Vous êtes sur la page 1sur 8

The Nonexistence of the Trace-Binding Algorithm Author(s): Geoffrey K. Pullum Reviewed work(s): Source: Linguistic Inquiry, Vol.

10, No. 2 (Spring, 1979), pp. 356-362 Published by: The MIT Press Stable URL: http://www.jstor.org/stable/4178113 . Accessed: 19/06/2012 11:14
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to Linguistic Inquiry.

http://www.jstor.org

356

SQUIBS

AND

DISCUSSION

toral dissertation, University of California,San Diego, California. Sadock, J. M. (1972)"Speech Act Idioms," in P. M. Peranteau, J. N. Levi, and G. C. Phares, eds., Papers from the
Eighth Regional Meeting of the Chicago Linguistic Soci-

ety, University of Chicago, Chicago, Illinois.


Sadock, J. M. (1975) Toward a Linguistic Theory of Speech

Acts, Academic Press, New York.

THE NONEXISTENCE OF THE TRACE-BINDING ALGORITHM*

Geoffrey K. Pullum, University College London

Recent proposalsby Chomsky(1975; 1977a;1977b)and Fiengo (1977)underthe heading"trace theory" claimthat explanations for various facts in the syntax of English can be derived from in shallow strucgeneralconditionsimposedon the distribution ture' of phrase nodes and the "traces" they leave behindwhen they are moved by transformational rules. For a simple example, consider (1). (1) a. b. C. d. John was seen by [NPe] *Was seen by John. [NP e] was seen by [NP John] *Himselfwas seen by John.

From an initial phrase markerlike (la), the extremely general transformation of NP Movementcould form (lb) by postposing John. But (lb) is ungrammatical. The claimed explanationis that the derived structureof (lb) would be like (lc), in which some indexing procedure is assumed to "bind" the moved phrase John to the empty NP node that is its "trace". Then (lc) is bannedby the same principlethat bans (ld): the ""binding" of traces counts as intrasententialanaphoraof the same sort as is found with reflexive pronouns, and one of the things that is forbiddenin Englishreflexivizationis for the anaphoric element to precede and command the antecedent. (Moved phrases are analogousto antecedents, and traces to anaphoric elements.) Clearly, the algorithmby which one may determinefrom a shallow structure whether the traces in it are "properly bound" (Fiengo (1977, 45)) is the crucial component of this hypothesis, for the explanationfor facts like those above is
* I am gratefulto MichaelBrody, RobertFreidin,JaninaGiejgo, Alan Prince, and Edwin Williamsfor comments and discussion concerning an earlier version of this squib. They do not by any means necessarilyagree with what this version says. I Chomsky(1977b)uses this term for the level at which semantic interpretation is carriedout in his theory. It is much preferableto the earlier term "surface structure", which is better kept for the more superficiallevel at which surfacefilters apply(afterdeletionrules have applied).

SQUIBS

AND

DISCUSSION

totally dependent on it. Two questions need to be answered: (A) under what conditions we are to take the relationbetween a moved phrase and its trace to be that of "properbinding", and (B) more fundamentally,how we are to determinethat a given trace is the trace of a given moved phrase. Fiengo answers(A) by stating(p. 45) thata phraseproperly binds its trace if it precedes that trace, introducinga possible qualificationof this later (p. 53). AlthoughI believe it can be shown that there are real problems with answering this first question, I shall not discuss them here. I shall assume that at the very least we can take it as a sufficientconditionfor proper binding that the moved phrase precedes and commands the trace, and, in all my examples, this conditionwill be met where appropriate. WhatI shall do is to show that question(B), on which (A) depends, has not been answered in the publishedliterature.I shall furthershow that no answer seems to be availablefor it given current assumptions, for a general problem seems to arise, namelythat no simpleindexingproceduredeals correctly with multiple movements of a phrase if movement rules are stated in the maximallygeneral way Chomskyemploys. Chomsky(1977b,6-8) makes the operationof the indexing procedurehe assumes quite explicit. Let us call a phrase that moves a mover, and the empty node to which it is moved, the target. Chomsky's diagramsclearly show he is assuming that phrase nodes are all separately indexed with integers in the and that the principlegoverningindexing base phrase marker,2 after movementis (2). (2) A movercarriesits own index to the targetand leaves a copy of its own index behind. Fiengo (1977, 47, fn. 18) cites J. R. Ross as having observed that derivations like the one I give in (3) would not result in the blockingof starredsentences like (3d) *Whowas seen Bill by?, because under the indexing procedure(2), both the traces involved would be properlybound. (3) a. b.
c.

e]] [s[NP2 who] was seen [p3 Bill] by [COMP[NP1 e]] [COMP[NP, ell [s[NP2 e] was seen [Np3 Bill] by who]] [NP2 who]] [S[NP2 e] was seen [NP3 Bill] by [coMP[NP2
[NP4

e]] [NP2

d.

*Whowas seen Bill by?

2 Formally, the assumptionseems to be that a terminalnode in shallow (or initial) structureis either a lexical or a grammatical formative or the identityelemente, anda preterminal node is eithera lexical or a grammatical categorylabel or an orderedpair (N, i) where N is a phrase node and i is an integer.

358

SQUIBS

AND

DISCUSSION

The (optional)rule "Move NP" derives (3b) from (3a), and the rule "Move wh-phrase"then derives (3c), in which both traces are preceded and commandedby a node with the same index. It might be thoughtthat the problemhere is cruciallytied "Agent Postposto the acceptance of the rightward-movement ing" derivationfor passives, and.thatif such derivationswere outlawed, passive agent phrases being generatedin situ by the base rules, the argumentwould be defeated. This is not so. Considerthe following case: (4) a.
b.
C. [COMP[NP,

ell

[SiNP2

e] hit [NP3 who]]


who] hit
[NP3

[COMP[NP1 ell [s[NP3 [COMP[NP3

el]

d.

who]] [S[NP3 e] hit [NP3 ell *Whohit?

Here no agent phrase or passive constructionis involved, and all movement is to the left (though the output (4d) could also be derived via rightwardNP Movement from the structure COMP [s[p who] hit [NP el]). The object NP is moved leftward into the empty subjectNP node by the rule "Move NP", before being moved into COMPby Wh Movement. Anotherobjectionmightbe that in some way the semantic component would dispose of such cases. But this does not seem to be true. Chomsky's interpretiverule for interrogative constructions (1977, 84, rule 38) will apply to (4c) to replace both the traces by variables, yielding (4e). (4) e.
[COMP

for whichx, x a person], [s x hit x]

This is a coherent proposition in the quasi-logical language Chomskyrefers to as logicalform, as far as one can make out. It asks for the identity of the personx who hit herself/himself. And nothing that has been given in the literature will independently prevent this outcome. Chomsky and Lasnik (1978, con268, fn. 1) imply that variablesubstitutionin interrogative structionsapplies to a unique trace, and they assert that there is a way to find the righttrace: "there is a simple algorithmto
determine this from surface structure. . ." But they are wrong

about the algorithm;in a case like the present one, assuming the indexingprocedureunder discussion here, there is no way to find the right trace, as one discovers when one tries to formulate the algorithmthat they claim exists.3 And with no
3Vergnaud (1974, 162-163), discussing French examples, recognizes the problemof findingthe righttrace in a structurethat contains more than one. He gives a solution which simplypicks out the maximally superiortrace in the S as the one to be replacedby a variable. But this can hardlybe taken seriouslyif properbindingis to be defined at shallow structure.Considera case like (i):

(i)

I wonder [t[COMP[NP, who]] [s you saw [NpI e]]l]]

[s

they think [sL[coMP[P, ell

SQUIBS

AND

DISCUSSION

way to pick out a unique trace, Chomsky's rule (38), as it is stated, will apply to both in (4c). Fiengo is right, therefore, to recognize that Ross has correctly identifieda class of undesirablepredictionsof trace theory. He is wrong, however, to think that his interpretiverule (82) (Fiengo (1977, 57)) provides an answer to the problem, because although under his assumptions the rule in question to (3d), it would would have the effect of denyinginterpretation not be relevant to (4d), since it only applies if the copula is
present.4

Before continuing,it is necessary to consider the question of whether some other interpretiveor filtering device might correctly exclude (4d) and examples like it. Alan Prince has
The most superiortrace in this structureis the one between think and you. (I have assumedhere, with Chomskyand Vergnaud,and against my own beliefs, that WhMovementappliessuccessively, hoppingfrom COMP to COMP.) This is surely not the trace to be replaced by a variable.That would give the logical form (ii): (ii) I wonder [S[coMP for which x, x a person], [s they thinkx you saw]]

One can speculate about how the desired results are intended to be achieved, but no solution that definitely works has yet been given explicitly by workerswithin "trace theory", or is known to me. 4 I do not believe Fiengo's rule (82) can explain anything. It is stated, in full, as follows: Rewritethe structure .[+N] ..COPULA. . .([+V])([+N])... as ([+V])([+N])(.. .+N.*.) This will apparently performoperationslike the following(of which the first two are given explicitly by Fiengo): (i) (ii) (iii) (iv) (v) (vi) Judy is a good person -- good person (Judy) Antarctica -- uninhabited Antarctica (*)Manis uninhabited (man) (*)Whowas hit Bill by -o hit Bill (who by) That child is annoyingme -- annoyingme (that child) John is easy to please -* easy (Johnto please) Susan was given nothing -- given nothing(Susan)

Among the many questions that arise concerningwhat this reordering process is supposed to achieve, and what kind of "semanticinterpretation" is effected by it, the most relevantone is: by what independent constraintson the outputof Fiengo's ruleare the outputscorresponding to the two starredexamples predicted to be ill-formed?What distinguishes (ii) and (iii) from (i) and (vi), for example? Since Fiengo's conceptionof Englishsyntax characterizesthe lefthandversions of (ii) and (iii) as grammatical,it is only the claimed ill-formednessof the reordered("logical form"?) versions of them that characterizesthem (in accord with the facts) as bad. Fiengo merely says of (ii), "The
expression uninhabited Antarctica. . .is not a possible property, . . .at

least not a possible propertyof man" (1977, 56). Quite so. But in what way does his grammar for Englishpredictthis fact?

360

SQUIBS

AND

DISCUSSION

made the suggestion that structuresof the form (4d) might be blocked by the constrainton anaphorathat excludes coreference between who and he in (5): (5) Who did he hit t? Under the assumptionsof trace theory, this would be an interesting solution, because it would not be ad hoc, but would reinforce the parallelism between anaphor/antecedentand trace/moverrelations (cf. Chomsky (1975, 99-100)). But it is not correct. The constraintthat blocks (5) is that an anaphoric element cannot have as its antecedenta boundvariablelocated to its right. But this could not possibly be made to block (3) or (4), because it must not be allowed to block (6), which has the same generalform: (6) a. e] [COMP[NP, ell [sbNp2
[COMP[NP3 who]] [S[NP3

b.
C.

e]] [SNP3 who] [COMPrNP, Who was murdered?

was murdered[Np3 who]] was murdered[NP3 e]]


e] was murdered [Np3 e]]

d.

Since the shallow structure Who t was murdered t? must be both allowable and interpretable,it would appearthat *Whot was seen Bill by t? and *Whot hit t? will similarlybe treated as well-formedby a grammar of the sort Chomskyand Fiengo advocate. And again, the answer cannot lie in a difference between the "logical forms" assigned in each case, because there is no basis for assigningthem relevantlydifferentlogical forms; (3), (4), and (6) each have shallow structuresof the form
*** I.ti There is no basis for predicting either the well-formedness or . [CoMP. . [wh-]i .*
] [s .

. . .

the correct logical form for any of them. Given that the indexing procedurein (2) has been shown to give undesirable results, let us now consider a published alternativeto (2). Fiengo (1977, 48, fn. 18) states: A reviewerof this articlehas perceptivelypointedout that this result [i.e. the problemillustratedin (3)/GKP]can be avoided if trace theory is altered in such a way that the index of a trace left by movementis the same as the index of the NP position to which movementtakes place. The indexingprocedurebeing suggestedhere is (7). (7) A mover assumes the index that the target has, and
creates a copy of this in the vacated site.

The mechanismin (7) would solve the problem of (3). If the derivationis done accordingto (7), the outcome is (3c'):

SQUIBS

AND

DISCUSSION

(3) c'.

who]] [S[NP4 e] [COMP[NP,


[NP3

was seen

Bill] by [NP, el]

In (3c'), the trace in subject position, NP4, is not properly bound, so the ungrammaticality of (3d) is correctly predicted. And (4d) is predictedto be ungrammatical in a similarway. But, unfortunately, (7) makes the derivationof simple passives like (6d) impossible. Under (7), (6d) would have the following derivation:
(6') a.
[COMP[NP1 e]]
[s[NP2

e] was murdered

[Np3

who]]

b.
c.

[COMP[NP1 e]] [s[NP2who] was [COMP[NP1 who]] [S[NP,

murdered[NP2 e]] e] was murdered[NP2 e]]

Here the first trace is properly bound (for it is preceded and commandedby a mover with the same index) but the second is not. Under the mechanism(7) suggestedby Fiengo's referee, this will always happen in cases of multiplemovement to the left-e.g. in cases of successive cyclic Wh Movement across several clause boundaries.Whenevera mover makes a second move, the trace left by its first move will no longer have the same index as the mover associated with it. The latter observationsuggests a modificationto (7): (8) A moverassumes the index that the targethas, leaves a copy of the target's index in the vacated site, and simultaneously all other phrases with the mover's formerindex are given the target'sindex instead. This will affect (6') by changingthe index of the rightmostNP in (6'c) to 1, so that [NP1 who] in the COMPposition will count as a properbinderfor both of the traces. But if the implicationsfor (3) and (4) are now examined, it will be found that (8) makes them all well-formed.The modification that turns (7) into (8) makes its consequences essentially the same as those of (2) in the cases under examination here. What, then, is the indexing procedure that defines the trace/mover relations in trace theory? How can indexing be controlled duringderivations so that at shallow structurethe generalizationsthat Chomskyand Fiengo seek can be stated? Until a clear and explicit answer is given to these questions, it cannot truly be said that trace theory provides an explanation for anything-not even the very simple example (lb).
References Chomsky, N. (1975) Reflections on Language, Pantheon, New York. Chomsky, N. (1977a) "On Wh-Movement," in P. W. Culi-

362

SQUIBS

AND

DISCUSSION

cover, T. Wasow, and A. Akmajian,eds., Formal Syntax, Academic Press, New York. Chomsky, N. (1977b)"Introduction,"in Essays on Form and Interpretation,AmericanElsevier, New York. Chomsky, N. and H. Lasnik (1978) "A Remarkon Contraction," Linguistic Inquiry 9, 268-274.

Fiengo, R. (1977) 'On Trace Theory," Linguistic Inquiry 8,


35-61. Vergnaud, J.-R. (1974) French Relative Clauses, Doctoral dis-

sertation, MIT, Cambridge,Massachusetts.

Is LACHMANN'S LAW A RULE?

Consider the following Latin verbal data: Present Active Passive legit legitur editur edit agit agitur frangit frangitur vincit vincitur facitur facit rumpit rumpitur capit capitur Active legit edit egit fregit vicit fecit rupit cepit Perfect Passive lectum est esum est actum est fractum est victum est factum est ruptum est captum est Gloss 'read' 'eat' 'act' 'break' 'conquer' 'do' 'break' 'capture'

Jurgen Klausenburger, University of Washington

(1) (2) (3) (4) (5) (6) (7) (8)

All eight verbs have a morphological rule of root vowel lengthening in the perfect active, e.g. legit (pres.) -l 1ejgit (perf.). In the passive perfect participle, however, only examples (1) through (4) show a lengthened root vowel. They illustrate Lachmann's Law, which, according to Perini (1978, 145), is not a separate rule, but a simplification of the vowel lengthening rule of the perfect active, (9), to (10). (9) V (10) V [+long] / [+long] / C0 + [+affix +perfect +active

Co + [+perfect]

Examples (5) through (8) contradict this analysis. If Lachmann's Law was entirely a morphological process (cf. Watkins (1970)), why did the extension of vocalic length from the perfect active to the perfect passive not occur in all verbs with distinctive lengthening of perfect actives, including those with voiceless root closing consonants, e.g. examples (5) through (8)?
1 A similar analysis, under the rubric "partialde-morphologization", is given in Klausenburger (1976, 318).

Vous aimerez peut-être aussi