LINGUIST List 4.1063

Thu 16 Dec 1993

Disc: Syntax of Scope

Editor for this issue: <>


Directory

  1. Maryellen MacDonald, Aoun & Li and psycholinguistics

Message 1: Aoun & Li and psycholinguistics

Date: Mon, 6 Dec 93 15:39:38 PSTAoun & Li and psycholinguistics
From: Maryellen MacDonald <mcmgizmo.usc.edu>
Subject: Aoun & Li and psycholinguistics

In his review of Aoun & Li, SYNTAX OF SCOPE, Tony Kroch asks about the
extent of cross-linguistic variation in quantifier scope judgments.
His question is motivated by a) Aoun & Li's argument for universal
scope assignment principles in the grammar, yet b) Aoun & Li report cross-
linguistic differences in Chinese and English, namely that both (1) and
(2) are ambiguous in English, whereas only (2) is ambiguous in Chinese:

1. Every man loves a woman. (Kroch's (1))
2. Everyone was arrested by a woman. (Kroch's (3))

Like many linguists, Aoun & Li interpret scope judgments as owing to
the grammar; Kroch nicely summarizes their particular analysis of
English and Chinese. In contrast, my argument below might be termed
"the psycholinguistics of scope." I suggest that a substantial amount
of the variation in scope judgments in English (e.g. Kroch's examples
(5-9) and others I discuss below) are attributable at least in part to
psycholinguistic factors, namely the nature of ambiguity resolution
processes during language comprehension, so that it is a mistake to
assume that a particular pattern of scope judgments necessarily reveal
something about syntax/LF. I then speculate on some ways in which
cross-linguistic variation in scope judgments might emerge. Although
I use a small sample of Aoun & Li's data in this discussion, it should
be clear that I am not offering an alternative to their account,
which goes far beyond the few sentences I discuss here. Instead I am
making a more general claim that ANY scope-ambiguity judgment data may
reflect the parser at least as much as the grammar. Though I try to
be brief, laying out the relevant pieces of psycholinguistic theory
takes time; you are hereby warned that this is a long message.

When speakers of a language report that they can't assign wide scope
to X--some quantified NP in a multiply-quantified sentence--linguists
often interpret this judgment as evidence that a structure in which X
receives wide scope violates some principle in the grammar. A
psycholinguistic account would place the explanation on the nature of
ambiguity resolution processes. Work in syntax and syntactic
ambiguity resolution offers an analogy: Speakers of English generally
report that they cannot recover a well-formed interpretation for "The
horse raced past the barn fell," but neither psycholinguists nor
syntacticians would say that the sentence is ungrammatical. Instead,
we believe that there is something about the parsing process,
specifically the ambiguity resolution process, that makes it nearly
impossible to assign an interpretation to this ambiguity. Indeed,
much psycholinguistic work has revealed a great deal about what it is
that makes ambiguity resolution for this sentence impossible, while
other sentences with the identical structure don't pose a problem
(e.g. "The men arrested in the raid were taken downtown."). The scope
equivalent of this example is that the interpretation with wide scope
on the quantified NP "X" is perfectly grammatical, but there is something
about the nature of ambiguity resolution processes such that speakers
cannot resolve the ambiguity in a way that yields wide scope on X.

What does the theory of ambiguity resolution have to offer treatments of
scope? Scope ambiguity is largely neglected in studies of ambiguity
resolution, but there are some things we can note about how ambiguity
resolution appears to work in this case. The following principles
concerning how comprehenders resolve quantifier scope ambiguity are
stated informally here and are meant to be only descriptive; Kurtzman &
MacDonald (1993) offer a somewhat different treatment and discuss other
principles proposed in the literature. Another good source for discussion
of processing principles is Fodor (1982).

I. Lexical Preference (as noted by a number of linguists, including
Kroch in his dissertation). As part of its lexical representation,
each quantifier term has a particular "strength" in which it tends to
take wide scope over other quantifiers. Quantifiers such as "one,"
"each" and "someone" strongly tend to take wide scope; "every" and
"everyone" are weaker.

II. Structural Preference. Quantified NPs in certain positions are more
likely to be assigned wide scope than are other QPs. The specific variants
proposed in the literature suggest that the ambiguity resolution
mechanism prefers to assign wide scope to the following sorts of quantified
NPs: Quantified Subjects (more likely to take wide scope than objects),
topics (compared to non-topics), first QPs (the first quantified NP the
comprehender encounters in the sentence preferentially gets wide scope
over subsequently encountered QPs in the sentence), c-commanding NPs,
Agents, External Arguments, etc. For most of this discussion, we don't need
to be more specific than this vague description of structural constraints.

III. Weighing constraints. Constraints like (I), (II), and others can
converge and conflict. A comprehender's preferences for one scope
interpretation is strong when multiple constraints converge to promote
wide scope on a particular phrase, but when several constraints conflict,
speakers are less likely to have a clear preference in scope interpretation.
This effect of multiple constraints on the degree of ambiguity is known to
be a general property of ambiguity resolution processes and is not about
scope in particular.

Next we consider three variations in scope judgments in English which can
be attributed to these processing constraints.

First, we know from constraint (I) that "someone" tends to take wide scope
over "everyone," and from (II) that quantified subjects (and/or Agents,
etc.) tend to take wide scope over quantified non-subjects (non-Agents,
etc.). From III, we'd expect stronger scope interpretation preferences
when I and II converge (i.e. when "someone" is the subject, as in (3)) and
weaker preferences when (I) and (II) conflict, as in (4).

3. Someone loves everyone.
4. Everyone loves someone.

A number of linguists (e.g. Fodor, 1982, for similar sentences) have noted
that in (3) "someone" very strongly takes wide scope, whereas (4) seems
much more ambiguous, and the joint effect of quantifier terms and
structural position has been confirmed in psycholinguistic experiments
with large numbers of sentences and experiment participants (Kurtzman &
MacDonald, 1993).

Of course it's not surprising to find processing considerations
affecting the choice between two grammatical alternatives such as
(3-4), but preferences in the next example have been attributed to the
ungrammaticality of one of the alternatives. This example comes from
sentences (6) and (8) in Kroch's review of Aoun & Li, repeated below.

Kroch's (6) John assigned one student every problem.
Kroch's (8) Mary assigned every student one problem.

Aoun & Li note that (6) appears to be unambiguous with wide scope on
"one student," but Kroch notes that (8) is ambiguous, an unexpected
outcome from Aoun & Li's perspective. Recall from (I) that "one" takes
wide scope more strongly than "every," and that one of the formulations of
(II) is for surface linear order--the first-encountered quantified phrase
preferentially gets wide scope over subsequent ones. We see these two
constraints converging to promote wide scope on "one student" in Kroch's
(6), and they conflict in Kroch's (8), yielding the judgment of ambiguity.
On this analysis, the extent to which these examples are / are not
ambiguous is not attributed to the syntax/LF but rather emerges from the
weighing of constraints during the ambiguity resolution process.

[To forestall some replies from List subscribers, I should say here that I
offer the linear order constraint for the sake of simplicity and don't believe
it is actually the correct description of the processing constraint(s) that
operate here. There are clear counterexamples to linear order discussed in
Kurtzman & MacDonald, and linear order certainly fails to capture the fact
that there is something special about scope assignment in double object
constructions such as Kroch's (6 & 8). For example, compare Kroch's (6)
with a non-dative sentence with the same quantifiers and linear order:

5. John praised one student every day.

Unlike Kroch's (6), (5) seems quite ambiguous to me. The correctness
of linear order aside, I do want to embrace the processing nature of
the argument. Kroch's contrast between (6&8) is analogous to the
contrast between the garden path "horse raced" sentence and the easily
parsed "The men arrested in the raid were taken downtown." When two
sentences with the same structure have different ambiguity resolution
outcomes as a function of the lexical items in the sentences, one
should start looking for explanations in the processor, not (only) in
the nature of syntax or LF.]

Finally, consider actives and passives, in (1-2), repeated here:

1. Every man loves a woman. (Kroch's (1))
2. Everyone was arrested by a woman. (Kroch's (3))

Kurtzman & MacDonald note that several linguists have reported that
passives seem more ambiguous than actives, and psycholinguistic
experiments strongly confirmed this pattern. Our experiment
participants treated the passives as completely ambiguous, but for
actives they were generally willing to accept wide scope on the object
(e.g. "a woman" in (1)) only 20-30% of the time (the exact percentage
varied with the wide scope tendencies of the quantifier terms, as
addressed in (I)). One way to account for this pattern is to say that
the structural constraint sketched in (II) is actually two
constraints, perhaps one that favors wide scope on subjects and
another that favors wide scope on Agents. These two converge in (1)
and conflict in (2), yielding stronger scope interpretation
preferences for actives than passives. [Again I don't intend to
advocate this particular solution, and I don't want to leave the
impression that psycholinguists name a constraint for every new
judgment. We expect the scope ambiguity resolution constraints to be
quite general, just as they appear to be for ambiguity resolution at
other levels of linguistic representation that have received more
study in psycholinguistics.]

CROSS-LINGUISTIC VARIATION.

Given these data from both judgments and psycholinguistic experiments
that passives are far more ambiguous than actives for English
speakers, Aoun & Li's contrast between Chinese and English can be
viewed not as qualitatively different but rather as differing in
degree. In English, passives are ambiguous and actives are slightly
ambiguous, and in Chinese, passives are ambiguous and actives are
apparently not ambiguous at all. One possible account of these
differences could preserve the universal nature of scope assignment
processes (via the grammar or the processor) by relying on
cross-linguistic variation in quantifier terms. Like any lexical
item, quantifier terms may not have exact translations from one
language to the next, and it is certainly possible that
near-translations of some quantifiers may not be exactly equivalent in
(I), the strength to which they take wide scope in different
languages. For example, a native speaker of Lebanese Arabic tells me
that the Arabic equivalent of "everyone" has a stronger tendency to
take wide scope in her language than in English, so that the relative
ambiguity of Arabic equivalents of (3-4) are not the same as in
English. Given the demonstrated effects of quantifier terms on the
degree of ambiguity, we might speculate that the Chinese-English
differences for (1-2) reflect differences in the wide scope tendencies
of the quantifier terms in (1-2). I have absolutely no data about the
wide-scope tendencies of quantifier terms in Chinese, and so this
suggestion is truly speculative. Perhaps some Chinese speakers on the
List can address this issue, e.g. by observing to what if any extent
the judgments of ambiguity in Chinese versions of (1-2) vary as a
function of the particular quantifier terms used in the sentence.

In sum, I have suggested that in using judgments of ambiguity as a
reflection of the nature of the grammar, linguists should not overlook
the contribution of ambiguity resolution processes that are part of
language comprehension. I argue that in the same way that we do not
want to invoke grammatical explanations for the incomprehensibility of
multiply center-embedded sentences and garden path sentences, we may
not want to ascribe grammatical explanations to some scope
interpretation preferences, including preferences so strong that one
interpretation appears impossible/ungrammatical. Of course I am not
saying that there is NO role for the grammar here; instead I mean that
one needs to determine which phenomena are best attributed to the
grammar and which to the processor. As the field of ambiguity
resolution is one of the largest and most productive in
psycholinguistics, I suggest that the process of sorting out
grammatical vs. processing effects would proceed more rapidly if
linguists did not evaluate these issues entirely on (syntax and LF)
theory-internal grounds but rather investigated what ambiguity
resolution theory has to say about how ambiguity interpretation
preferences emerge.

Maryellen C. MacDonald
University of Southern California
mcmgizmo.usc.edu

REFERENCES

Fodor, Janet D. (1982). The mental representation of quantifiers. In S.
Peters & E. Saarinen (Eds.), PROCESSES, BELIEFS, AND QUESTIONS.
Dordrecht: Reidel.

Kroch, Anthony. (1975). THE SEMANTICS OF SCOPE IN ENGLISH. Doctoral
dissertation, MIT. Available from MITWPL & also published by Garland.

Kurtzman, Howard S. & MacDonald, Maryellen C. (1993). Resolution of
quantifier scope ambiguities. COGNITION, 48(3), 243-279.
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue