Cambridge University Press is collaborating with JSTOR to digitize, preserve and extend access to Journal of
Linguistics.
http://www.jstor.org
J. Linguistics 23 (I987),
109-I32.
Zwicky on heads'
RICHARD A. HUDSON
Department of Phonetics and Linguistics, University College London
INTRODUCTION
Hudson, I984; Nichols, I986). All these treatments agree not only in using
the term 'head', but also in using it to refer to the element in some
constructionto which all the other parts of that constructionare (in some
sense)subordinate.
It is reasonableto be suspiciousof notionswith as chequereda historyas
'head' on the groundsthat it developedin the days beforeformaltheories
of syntacticstructurewereavailable,as a rathermetaphoricalandvagueway
of referringto notionswhichwe maynow be ableto definemoreeconomically
and insightfullyin termsof other, more primitivenotions. It is sometimes
temptingin such cases to reintroducethe traditionaltermjust becauseit is
traditional.Thereis then a dangerof multiplyingentitiesunnecessarily- as
if one were to try to find a place for notions like 'guttural' or 'lilting' in
modernphonetics.Worsestill, a traditionaltermmay be used to referto a
multiplicityof notions whichare more or less closely relatedto the (vague)
traditionalconcept,but whichare in fact independentof each other.Arnold
[i]
Arnold Zwicky gave me most valuable comments on an earlier version of this paper, for
which I should like to thank him. I am also grateful to Nigel Vincent for his help with the
present version.
IO9
RICHARD
A. HUDSON
ANALYSIS
ZWICKY
ON HEADS
C. The MORPHOSYNTACTICLOCUS - as explainedearlier,this is the constituent on which any inflexionswhich are relevantto the motherare located
(or on which they would have been located had the languageincludedany
such inflexions).
D. The SUBCATEGORIZAND- the constituentwhich is subcategorizedwith
respectto its sisters,in the familiarsense.
E. The GOVERNOR- the constituentwhichdeterminesthe morphosyntactic
form of some sister.
F. The DISTRIBUTIONALLYEQUIVALENTconstituent- the constituentwhose
distributionis similarto that of the mother.
G. The OBLIGATORYconstituent- the one which has to be presentif the
motheris to be categorizedas it is.
analysis,the
H. The RULER of dependencytheory- in a dependency-based
'ruler' (Zwicky'stermfor what is often called 'head' in actualdependency
analyses)is the word on whichother wordsdepend.
Zwickydiscussesthesecategoriesin relationto six Englishconstructions:
e.g. control those penguins
(i) V + NP
towardthose penguins
(ii) P + NP
we control those penguins
(iii) NP + VP
those penguins
(iv) Det + N
must control those penguins
(v) Aux+ VP
that we control those penguins
(vi) Comp+ S
Given the analysesthat Zwicky assumesfor these constructions,his eight
categoriesapply to them as shownin Table I. The rows show the head-like
notions listed above (A to H), and the columns show the six English
constructions.Eachentryin the tableshowswhichof theconstituentsin these
constructionsis taken,in Zwicky'sanalysis,to be the head-likeconstituent,
but theseentriesare givenin termsof the correspondingentriesfor an extra
row whichI havesuppliedat the top of the table,for 'semanticfunctor'.This
is of coursejust the converseof the row for 'semanticargument',as can be
seen from the row of stars in the latter, but it will be helpfulfor the later
discussionto take the semanticfunctorentriesas a standardagainstwhich
we can measureotherentries.
If all the entriesin Table I had been either'=' or blank then we could
have claimedthat all of the eight head-likecategorieswerein fact the same
category,presentedwith respectto differentproperties.A naturalname for
this super-categorywould have been 'head', and the value- indeed,
indispensability- of 'head' would have been established.The big attraction
of the notion 'head', for those of us who believein it, is preciselythat it
integratesa wide rangeof differentphenomenasuchas the eight 'head-like'
concepts.(It goes withoutsayingthat it is immaterialwhat we actuallycall
thissuper-concept;if one linguistcallsit 'head' andanothercallsit something
else the disputeis about nothingbut terminology.)But of course the main
III
RICHARD
A. HUDSON
thrust of Zwicky'spaper is that this is NOT how things are- the various
concepts in fact pick out different ranges of items across the various
constructions,as shown by the distributionof = and * in Table i.
Semantic functor...
(A)
(B)
(C)
(D)
(E)
(F)
(G)
(H)
Semantic argument
Determinant of concord
Morphosyntactic locus
Subcategorizand
Governor
Distributional equivalent
Obligatory
Ruler
V+NP
(V)
P+NP
(P)
NP+VP
(VP)
Det+N
(Det)
*
*
*
*
*
=
=
=
*
(*)
Aux+VPComp+S
(Aux) (Comp)
=
=
ZWICKY
ON HEADS
ARGUMENTS
control.This interpretationof 'head' applieseven more obviouslyto compoundslikejam sandwich,whose head is sandwichbecausethe wholedefines
a kind of sandwich,and not a kind of jam. Thus the head'ssistershave the
functionof modifyingthemeaningof thehead,in theeveryday,non-technical
meaningof 'modify'.
Having given this excellentintroduction,however,Zwicky immediately
suggests 'a sharpening(and extension)' of it in terms of the semantic
distinction between functors and arguments.The train of thought is as
follows:let us assumethat in Det + N the semanticheadis N because'those
penguinsdescribesa kind of penguin[sic]'. Now in standardfunctor/argument analysesit is the N which is taken as the argument,so let us assume
that the notion 'argument'is just a sharpened-upequivalentof the element
of which the whole defines 'a kind'. I shall suggest in Section 6 that the
standardanalysisof Det+N on which this conclusionrestsis faulty,but it
is surelyeasy to see that the analysisdoes not in fact generalizeto the other
constructions.
Takethe V + NP construction,to whichwe havealreadyreferred.According to the standardanalysesthe semanticargumentis clearlythe NP, but
we have seen that controlthosepenguinsrefersto a kind of controlling,and
not to a kind of penguins.Similarremarksapply to NP +VP examples:as
I suggestedabove, Wecontrolthosepenguinsrefersto a kind of controlling,
namelyone in whichwe and the penguinsare involved.Of course,thereis
a long traditionof takingNP + VP as a semanticallyexocentricconstruction,
in which neitherconstituenthas a uniquelyprivilegedstatus. This seems,
however,to confuse two questions.One is about the relationof the NP to
the VP - does the wholereferto an instanceof whateveris referredto by the
VP? Consider the well-establishedlogical tradition of representingthe
meaningof John loves Mary as 'Loves (John, Mary); here the respective
II3
RICHARD
A. HUDSON
ZWICKY
ON HEADS
RICHARD
A. HUDSON
ZWICKY
ON HEADS
Semantic functor...
(C) Morphosyntactic
locus
V+NP
(V)
P+NP
(P)
NP+VP
(VP)
(D) Subcategorizand
(E) Governor
(G) Obligatory
(H) Ruler
Det+N
(Det)
Aux+VPComp+S
(Aux) (Comp)
Table2
Zwicky's analysis of six constructioas in terms of the six genuinely head-like categories
AUXILIARIES
AND COMPLEMENTIZERS
RICHARD
A. HUDSON
I2)
ZWICKY
ON HEADS
in thisrespect.Takea complementizer
thatis a veryuntypicalcomplementizer
like since,whichis muchmore typical.Hereit is possibleto omit the clause,
by anaphoric ellipsis similar to the process responsiblefor VP ellipsis
(compare I haven'tseen him since we had that argumentand I haven'tseen him
RICHARD
A. HUDSON
Semanticfunctor...
(C) Morphosyntactic
locus
(D) Subcategorizand
(E) Governor
(F) Distributional equivalent
(G) Obligatory
(H) Ruler
V+NP
(V)
P+NP
(P)
=
=
Table3
Zwicky'sanalysisof six constructionsin termsof the six genuinelyhead-likecategories,
showingfirstrevisions
The conclusionso far reachedin this section,then,is that Aux and Comp
are the distributionalequivalents and the obligatory elements in their
in P + NP. Table3
respectiveconstructions,andthatP is the subcategorizand
is a revisedversionof Table 2, in whichthesechangesare included.
Two otherquestionsarisein connexionwithAux+ VP and Comp+ S. The
firstis about the morphosyntacticlocus in Comp+ S. Accordingto Zwicky
it is the S, and the reasonfor this choiceis that S is wherethe morphological
markersof tense are. I take it that he accordsthis criterialstatus to tense
becausehe believesthat tense should percolateup from S on to the node
dominatingComp+ S, so presumablyhe has reasonsfor believingthat this
node shouldcarrya tense feature- perhapssomethingto do with subcategorizationof matrixverbs. However,since he does not explainhis reasons
it is hardto commenton the choice. What is clear,however,is that at least
as good a case can be made for takingComp as morphosyntacticlocus, on
preciselythe samegroundsas he offersfor takingP as morphosyntacticlocus
in P+NP: that the selectionof particularprepositionsis analogousto case
features(a verytheory-particular
decisionassociatedwithGeneralisedPhrase
Structure Grammar (Gazdar et al., 1985: 23) and Lexical-Functional
Grammar(Kaplan& Bresnan,1982: 197)). Sincethereare verbsin English
which select particular prepositions (e.g. informof, tell to), we must show this
ZWICKY
ON HEADS
DETERMINERS
RICHARD
A. HUDSON
ZWICKY
ON HEADS
RICHARD
A. HUDSON
HARMONIOUS
ANALYSIS
ZWICKY
ON HEADS
Semantic functor...
(C)
(D)
(E)
(F)
(G)
(H)
Morphosyntactic locus
Subcategorizand
Governor
Distributional equivalent
Obligatory
Ruler
V+NP
(V)
P+NP
(P)
=
=
NP+VP
(VP)
Det+N
(Det)
Aux+VPComp+S
(Aux) (Comp)
=
-
=
-
Table 4
Final revised version of Zwicky's analysis of six constructions in terms of the six genuinely
head-like categories
RICHARD
A. HUDSON
ADJACENCY
ZWICKY
ON HEADS
PRINCIPLE
praise'.)
To see how the principleworks let us look at the dependencystructures,
using the Word Grammarnotation in which arrowspoint from the head
towardsits dependents:
I27
RICHARD
(a)
A. HUDSON
'
(b)
with greatdifficulty
greatwith difficulty
keeps
talking
withdifficultybecausetheinterveningwordwithis headratherthandependent
of difficulty.If thedirectionof dependency- i.e. thedifferencebetween'head'
and 'dependent'- had been irrelevant,therewould have been no difference
in statusbetweenthesetwo examples,becausein both cases the dependency
I28
ZWICKY
ON HEADS
relations are in other respects the same (each of great and with has a
dependencyrelationto difficulty).
As I promisedin the discussionof determiners,the AdjacencyPrinciple
throwsextra light on the relationbetweenDet and N in Det+N. Imagine
a sequenceof determiner+adjective+common noun, such as the big book.
Now if the common noun were the head of the determiner,as in the
traditionalanalysis,thedependencyrelationbetweenthemwouldbe precisely
the same as that betweenbig and book,wherebookis clearlythe head. But
if that were the case, there would be no explanationfor the fact that the
determineralways precedesthe adjective;why should it not be allowedto
follow, possiblysubjectto some pragmaticconstraintssuch as those on the
relativeorderof adjectives?E.g. why should*bigthebookbe so muchworse
thana phraselikeredbigbooks(comparedwithbigredbooks)?If on the other
hand we adopt the analysis which I assumed in the earlier discussion,
accordingto whichit is the determinerwhichis head of the commonnoun,
then the word-orderrestrictionsfollows absolutelyautomaticallyfrom the
AdjacencyPrinciple.Thatis, *bigthebookis bad for exactlythe samereason
as *great with difficulty.
The discussionin this section has shown the crucial importanceof the
notion'head' in dealingwiththe continuityof phrasesin a dependency-based
analysis.A good deal of researchis neededbeforewe can be sure that the
AdjacencyPrincipleis compatiblewith all we know about English,not to
mention other languages,and of course even more researchis needed to
evaluate the general principlesof dependencygrammar,but so long as
dependencytheoriesinvoke the AdjacencyPrinciple,they must also invoke
the notion 'head'.
9.
GRAMMATICAL
RELATIONS
head: V
P
VP
Det
Aux
Comp
129
LIN 23
RICHARD
A. HUDSON
Head
Dependent
Pre-dependent
Subject
Visitor
Post-dependent
Pre-adjunct
Object
Complement
Incomplement
Free
Post-adjunct
Oblique
Figure i
ZWICKY
ON HEADS
REFERENCES
Anderson, J. M. (197
I).
UniversityPress.
Anderson, J. M. (1976). On serialization in English syntax. Ludwigsburg: Ludwigsburg
University.
Anderson, J. M. (1977). On case grammar. Prolegomena to a theory of grammatical relations.
London:CroomHelm.
Anderson,J. M. & Durand,J. (1986).Dependencyphonology.In Durand,J. (ed.),Dependency
and non-linearphonology. London: Croom Helm. I-54.
Atkinson, M., Kilby, D. & Roca, I. (i982). Foundations of general linguistics. London: Allen
& Unwin.
Bloomfield, L. (1933). Language. London: Allen & Unwin.
Bresnan, J. (1982a). Control and complementation. In Bresnan, J. (ed.) i982b. 282-390.
Bresnan, J. (ed.) (i982 b). The mental representationof grammatical relations. Cambridge, MA:
MIT Press.
Chomsky,N. (1970). Remarkson nominalization.In Jacobs,R. A. & Rosenbaum,P. S. (eds),
Readings in English transformationalgrammar. London: Ginn. 184-221.
Durand, J. (ed.) (1986). Dependency and non-linearphonology. London: Croom Helm.
Oxford:Blackwell.
Gazdar,G. & Pullum,G. (198i). Subcategorization,
constituentorderand the notion 'head'.
In Moortgat,M., Hulst,H. v.d. & Hockstra,T. (eds), Thescopeof lexicalrules.Dordrecht:
Foris, I07-I23.
Hudson,R. A. (I984). Wordgrammar. Oxford:Blackwell.
Hudson,R. A. (I985a). Grammaticalrelations.Mimeo.
Hudson,R. A. (I985b). The limitsof subcategorisation.
Linguistic Analysis 15. 233-255.
Hudson,R. A. (I986). The Comp-traceand that-traceeffects.Mimeo.
Jackendoff,R. (1983). Semantics and cognition. Canbridge,MA: MIT Press.
Jacobs,R. A. & Rosenbaum,P. S. (eds).(I970). Readings in English transformationalgrammar.
London:Ginn.
Kaplan,R. & Bresnan,J. (I982). Lexical-functional
grammar:a formalsystemfor grammatical
representation. In Bresnan, J. (ed.), The mental representation of grammatical relations.
5-2
RICHARD
A. HUDSON
132