THE RISE OF FUNCTIONAL CATEGORIES
Elly vand Gelderen
0.1 Word Order
Word order has always been a phenomenon of interest to linguists. In the past, it was sometimes suggested that languages had either a free or a fixed word order; the free word order languages marking grammatical relations by means of Case and the fixed order ones indicating such relations through linear order. With the introduction of the transformational framework in the sixties, however, word order was assumed to be relevant for every language. More importantly, it was shown that hierarchical order must be indicated to account for phenomena such as antecedent-anaphor relations. Phrase Structure (PS)-rules became a way to represent both the linear and the hierarchical order.
Since the early years of transformational grammar, PS-rules have become more generalized. At present, their form is most often seen as (1) and (2), cf. Chomsky (e.g. 1986; 1992) where X is no longer restricted to categories of the form [±V][±N] but can be C or I as well (Fukui 1986):
(1) XP --> X' (ZP)
(2) X' --> X (YP)
In (1) and (2), X, Y and Z stand for any lexical category (Verb, Noun, Preposition, Adjective) or functional category (Inflection, Complementizer, Determiner); in (1), ZP is the Specifier (or subject) and in (2), YP is the complement. The basic relations are thus between a head and its specifier and a head and its complement. Rules (1) and (2) can be regarded as rules of Universal Grammar, perhaps to be derived from more basic principles. The order between X' and ZP and between X and YP is free and must be specified in each language, i.e., there is a parameter that must be set by the language learner. There are restrictions on the way PS-rules are used. For instance, an intransitive V will not have a complement, but these restrictions have no place in the PS-rules; they belong in the lexicon.
Recently, Chomsky (1989; 1992) and Pollock (1989) have proposed changes in the PS-rules by using more (functional) categories. What was formerly I(nflection), i.e. the node with tense and agreement features, is now split into AGRs (agreement with the subject) and T(ense). AGRs contains the agreement features responsible for nominative Case assignment and T contains tense features. In addition, other categories have been argued for, Neg(ation), AGRo (agreement with the object), Num(ber)P and ASP(ect). Therefore, the title of this book marks a direction of current research in Government Binding/Minimalism. This direction will be examined critically. However, once one makes the distinction between lexical and functional categories, it also becomes possible to see the change between Old English and modern English as involving an increase, a rise, in the number of functional categories. The title can also be interpreted that way.
0.2 Functional Categories and Features
Thus, the issue of word order is now increasingly seen as one centered on functional categories (C, T or I, AGRs, AGRo and others) and functional categories will be my main concern also. I examine their existence in some languages, their non-existence in others and the introduction of some of them in Middle English partly within a Minimalist framework (i.e. Chomsky 1992). My main points are that parameters exist in Universal Grammar that allow languages (a) to select/incorporate functional categories and (b) to select the node in which the sets of features (e.g. tense and agreement in a pre-1992 approach; V and NP features in a Minimalist one) in that language must be placed. These parameters must be set in each language. Thus, the absence of functional categories does not mean an absence of those elements associated with functional nodes such as tense and agreement features. I argue in favor of a separation of the tense and agreement features from their respective T and AGR nodes (see 0.3. for the connection between tense/agreement features and V/NP features). Languages must express tense, as evidenced by the difference between tensed and non-tensed clauses that exists in language, and agreement. However, tense and agreement should be seen as sets of features not necessarily connected with the head of TP or the head of AGRsP. In some languages, the tense and agreement features may in fact occupy these positions; in others (English), they occupy T; and in some (Dutch and Middle English), they occupy V or C. Contrary to tense and agreement features, functional categories and their projections will only be included in the structure of a language if there exists positive evidence, such as elements that exist outside VP. I call this structural evidence.
That languages lack functional categories has been claimed for languages as diverse as Old High German and Japanese. For instance, Lenerz (1985) argues that Old High German lacks a C and that the non-occurrence of Verb-movement in certain constructions can be explained in this way; Fukui (1986) argues that Japanese lacks C and D and others have claimed it lacks I. My proposal is an alternative to proposals for Dutch and German such as those of Reuland & Kosmeijer (1988), Bayer & Kornfilt (1990), Reuland (1990) and Abraham (1991) where mixed categories of V and I are used for the seeming non-existence of I and IP. It is also an alternative to the COMP/INFL analysis that has been proposed for a number of Germanic languages by e.g. Platzack (1983).
It might be argued that functional categories do not `cost' anything: if features are present, they project automatically onto categories and maximal projections. This cannot be maintained given the evidence of languages in which e.g. tense features occur in C and not in T (den Besten 1985 and chapter 3 below) and where agreement features occur in C and not in AGR (Rizzi 1990: 52 and chapter 6 below). In these languages, there would be no automatic projection of the tense and agreement features onto a T and AGR position respectively. These are typically languages that have some Case assignment/checking under government by an element in C and V.
The model of syntactic change I assume is the one developed by Lightfoot (e.g. 1979; 1991): each language learner must internalize her or his grammar on the basis of the data available. These data may be different from the data a previous generation encountered. For instance, the change of modals and to from lexical Verb and preposition respectively to auxiliary (that is, their grammaticalization) may make it necessary for a child to set a parameter differently from the way its parents set it. I argue that in late Middle English, a functional category is introduced. This change should not just be seen as a reanalysis of some words from lexical to functional categories (e.g. Roberts 1992 in relation to the future in Romance) but also as a resetting of the parameter ±T(P), a reaction by the language learner to the grammaticalization or `bleaching' of to and modals.
My proposal has consequences for subcategorization and the position of subjects. I will argue that subcategorization is in terms of categories and sets of features. In English, tense features are in T or I whereas those languages (Old English and Dutch) that do not select a TP or IP have tense features in other categories. These differences should be noticeable in subcategorization. I show they are and formulate a theory of subcategorization making crucial use of such features. The other consequence of disassociating tense features from the T-position concerns the subject. If certain languages do not make use of TP or IP and the head or Specifier of TP or IP, the question arises what position the subject is generated in. This problem is solved if, as e.g. in Koopman & Sportiche (1991), it is argued that subjects are base generated in the Spec of VP. Then, the Spec of TP/IP is no longer needed as a position in which to base generate the subject.
0.3 The Minimalist Framework: Chomsky (1992)
As mentioned, the framework used is a modified `Minimalist' one based on earlier work such as Chomsky (1981; 1986). As I go along, relevant aspects of this theory will be spelled out. In this section, I highlight the points most notable to my topic.
Relevant first of all is the possibility of unlimited numbers of functional categories. In relation to language learnability, there are several approaches one might take. Assuming that Universal Grammar makes functional categories available to the language learner, one could say (a) they are available in all languages even though they may be `inactive', or (b) a language learner, on the basis of contact with the language surrounding her or him, selects certain functional categories to be part of the analysis for that language. I argue in favor of (b) in this book.
A second striking aspect of the Minimalist theory is that Case assignment no longer takes place. Neither do Verbs any longer move to T or AGRs to `pick-up' their inflections. Instead, elements such as Nouns and Verbs "are drawn from the lexicon with all of their morphological features, including Case and [phi]-features, and [...] these [...] must be checked in the appropriate position" (Chomsky 1992: 41). The appropriate position for an NP is the specifier of a functional category and for a Verb, the position is adjoined to the head. The functional categories have V and NP features and after the V and NP move, a checking takes place with the relevant features of the functional head. If the features match, the features of the head disappear (p. 39) and SPELL-OUT takes place; if the features do not match the V or NP, the features remain and will `crash' at Phonological Form (hence PF).
A possible problem is that an item must be selected at random from the lexicon and after checking at Logical Form (hence LF), the result may be ruled ungrammatical. This is a cumbersome process even though the processor may be very fast and checking may therefore be non-problematic. Alternatively, there is perhaps feedback to the lexicon of some kind. I will not examine this particular problem.
Differences between languages in word order will reduce to whether checking must occur at LF or before LF, i.e. covertly or overtly. The reason for overt movement would be that the features in the functional head are strong and that if they are not erased by PF, the construction will `crash'. If the features are weak, checking waits till LF because that is less `costly' on the system. Thus, if a language has strong V-features, Verb-movement takes place; if it has weak features, Verb-movement does not occur. The same holds for NP-movement. In this theory, differences between languages with respect to movement follow from whether features in the functional head are weak or strong.
It is not spelled out in Chomsky (1992: 40-4) what the relation of V and NP-features is to what used to be called tense, agreement and Case features. I will assume they are similar, i.e. V-features are like tense and agreement and NP-features involve Case.
A third major aspect, related to the second, of the above system is that all Case relations are Spec-Head relations. I will, however, follow work by, for instance, Koopman & Sportiche (1991) and argue that structural Case can be checked/assigned under government as well. (For the sake of convenience, I refer to Case checking as `checking/assignment'). In addition, there will be inherent Case, about which Chomsky (1992) says virtually nothing.
The outline is as follows. Chapter 1 points out that in English there is evidence for one functional node between C and Neg(ation), i.e. T (or I). The evidence derives from modals, do, to and from adjunction. Chapter 2 provides evidence that there are languages without a T/I-position. The evidence is based on the non-existence of elements such as modals in English that seem separate from the VP and on adjunction. Putting this in terms of Universal Grammar, it can be said that there is a parameter ±TP or ±IP (but see note 2): Universal Grammar makes available functional categories and projections but they are not selected in every language. The related second parameter is that the positions must be selected in which the agreement, tense and other features (also referred to as V and NP-features) are placed: these can be any functional position (C, T, or AGR) and V.
Chapter 3 shows that a T/I position is introduced in English by 1380 and that this position is then occupied by to, tense features, do, and by modals successively. This change makes it possible for the main clause to be seen as a TP/IP, rather than as a CP (which it is in Old English). The decline of Verb-movement (V-to-C) can be seen in this way: instead of arguing that V-to-C gradually disappears, it is possible to say that adverbials and topicalized elements are no longer seen as occupying Spec CP but as adjoined to the matrix. This is true already in Old English since many adverbials and topicalized elements do not trigger inversion. In Old English, however, there is evidence for a CP main clause; in late Middle English the matrix is ambiguous between CP and TP/IP. Chapter 4 gives evidence that the position in which tense features are generated shifts: initially they are on the V, then in C and finally, they occur in a separate node, namely T or I. This fits with the second parameter that features do not automatically come with a particular node but that each language selects whether they occupy C, T, or V.
In chapter 5, I discuss the use of tense features in subcategorization. This shows that features are needed independently of categories. It provides additional evidence for splitting up categories such as T from features.
In chapter 6, it is shown that agreement features occupy different positions in different languages, again in accordance with the parameter that features must be placed in either a functional category (C, T or AGR) or V. In recent publications, it is argued that agreement features are placed in the AGR-position. However, even in languages such as modern English, I argue the evidence for AGRs and AGRsP is non-existent. In this language, the agreement features are in T, together with the tense features, as in Chomsky (1986). I show that in Dutch, Old and early Middle English the features are in C and V (just as the tense features are). There is also evidence that modern English has fewer agreement features than Old English. The strength of the agreement features as well as the non-existence of special functional categories may account for the difference in Case assignment. Dutch does not have the same nominative that English has. Finally, I explain some differences between Old English and modern Dutch nominative Case marking.
In chapter 7, AGRo(P) and other functional categories such as PerfP, ASPP and PassP are examined. I show that in none of the languages under discussion, i.e. the various stages of English and Dutch, is there evidence for an AGRo(P). If AGRoP is not present, objective Case cannot come about via Spec-Head agreement. I examine other ways. There is, in modern English, evidence for three other functional categories besides T, namely PerfP, ASPP and PassP. In chapter 8, I look in detail at one of these, namely ASP. Again, it is shown that features, e.g. aspect features, may occupy a category other than the ASP position.
The increase of functional categories in English is not necessarily an irreversible change. One might imagine the reverse happening again. For instance, tense features might again come to be situated in C in modern English and the T-position might disappear. The parameters in Universal Grammar were no different in Old English. Nor are they different in Dutch or German, only their parameter settings are.