danglers

[Arnold Zwicky]

        One of my summer projects concerns sentence-initial "dangling
modifiers" of various kinds (especially of the sort "as" + NP, though
here i'll talk about various types).  Dangling modifiers are instances
of (subjectless) "free adjuncts" (in the terminology of Stump and of
Kortmann) - instances that, it is claimed, present difficulties in
interpretation.  This is a (not so brief) progress report.

        The unexpressed subjects of free adjuncts are very much
parallel to overt personal pronouns.  In both cases, the usual
situation is that there is a trigger for positing a discourse
referent, which then is to be identified with a discourse referent
otherwise set up, usually by a NP in the discourse.  Overt personal
pronouns serve as such triggers; so do unexpressed subjects in free
adjuncts.  Personal pronouns are predominantly *ana*phoric - the
relevant discourse referent is usually set up by a NP in preceding
discourse - but they can also be cataphoric:

  A: After Mary finished _Moo_, she became interested in dinner.
  C: After she finished _Moo_, Mary became interested in dinner.

Unexpressed subjects in free adjuncts are mostly cataphoric, but they
can be anaphoric as well:

  C: Having finished _Moo_, Mary became interested in dinner.
  A: Mary became interested in dinner, having finished _Moo_.
     Mary, having finished _Moo_, became interested in dinner.

(There's a general preference for anaphora over cataphora, but this is
overridden for free adjuncts, since these serve as framing sentence
adverbials, for which sentence-initial position is in general
preferred.  For free adjuncts, initial position necessarily means
cataphora.)

        In my earlier postings on anaphoric pronouns (in connection
with the Possessive Antecedent Proscription), I emphasized the
importance of topicality/foregrounding/salience in their
interpretation.  Examples that, as isolated sentences, are confusingly
ambiguous with respect to the antecedent of a pronoun or have a
hard-to-find antecedent for one become just fine when appropriate
(discourse and/or real-world) context is supplied.

        For cataphoric pronouns, the importance of context, especially
preceding discourse context, is especially great.  It's possible for a
cataphoric pronoun to occur at the beginning of a discourse:

  When she finished _Moo_, Mary became interested in dinner.
  When she finished _Moo_, Mary's interest turned to dinner.

(Note that the second of these examples violates one version of the
PAP.)  but, as a number of linguists have noted over the years, in
actual speech and writing the discourse referent that's relevant for
the interpretation of the pronoun has usually been established already
in preceding discourse.

        All this is important for free adjuncts, especially with
respect to the way they're discussed in manuals of grammar and usage.

        To start with, as with possessive antecedents, the manuals
deal almost exclusively with isolated sentences.  so they look forward
from a sentence-initial free adjunct and search for some NP (well,
mostly they look for a *noun*) that can supply the needed discourse
referent - what is sometimes called the "controller" for the
unexpressed subject.  The manuals mostly stipulate that the controller
should be the subject of the main clause.

        Now, certainly, the controller quite often is the subject.
and sometimes when it is not, or when there is no overt controller in
the main clause, an ambiguity or difficulty of interpretation results;
we get one of the standard hilarious or baffling "dangling modifiers"
of the handbooks:

  Looking through his field glasses, the bird flew away.
    (Barron's _Essentials of English_)

But many examples that are awkward out of context improve enormously
when context is taken into account.  two recent examples: first, out
of context: from a letter to the Palo Alto Daily News, 7/27/03, p. 8:

  As vice-chair, his words reflect the distorted thinking so
  evident in the political structure of Palo Alto.

Now, this is a pretty common type, involving a subject with a
possessive modifying some noun of "low referentiality" (yes, I know,
not a great term).  Even in isolation, these aren't bad.  But in fact
this is the beginning of the third paragraph of the letter.  The
second paragraph begins:

  As vice-chair of Palo Alto's Public Art Commission, Gerald
  Brett (in his guest opinion piece last Sunday) assumes the
  mantle of the holier-than-thou righteousness...

The whole letter is about Gerald Brett, and Brett as a discourse
referent is well established before the putative dangler comes up.

        Second, out of context: from a KQED Morning Edition piece,
7/25/03:

  If passed, California would be the fourth state in the nation
  with such a law.

(The law would ban discrimination on the basis of transgendered
status.)  Presented like this, the sentence sounds just dreadful.  But
in fact everything that led up to it was about this particular law,
which had been mentioned several times.  Probably only someone hot on
the dangler trail, like me, would even have noticed it.

        As with possessive antecedents, many manuals elevate
reasonable advice, about avoiding ambiguity or hard-to-find referents,
into a purported rule *of grammar*: the subject of the main clause
*must* be the controller.  Anything else - no matter how frequent the
type is, or who the writers are who use it, or how easily interpreted
it is, or how many speakers judge it to be acceptable - is just wrong.
>From published sources, two examples that would be proscribed, though
they strike me as entirely felicitous:

  As a linguist, what struck me especially about his work on meaning
    in natural language was his belief that... (Stanley Peters)
  As a mother, my job is to take care of what is possible and trust
    God with the impossible.  (Ruth Bell Graham)

(A very few manuals - Evans & Evans, in particular - simply refuse to
excoriate "dangling modifiers" in general as grammatical errors.
Others admit that such modifiers don't always present difficulties of
interpretation, but since they sometimes do, they should be avoided:
Just Say No.)

        Kortmann, a linguist who works with corpora, merely claims
that the subject is the default controller.  But even he puts this
forward as a *syntactic* condition.  A reasonable alternative is that
subjects are preferred as controllers because subjects usually denote
topical/foregrounded/salient entities.  The controller could then be a
possessive modifier of a (low-referentiality) head noun in the
subject, the object of a verb with dummy subject "it", a NP *within* a
clausal subject, etc.; any subject of low referentiality, and hence
low topicality, will be less available as a controller than a more
topical NP elsewhere in its clause, or within the subject.  A lot
would depend on the details of the main clause, not to mention the
discourse and real-world context.

        But there's *something* going on here between the unexpressed
subject of the free adjunct and some element in the main clause.  To
see this, note that sentence-initial adverbials with overt cataphoric
subjects have a much easier time of it than free adjuncts do:

  Mary read most of the afternoon.
    When she finished _Moo_,
                                   John started cooking dinner.
  ??After/On/Upon finishing _Moo_,
  ??Having finished _Moo_,

(on the reading in which Mary is the one who finished the novel).
What seems to be going on here is that
topicality/foregrounding/salience as established by preceding
discourse isn't, usually, enough to license an unexpressed subject of
a free adjunct (though it helps); free adjuncts normally signal that
there is a controller in the clause that follows, while overt pronouns
do not.

        Some manuals go further, and in effect claim that the rule of
grammar subjects-must-supply-unexpressed-subjects is *necessitated* by
the nature of free adjuncts.  The proposal (usually presupposed,
rather than argued for) is that free adjuncts are in fact modifiers of
their controlling NPs - that is, adjectivals (though not AdjPs, of
course; they are mostly VPs or PPs) - and so, like modifiers in
general, have to be adjacent to the heads they modify, except in very
special situations (this claim about modifier-modified adjacency is
also implicit).  Manuals that take this position will suggest, as one
possible correction for a dangling modifier, rearranging things so
that the modifier is as close as possible to the modified NP:

  NOT  Stunned by the blow, his opponent easily knocked him out.
  BUT  Stunned by the blow, he was easily knocked out by his opponent.

        There is something right here: free adjuncts do indeed have a
special relationship with a controller in the following main clause
(see above), and this relationship is much like that of an appositive
to its NP.  In fact, NP-headed predicates can serve as free adjuncts
as well as appositives, and such predicates have the same
(predicative) interpretation as appositives:

  A scholar of ancient languages, Mary quickly deciphered the
    inscription.
  Mary, a scholar of ancient languages, quickly deciphered the
    inscription.

  Not a scholar of ancient languages, Mary was baffled by the
    inscription.
  Mary, not a scholar of ancient languages, was baffled by the
    inscription.

        However, this alternation between a sentence-initial expression and
one located immediately after the subject is not some special property
of free adjuncts; sentence adverbials of many types exhibit the same
possibilities:

  Once John had finished _Moo_, Mary began reading it herself.
  Mary, once John had finished _Moo_, began reading it herself.

Syntactically, free adjuncts are a subspecies of (loosely adjoined)
sentence adverbials.  What they share with appositives is merely their
interpretation as semantic predicates - an interpretation that
requires the positing of a discourse referent of which something can
be predicated.

        Talk of free adjuncts as "modifiers" of their controllers is
undoubtedly encouraged by the careless - often unclear, sometimes
incoherent, rarely examined - use of grammatical terminology in many
of the manuals.  Part of the problem is that the same verb, "modify",
is used for the relationship between adnominal expressions, AdjPs in
particular, and their head Ns, and for the relationship between
predicative expressions, AdjPs in particular, and their subjects.  In
addition, the manuals are often unclear as to whether the
relationships in question are syntactic or semantic.  Once these
conceptual matters are cleared up, there is no reason to say that free
adjuncts "modify" their controllers and hence should be subject to a
condition requiring that modifiers of nouns be adjacent to those
nouns.

        To sum up: we see the same features as in the manuals'
treatment of dangling modifiers as in their treatment of the PAP: (1)
the manuals judge examples as erroneous or infelicitous out of
context, from which (2) a general principle is formulated to cover the
core set of examples and (3) then elevated to a (purported) rule of
grammar; (4) the rule is then justified on the basis of some external
consideration, having to do with (5) dubious definitions of linguistic
terms, universals of grammar, or principles of logic; and (6) this
discussion proceeds with little regard for the practice of skilled
writers of established standard formal written english, when this
departs from the strictures of the purported rule.

        There is in fact a possible interaction between the PAP and
the immediate adjacency condition for free adjuncts.  If the subject
of the main clause is a NP of the form possessive + head, the
immediate adjacency condition, interpreted literally, requires that
the *possessive* be the controller for a free adjunct.  So, in both
of the following examples, "Mary" would be predicted to be the
controller NP - incorrectly in the first, correctly in the second:

  As a mother, Mary's sister gives first priority to the children.
  As a mother, Mary's first concern is for the children.

The PAP comes into the story because, if it extends to zero cataphora
(as well as to the cases of overt anaphora that seem to have motivated
it), it absolutely bars the possessive as the controller (correctly
in the first example, incorrectly in the second).  A droll standoff
between two purported rules of grammar.

        Forget about the immediate adjacency condition.  There's
still the rigid requirement that the subject be the controller (which
makes the same predictions as the PAP).  Like the PAP, this is wrong
as a description of the practice of good writers.  (Some manuals that
insist on the subject as controller nevertheless note that dangling
modifiers do occur in the works of celebrated writers - the
implication being that it's ok for the pros to break the rules for
the sake of some effect, but ordinary folks shouldn't try this sort
of thing, at least not in formal writing.)

        And, like the PAP (which seems to have been invented within
the past hundred years and then spread through textbooks, though not
in the practice of most writers), the subject controller requirement
has a history.  I have to confess I don't know much about this
history, but at least two manuals claim that dangling modifiers were
frequent in 18th century English.  (Geoff Pullum has pointed out to me
that there's one in the Declaration of Independence!)  What happened?
The obvious suggestion is that explicit instruction made writers
self-conscious about their free adjuncts and editors sensitive to
them.  It would be interesting to see the history in some detail.

(With thanks to Micha Rinkus for going through manuals, collecting
danglers, and discussing their analysis with me.)

arnold (zwicky at csli.stanford.edu)