideas arouses and the clear style in which the arguments are made.


Chomsky, N. (1992) “A minimalist program for linguistic theory”. MIT Occasional Papers in Linguistics 1. MIT, Cambridge.
Vries, W. de (1911-1912) “Dysmelie. Opmerkingen over Syntaxis II”. Programma van het Stedelijk Gymnasium te Groningen. B. Jacobs, Groningen.

1 I would like to thank Marcel den Dikken and Jan-Wouter Zwart for comments and discussion.

2 The idea that the familiarity clause must be tied down to agreement is the result of research conducted within the framework of HPSG, surprisingly.

3 This is not to say that within linguistics short theses are more exact than longer ones, but there surely is a correlation between the length of a thesis and the "exactness" of the discipline from within which it originates.