THE RETURN OF THE SUBSET PRINCIPLE
Theresa Biberauer and Ian Roberts
University of Cambridge
mtb23@cam.ac.uk
igr20@cam.ac.uk
1.
Introduction and general preliminaries
In the context of generative theory, it has often been argued that aspects of
language acquisition and/or constraints on learnability are a cause of change (see in
particular Lightfoot 1979, 1991). In particular, the abductive aspect of language
acquisition has been thought to lead to reanalysis and associated parametric changes.
In this context, it is desirable to attempt to relate principles of learnability to principles
of change. To the extent that this can be done, the general programme for
understanding language change in terms of acquisition and learnability can be
advanced. The goal of this paper is to make a contribution in this direction.
The particular principle that we will be concerned with here is the Subset
Principle (SP henceforth). Originally put forward by Berwick (1985), we can give the
following informal statement of this principle (cf. also and Manzini & Wexler 1987):
(1) “the learner must guess the smallest possible language compatible with the
input at each stage of the learning procedure” (Clark & Roberts 1993:304-5)
The conceptual interest of the SP lies in the fact that it arguably derives from a
widely recognised fact about language acquisition: that children do not seem to make
use of (direct) negative evidence. In other words, information that certain parts of the
input text (sentences or strings), are ill-formed is not available, or at least not made
use of by acquirers. It follows from this that acquirers run the risk of falling into
“superset traps”: if a grammar which generates a language which is a superset of the
target language is posited, no positive evidence can disconfirm this hypothesised
system. Hence acquirers must always posit the grammar which generates the
“smallest language” consistent with the trigger experience; in this way positive
evidence can be maximised in the process of convergence towards the target grammar
in the sense that evidence of this type may be sufficient to cause the acquirer to revise
hypotheses.1
If the SP, or something like it, is indeed relevant to language acquisition, then, to
the extent that syntactic change is driven by acquisition, the SP may be relevant in
accounting for change (this point was also made by Clark & Roberts 1993:305-307).
However, as has frequently been noted (cf. i.a. MacLaughlin 1995, LaFond 2001),
there is a problem with the SP as formulated along the lines in (1). This is that many,
perhaps all, of the parameters that have been independently proposed in the literature
A grammar G generates a language L which is smaller than another language L’ generated by G’
iff G’ generates at least one grammatical string Si which is not generated by G and G generates no
grammatical string Sj≠i which is not generated by G’. Taking grammars to be defined as sets of values
of parameters of UG, we can then say that a given parameter-value vi of parameter Pi is a subset value
of Pi where Pi defines a grammar of type G as just defined and value vj defines a grammar of type G’,
assuming (crucially) all other parameters are set to the same values in both G and G’.
Regarding the determination of subset and superset parameter values, clearly this cannot be done
by inspection of the sets of strings generated by the different grammars defined by the different
parameter values, since these will presumably always be infinite. This is an aspect of what we call the
“implementation problem” associated with the SP – see Note 2.
1
on comparative syntax seem to define intersection relations rather than inclusion
ones.2 Consider, for example, the parameter which determines OV as opposed to VO
order in transitive VPs, which, for convenience and following tradition, we can call
the Head Parameter. Set to OV, it generates the grammatical strings in (2a,b) and not
the ungrammatical one in (2c), and set to VO it generates the strings in (3b.c) and not
the ungrammatical one in (3a):
(2) a.
b.
c.
John Sue loves.
John walks.
*John loves Sue.
(3) a.
b.
c.
*John Sue loves.
John walks.
John loves Sue.
Clearly, the OV grammar and the VO grammar are in an intersection relation, as
shown in (4):
(4)
John Sue loves
G1 (OV)
John walks
John loves Sue
G2 (VO)
Even the classical Null Subject Parameter of Rizzi (1982), which Berwick (1985)
took as an example to illustrate the SP, defines an intersection relation if one takes
into consideration null-subject languages of the Italian type lack overt expletives
while non-null-subject languages lack null pronouns:
(5) a.
b.
c.
He speaks.
*Speaks.
It seems that John speaks.
(6) a.
b.
c.
Lui parla.
Parla.
*Ciò sembra che Gianni parla
2
The intersective nature of many parameters is an empirical problem stemming from comparative
linguistics, and represents in a way the main focus of this paper. There is also a conceptual problem
with the SP, which may have a psycholinguistic dimension: the “implementation problem” alluded to
in Note 1. The issue is how the learning device (whether a child or an algorithm) “knows” which are
the subset and superset values of a given parameter. Since this cannot be directly determined by
inspection of the trigger experience, we must, as it were, “build it in” to the learning process or the
structure of parameters. One possibility is to assume that the learning device is equipped with an
inherent system of default values for parameters which correspond to the subset values (as proposed by
Manzini & Wexler 1987). Alternatively, the relevant notion of markedness may be more abstract, being
a property of systems of parameters rather than of individual parameter settings: in this case, too, the
learning device could be structured so as to inherently favour certain combinations of parameter values
over others, and these could be combinations yielding subset grammars. In this connection, the
discussion of the SP in relation to “shifted” parametric systems in Clark (1992) is relevant. Deciding
amongst these and other imaginable options is well beyond the scope of this paper; what seems clear,
however, is that some further specification of the nature of UG and/or of the learning device is
required. On this, see Clark & Roberts (1993), Hale & Reiss (2003), Fodor & Sakas (2005).
2
Again, we have an intersection relation, as (7) shows:
(7)
Null subjects
G1 (NSL)
Lexical subjects
Expletive subjects
G2 (non-NSL)
For this reason, the SP has been thought to be of limited value. In fact, if all
parameters turned out to define intersecting grammars, which appears likely on the
basis of the above discussion, the SP would be of no real value at all. The point of this
paper is to argue that this is not the case, once the role of true formal optionality (i.e.
syntactic “free variation”) is fully taken into consideration. We will show that there
are parameter settings which give rise to grammars which generate languages which
are in inclusion, rather than intersection, relations, and therefore the SP is relevant
both for language acquisition and in language change.
Our central idea will be that, where the evidence for the grammar which generates
the larger language is not sufficiently robust, acquirers “default” to a grammar
generating a smaller language. This, as noted above, is ultimately a consequence of
the absence of negative evidence in language acquisition. So here we see a case where
the conditions of language acquisition lead directly to change, and hence the SP is
relevant in both domains.
More specifically, we will show that the SP is relevant to two types of change.
First, where a system which at one stage allows a pied-piping option alongside a
“stranding” option at a later stage allows only the stranding option (we will give a
precise characterisation of these notions of pied-piping and stranding in Section 2).
Second, the SP is relevant in the case of restriction of function (as introduced by
Roberts & Roussou 2003), i.e. the narrowing down of an operation to a subset of the
contexts in which it formerly applied.
The paper is structured as follows: section 2 presents the theoretical background
underlying our assumptions about pied-piping, word-order variation and “true”
optionality. Sections 3 and 4 then focus on changes that took place in the history of
English that, we believe, illustrate the operation of the SP in the diachronic context.
Firstly, section 3 considers the loss of pied-piping options that previously co-existed
with “stranding” structures, thereby bringing about word-order change in earlier
English. Section 4 considers two instances of the second case, restriction of function:
firstly, the restriction of object movement during Middle English and secondly, the
restriction of V-to-v-to-T movement that led to the rise of the modern English system
of do-support. Section 5 concludes the paper.
2.
Theoretical preliminaries
We assume the Probe/Goal/Agree system of Chomsky (2000 et seq.). In terms of this
system, heads are syntactically active as long as they bear one or more unvalued
features. To see how it works, consider (8) below:
3
(8)
(a)
… XPROBE … [YP … ZGOAL … ] …
(b)
XP
ru
XPROBE
YP
ru
Y’
ru
ZP
…ZGOAL…
…
Y
Here, a head X bearing an unvalued feature (or features; for expository
convenience, we will present our discussion in terms of single unvalued features) acts
as a probe in search of a valued counterpart of its unvalued feature. This it finds on
the goal, Z, which is located in its c-command domain and which, in turn, bears an
unvalued feature of its own, which renders it active and therefore visible to the probe.
Agree takes place between the probe and goal, with the result that the unvalued
features on both heads are valued. Feature valuation, which corresponds to the feature
checking of earlier versions of the theory (cf. Chomsky 1995), is therefore achieved
without movement. For movement to take place, a probe must bear an additional
movement-triggering feature, conventionally referred to as an EPP-feature (although
see Chomsky 2005 et seq. on so-called Edge Features). Following Richards &
Biberauer (2005) and Biberauer & Richards (2006), we assume that UG offers various
parametric options in respect of the “size” of the constituent that undergoes movement
wherever X probes Z and also bears an EPP-feature (i.e. movement diacritic). These
are given in (9):
(9)
a.
b.
c.
d.
[XP Z - X [YP … [ZP … (Z) …]]]
[XP ZP X [YP … (ZP) … ]]
[XP [YP … ZP … ] X (YP) ]
[XP ZP X [YP … (ZP) … ]] AND
[XP [YP … ZP … ] X (YP) ]
Z-movement only:
ZP-movement only:
Obligatory pied-piping:
Optional pied-piping:
In the case where the probe X = T, the goal Z = a D-feature (or, more
probably, the phi-features associated with an active DP) and Y = v, we therefore
expect the following range of movement possibilities:
(10)
a.
b.
c.
d.
[TP D - T [vP … [DP … (D) …]]]
[TP DP T [vP … (DP) … ]]
[TP [vP … DP … ] T (vP) ]
[TP DP T [vP … (DP) … ]] AND
[TP [vP … DP … ] T (vP) ]
D-movement only:
DP-movement only:
Obligatory pied-piping:
Optional pied-piping:
(10a) involves adjunction of the D-head of the DP-subject to the probing T-head, a
movement operation that has been argued to take place in various VSO languages (cf.
i.a. Baker & Hale 1990, Guilfoyle et al. 1992 and Travis 2006 for discussion). As this
mode of EPP-satisfaction does not play a role in the specific contexts with which we
will be concerned, we will leave it aside here. (10b – d) are, however, directly relevant
to our concerns in this paper.
4
(10b) instantiates the mode of EPP-satisfaction standardly assumed for Dseeking T in modern English (and many other languages), i.e. subject DP-raising to
Spec-TP. (10c) represents a mode of EPP-satisfaction that Richards & Biberauer
(2005) and Biberauer & Richards (2006) argue to differ minimally from those more
standardly assumed: instead of a D-head (which could also be associated with finite
V; cf. Alexiadou & Anagnostopoulou 1998) or DP undergoing raising to Spec-TP, it
is also possible that either of these D-bearing categories may constitute the goal of
Agree3, while a larger category properly containing this goal is targeted for
movement. In other words, the authors exploit the distinction between two related
properties that a probe may have in the context of Chomsky’s Probe/Goal/Agree
system, namely that of being a probe on the one hand and that of being associated
with one or more EPP-features on the other: in terms of this system, it does not follow
that the category that a given head probes has to correspond to the category that
ultimately undergoes movement under the influence of the probe’s EPP-feature; as
long as the moving category contains the goal, the computational system will “not
mind” (cf. also Biberauer & Roberts 2005, 2006a for further discussion of this point).
Against that background, it can be seen that (10c) differs from (10a,b) only in
respect of the “size” of the moved category: the goal in this case may once again be
either an active D or DP in the vP-domain, but the category that undergoes movement
is vP, i.e. the category that immediately contains the D(P)-goal. Richards & Biberauer
(2005) and Biberauer & Richards (2006) propose that inflectionally rich German
represents a language in which T Agrees with D-on-V (cf. Alexiadou &
Anagnostopoulou’s (1998) proposal for Greek) and then moves the entire vP to SpecTP. Inflectionally more impoverished Dutch and Afrikaans, by contrast, are argued to
be languages which target the subject DP, but then move the whole containing vP (cf.
also Biberauer 2003; Jouitteau 2005 proposes that Breton is another language that
employs vP-raising as a means of satisfying T’s EPP-requirements, and cf. also
Chomsky 2001:38).
The important point about the pied-piping modes of EPP-satisfaction in the
present context is that they, under currently rather poorly understood circumstances,
appear to permit formal optionality which is not available to languages employing
non-pied-piping modes of EPP-satisfaction (i.e. D- or DP-raising). In particular, there
appear to be synchronically attested languages that allow EPP-satisfaction via vPraising (i.e. pied-piping) to alternate with satisfaction via DP-raising (i.e. stranding),
modern spoken Afrikaans and various dialects of German and Dutch being cases in
point (cf. Biberauer 2003, Biberauer & Roberts 2005, 2006a). The crucial
consideration for our purposes is that a grammar which permits this optionality would
correspond to (10d) above, i.e. to a grammar which generates both the strings
generated by (10b) and those generated by (10c). (10d) therefore represents a superset
language in relation to both (10b) and (10c). As such, we assume that it must be very
robustly triggered by the PLD as a system of this kind is inherently disfavoured by the
SP (cf. discussion in section 1). Biberauer & Roberts (2005; henceforth: B&R)
propose that earlier stages of English permitted pied-piping modes of EPP-satisfaction
3
Biberauer & Roberts (2006a) speculate on the factors that may potentially determine the nature of the
category that counts as the goal for a given probe, suggesting that the distribution of the features being
probed, as “signaled” by morpho-lexical cues, may play a crucial role. Thus, if a probe seeks a
complete set of phi-features, as is usually assumed for finite T, and these features are consistently
present on a D-head in T’s c-command domain, D may become the designated goal for phi-seeking T
in that language. Similarly, if these features are spread across various DP-internal heads, the entire DP
may count as the goal.
5
that are no longer available in modern English. According to B&R, various nonsyntactic factors interacted with existing syntactic properties to bring about the loss of
the relevant pied-piping operations. In the following section, we will show how the
loss of these options can in fact be understood as following from the operation of the
SP, understood as a guiding principle in language acquisition.
3.
Case study I: word-order change in Middle English (ME)
As noted above, B&R propose that Old and Middle English (henceforth: OE and ME)
represent stages of English that feature pied-piping modes of EPP-satisfaction in
domains in which this is no longer synchronically possible. More specifically, B&R
propose that OE differed from modern English (NE) in that both T and v bore
movement-triggering EPP-features. As such, they did not only, as in NE, act as probes
for D-features (those associated with the subject and the direct object respectively),
but they also brought about movement of categories containing these D-bearing goals.
Let us firstly consider OE. At this stage, both v and T were associated with
EPP-features which could be satisfied either by movement of the goal alone (i.e. DPmovement) or by movement of the category immediately dominating the goal-DP (i.e.
VP- or vP-movement). In respect of the satisfaction of the EPP-features on v and T,
therefore, OE was an optional pied-piping language of the kind schematised in (9d)
above, i.e. a superset language relative to one permitting only one or none of these
modes of EPP-satisfaction.4 More specifically, it permitted the following EPPsatisfaction options in the vP- and TP-domain respectively (following B&R, we
assume V-to-v raising in OE and ME; cf. section 4.2 for further discussion of this
point):
(11)
a.
b.
VP pied-piping: [vP [VP … DP-OBGoal … ] V+vProbe (VP) ]
VP “stranding”: [vP DP-OBGoal V+vProbe [VP .. (DP) ..]]
(12)
a.
b.
vP pied-piping:
vP “stranding”:
[TP [vP DP-SUGoal VP V+v ] TProbe (vP) ]
[TP DP-SUGoal TProbe [vP (DP) … ]]
Let us firstly consider the v-related movement operations schematically
represented in (11) above and illustrated in (13) below (goal underlined and moved
category boldfaced in each case; lower, ultimately unpronounced copies indicated in (
)):
(13)
Satisfaction of v’s D(object)-oriented EPP-feature:
a.
VP pied-piping:
… þæt ic [VP þas boc to Engliscre spræce (V)] awende+v (VP)
... that I
this book to English tongue
translate
“... that I translate this book into English”
[AHTh, I, pref, 6; van Kemenade (1987:16)]
4
It is worth noting that a system consistently requiring a particular movement operation (e.g. a (9a), (b)
or (c)-type system) and one consistently lacking this movement operation are not in a superset-subset
relationship: a movement-requiring grammar will not generate more strings than a movement-lacking
one; the two grammars will simply consistently deliver different output strings wherever the movement
operation applies in the former, but not the latter. From the perspective of the SP, then, movement per
se is not associated with any “cost”.
6
b.
VP pied-piping:
swa þæt se scinenda lig [VP his locc up(V)] ateah+v (VP)
so that the shining flame his locks up- drew
“… so that the shining flame drew his locks up”
[ÆCHom II, 39.1.295.241; Fischer et al. 2000:189]
c.
VP stranding (i.e. sole movement of the goal) in the vP-domain:
Dunn hafað [DP þas boc ]
gesald+v [VP his wife (V) (DP)]
Dunn has
this title-deed given
his wife
“Dunn has given this title-deed to his wife”
[ch1514(Rob9)1); Koopman 1994:59]
d.
VP-stranding:
ðe he [DP hine] ætbræd+v [VP fram flæsclicum lustum (V) (DP)]
that he himself withdrew
from fleshly
lusts
“… because he had withdrawn himself from fleshly lusts”
[AHTh, I, 58; van Kemenade 1987:34]
In (13a), the goal probed by v is the direct object þas boc. In moving to SpecvP under the influence of v’s EPP-feature, this goal pied-pipes the additional material
contained in VP (i.e. to Engliscre spræce (V)), with the result that the lexical verb
ultimately surfaces string-finally.5 In (13c), by contrast, the same goal (þas boc)
moves independently of the remaining material in VP, giving the surface effect of
“leaking”/extraposition. Both structures converge because an XP containing the goal
moves to the specifier of the probing v. (13b) is like (13a) in being the output of a
derivation involving VP pied-piping: in the context of the analysis proposed in B&R,
in terms of which V consistently raises to v (cf. section 4.2 for further discussion), it is
clear that orders like (13c) must involve movement not only of the probed object-DP,
but also of the particle; B&R propose that these elements raise together as another
case of VP pied-piping (cf. Elenbaas 2007 for further discussion). Particle-verb
structures, alongside rigidly verb-final structures such as that illustrated in (13a),
therefore constitute a key piece of data indicating the availability of VP pied-piping as
a means of satisfying v’s EPP-feature. (13d), in turn, represents further evidence
(alongside (13c)-type structures) that VP-stranding is also an option: here the
reflexively used personal pronoun constitutes the goal probed by v, with only this
element undergoing raising under the influence of v’s EPP-feature and the PP being
stranded when the object undergoes movement.
The T-related modes of EPP-satisfaction available in OE are illustrated in
(14):
(14)
Satisfaction of T’s D(subject)-oriented EPP-feature:
a.
vP pied-piping:
þæt [vP þæt folc Gregorium to papan gecoran ] hæfde (vP)
that
the people Gregory
to pope elected had
“… that the people had elected Gregory as pope”
On the approach advocated in B&R, the “head-final” orderings in OE and ME therefore fall out from
a consistently head-initial underlying structure (cf. Biberauer & Roberts 2006a,b for more discussion).
5
7
[AHG, IX, 104; van Kemenade 1987:34]
b.
vP stranding:
þæt [DP he ] mehte [vP (DP) his feorh
generian ]
that
he could
his property save
“so that he could save his property”
[Orosius, 48, 18; van Kemenade 1987:59]
(14a) illustrates the structure that results when the subject-goal of the T-probe piedpipes the vP within which it is contained. Note that this typically West Germanic
SOVAux ordering also requires pied-piping satisfaction of v’s EPP-feature, i.e.
although Gregorium is the goal of v, it must pied-pipe the rest of VP to deliver [vP [VP
Gregorium to papan] gecoran+v] prior to the merger of the subject-DP. In (14b), only
the subject-goal, he, undergoes movement to Spec-TP, with the consequence that the
remaining vP-internal material (his feorh generian or his feorh, depending on one’s
analysis of the structure; cf. Biberauer & Roberts 2006a for further discussion)
surfaces in post-T position.
B&R propose that VP pied-piping as in (11a)/(13a,b) was lost as an option
alongside VP-stranding (i.e. sole movement of the goal) in early ME. Biberauer &
Roberts (2006b) ascribe this to a number of factors, including:
(i) the large incidence following the Norman invasion of French borrowings
which i.a. replaced particle verbs and thereby removed from the input OPart-V orders of the type illustrated in (13b) (cf. Spasov 1966, cited in
Kroch & Taylor 2000:146, who notes that particle verbs are vanishingly rare
during the 12th and 13th centuries; see also Fischer 1992:386 and the
discussion in Biberauer & Roberts 2006b)
(ii) the relatively low incidence of compound tense-containing structures in OE
and early ME (cf. Traugott 1972). This consideration is relevant because the
majority of OE and early ME main clauses were V2 structures in which the
finite verb surfaced in clause-second position (cf. Fischer et al. 2000:118ff
for discussion of structures featuring personal pronouns). As such, these
clauses, which are usually assumed to play a crucial role in the acquisition
context (cf. i.a. Lightfoot’s (1991) ‘degree zero learnability’ proposal),
would have been ambiguous in respect of the nature of the operation that has
taken place to satisfy v’s EPP-feature: once the lexical verb has raised to
clause-second position, it can no longer serve as “signpost” as to the size of
the category that has undergone raising to Spec-vP – cf. (15) where the XPVf-Subj-Obj-PP output string does not allow one to detect whether the string
has in fact been derived via VP pied-piping as in (a) or VP-stranding as in
(b):
(15)
a. VP pied-piping in an auxiliary-less V2 clause:
[CP XP V+v+T +C [TP Subj (V+v+T) [vP Subj [VP Obj PP (V)] (V+v)] (VP)
b. VP-stranding in an auxiliary-less V2 clause:
[CPXP V+v+T +C [TP Subj (V+v+T ) [vP Subj [VP Obj (V)] (V+v)] PP (V)
(Obj)
(iii)
the loss of dative case (cf. Allen 1995:441, Table 10.1) and the
concomitant rise of indirect object-PPs. This resulted in a rise in the
8
number of argumental (as opposed to adjunct) PPs surfacing in “leaking”
configurations of the kind illustrated in (13d) above at the expense of the
IO-DO-V orders which had formerly triggered the VP pied-piping option
in (11a). Thus structures like that in (16) featuring a dative-marked indirect
object where replaced by structures like (17) in which the indirect object is
realised as a PP:
(16)
gif se sacerd ne mæg [VP ðam læwedum mannumIO larspelDO ] secgan+v
if the priest not can
theDAT layDAT
peopleDAT homilyACC say
“if the priest cannot say a homily to the lay people”
[Ælfric’s Homilies II, 41.306.66; Koopman & van der Wurff 2000:259, 1a]
(17)
He slewe his broder Amon that [DP suche desloyaltee and vntrouth ] had
done to his sister
He slew his brother Amon that such disloyalty and untruth had done to
his sister
[Caxton Knight of Tower 87.15]
Biberauer & Roberts (2006b) suggest that the lexical and morphological
considerations in (i – iii) conspired to affect the PLD to which early ME children were
exposed in such a way that the VP pied-piping operation in (11a) became
insufficiently robustly triggered. In other words, they propose that independent and
contingent non-syntactic factors were behind the syntactic change that took place in
early ME. Since the SP inherently favours a grammar generating fewer strings (cf.
Note 1), the drop in sufficiently unambiguous triggering evidence for VP pied-piping
led to the loss of (11a) as a means of satisfying v’s EPP-feature. In respect of this
parametric setting, early ME therefore became a subset language of the (9b)-type
relative to OE which was a (9d)-type language.
Let us now consider the changes that affected T during ME. Following the SPmediated reanalysis of the ways in which v’s EPP-feature might be satisfied (i.e. the
loss of VP pied-piping just discussed), object movement became more and more
restricted during the ME period, ultimately being lost for non-quantified objects (cf.
the discussion of this particular instance of restriction of function in section 4.1
below). The restriction of object movement to only a specific subset of objects had an
important consequence for the PLD ME children received regarding the manner in
which T’s EPP-requirements could be satisfied. Recall that both exclusive movement
of the goal (i.e. subject DP-raising) and pied-piping of the category containing the
goal (i.e. vP-raising) were available options in OE. Once object movement became
restricted, however, the PLD contained an increased number of structures in which vP
pied-piping as in (12a) was indistinguishable from vP-stranding as in (12b). Let us
consider why this is so.
Wherever an object fails to undergo EPP-driven raising into the vP-domain, it
remains VP-internal. In the context of the phase-based Probe/Goal/Agree system of
Chomsky (2000 et seq.) and, in particular, (the strict version of) the Phase
Impenetrability Condition (PIC), material located within the VP-domain is subject to
spellout upon completion of the vP phase. B&R propose a radical interpretation of
“sending to Spellout” in terms of which elements that have been sent to Spellout are
not available for any further syntactic operations, including movement. Thus an object
which has not undergone raising to Spec-vP is sent to Spellout upon completion of the
9
vP phase and will be spelled out in its VP-internal position (along with any other VPmaterial). In the context of the ME structures that we are considering here, this
proposal has the specific consequence that unmoved objects will no longer be part of
the vP which is available for raising to Spec-TP. In other words, the loss of
generalised object movement in later ME had the consequence that structures in
which T’s EPP-requirements were satisfied by vP-raising became harder to
distinguish from those in which DP-raising satisfied these requirements. In a variety
of structures, including matrix and embedded clauses lacking an auxiliary, it would in
fact have become impossible to determine which raising operation had taken place as
the output string in both cases will be S-V-O. Consider the embedded clause case
illustrated in (18) (material sent to Spellout indicated in outline font):
(18)
a.
[TP [ Subj ] T [vP (Subj) V+v ]
-- DP-raising
b.
[TP [vP Subj V+v] T ] (vP)
-- vP-raising
In (a), DP-raising results in an S-V-O string in which the finite verb is located in v
within the unraised vP, while vP-raising in (b) produces a surface-identical string in
which the finite verb is once again located in v, but vP is located within the TPdomain. V2 matrix clauses likewise constitute ambiguous input for reasons similar to
those outlined for (15) above. Clearly, then, the changes affecting the manner in
which v’s EPP-feature could be satisfied had a knock-on effect in the TP-domain.
This can be schematised as follows:
(19)
a. Changes in the satisfaction of v’s EPP-feature
Reanalysis I:
Early ME – loss of the VP pied-piping mode of satisfaction
[vP [VP (V) O ] V+v
] → [vP O V+v
]
Later ME – restriction of the presence of EPP on v
[vP O V+v
] → [vP ONeg V+v
Thus, for non-Neg O: [vP V+v
]
]
b. Changes in the satisfaction of T’s EPP-feature
Reanalysis II:
Late ME, around 1450 - loss of the vP pied-piping mode of satisfaction
[TP [vP S V+v ] T ([vP…
]) ] → [TP S T [vP (S) V+v
]]
Reanalysis II effectively created a “canonical subject position” in Spec-TP (cf.
Biberauer 2006), with consequences for expletive distribution, raising-to-subject in
passive and unaccusative contexts and so-called “Stylistic Fronting” structures (cf.
B&R and Biberauer & Roberts 2006a,b for further discussion). In the context of our
present discussion, the point worth noting is that the T-related reanalysis, like its vrelated counterpart, involved the loss of a pied-piping option that had, for independent
reasons, become difficult to discern on the basis of the PLD. In both cases, therefore,
the operation of the SP resulted in a grammar permitting both pied-piping and
stranding being replaced by an innovative one generating a smaller language.
10
4.
Case Study II: “Restriction of Function”
By “restriction of function” we mean the case where a change operates in such a way
as to limit the set of contexts in which a movement operation applies. More
specifically, restriction of function denotes those cases where, at an earlier stage, a
movement operation affects a relatively large class E of elements and, at a later stage,
only a subset D  E. The SP is relevant to this kind of change at the stage where the
movement process applies optionally to E – D (the part of E that is not in D). A
grammar where a movement operation applies optionally in E – D is a superset both
of the one requiring movement in the larger set of contexts E and of the requiring it
only in the smaller set of environments D. Again, optionality gives rise to the superset
grammar.6
We consider two cases here: first, object movement in ME (which follows on
directly from the discussion of word-order change in the previous section) and the
more complex case of the development of auxiliary do in ENE.
4.1
Object Movement in ME
Recall that in Section 3 we stated that the following reanalysis took place in Early
ME:
(20)
Reanalysis I
[vP [VP (V) O ] V+v (VP) ]
→
[vP O V+v [VP (V) (O) ]]
This change affected definite and indefinite objects differently (in fact, according to
B&R, Kroch & Taylor (2000), Pintzuk 2002 and Pintzuk & Taylor 2003, definite and
indefinite direct objects underwent movement under slightly different conditions in
OE too). After this change, movement of indefinite objects remained obligatory, as we
shall see below. Movement of definite objects also remained, but in this case
apparently as an option until c.1400 (van der Wurff 1997, 1999 and Foster & van der
Wurff 1997). In terms of our theoretical assumptions, this means that the EPP-feature
responsible for triggering movement of definite direct objects was optional after the
reanalysis in (20). Chomsky (2001:34) proposes that optional EPP-features must be
associated with a semantic effect of some kind. Hence, a consequence of the
movement of definite direct objects becoming optional was that it became associated
with a specific discourse effect, namely defocusing. Around 1400, this movement of
definite direct objects was lost, while, among indefinites, at least negatively
quantified DPs continued to move.
Here are some examples of pre-1400 object shift of definite direct objects:
(21)
a.
b.
þæt ic nule þe forsaken
that I will-not you forsake
[St Juliana (Bod) 278; Koopman & van der Wurff 2000:269]
ðat we moten … swa þis scorte lif her laden
Roberts & Roussou (2003:162) use the term ‘restriction of function’ in a slightly different
way, referring for example to the change in distribution of Ancient Greek indefinites, which in Modern
Greek can only function as wh-words. The two notions are related if we think of the “function” in
question as the environments in which (first or second) Merge is possible.
6
11
c.
that we may … thus this short life here lead
[Vices & Virtues 21.23; Koopman & van der Wurff 2000:269]
þet heo ne schal þene stude neauer mare changing bute for need ane
that she NEG shall the abode never more change but for need alone
[CMANCRIW, I.46.52; Kroch & Taylor 2000:148]
Examples of this type are only found with defocusing after 1400 (see B&R), and are
not found at all after approximately 1500. As mentioned above, in later Middle and
Early Modern English, i.e. in the later 15th and early 16th centuries only negative
direct objects shift (van der Wurff 1997, 1999, Kroch & Taylor 2000, Moerenhout &
van der Wurff 2000, Pintzuk 2002, Ingham 2002, 2003):
(22)
a.
b.
c.
I … wist not where I myghte any such fynde
I knew not where I might any such find
[More Apology 54.12]
þei shuld no meyhir haue
they should no mayor have
[Capgrave Chronicles 62.23]
and as ffor any hodyr tydyngys I con none wrytt vnto yow as zett
and as for any other tidings I can none write unto you as yet
[Cely Letters 45.9]
As van der Wurff (2000:527) comments “… the pattern with an auxiliary (usually a
modal) and a negative object is predominant”. See Biberauer & Roberts (2006a) for a
speculation as to why object shift occurred in precisely these contexts at this period.
We can schematise the changes affecting object shift during ME as follows:
(23)
Pre-1400 ME:
a.
…O V…
b.
…V O …
(24)
Post-1400 ME:
a.
… Oneg V …
b.
…VO…
In (23), “O” is intended to designate objects in general, although this glosses over the
fact that definite and indefinite objects are in fact subject to distinct movement
triggers, with movement of definite objects being optional, i.e. the consequence of an
optional EPP-feature, as just pointed out, while indefinite objects moved obligatorily.
In the later grammar (24), however, only negatively-quantified objects could move.
The instance of restriction of function in question concerns the precise nature of the
category whose movement is triggered by the EPP-feature, as follows (where [+neg]
means a negative determiner, i.e. a negative quantifier):
(25)
EPPD → EPPD[+neg]
It is clear that the set of negative D-elements is a subset of the set of D-elements;
hence we have a paradigm case of restriction of function, as characterised at the
beginning of this section. In other words, we can take the set of direct objects to
12
correspond to E in the above discussion, and the set of negative direct objects to
correspond to the subset D  E. Now, the earlier grammar, that of (23), was one
where non-negative direct objects underwent optional movement. Thus we have an
optional rule applying to E – D; as pointed out above, a grammar of this type is a
superset both of the one requiring movement in the larger set of contexts E and of that
requiring it only in the smaller set of environments D. Again, optionality therefore
gives rise to the superset grammar. So we see that the grammar in (24) would, in the
absence of robust input data signalling the contrary, be favoured by the SP.
In the ME case, optional movement of definite objects for defocusing was
disfavoured since it was hard to detect in certain rather common types of
constructions. For example, in a V2 clause where the verb is in a simple tense (i.e. no
auxiliary is present, with the result that the verb is second), it is impossible to tell
what structural position the object occupies, since, moved or not, it is final in the
string. Thus, a sentence like (26a) could have the either the structure in (26b) or that
in (26c):7
(26)
a.
In þus many maners touches þe ymage of dremes men
in so many ways touches the image of dreams men
[CMROLLEP, 93.499; Trips 2002:251]
b.
[CP XP V+v+T+C [TP S (V+v+T) [vP O (V+v)
c.
[CP XP V+v+T+C [TP S (V+v+T) [vP (V+v)
]]]
]]]
Verb Projection Raising is another context where the surface order S-Aux-V-O is
structurally ambiguous, as (27) shows:
(27)
a.
þat 3e mahen ane pine me here
that you may alone torture me here
[St Juliana; Fischer et al. 2000:161]
b.
VR [TP PRO V+v+T [vP O (V+v) [VP (V) (O) ]]]
c.
VR [TP PRO V+v+T [vP (V+v) [VP (V) O ]]]
The ambiguity of cases like (26) and (27) meant that the trigger for the optional EPPfeature was often obscured. We take it that it was therefore insufficiently robust to
support the postulation of the superset grammar, since this is inherently disfavoured
by the SP.
4.2
Do-support
The modern do-support system arose in the 17th century from the 16th century system
in which do was a freely inserted auxiliary. Auxiliary do itself arose from the
reanalysis of an earlier verb which had an ECM, causative and raising interpretation
with an infinitival complement (see Denison 1985, 1993, Roberts 1993); quite
possibly this reanalysis is exactly the one which affected English modals in the first
7
Here XP could also be the subject; this does not affect our main point regarding the position of
the object.
13
half of the 16th century (Lightfoot 1979, Roberts 1985, 1993, Warner 1983, 1993,
1997). In the second half of the 16th century, do was an auxiliary with a distribution
somewhat different from that of Present-day English, in that it was able to appear in
positive declarative clauses without needing to bear emphatic stress. This use of do
was referred to by Jespersen (1909-1949) as “exuberant do”. The following quotation
from Palsgrave’s (1530) manual of French grammar for English speakers illustrates
this aspect of the distribution of do at around this time:
I do is a verbe muche comenly used in our tonge to be before other verbes, as, it is all
one to say [Visser’s emphasis] I do speake and such like, and I speake …
[Palsgrave (1530) Esclaircissement de la langue françoyse 84, 380, 523; in
Visser (1963-73, §1419), Roberts 1993:294].
Also, examples of the following kind show that do in positive declarative clauses did
not have to be stressed, as here the line of verse does not scan correctly if do is
stressed, as the accents are intended to indicate:
(28)
Rough wínds do sháke the dárling búds of Máy
[Shakespeare, Sonnet 18]
It is quite well-established that 16th-century English had general movement of the verbal
complex v-V to T in finite clauses (cf. Roberts 1985, 1993, 1999; Pollock 1989), and that
this was lost around 1600, perhaps slightly earlier or later (see Warner 1997 on the
chronology of this change). On this basis, although it has been discussed far less, it is
usually assumed that ME had verb-movement to T. But this assumption is doubtful. OE
was a V2-OV language like Modern German and Dutch,8 and as such it arguably lacked
V-to-T, since, as shown by Vikner (2005), the latter languages do not have V-to-T
movement (cf. the analysis of OE word order in B&R). If this is the correct analysis for
OE and the OV stage(s) of ME (see §4.1), the question of when verb-movement to T was
introduced into English arises.
Following Biberauer & Roberts (2006), we propose that V-to-T was introduced in
English in the 15th century, as a consequence of the reanalysis of subject-initial V2 orders
along the lines shown in:
(29)
[CP DP [C V ] [TP tDP [T tV ] [vP tDP [v tV ] VP ]]]
→ [TP DP [T V ] [vP tDP [v tV ] VP ]]
(See also Adams 1987 and Roberts 1993 on Old French, and Willis 1998 on Middle
Welsh). This reanalysis gives rise to the system that is attested for Late ME and 16 thcentury NE, where V undergoes “French-style” raising to T, and possibly beyond (see
Roberts 1985, Pollock 1989):
(30)
a.
b.
My wyfe rose nott [Roberts 1985: 23]
Se ye not how his herte is endured .. ?
8
Both of these assertions need to be somewhat qualified. OE admitted a well-known class of
exceptions to V2 involving preverbal pronouns (van Kemenade 1987, Roberts 1996, Koopman 1997,
Haeberli 1999/2002, Fischer et al 2000, Kroch & Taylor 2000). Also, OE was considerably more
liberal than Modern Dutch or German regarding verb-final orders in non-V2 environments (see
Stockwell 1977, van Kemenade 1987, Lightfoot 1991, Pintzuk 1991/1999, Roberts 1997, Fischer et al
2000 and B&R.)
14
[Anon. The Examination of Master William Thorpe, 44; Roberts 1993:239]
c.
Looks it not like the king?
[Hamlet I, I, 43; Roberts 1993:293]
d.
And gif he be noght so, then …
and if he be not so, then …
[1420s: James I, Kingis Quair, 62; Gray, 1985:73; Roberts, 1993:323]
(Note that (30d) shows this operation cannot be analysed as CP-recursion, assuming
standardly that if-clauses resist V2).
We follow the analysis of cross-linguistic variation in verb-movement to T put
forward by Biberauer & Roberts (2006c). They propose as a universal property of simple
finite clauses that T and V enter into an Agree relation. More precisely, T has an
unvalued V-feature since is inherently verbal but lacks the fundamental semantic
property of verbs, namely argument structure. Since finite verbs may bear tense
morphology but have no temporal content of their own, V has an unvalued Tense feature.
On the other hand, (main) verbs have argument structure, which Biberauer & Roberts
take to be intrinsically connected to having a valued V-feature (see Baker 2003).
These universal properties of T and V suffice to cause T to function as a Probe and V as
a Goal in a simple clause. How the Probe-Goal relation results in feature-valuing varies
parametrically. In Present-day English (since roughly 1600), Agree licenses V’s tense
morphology; this is how Biberauer & Roberts conceive the traditional “Affix Hopping”
relation of Chomsky (1957) and much subsequent work. Crucially, Biberauer & Roberts
assume that in Continental West Germanic V only moves in verb-second environments,
and therefore only to C and never just to T. Therefore, in non-V2 environments in these
languages, T and V Agree with no V-movement taking place, as mentioned above.9 In
Romance languages (both in French and the null-subject languages Italian, Spanish, etc.),
this Agree relation is associated with an EPP-like feature on T which triggers Vmovement.
Biberauer & Roberts propose that this difference between Germanic and
Romance is correlated with the richness of the inflectional (i.e. synthetic) marking of
tense distinctions. The Germanic and Romance languages differ noticeably in the
number of synthetic tense paradigms that are typically found. The Romance languages
are considerably richer, in a clear intuitive sense, than the Germanic ones in this respect,
as the following contrasts show:
(31)
Romance:
French: parle (present indicative/subjunctive), parlerai (future), parlerais
(conditional), parlais (imperfect), [parlai (preterit), parlasse (past subjunctive)];10
Italian: parlo (present), parlerò (future), parlerei (conditional), parlavo
(imperfect), parli (present subjunctive), parlassi (past subjunctive)
(32)
Germanic:
German: spreche (present indicative/subjunctive), sprach (past), spräche (past
subjunctive)
English: speak (present), spoke (past)
9
Hence Biberauer & Roberts disagree with the view that in subject-initial V2 clauses V moves
only to T (cf. Travis 1984, Zwart 1997). See Schwartz and Vikner (1996) for arguments for why V
must leave TP in V2 clauses.
10
The last two forms are in parentheses since they are not part of spoken French, existing only in
the literary language and mainly learnt in school.
15
Biberauer & Roberts thus postulate that there are two quite distinct types of
“richness” of verbal inflection agreement inflection and tense inflection. “Rich”
agreement inflection has many of the properties which are standardly attributed to it: it
triggers movement of a D-bearing category (perhaps V in a null-subject language, but
crucially not in a non-null-subject language); and it licenses null subjects. “Poor”
agreement inflection, on the other hand, determines the presence of overt expletives,
perhaps for the reasons given in Richards & Biberauer (2005). On the other hand, “rich”
tense inflection triggers V-movement and is irrelevant to subject-licensing.
Biberauer & Roberts’s proposals entail a new typology of the relation between
tense and agreement marking and various syntactic properties. The typology is given in
(33):
(33)
a.
b.
c.
d.
Rich agreement and rich tense inflection: hence V-to-T and null subjects,
e.g. Italian, Greek, Spanish, etc.
Poor agreement but rich tense: hence V-to-T, but no null subjects, e.g.
French (and see below on ME).
Poor tense and poor agreement: hence no V-to-T and no null subjects, e.g.
Modern English, Mainland Scandinavian.
Rich agreement and poor tense: null subjects, but no V-to-T; no clear
example.11
With this background, we can look again at the status of verb-movement to T in the
history of English. Until the 15th century, when V2 eroded significantly (see Fischer et al
(2000:128-129) on this, English was essentially like West Germanic in having V2, but no
independent verb-movement to T (and, of course, no null subjects). With the loss of V2
and the reanalysis of subject-initial clauses shown in (29), English developed verbmovement to T. However, there is no sense in which (late) 15th-century English had or
developed a rich tense inflection; tense-marking at this period was more or less what it is
in Modern English. Given the analysis of the relation between “rich” tense-marking and
verb-movement to T in Biberauer & Roberts (where further technical details and
motivation are provided), a Romance-style V/T-Agree system could not be supported by
the feeble tense-inflection of English. B&R suggest that this contributed to the reanalysis
of modals and do as auxiliaries in early 16th century.
As already mentioned, it is well-known that “Pollockian” verb-movement to T
was lost around 1600 (ca. 1575 according to Kroch 1989, but see Warner 1997). In terms
of the clause structure in Chomsky (1995, 2001), this must be treated as the loss of V-tov-to-T movement. We propose that, after the loss of V-to-v-to-T movement, v-to-T
movement remained (i.e. T retained a V-oriented EPP-feature). In other words what was
lost was V-to-v movement. Two innovations in the English of this period, both of which
also hold of Present-day English, support this idea. The first is that (non-heavy) direct
11
Marc Richards (personal communication) points out that Icelandic may be an example of this
kind of language. Certainly its morphological properties seem to indicate that it fits into this category:
it has often been pointed out that Icelandic has an agreement paradigm which is as “rich” as that of
some null-subject languages, e.g. Romanian, and it only has four synthetic tenses, like German.
Icelandic is not, however, a null-subject language and has been argued to have V-to-T movement (see
Vikner 2001, and the references given there). If, however, we assume that the V2 property of Icelandic
somehow renders null subjects impossible for independent reasons and that the verb-movement pattern
is the same as in other Germanic languages in not featuring movement to T but “symmetric” movement
to C, then it may fit into this fourth type, with the usual pattern of behaviour associated with this type
being obscured by the interaction of other parametric settings.
16
objects and finite main verbs are required to be rigidly adjacent from this period on. This
observation has also been made by Zwart (2005) (who interprets it rather differently).
Orders of the following kind, where material intervenes between the finite main verb and
the direct object, are not found after approximately 1600 (intervening adverb in bold):
(34)
Here men vndurstonden ofte by this nyght the nyght of synne
[Wycliffite Sermons I,477.605: Han & Kroch 2000:2]
The second observation has to do with inflectional morphology. The loss of almost all
agreement marking (except 2sg –st and 3sg present –s/-th) by the early 16th century and
the loss of infinitival –en, combined with the loss of subjunctive inflection and the
general absence of a dedicated present-tense marker had the consequence that verbs
frequently appeared as bare roots. From the 16th-century onwards, verbs in the present
tense, both indicative and subjunctive, the infinitive and the imperative all appear as bare
verbal roots. This contrasts markedly with the situation in the Romance languages and in
most other Germanic languages (in fact, Afrikaans is the only other Germanic language
which systematically allows bare roots to appear; although the Mainland Scandinavian
languages lack agreement marking, they have both an infinitive ending and a presenttense marker). If we think of V-to-v movement as creating a verbal category from an
acategorial root, then it is natural to think that a reflex of this would be the marking of
verbs as such with specifically verbal inflection; the “theme vowels” of the Romance
conjugation classes are arguably realisations of features of v (see for example Kayne
1994:42-46). Conversely, the fact that words which are interpreted as verbs regularly
appear as roots can be interpreted as indicating the lack of v-to-V movement. In this
connection, a further innovation of 16th-century English is worth mentioning here: the
development of zero-conversion of verbs to nouns and nouns to verbs as the productive
way in which to derive new elements in each category (see for example Görlach
1991:178). The development of zero-conversion as a productive process is a natural
consequence of the absence of verb-marking inflectional morphology, which in turn, we
suggest, is linked to the loss of V-to-v movement at this period.
Concerning auxiliaries, we propose, still following Biberauer & Roberts, that at
this period auxiliaries were merged in v and raised to T. Moreover, this operation
consistently applied in all finite contexts, and so, where there was no overtly realised
auxiliary, “null” v (i.e. minimally specified v which was not phonologically realised)
raised to T. This accounts for a further property of ENE that is distinct from Present-day
English: the fact that do was not required in negative clauses, even in the absence of
main-verb movement. We thus find examples with a superficially “Scandinavian”
negation pattern such as those in (35):
(35)
a.
Or if there were, it not belongs to you.
[1600: Shakespeare 2 Henry IV, IV, i, 98; Battistella & Lobeck 1988:33]
b.
Safe on this ground we not fear today to tempt your laughter by our
rustic play.
[1637: Jonson Sad Shepherd, Prologue 37; Kroch 1989:33]
The difference between examples like these and comparable examples with “do-support”
is simply that in the latter type, do (or rather, the features ultimately lexicalised as do –
see Note 12) is merged in v and raised to T over not; this operation was not yet
obligatory in the English of this period, with the only obligatory operation at the time
17
being the raising of the feature-bundle located in v to T, regardless of whether this would
ultimately be spelled out or not.12
A crucial step in the development of the modern system of do-support, where do
is of course obligatory in negative clauses like (35), from the earlier optional system had
to do with the development of negative auxiliaries. In Modern English, negative
auxiliaries such as won’t, can’t, shan’t, and, crucially, don’t are lexically derived by
suffixation of n’t to the auxiliary, rather than any syntactic operation of cliticisation or
the like. One reason to think this comes from the fact that n’t triggers synchronically
unpredictable stem allomorphy in the cases just mentioned. Zwicky & Pullum (1983)
show that inflections but not clitics trigger stem allomorphy, and therefore n’t must be
analysed as an inflection which attaches to auxiliaries. Another reason, due to Williams
(1994:68), is that the relative scope relations of modals and negation are unpredictable,
and must therefore be lexically specified, in the case of negative auxiliaries (compare for
example mustn’t and needn’t).
Presumably Aux + n’t originates from the simple phonological reduction of not.
At some point the sequence of Aux + n’t must have been reanalysed as a negative
auxiliary merged in T, if the analysis of these sequences as lexical items is correct for
Present-day English. Contraction of not to n’t was certainly found by 1600, as the
following observation of Jespersen’s shows:
The contracted forms seem to have come into use in speech, though not yet in writing, about
the year 1600. In a few instances (extremely few) they may be inferred from the metre in
Sh[akespeare], though the full form is written.
[Jespersen 1909-49, V:429, cited in Roberts 1993:305]
As soon as these negative auxiliaries are lexically formed, we have a class of Telements, i.e. auxiliaries which do not undergo raising from v to T. Let us suppose
that this reanalysis took place early in the 17th century: Roberts (1993:305) follows
Jespersen in assuming that negative auxiliaries were established by the 1660s at the
latest. A further relevant consideration is that there are no attested examples of a main
verb combining with n’t, implying that contracted negation, and thus lexical
combinations of Aux + n’t, are only found after the complete loss of main-verb
movement, i.e. after 1600 (this observation is due to Plank 1984).
A consequence of the development of negative auxiliaries (i.e. elements firstmerged in T) is that, from the first half of the 17th century, T was readily able to be
lexicalised independently of the presence of an EPP-feature. It is also possible that
some modals which were formerly merged in v were reanalysed as T-elements at this
stage, thus removing further cases of v-to-T movement. So these changes meant that T
no longer had to have an obligatory EPP-feature in order to be lexicalised; instead, T’s
EPP-feature could become optional, with the result that attraction of the remaining vThe question that obviously arises here is what features distinguish do from “null” v at this
stage. In Present-day English, do differs from the most minimally specified v by bearing a polarityrelated feature (cf. its obligatory occurrence in strongly affirmative contexts that would otherwise lack
an overtly realized auxiliary). At the period in question, however, it is hard to be sure what the situation
was. It has been suggested that do had a perfective interpretation; extrapolating from the facts of
contemporary South-Western dialects of British English which allow positive declarative do with a
habitual interpretation (Chalcraft 2006, Trudgill & Chambers 1991), we might ascribe an imperfective
feature of some kind to ENE do. Given that auxiliary do arose from the reanalysis of an ECM,
causative and raising verb, as mentioned above, there may have been various homophonous auxiliaries
do in ENE. The important thing is that these elements all had some extra feature in addition to those
borne by “null” v.
12
18
elements began to have a discourse effect (see again Chomsky’s 2001 proposal that
optional EPP-features must have an effect on output). Under these conditions, moved
v clearly has to be overt. Minimally specified v, which lacks phonological realisation,
can therefore no longer move to T, with the result that it consistently remains in situ
and do becomes the minimally specified T-element. In contrast to the earlier system,
then, v-to-T movement in NE always and only applies when v is realised as an overt
auxiliary. The discourse effects associated with merger of do in v include
interrogativity and emphasis, as well as the licensing of VP-fronting and VP ellipsis,
both of which are also associated with emphatic readings (given our proposals, it is
worth noting that the VP-constructions just mentioned are constructions which only
emerged from ca. 1600 on; cf. Plank 1984, Roberts 1993):
(36)
a.
b.
c.
d.
Did he smoke?
John DOES (so/too) smoke
He threatened to bring some sandwiches and [bring some sandwiches] he
did -I hoped he would bring some sandwiches and he did [bring some
sandwiches]
It is easy to see that the replacement of general v-to-T movement by v-to-T only under
specific conditions represents another case of restriction of function as defined at the
beginning of this section. The SP is relevant here since the earlier grammar permitted
minimally specified v to be realised as either do or zero whereas the later grammar
requires it to have a null realisation when it does not move and to be realised as do
when it does. The earlier grammar, then, allows a superset of the strings allowed by
the later grammar because it allows strings of the type [v (do)+T] X vP where X is an
adverb, floated quantifier or negation and VP contains the main verb and relevant
complements. The later grammar requires the realisation of do where T is associated
with a discourse effect or is negative and disallows it otherwise; in other words, where
minimally specified v moves to T in Present-day English, this is always realised as do
+ T, i.e. the resulting string is necessarily [v do+T] X vP. So we see that do is never
optional.
This instance of the SP does not involve optional movement, as we have seen,
but instead involves a change in the optional realisation of the minimally specified v
to obligatory realisation as do whenever movement takes place. So we see how the SP
can play a role in eliminating optional realisation of formatives just as it can play a
role in eliminating optional movements. In other words, it influences the diachronic
incidence of both movement and realisation by influencing how acquirers set the
parameters relating to these options as they apply to functional heads.
Conclusion
This paper has tried to show that, contrary to what is often thought given the
intersective nature of many parameters, the SP does appear to be relevant as a causal
factor in syntactic change. We have seen two general types of effect of the SP
resulting in the elimination of formal options: in one case, this directly affected
movement; in the other case, realisation of a formative. It is entirely possible that the
SP has other effects in diachrony. For example, one issue that we have touched on
here without developing is the characterisation of the “robustness” of the trigger
experience necessary both to preserve and to cause a system to move to a superset
19
grammar. Notice that the issue of preservation of superset systems relates to the
Inertia Principle of Keenan (2002) and Longobardi (2001). In fact, there may be a
tension between the simplest understanding of the notion of inertia and the SP; until
the question of robustness is fully clarified, we cannot really tell. The question of how
systems can move from a relatively unmarked to a relatively marked state (i.e. from a
subset to a superset grammar in the present case) is much more difficult; it is clear,
though, that this must be driven above all by the nature of the primary linguistic data.
The conclusion that the SP is after all relevant to syntactic change is positive, since, as
we saw in the introduction, the SP is based on a recognised fact about language
acquisition: the fact that language acquirers do not have access to negative evidence.
More generally, in showing how the SP plays a causal role in syntactic change, we see
one small way in which the study of syntactic change and language acquisition may
begin to converge.
References
Adams, M. (1987). From Old French to theory of pro-drop. Natural Language and
Linguistic Theory 5: 1 – 32.
Alexiadou, A. & E. Anagnostopoulou (1998). Parametrizing AGR: word order, Vmovement and EPP checking. Natural Language and Linguistic Theory 16: 491 –
539.
Allen, C. (1995). Case Marking and Reanalysis. Grammatical Relations from Old to
Early Modern English. Oxford: Oxford University Press.
Baker, M. (2003). Lexical Categories. Verbs, Nouns, and Adjectives. Cambridge:
CUP.
Baker, M. & K. Hale (1990). Relativized Minimality and Pronoun Incorporation.
Linguistic Inquiry 21: 289 – 297.
Battistella, E. & A. Lobeck (1988). An ECP account of verb second in Old English.
Proceedings of the Conference on the Theory and Practice of Historical
Linguistics.
Berwick, R. (1985). The Acquisition of Syntactic Knowledge. Cambridge, Mass.: MIT
Press.
Biberauer, T. (2003). Verb Second (V2) in Afrikaans: a Minimalist investigation of
word-order variation. Ph.D. dissertation: Cambridge University.
Biberauer, T. (2006). Changes at the edge: tracking the rise of a canonical subject
position in Germanic. Paper presented at Edges in Syntax (Nicosia).
Biberauer, T. & M. Richards (2006). True optionality: when the grammar doesn’t
mind. In: Boeckx, C. (ed.). Minimalist Essays. Amsterdam: Benjamins, 35 – 67.
Biberauer, T. & I. Roberts (2005). Changing EPP-parameters in the history of
English: accounting for variation and change. English Language and Linguistics
9: 5 – 46.
Biberauer, T. & I. Roberts (2006a). The loss of “head-final” orders and remnant
fronting in Late Middle English: causes and consequences. In: Hartmann, J. & L.
Molnárfi (eds.). Comparative Studies in Germanic Syntax: From From
A(frikaans) to Z(ürich German). Amsterdam: John Benjamins, 263 – 297.
Biberauer, T. & I. Roberts (2006b). Cascading parameter changes: internally driven
change in Middle and Early Modern English. To appear in: Eythórsson, T. & J.T.
Faarlund (eds.). Grammatical Change and Linguistic Theory: the Rosendal
Papers. Amsterdam: John Benjamins.
20
Biberauer, T. & I. Roberts (2006c). Subjects, Tense and Verb-movement in Germanic
and Romance. Paper presented at GLOW in Asia V (Delhi).
Breivik, E. (1990). Existential ‘there’: a synchronic and diachronic study. Oslo:
Novus Press.
Chalcraft, F. (2006). Forthcoming Cambridge Ph.D. dissertation.
Chomsky, N. (1957). Syntactic Structures. The Hague: Mouton.
Chomsky, N. (1995). The Minimalist Program. Cambridge, Mass.: MIT Press.
Chomsky, N. (2000). Minimalist Inquiries: The Framework. In: Martin, R., D.
Michaels & J. Uriagereka (eds.). Step by Step: Essays on Minimalist Syntax in
Honor of Howard Lasnik. Cambridge, Mass.: MIT Press, 89 – 156.
Chomsky, N. (2001). Derivation by phase. In: Kenstowicz, M. (ed.). Ken Hale: a life
in language. Cambridge, Mass.: MIT Press, 1 – 52.
Chomsky, N. (2005). On phases. To appear in: Otero, C. (ed.). Foundational Issues in
Linguistic Theory. Cambridge, Mass.: MIT Press.
Clark, R. & I. Roberts (1993). A Computational Approach to Language Learnability
and Language Change. Linguistic Inquiry 24: 299 – 345.
Denison, D. (1985). The origins of periphrastic do: Ellegård and Visser reconsidered.
In: Eaton, R. et al. (eds.). Papers from the 4th International Conference on
Historical Linguistics. Amsterdam: John Benjamins, 45 – 60.
Denison, D. (1993) English Historical Syntax. London: Longman.
Elenbaas, M. (2007). The synchronic and diachronic syntax of the English verbparticle combination. LOT Dissertations in Linguistics.
Fischer, O. (1992). Syntax. In: Blake, N. (ed.). The Cambridge History of the English
Language, vol 2. Cambridge: CUP, 207-408.
Fischer, O., B. Los, A. van Kemenade & W. van der Wurff (2000). The Syntax of
Early English. Cambridge: CUP.
Fodor, J. & W. Sakas (2005). The Subset Principle in syntax: costs of compliance.
Journal of Linguistics 41(3): 513 – 570.
Foster, T. & W. van der Wurff (1997). From Syntax to Discourse: the Function of
Object-Verb Order in Late Middle English. Fisiak, J. (ed.). Studies in Middle
English Linguistics. Berlin: Mouton, 135 – 156.
Görlach, M. (1991). Introduction to Early Modern English. Cambridge: CUP.
Guilfoyle, E., H. Hung & L. Travis (1992). Spec of IP and Spec of VP: Two Subjects
in Austronesian Languages. Natural Language and Linguistic Theory 10: 375 –
414.
Haeberli, E. (1999). Features, categories and the syntax of A-positions. Synchronic
and diachronic variation in the Germanic languages. Ph.D. dissertation:
University of Geneva. . Published as: Haeberli, E. (2002). Features, categories
and the syntax of A-positions. Cross-linguistic variation in the Germanic
languages. Dordrecht: Kluwer.
Hale, M. & C. Reiss (2003). The Subset Principle in phonology: why the tabula can’t
be rasa. Journal of Linguistics 39(2): 219 – 244.
Han, C-H. & A. Kroch (2000). The rise of do-support in English: implications for
clause structure. Proceedings of the 30th Annual Meeting of the North East
Linguistics Society (NELS 30). Amherst, MA:GLSA, 311 – 325.
Ingham, R. (2002). Negated subject and object positions in 15th century non-literary
English. Language Variation and Change 14: 291 – 322.
Ingham, R. (2003). The changing status of Middle English OV order: evidence from
two genres. Unpublished ms.: University of Reading.
21
Jespersen, O. (1909-1949). A Modern English Grammar on Historical Principles IVII. London/Copenhagen: Allen and Unwin.
Jouitteau, M. (2005). Nominal properties of vPs in Breton. A hypothesis for the
typology of VSO languages. In: Carnie, A., H. Harley & S. Dooley (eds.). Verb
First. On the syntax of verb-initial languages. Amsterdam: Benjamins, 265 – 280.
Kayne, R. (1994). The antisymmetry of syntax. Cambridge, Mass.: MIT Press.
Keenan, E. (2002). Explaining the Creation of Reflexive Pronouns in English. In:
Minkova, D. & R. Stockwell (eds.). Studies in the History of English: A
Millennial Perspective. Berlin: Mouton de Gruyter, 325 – 355.
van Kemenade, A. (1987). Syntactic Case and Morphological Case in the History of
English. Dordrecht: Foris.
Koopman, W. (1994). The order of Accusative and Dative Objects in Old English.
Unpublished ms.: University of Amsterdam.
Koopman, W. (1997). Another look at clitics in Old English. Transactions of the
Philological Society 95: 73 – 93.
Koopman, W. & W. van der Wurff (2000). Two word order patterns in the history of
English. In: Sornicola, R., E. Poppe & A. Shisha-Halevy (eds.). Stability,
variation and change of word-order patterns over time. Amsterdam: John
Benjamins, 259 – 283.
Kroch, A. 1989. Reflexes of grammar in patterns of language change. Language
Variation and Change 1: 199 – 244.
Kroch, A. & A. Taylor (2000). Verb-object order in Middle English. In: Pintzuk, S.,
G. Tsoulas & A. Warner (eds.). Diachronic syntax: models and mechanisms.
Oxford: OUP, 132 – 163.
LaFond, L. (2001). The pro-drop parameter in second language acquisition revisited:
a developmental account. Ph.D. dissertation: USC.
Lightfoot, D. (1979). Principles of Diachronic Syntax. Cambridge: CUP.
Lightfoot, D. (1991). How to set parameters: arguments from language change.
Cambridge, Mass.: MIT Press.
Longobardi, G. (2001). Formal Syntax, diachronic minimalism, and etymology: the
history of French chez. Linguistic Inquiry 32: 275 – 302.
MacLaughlin, D. (1995). Language acquisition and the Subset Principle. Linguistic
Review 12(2): 143 – 191.
Manzini, R. & K. Wexler (1987). Parameters, Binding Theory, and Learnability.
Linguistic Inquiry 18: 413 – 444.
Moerenhout, M. & W. van der Wurff (2000). Remnants of the old order: OV in the
Paston letters. English Studies 6: 513 – 530.
Pintzuk, S. (1991). Phrase structures in competition. Variation and change in Old
English word order. New York: Garland.
Pintzuk, S. (2002). Verb-Object order in Old English: variation as grammatical
competition. In: Lightfoot, D. (ed.). Syntactic effects of morphological change.
Oxford: OUP, 276 – 299.
Pintzuk, S. & A. Taylor (2004). Objects in Old English: why and how Early English
is not Icelandic. York Papers in Linguistics 2(1): 137 – 150.
Plank, F. 1984. The modals story retold. Studies in Language 8, 305-364.
Pollock, J-Y. (1989). Verb movement, Universal Grammar and the structure of IP.
Linguistic Inquiry 20: 365 – 424.
Prince, A. & Tesar, B. (2004). Learning phonotactic distributions. In: Kager, R., J.
Pater & W. Zonneveld (eds.). Constraints on Phonological Acquisition.
22
Cambridge: CUP. Also available from http://roa.rutgers.edu/files/353-1099/3531099-PRINCE-0-0.PDF
Richards, M. & T. Biberauer (2005). Explaining Expl. In: den Dikken, M. & C.
Tortora (eds.). The function of function words and functional categories.
Amsterdam: Benjamins, 115 – 154.
Rissanen, M. (1994). The position of not in Early Modern English questions. In:
Kastovsky, D. (ed.). Studies in Early Modern English. Berlin: Mouton de Gruyter,
339 – 348.
Rizzi, L. (1982). Issues in Italian Syntax. Dordrecht: Foris.
Roberts, I. (1985). Agreement parameters and the development of English modal
auxiliaries. Natural Language and Linguistic Theory 3:21 – 58.
Roberts, I. (1993). Verbs and diachronic syntax. Dordrecht: Kluwer.
Roberts, I. (1996). Remarks on the Old English C-system and the Diachrony of V2.
Linguistische Berichte: 154 – 167.
Roberts, I. (1999). Verb movement and markedness. In: deGraff, M. (ed.). Language
Creation and Change. Cambridge, Mass.: MIT Press, 287 – 328.
Roberts, I. & A. Roussou (2003). Syntactic Change. A Minimalist Approach to
Grammaticalization. Cambridge: CUP.
Schwartz, B. & S. Vikner (1996). The verb always leaves IP in V2 clauses. In:
Belletti, A. & L. Rizzi (eds.). Parameters and Functional Heads. New York: OUP,
11 – 62.
Stockwell, R. (1977). Motivations for exbraciation in Old English. In: Li, C. (ed.).
Mechanisms of Syntactic Change. Austin: University of Texas Press, 291 – 314.
Traugott, E. (1972). A History of English Syntax. New York: Holt, Rhinehart &
Winston.
Travis, L. (1984). Parameters and Effects of Word Order Variation. Ph.D.
Dissertation: MIT.
Travis, L. (2006). VP, D0 movement languages. In: Zanuttini, R., H. Campos, E.
Herburger & P. Portner (eds.). Negation, Tense and Clausal Architecture: crosslinguistic investigations. Washington, D.C.: Georgetown University Press, 127 –
147.
Trips, C. (2002). From OV to VO in Early Middle English. Amsterdam: John
Benjamins.
Trudgill, P. & Chambers, J. (1991). Dialects of English: Studies in grammatical
variation. London: Longman.
Vikner, S. (2001). Verb Movement Variation in Germanic and Optimality Theory.
Habilitationschrift: University of Tübingen.
Vikner, Sten. (2005). Immobile Complex Verbs in Germanic. Journal of Comparative
Germanic Linguistics 8 (1/2): 83 – 115.
Visser, F. (1963-73) An Historical Syntax of the English Language. Leiden: Brill.
Warner, A. (1983). Review of Lightfoot (1979). Journal of Linguistics 19: 187 – 209.
Warner, A. (1993). English Auxiliaries: Structure and History. Cambridge: CUP.
Warner, A. (1997). The structure of Parametric Change and V-movement in the
history of English. In: van Kemenade, A. & N. Vincent (eds.). Parameters of
morphosyntactic change. Cambridge: CUP, 380 – 393.
Williams, E. (1994). Thematic Structure in Syntax, Cambridge, Mass.: MIT Press.
Williams, A. (2000). Null subjects in Middle English existentials. In: Pintzuk, S., G.
Tsoulas & A. Warner (eds.). Diachronic Syntax. Models and Mechanisms.
Oxford: OUP, 164 – 187.
23
Willis, D. (1998). Syntactic Change in Welsh. A study of the loss of Verb-Second.
Oxford: OUP.
van der Wurff, W. (1997). Deriving object-verb order in late Middle English. Journal
of Linguistics 33: 485 – 509.
van der Wurff, W. (1999). Objects and verbs in Modern Icelandic and fifteenthcentury English: a word order parallel and its causes. Lingua 109: 237 – 265.
Zwart, J-W. (1997). The Morphosyntax of Verb Movement: A Minimalist Approach to
the Syntax of Dutch. Dordrecht: Kluwer.
Zwart, J-W. (2005). A comparative approach to syntactic change in the history of
English. English Language and Linguistics 9: 157 – 179.
Zwicky, A. & G. Pullum (1983). Cliticisation vs. inflection: English n’t. Language
59: 502 – 513.
24