Weak Crossover and the Absence of Traces

[Mary Dalrymple]

We would be very interested in comments on the proposal sketched below.

 - Mary, Ron, and Tracy


	       Weak Crossover and the Absence of Traces

Examples such as (1) illustrate the phenomenon of weak crossover,
originally discussed by Wasow (1979):

(1) *Who(i) do his(i) friends adore?

Transformational analyses of weak crossover assumed that a violation
ensues when a wh-phrase like "who" moves to the front of the sentence
and "crosses over" a coindexed pronoun; in (1), "who" crosses over the
coindexed pronoun "his".  Nontransformational accounts assume that
such examples violate prominence requirements between the pronoun and
the quantifier that binds it.  Bresnan (1995) proposes a theory of
prominence for anaphoric binding that accounts for examples like (1)
by assuming the presence of a trace.

We propose a new definition of prominence for anaphoric binding that
accounts for the ungrammaticality of examples involving weak crossover
without assuming traces in wh-constructions.  Since our account
eliminates the need to posit traces in the analysis of weak crossover,
and we know of no other evidence that can only be explained by
assuming traces in LFG, we are optimistic that the earlier LFG
assumption that there are no traces can in fact be maintained.

Bresnan (1996) shows that binding of pronouns by operators such as
quantifiers and wh-phrases is constrained by prominence relations: an
operator cannot bind a pronoun that is more prominent than it.  Under
her account, prominence has three aspects: thematic prominence,
syntactic rank, and linear order.  Different aspects of prominence are
relevant for different languages.  We provide revised definitions of
these three aspects of prominence, in which a basic relation of
prominence holds between coarguments, and a derived relation of
prominence holds between elements contained in coarguments.  Our
analysis is inspired by the proposal of Sag (1998) for analyzing weak
crossover in a framework without traces.  The intuition behind his
approach is to consider the element which contains the operator as the
important one in determining linear order constraints on quantifier
binding.

We define prominence among the arguments of a predicate in the
following way:

  Coargument thematic prominence: following Bresnan (1996) and
  references cited there, a relation of prominence among coarguments
  defined by the projection relation between f-structure and argument
  structure (e.g., an agent of a predicate is more prominent than a
  coargument theme).

  Coargument syntactic prominence: following Bresnan (1996) and
  references cited there, a relation of prominence among coarguments
  defined at functional structure on the functional hierarchy (e.g., the
  subject of a predicate is more prominent than a coargument object).

  Coargument linear prominence: a relation of prominence among
  coarguments defined in terms of f-precedence; an argument A of a
  predicate is more prominent than its coargument B if A f-precedes B.

It is generally accepted that basic prominence relations are defined
on coarguments for thematic prominence and syntactic rank, which are
traditionally defined only for coarguments of a single predicate.  Our
proposal differs from previous ones in that linear precedence between
coarguments is taken to be the relation that is primarily relevant for
defining linear prominence.  Relative to the basic coargument
relations of prominence, we define indirect prominence relations for
elements contained in coarguments:

  Thematic/syntactic/linear prominence: A is more
  thematically/syntactically/linear prominent than B if the f-structure
  for A is the same as or contained in some A', the f-structure for B is
  the same as or contained in some B', and A' outranks B' in coargument
  thematic/syntactic/linear prominence.

We claim that all cases of weak crossover can be analyzed in terms of
these definitions of precedence without the need to assume traces,
including data from German, Hindi, Malayalam, and Palauan discussed
by Bresnan (1996).

We illustrate our proposal by considering weak crossover and pronoun
binding in German.  Following Choi (1995), Bresnan (1996) shows that a
quantifier or wh-phrase in German may not corefer with a pronoun that
outranks it in both syntactic and linear prominence (see also Berman
2000).  The following three examples are grammatical:

(4) dass [jeder]        [seine Mutter] mag
    that  everyone-NOM   his   mother likes
    `that everyone(i) likes his(i) mother'

(5) dass [jeden]        [seine Mutter] mag
    that  everyone-ACC   his   mother likes
    `that his(i) mother likes everyone(i)'

(6) dass [seine Mutter] [jeder]        mag
    that  his    mother  everyone-NOM likes
    `that everyone(i) likes his mother(i)'

In determining whether the prominence conditions are met in these
examples, we consider the subject and object coarguments of the verb
"mag" `likes', which have been enclosed in brackets.  In (4), the
subject quantifier outranks the object, which contains the pronoun, in
syntactic and linear prominence, and the sentence is grammatical.  In
(5), the object quantifier outranks the subject, which contains the
pronoun, in linear prominence, though not in syntactic prominence; in
(6), the subject quantifier outranks the object, which contains the
pronoun, in syntactic prominence.  In all situations the sentences are
grammatical.  The proposals of Bresnan (1996) and Berman (2000) make
similar predictions, since they assume there is no trace in scrambling
and clause-internal topicalization in German.

If the pronoun outranks the quantifier in both syntactic and linear
prominence, the sentence is ungrammatical:

(7) *dass [seine Mutter] [jeden]        mag
     that  his   mother   everyone-ACC likes
    `that his(i) mother likes everyone(i)'

In (7), the subject of "mag" `likes' contains the pronoun, and it
outranks the object quantifier both in syntactic and linear
prominence; this situation is disallowed.  Similar facts hold for
wh-phrases, as Bresnan (1996) and Berman (2000) show.

In example (8), a long-distance dependency is involved, and the same
rules of prominence apply:

(8) *jeden/wen    meinte [seine Mutter], [habe sie getroested]
     everyone/who said    his   mother    has  she consoled
    `Everyone(i), his(i) mother said that she consoled./
     Who(i) did his(i) mother say that she consoled?'

Here we consider the coarguments that contain the quantifier and the
pronoun at f-structure.  These are the subject and sentential
complement of "meinte" `said', which, as above, have been enclosed in
brackets.  Like example (7), the element containing the pronoun
outranks the coargument f-structure containing the quantifier in both
linear and syntactic prominence, and the binding relation is
disallowed.  The proposals of Bresnan (1996) and Berman (2000) account
for this example by assuming that there is a trace of the fronted
quantifier or wh-word in the subordinate clause which is outranked by
the possessive pronoun in linear prominence.

In contrast, sentence (9) is fully grammatical:

(9) [jeden/wen]   sagte sie, habe [seine Mutter] getroested
     everyone/who said  she  has   his   mother  consoled
    `Everyone(i), she said that his(i) mother consoled./
     Who(i) did she say that his(i) mother consoled?'

Bresnan (1996) and Berman (2000) account for the grammaticality of (9)
by assuming that the trace in the subordinate clause must appear in
clause-initial position so that it outranks the pronoun "seine" `his'
in linear prominence.  On our approach, no appeal to traces is
necessary: the coarguments that contain the pronoun and the quantifier
are the subject and object of the verb "getroested" `consoled'.  The
quantifier outranks the f-structure containing the pronoun in linear
prominence, and the binding relation is allowed.

Thus, our new definition of prominence accounts for weak crossover
without assuming traces, allowing a return to the earlier traceless
LFG account of long-distance binding.



			      References

Berman, Judith. 2000. Topics in the Clausal Syntax of German.  PhD
thesis, University of Stuttgart.

Bresnan, Joan.  1995.  "Linear order, syntactic rank, and empty
categories: On weak crossover".  Formal Issues in Lexical-Functional
Grammar, ed. by Mary Dalrymple; Ronald M. Kaplan; John T. Maxwell; and
Annie Zaenen, 241-274. Stanford University: CSLI Publications.

Bresnan, Joan. 1996. "Morphology Competes with Syntax: Explaining
Typological Variation in Weak Crossover Effects."  To appear in Is the
Best Good Enough? Proceedings from the Workshop on Optimality in
Syntax, MIT, MAY 19-21, 1995, edited by Pilar Barbosa, Danny Fox,
Paul Hagstrom, Martha McGinnis, and David Pesetsky and published by
The MIT Press and MIT Working Papers in Linguistics.

Choi, Hye-Won.  1995.  Weak Crossover in Scrambling Languages.  Paper
presented at the 1995 LSA meeting.

Sag, Ivan A. 1998. Without a Trace.

Wasow, Thomas. 1979.  Anaphora in Generative Grammar.  Ghent:
E. Story.


Mary Dalrymple <dalrymple at parc.xerox.com>
Ronald M. Kaplan <kaplan at parc.xerox.com>
Tracy Holloway King <thking at parc.xerox.com>
Xerox PARC, 23 October 2000