 

 

lHllllllHHIIllllllllllllllllllllUlllllllzlllzlllllLlIHHl

was“ - 1 31293 013992

(M ii 633

This is to certify that the

thesis entitled

The Effects of Context on
Lexical Ambiguity Resolution

presented by

Phillip A. Weeks, Jr.

has been accepted towards fulfillment
of the requirements for

M.A. degree in Psychology

\3/ /
' ’ Major professor
Date [/3 « 7r
/ ,

   

 

0-7639 MS U is an Afﬁrmative Action/Equal Opportunity Institution

 

LIBRARY
Mlchlgan State
University

 

 

 

--~-A

PLACE N RETURN BOX to mow-thin checkout from yam record.
TO AVOID FINES Mum on or bdon dd. duo.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

MSU chn W AdlonlEcpol Oppoﬂunlty Institution
Wane-m

THE EFFECTS OF CONTEXT ON LEXICAL AMBIGUITY RESOLUTION

by
Phillip Anthony Weeks, Jr.

A THESIS

Submitted to
Michigan State University
in partial fulﬁllment of the requirements

for the degree of

MASTER OF ARTS

Department of Psychology

1994

ABSTRACT
THE EFFECTS OF CONTEXT-ON LEXICAL AMBIGUITY RESOLUTION

by
Phillip Anthony Weeks, Jr.

The effects of context on lexical ambiguity resolution were examined in
two experiments as a replication and extension of Simpson and Krueger (1991),
who found evidence for selective access of word meanings. In a cross-modal
naming paradigm, ambiguous words were placed in sentences that biased the
dominant meaning, the subordinate meaning, or no meaning (neutral context) of
the ambiguous mrd. Subjects named target words that were either related or
unrelated to the ambiguous mrd after 0, 200, 600 ms S.O.A. (Experiment 1); or
0, 300, 700 ms I.S.I. (Experiment 2). In Experiment 1, no effects of context were
found at any level of S.O.A but there was an effect of dominance at 600 ms
S.O.A. supporting a modular theory of lexical access. In Experiment 2, there
were no effects of context on meaning access at any level of I.S.l. The results of
these experiments are discussed in terms of recent theories of lexical ambiguity

resolution.

List of Tables
Introduction
Conclusion
Appendix A

Appendix B
Appendix C

Bibliography

Table of Contents

page 4
page 5
page 36
page 44
page 50
page 54

page 55

III

List of Tables

Table 1.
Mean Reaction Times (RT) in ms for S.OA X Target Ambiguity Strength
X Target Relation X Context.

Table 2.

Mean Reaction Times (RT) in ms for l.S.l. X Target Ambiguity Strength X
Target Relation X Context.

IV

INTRODUCTION

Lexical ambiguities can provide an insight into human mental processes.
By examining how a particular meaning of an ambiguous word is accessed,
psychologists can gain insight into how words are stored in memory and how
mrds with different meanings are accessed. Many English words have multiple
meanings, but listeners have very little difﬁculty determining which meaning is
appropriate. This lack of difﬁculty results from the disambiguating effects of
sentential context.

Traditionally, there have been two broad theories about the nature of this
process, multiple access and selective access. Multiple access theories posit
that sentence context has no prior effect on the access from memory of the
relevant meaning of an ambiguity. (Multiple access theories are also called
exhaustive models because all possible meanings are exhaustively accessed
before a decision is made.) These models are consistent with a highly modular
view of the language processing system, with higher levels in the system having
inﬂuence on selection only after all meanings of the ambiguous word have been
accessed. In mis case, semantic and syntactic information have no inﬂuence on
the bottom-up processing of lexical information during sentence processing.

On the other hand, selective access theories propose that sentence
context can directly inﬂuence which meaning of an ambiguous word is accessed.

This inﬂuence is the result of the overall meaning of the sentence serving to

2
determine which meaning of the ambiguous word is appropriate given the
meaning conveyed by the sentence. Thus, selective access theories support an
interactive view of the language processing system, with higher levels in the
system inﬂuencing the selection of appropriate meaning as the sentence is
being processed. I will new review some studies that examine these issues.
I 'l I B .

Tanenhaus, Leiman & Seidenberg (1979) found support for a multiple-
access model of meaning for noun-verb lexical ambiguities. Subjects listened to
sentences and then named a visually presented target as quickly as possible
(cross-modal naming). The last word in the sentence was both semantically and
syntactically ambiguous: One meaning corresponded to a noun and the other to
a verb (e.g. rose). The sentences provided contextual information that favored
only one of the possible meanings of the ambiguous word. The target word's
meaning could either be related to the meaning of the target biased by the
sentence context, related to the meaning unbiased by the sentence context, or
unrelated to either meaning. Target words appeared with either 0, 200, or 600
msec Stimulus Onset Asynchrony (S.O.A.). According to Tanenhaus et al.'s
predictions, naming latencies should be faster for the target word related to the
biased meaning of the ambiguous word relative to controls if context inﬂuences
access of meaning. However, both meanings of the ambiguous word should
show facilitation relative to controls if context exerts no inﬂuence on access of

meaning.

3

Tanenhaus et al. (1979) found that target naming latencies varied not
only with the congruency of the target word meaning and the meaning of the
ambiguous word biased by the context, but also with the amount of time between
presentation of the ambiguous word and target. At 0 msec S.O.A. there was
facilitation for target words related to both contextually appropriate and
inappropriate meanings of the ambiguous word. At 200 msec S.O.A, however,
there was only facilitation for the target word related to the contextually
appropriate meaning. They concluded that both meanings of an ambiguous
word are accessed initially and one of the meanings is selected within 200 msec.
They further posit that the selection is based on the meaning of the context. The
results at 600 msec S.O.A were complex Targets related to the verb meaning
of the ambiguous word were facilitated only when the context biased the verb
meaning of the ambiguous word. However, targets related to both verb and
noun meanings showed facilitation when the context was biased toward the
noun meaning of the ambiguous word. This pattern suggests that the verb
meaning of an ambiguous word is only minimally affected by context at 600
msec (T anenhaus, et al. 1979). Thus, for verb ambiguities, both meanings are
highly activated at some early stage,and later context plays a small role in
choosing the appropriate meaning.

Swinney (1979) found similar results using the same cross-modal

priming paradigm, but with lexical decision times instead of naming latencies as

the dependent measure. Also, the alternative meanings of the ambiguous word

4
were always from the same syntactic category. In Experiment 1, subjects
listened to sentences that contained either an ambiguous noun or an
unambiguous control. The sentence context was either strongly predictive of
one of the meanings of the ambiguous word, or provided no disambiguation.
Following the auditory presentation of the sentence, subjects performed a lexical
decision task on word and nonword targets presented at the onset of the
ambiguous word (0 msec S.O.A) Word targets were either related to the
contextually appropriate meaning of the ambiguous word, related to the
contextually inappropriate meaning, or unrelated to either meaning. Swinney
(1979) found that even in the presence of a strongly biased context, both
meanings of an ambiguous word are accessed, suggesting that lexical access is
fairly autonomous from the inﬂuences of contextual information.

In Experiment 2, Swinney (1979) examined the time-course of activation
for appropriate and inappropriate meanings of an ambiguous word. The target
word appeared three syllables after the presentation of the ambiguous word;
otherwise, all other procedures remained the same as in Experiment 1. Swinney
found facilitation only for the contextually-related target word. Swinney thus
concluded that, initially, lexical access is autonomous with respect to sentential
context in that "the entire inventory of information stored for a lexical form is
made available to the sentence comprehension device" (p. 657). A later
postaccess decision process uses context to select the relevant meaning of the

ambiguous word, and this process is complete three syllables after the

5
ambiguous word has been encountered.

Research by Seidenberg, Tanenhaus, Leiman and Bienkowski (1982)
also largely supported a multiple-access view of meaning for ambiguous words.
They distinguished between two classes of ambiguous words (noun—noun and
noun-verb) and three types of contexts (priming, nonpriming, and neutral).
Noun-noun ambiguous words are those that have two unrelated meanings that
are both nouns, (e.g. bank), and noun-verb ambiguities are those that have two
unrelated meanings where one is a noun and the other is a verb, (e.g. rose).
Seidenberg, et al. deﬁned priming contexts as those that contain a word that is
highly associated with the biased meaning of the ambiguous word, and
nonpriming contexts as those that bias the meaning of the ambiguous word
through syntax or pragmatics but contain no semantic associates of the
alternative interpretations. Thus, the overall meaning of the context serves to
reduce the appropriateness of one of the meanings.

Seidenberg et al. found that for noun-noun and noun-verb ambiguities in
neutral contexts, both meanings of the ambiguous word were accessed initially,
but just one of the meanings was available 200 ms later. For noun-noun
ambiguities in contexts that contained a word that was highly associated to a
particular meaning of the ambiguous word, only that meaning of the ambiguous
word was accessed, even at the earliest probe point. Seidenberg, et al. argued
that this result is due to lexical priming. A word in a context that is highly

associated with one of the meanings of the ambiguous word is linked by

6
preexisting associations within the lexicon. As a result, intra-lexical priming can
restrict the search within the lexicon to those items related to the biased
meaning. Seidenberg et al. found converging evidence for the lexical priming
theory when they failed to ﬁnd selective access for noun-noun ambiguities.
They occurred in contexts that favored one meaning of the ambiguous word, but
contained no semantically associated word. They continued to ﬁnd multiple
access for noun-verb ambiguities even in contexts that contained a semantically
related word, indicating that noun-noun and noun-verb ambiguities are resolved
differently.

All of the research discussed to this point has assumed that the
alternative meanings of an ambiguous word are equally salient. However, for
many ambiguous words, one meaning is more frequent than the others. As a
result, some ambiguous words have what is called a dominant and subordinate
meaning, where the former is the most frequently occurring meaning and the
latter is the second most frequent meaning. Some recent research has shown
support for selective access of the dominant meaning of an ambiguous word
even in contexts without lexical associates. In a cross-modal priming study,
Tabossi (1988) had subjects listen to sentences and then make a lexical
decision on a target word. Three conditions were employed. In the ﬁrst
condition, the sentence biased the dominant meaning of the ambiguous word
without strong constraints on that meaning. The target word was either related

to one of the meanings of the ambiguous word without denoting any

7
characteristic of it, or was a control (For example: During the walk, the boy
noticed on the trees many buds, although it was already late in the season. The
related and unrelated target words were Flower and Chair, respectively.
Ambiguous word in italics.) In the second condition, the same context was used,
but the target word either denoted a speciﬁc characteristic feature of one of the
meanings of the ambiguous word or was a control (e.g. During the walk, the boy
noticed on the trees many buds, although it was already late in the season.
Delicate-Order, related and unrelated target mrds, respectively). In the third
condition, the sentence biased the dominant meaning of an ambiguous mrd by
making salient a particular characteristic of that meaning. In this case the target
word was associated either with the dominant meaning, the subordinate
meaning or unrelated (control) (e.g. In the spring, a light wind among the
branches of the tree was sufﬁcient to damage the ﬁrst young bud, just opened.

F Iower- Precious-Order, dominant, subordinate, and control target words,
respectively).

Tabossi (1988) found that in the ﬁrst condition, lexical decisions were
faster for both the appropriate and inappropriate meaning of the ambiguous word
relative to the control. This pattern of results also held for Condition 2 where the
target word denoted speciﬁc characteristics of one of the meanings. However, in
Condition 3, Tabossi found that lexical decisions on target words that are related
to the dominant meaning of the ambiguous word following sentences that bias

and constrain the dominant meaning of the ambiguous word are faster than

8
target words related to the subordinate meaning of the ambiguous word or target
words that were unrelated. Tabossi interpreted these results as showing
selective priming of the dominant meaning of an ambiguous word when the
ambiguity is embedded in sentential contexts that "render a central aspect of its
dominant meaning particularly salient" (p. 333). These results also indicate that
lexical access can be inﬂuenced by dominance and context, but the actual
mechanism responsible for this type of priming has yet to be determined.

Duffy, Morris and Rayner (1988) found a similar pattern of results and
interpreted them as favoring a reordered-access model of lexical access for
ambiguous words instead of a selective-access or purely exhaustive model. In
their study, subjects' eye movements were monitored as they read sentences
containing lexically ambiguous words that were either equibiased (both
meanings equally frequent) or non-equibiased. These ambiguous words
appeared in sentences that had a disambiguating clause either preceding or
following the ambiguous word. Duffy et al. used an eye movement paradigm to
measure gaze durations (all consecutive fixations made on the target word
beginning with the ﬁrst ﬁxation on the word and ending with the last ﬁxation
before the eyes moved to another word, p. 436) on the ambiguous word as well
as on later parts of the sentence. For sentences that had a non-equibiased
ambiguous word, the disambiguating clause always favored the less frequent
meaning. Previous studies (Duffy and Rayner, 1988) had shown that gaze

durations were longer for equibiased but not for non-equibiased ambiguous

9

words when they were preceded by a context that did not decide between the
two meanings. Duffy et al. predicted, then, that if an autonomous access model
is correct, gaze durations should be longer for both equibiased and
nonequibiased ambiguous words when the disambiguating clause precedes the
target word. However, if the reordered-access model is correct, gaze durations
should differ depending on whether the ambiguous word is equibiased or non-
equibiased. For equibiased ambiguities, when the disambiguating clause comes
before the target word, the appropriate meaning should become available before
the inappropriate meaning and should be incorporated into the sentence without
any difficulty. As a result, the equibiased ambiguity will resemble a non-
equibiased word when the dominant meaning is correct, with no difference in
gaze durations for equibiased ambiguous words and their controls. However,
with non-equibiased ambiguous words, when the disambiguating clause comes
before the target word, because the clause supports the less frequent meaning
of the ambiguous word, the less frequent meaning should become more
available than normal and thus require the system to select between the two
meanings. As a result, the non-equibiased ambiguities should resemble
equibiased ambiguous words with longer gaze durations compared to their
controls.

The pattern of gaze durations supported the reordered-access model.
For equibiased words, gaze durations on the target word in the after condition

(disambiguating information coming after the ambiguous word) were longer

10
compared to their controls, but in the before condition (disambiguating
information coming before the ambiguous word), gaze durations on the target
word and control were equivalent For non-equibiased words, gaze durations
were lengthened compared to controls in both the before and after conditions.
Gaze durations in the disambiguating region were longer when they followed an
ambiguous word compared to when they followed a neutral control word. Duffy
et al. concluded that the pattern of results in the before condition best ﬁt a
reordered model of lexical access. In the case of the equibiased ambiguous
words, the appropriate meaning becomes available ﬁrst and is easily
incorporated into the sentence without any competition from the inappropriate
meaning. When the ambiguous word is non-equibiased, because the less
frequent meaning becomes available sooner than normal, the system has to
select between two meanings, resulting in the longer gaze durations. These
results are incompatible with an exhaustive model that predicts that all meanings
should become available for both equibiased and non-equibiased ambiguous
words, thus resulting in longer gaze durations for both. Moreover, a purely
selective-access model cannot account for the data because of the pattern of
gaze durations in the post-target region of the sentence for non-equibiased
ambiguous words in the before condition. Here, gaze durations on the
ambiguous word were longer compared to controls. As a result, Duffy, et al.
concluded that the reordered-access model best describes the data. All

meanings of an ambiguity are accessed, but activation levels of alternative

11
meanings differ depending on the frequency of occurrence of the alternative
meanings and the semantic bias in any prior context.

In a follow-up study, Rayner, Pacht and Duffy (in press) attempted to
eliminate what they termed the subordinate-bias effect for non-equibiased
ambiguous words (longer gaze durations on non-equibiased ambiguous words
when the context favors the subordinate meaning) by increasing the contextual
bias of the sentence in favor of the subordinate meaning. In Experiment 1, Duffy
et al. attempted to inﬂuence the bias towards the subordinate meaning of non-
equibiased ambiguous words by having subjects participate in a paired-
associate task where an ambiguous word was paired with a word related to its
subordinate meaning. Thus, the pair-associate task should help to promote the
activation level of the subordinate meaning of the ambiguous word relative to the
dominant meaning. Following the paired-associate learning task, subjects read
sentences in which the context favored the subordinate meaning of an
ambiguous word while their eye movements were monitored. Rayner et al.
compared ﬁxation times on the ambiguous word when the subject learned the
paired-associate list compared to when they did not as well as ﬁxation times on
the ambiguous words relative to their controls for subjects that either had or had
not performed the paired-associate list task. Rayner, Pacht and Duffy found that
gaze durations did not differ for ambiguous words that were a part of the paired-
associate list compared to ambiguous words that were not Additionally, these

gaze durations did not differ from non-ambiguous control words, indicating that

12
the dominant meaning was still competing with the subordinate for ﬁnal selection
(p. 18).

In Experiment 2, Rayner, Pacht and Duffy again attempted to eliminate
the subordinate-bias effect by using a slightly different task. Concerned that the
effects of having performed the paired-associate list task did not carry over to
the episodically-dissimilar sentence-reading task, they designed a task where
the subjects were biased towards the subordinate meaning of the ambiguous
word while their gaze durations were measured. Subjects read a paragraph of
text which biased the subordinate meaning of an ambiguous word that occurred
towards the end of the paragraph. In some cases, the paragraph contained a
prior occurrence of the ambiguous word also disambiguated towards its
subordinate meaning. They then recorded gaze durations for the ambiguous
word at both the earlier and later encounters and compared them to their
controls. Rayner et al. found longer gaze durations on the ambiguous word at
both occurrences relative to non-ambiguous controls, again indicating that the
dominant meaning of the ambiguity was competing with the subordinate meaning
for ﬁnal selection. This research seems to support a view of lexical access that
is exhaustive, but is inﬂuenced by frequency of meaning and prior context.
Lexical access is exhaustive in the sense that until one meaning has been
integrated into the sentence, all meanings are accessed. However, degree of
activation of meaning is inﬂuenced by its frequency and by context, with higher

frequency occurring meanings being activated sooner than lower ones when the

13
context biases the higher frequency meaning, and with both meanings being
activated in a context that biases the lower frequency meaning. An important
point here is that a selective access account was not supported because the
subordinate bias effect was not eliminated.

Work by Serena, Pacht and Rayner (1992) favored a reordered or
integration model of lexical access. Concerned that subordinate meanings of
ambiguous words could be treated as low-frequency unambiguous words, they
compared ambiguous words to both high-frequency (HF) and low-frequency (LF)
control words. They found that initial gaze durations on biased ambiguous words
when the context biased the subordinate meaning of the ambiguous word did not
differ from the amount of time they spent on LF controls. However, total ﬁxation
time on ambiguous words relative to controls differed signiﬁcantly across the
three conditions (longest in the ambiguous condition, shorter in the LF condition,
and shortest in the HF condition). Thus, subjects were not treating ambiguous
words as low-frequency unambiguous words. They concluded that when the
subordinate meaning of an ambiguity is instantiated by context, subjects will
spend more time processing the ambiguous word relative to controls matched for
frequency, and that this pattern of results cannot be explained by a purely
selective or exhaustive model of lexical access. The longer ﬁxation times in the
subordinate bias condition are explained by the reordered model as competition
between me dominant and subordinate meanings because of the higher than

normal activation of the subordinate meaning. On the other hand, according to

14
the integration model, because the dominant meaning is incorrectly integrated
into the sentence, the longer gaze durations are the result of the reintegration of
the subordinate meaning of the ambiguous word. Sereno, Pacht and Rayner go
on to posit that the reordered model supports a more interactive view of the
memory system, while the integration model favors a more modular view, with
post-access ambiguity resolution.

Recent work by Tabossi and Zardon (1993) also suggests that both
context and frequency of occurrence inﬂuence access of meaning for ambiguous
words. In an experiment similar to Experiment 3 in Tabossi (1988) but with
target presentation at 100 ms before the end of the ambiguity, they found that
when context biases the subordinate meaning of the ambiguity, both meanings
of the ambiguity were activated. This is an interesting ﬁnding because in this
case, the effect of context causes both meanings to be highly active
simultaneously which resembles multiple access and adds further support to the
conclusions drawn by Serena, Pacht, and Rayner (1988).
W

The research I have described thus far supports the reordered access
model. However, in a recent study, Simpson and Krueger (1991) claim to have
evidence for a purely selective-access model, even for subordinate meanings.
Because my own research is an attempt to replicate and extend their work, I will
review their research in some detail.

Simpson and Krueger begin by assuming that in the absence of preceding

 

15

biasing context, the dominant meaning of an ambiguous word is accessed
before the subordinate meaning (in the case of non-equibiased words). In their
experiment, subjects named targets after reading a sentence that appeared on a
computer monitor. Sentences either biased the dominant meaning of the
ambiguous word, the subordinate meaning of the ambiguous word, or were
neutral with respect to meaning of the ambiguous word. Targets varied by either
being related to the dominant or the subordinate meaning of the ambiguous
word. Interstimulus interval (I.S.l.) was also varied: the target appeared at
either 0, 300, or 700 msec after presentation of the ambiguous word. Sentence-
type (dominant, subordinate, ambiguous) and I.S.I. were between-subject
variables, and target dominance and relatedness were within-subject variables.

Simpson and Krueger (1991) found that for neutral sentences and at 0
msec I.S.I., only the dominant meaning of the ambiguous word was activated as
seen in facilitation for targets related to the dominant meaning. However, at 300
msec. I.S.I., there was facilitation for both dominant and subordinate meaning
related targets. Finally, at 700 msec I.S.I., again there was only facilitation for
targets related to the dominant meaning of the ambiguous word. Simpson and
Krueger concluded that in the absence of biasing context, the dominant or most
frequent meaning of an ambiguous word will be accessed ﬁrst and subsequently
selected though the subordinate meaning is accessed (e.g. at 300 ms).

In biasing contexts, however, Simpson and Krueger (1991) found a

striking pattern of results. If the context biased the dominant meaning of the

16

ambiguous word, facilitation occurred at all lSI's for targets related to the
dominant meaning of the ambiguous word. Targets related to the subordinate
meaning of an ambiguous word showed no priming. Moreover, the pattern of
results reversed for sentences biasing the subordinate meaning of the
ambiguous word. In these cases, targets related to the subordinate meaning of
the ambiguous word showed priming whereas targets related to the dominant
meaning of the ambiguous word showed no such priming. This pattern also held
across the I.S.l. manipulation. Simpson and Krueger concluded that in some
cases, contextual information can immediately affect the access of the
appropriate meaning of an ambiguous word regardless of the frequency of that
meaning. These conclusions are inconsistent with the other work described
above, which shows clear effects of meaning dominance (T abossi, 1988; Duffy,
Morris, and Rayner, 1988; Sereno, Pacht and Rayner, 1992; and Tabossi and
Zardon, 1993).
Cl" [5' III {IEEIJ

Although Simpson and Krueger (1991) claimed to show selective access
of meaning for ambiguous words through sentence context in their paradigm,
there are some concerns about the design used in their study as well as the
implementation of their experiment which might account for their pattern of
results.

In Experiment 1 of their study, sentence-context and I.S.l. were between-

subject variables. Thus, subjects saw only one type of sentence: dominant

17

biased, subordinate biased, or ambiguous, at a particular I.S.I. (O, 300, 700
msec) on every trial. As a result, it is possible that subjects were able to form an
expectation about what type of sentence and target they would receive, enabling
them to develop the strategy of consciously accessing the appropriate meaning
of the ambiguous word. This tendency was likely exacerbated by the absence of
any ﬁller items in the stimulus presentation lists. If subjects were able to adopt
this type of strategy, Simpson and Krueger (1991) would be able to argue
against a modular (multiple access) view of the memory system because in this
case subjects would be using contextual information to select the appropriate
meaning of the ambiguous word. However, the use of I.S.l. in their experiment
suggests that maybe context was not inﬂuencing access, but that Simpson and
Krueger were probing after some period in time when both meanings of the
ambiguous word were accessed.

The second concern with Simpson and Krueger's (1991) manipulation is
the manner in which the target word was presented and the use of I.S.I.
Subjects read aloud a sentence that was presented to them on a computer
screen. As the subject ﬁnished pronouncing out loud the ambiguous word, the
experimenter pressed a button that presented the target word the subject was to
name at one of the I.S.l. times. What makes this type of paradigm questionable
is the fact that the 0 ms I.S.I. does not take into consideration naming latency
and pronunciation duration. Because subjects had to pronounce the word

before the target word appeared, on average, 400 to 500 ms (Siedenberg and

18
Waters, 1984) elapsed between the time when the subject ﬁrst saw the
ambiguous word and when he or she began pronouncing the target word. This
latency was then followed by an average pronunciating duration of 400 to 600
ms before the experimenter pressed the button to present the target word. As a
result, some 900 to 1,000 ms, on average, elapsed between the time when the
subject ﬁrst saw the ambiguous word and when the target word was presented in
the 0 ms I.S.l. condition. Thus, Simpson and Krueger's earliest probe time
probably does not tap into the earliest time at which lexical processing of the
ambiguous word is occurring. This fact, in combination with the blocked
presentation of context, leads to the possibility that subjects could consciously
generate expectations concerning which meaning of the ambiguity would be
probed.

Simpson and Krueger defended their use of 0 msec I.S.l. by the pattern of
data for the ambiguous sentences. Because ambiguous sentences seemed to
be following the same pattern of results at 0 msec I.S.l. as they did in an
experiment by Simpson and Burgess (1985), where they used 16-750 msec
S.O.A, they claimed that it is unlikely that they missed some earlier time when
both meanings of the ambiguous word were accessed in the present experiment.
However, this fact does not address the naming and pronunciation latency
problem caused by this subject naming paradigm. Thus, it may be that using
I.S.l. instead of S.O.A and a subject-paced target presentation paradigm missed

an earlier period during lexical ambiguity processing when both meanings are

19
simultaneously accessed.
Experiment 1

The purpose of the present experiment was two-fold. First, this
experiment was a replication of Simpson and Krueger (1991) using their stimulus
materials and a more controlled cross—modal priming paradigm. This paradigm
allowed for more precise control of target presentation using S.O.A instead of
I.S.l. and alleviated the need for experimenter-paced presentation of the target.
Additionally, the design of the experiment addressed the potential blocked
context problem. Subjects were randomly presented each of the three types of
sentence contexts (dominant, neutral, and subordinate). The second purpose
was to test the more recent theories of lexical access as put forth by Tabossi
(1988), Tabossi and Zardon (1993), and Rayner, et al. (1988, 1992) with the
results of Simpson and Krueger and the present experiment.

In Experiment 1, subjects listened to a sentence played through a speaker
and named a target presented on a computer screen. Sentence context either
biased the dominant, subordinate, or neither meaning of the ambiguous word
and targets were either related or unrelated to the dominant or subordinate
meaning of the ambiguous word. Sentence context and what I will call target
relation (related/unrelated) and target ambiguity strength
(dominant/subordinate) varied within subjects, and S.OA (0, 200, and 600
msec) varied between subjects. Moreover, ﬁller sentences occurred within all

presentation lists.

2 0

Using this experimental paradigm, if Simpson and Krueger (1991) are
correct and biasing context can allow selective access of a particular meaning of
an ambiguous word, then we should ﬁnd a similar pattern of facilitation of targets
related to the biased meaning of the ambiguous word in the dominant biased
and subordinate biased sentence context conditions at all S.O.A levels.

According to the multiple-access theory which does not incorporate a
mechanism to deal with the frequency of the alternative meanings of an
ambiguous word, both meanings of the ambiguous word will be accessed and at
some later time, selection of the appropriate meaning—mediated by sentence
context-will occur. Thus, I should ﬁnd facilitation for targets related to both
meanings at 0 msec S.O.A, and only facilitation for targets related to the
appropriate meaning of the ambiguous word at some later time. Again, because
the notion of dominant and subordinate meaning of ambiguous words is
irrelevant in this theory, the same pattern of results is predicted for both of the
biasing context conditions.

According to a reordered multiple-access theory of lexical ambiguity
which does make assumptions about the effects of meaning dominance, both
meanings of the ambiguous word should be accessed in a neutral context and at
0 S.O.A. with the dominant meaning having a higher activation level (level of
facilitation) than the subordinate meaning. At some subsequent selection point,
the dominant meaning should be selected with activation of the subordinate

meaning falling off. Thus, at longer I.S.l.'s, targets related to only the dominant

2 1
meaning should show facilitation.

In the dominant meaning biasing context condition, at 0 msec S.O.A. both
meanings of the ambiguous word will be accessed, again with the dominant
meaning having a higher level of activation. At the period of selection, only the
dominant meaning will show activation. This situation will be evident in
facilitation for target words related to both meanings in the former condition, and
only the dominant meaning in the latter condition.

In the subordinate meaning biasing context condition, targets related to
both dominant and subordinate meanings will show facilitation at 0 msec S.O.A.
However, because sentence context is posited to aid in the selection of the
appropriate meaning of the ambiguous word, at some selection time later, there
should be some competition between dominant and subordinate meanings. This
competition would be the result of the subordinate meaning consciously
remaining active while the dominant meaning retains its a priori activation. This
pattern of results could result in inhibition of target naming at some time later
than 0 msec S.O.A. with context effects ultimately resulting in activation of only
the subordinate meaning. Rayner and his colleagues (1988) have shown such
competition between dominant and subordinate meanings of ambiguous words
in their studies.

Method
Sum; One hundred and eight introductory psychology students at

Michigan State University participated as subjects in this experiment. Subjects

22
received partial credit in their introductory psychology courses for their
participation and all were native English speakers and had normal or corrected
to normal vision and hearing.

Bantam Response times to pronounce the target word were recorded
by a 486-33 IBM-PC clone micro computer with a VGA monitor. Digitized
speech sentences were presented to the subjects through a speaker using the
Computerized Speech Laboratory (CSL) from Kay Elemetrics Corporation
interfaced with the computer. Subjects used a button-box also interfaced with
the computer to start each trial. A voice-activated relay responded to subjects'
naming of the target word through a unidirectional microphone (Shure Prologue)
supplied by CSL.

Materials, Thirty-two experimental sentences were used from the
Simpson and Krueger (1991) study. In that study, 32 ambiguous words were
selected from the homograph association norms of Nelson, McEvoy, Walling and
\Nneeler (1980). Two associates were selected for each ambiguous word, one
representing the dominant and one the subordinate meaning. Three sentences
were created for each ambiguous word. One sentence biased the dominant
meaning of the word, another the subordinate meaning, and a third biased
neither (neutral). For the unrelated target condition, sentences were paired and
the related target words were interchanged between members of the pair. Four
additional experimental sentences were created to match the 32 experimental

trials from the Simpson and Krueger study to bring the total number of

23
experimental sentences to 36, a number divisible by the number of within subject
conditions. All of these sentences were equated for length and complexity, and
target words were balanced for length and frequency of occurence. These
sentences and their target words are included as Appendix A

Seventy-two ﬁller sentences were created where the last word in the
sentence was not obviously ambiguous. These sentences were equated in
length and complexity with the experimental sentences. For half of the ﬁller
sentences, a related target word was paired with the last word in the sentence
and for the other half, an unrelated target word was paired with the last word in
the sentence. Ten practice trials were created similar to the ﬁller sentences.
The last word in each of the practice trials was a neutral non-ambiguous word.
Half of the sentences had a related and half had an unrelated target word (see
Appendix A).

The 118 sentences were recorded by a native English male speaker and
digitized at a sampling rate of 10 KHz. A tag was placed at the onset of the last
word of each sentence using a waveform editor. The tag was used by the
computer as the point at which to display the target word on the monitor
depending on what S.O.A. had been selected for that subject. The tag was
inaudible to the subject.

Mme, Subjects were tested individually. Subjects sat in front of the
computer monitor, microphone, button-box and small speaker used to play the

digitized sentences. The experimenter read the instructions, explaining to them

24
their task for the experiment. They were told that they would hear a sentence
being played through the speaker and were to pronounce the word that
appeared on the computer monitor as quickly as possible without sacriﬁcing
accuracy.

During each trial of the experiment, subjects saw a prompt instructing
them to press a button on the button box to begin each trial. Next, a sentence
was played through the speaker and subjects listened to the sentence while
looking at the computer monitor. When the target word appeared on the
computer monitor, subjects pronounced the word as quickly as possible Into the
microphone. The target word appeared in the center of the monitor in capital
letters. Reaction time was measured from onset of the target word until the
subject pronounced the target word. After pronouncing the word, subjects were
again prompted to press a button on the button-box to proceed to the next trial.
If the subject did not pronounce the target within 5 seconds, the target word
disappeared, and the subject was prompted to push the button for the next trial.
Subjects were run in a practice block of ten trials to familiarize themselves with
the task and procedure. After the practice trials, the experimenter answered any
questions the subjects had about the procedure and then the subject proceeded
to the 108 experimental trials. After the experiment, subjects were debriefed by
the experimenter and thanked for their participation in the experiment. The

entire session lasted about 20 minutes.

W The design of the experiment was a 3 (Sentence

25
context) X 2 (Target relationship) X 2 (Target ambiguity strength) X 3 (Stimulus
onset asynchrony) factorial. Sentence context (neutral, biased dominant and
biased subordinate), target relationship (related and unrelated), and target
ambiguity strength (dominant and subordinate) were within subject variables.
Stimulus onset asynchrony (0ms, 200ms and 600ms) was a between subjects
variable. Each subject heard three sentences in each of the 12 within subject
cells and all 72 of the ﬁller items at one particular S.O.A An Omnibus ANOVA
that included all factors was conducted on mean naming latencies.
323ml;

Mean naming latencies at each S.O.A are shown in Table 1. Errors in
which the subject incorrectly named the target word never occurred. Trials that
were eliminated were those on which the subject made an extraneous noise
before pronouncing the target word which triggered the voice key, as noted by
the experimenter, or those trials on which the subject pronounced the target
word in less than 250 ms or more than 1500 ms. The analyses are based on

98.6% of the data.

 

Insert Table 1 Here

 

There was a main effect of S.O.A. F(2,105)=6.7858, MSe=56155, p<.005.
Naming latencies to targets decreased as S.O.A increased: 568 ms, 530 ms,

and 510 ms for 0, 200, and 600 ms S.O.A, respectively. There was a also main

2 6
effect of context, F (2,2)=9.7606, MSe=1839, p<.005. Naming latencies to
targets were fastest in the dominant context condition (530 ms), slower in the
subordinate context condition (535 ms) and slowest in the neutral context
condition (543 ms). There was also a main effect of target relation,
F(1,2)=11.2517, MSe=1962, p<.005. Naming latencies were faster for related
targets (532 ms) than for unrelated targets (540 ms). Finally there was a main
effect of target ambiguity strength, F (1 ,2)=8.7447, MSe=2005, p<.005. Targets
related to the dominant meaning of the ambiguous word had faster naming
latencies than targets related to the subordinate meaning of the ambiguous
word, 532 ms and 540 ms, respectively. The 3-way interaction of target relation
x target relation strength x S.O.A. was signiﬁcant, F(2,105)=3.4147, MSe=2089,
p<.05. As can be seen in Figure 1, at 0 and 200 ms S.O.A. the target related to
the meaning biased by the subordinate context shows greater priming than the
target related to the meaning biased by the dominant context. However, at 600
ms, the target related to the meaning biased by the dominant context shows
more priming than the target related to the meaning biased by the subordinate
meaning. No other interaction was signiﬁcant (all F's < 1).

Additional ANOVA's were conducted for each level of S.O.A. At 0 ms
S.O.A. there were no signiﬁcant main effects and no signiﬁcant interactions,
indicating that the 9 ms difference in priming between subordinate meaning
related targets and dominant meaning related targets as shown in Figure 1 was

not signiﬁcant. At 200 ms S.O.A. there was a signiﬁcant main effect of target

2 7
relation, F (1 ,35)=8.3226, MSe=1203, p<.005. Naming latencies were faster for
related targets (524 ms) than for unrelated targets (534 ms). There were no
signiﬁcant interactions, indicating that the 7 ms difference in priming between
the subordinate meaning related targets and the dominant meaning related
targets as shown in Figure 1 was not signiﬁcant. At 600 ms S.O.A. there was a
signiﬁcant main effect of target relation strength, F (1 ,35)=1 1 .266, MSe=1611,
p<.05. Naming latencies were faster for targets related to the dominant meaning
(503 ms) than for targets related to the subordinate meaning (516 ms). There
were no other signiﬁcant main effects. There was also a signiﬁcant interaction of
target relation x target relation strength F(1,35)=4.1964, MSe=1315, p<.05. By
600 ms, the 14 ms difference in priming between dominant meaning related
targets and subordinate meaning related targets is signiﬁcant, suggesting that
only the dominant meaning is accessed. There were no other signiﬁcant
interactions.
[1' .

The results of Experiment 1 contradicted the earlier ﬁndings of Simpson
and Krueger (1991) in that there was no evidence for selective access.
According to the selective access theory, a particular meaning of an ambiguous
word can be accessed in an appropriately biased context. Thus, in this
experiment, the dominant target should have shown priming in contexts that
primed the dominant meaning across all S.O.A. conditions. Likewise, the

subordinate target should have shown priming in contexts that primed the

2 8
subordinate meaning in all S.O.A. conditions. Whereas Simpson and Krueger
found this pattern of results in their experiments, there was no effect of context
at any of the S.OA conditions in the present experiment. Even though at 600
ms, the dominant meaning showed priming over the subordinate, this effect was
seen for both the dominant and subordinate contexts. Thus, it appears that all of
the effects were dependent on the dominance of meaning of the ambiguous
word. At 0 and 200 ms S.O.A. there was no signiﬁcant difference in priming for
the dominant or the subordinate target. However, at 600 ms, the dominant
related target was signiﬁcantly faster than the subordinate related target. This
pattern of data at 600 ms suggests an effect at the lexical level of
representation.

The pattern of data does not ﬁt well into a reordered theory of lexical
access, either. According to this theory, for nonequibiased ambiguous words,
the dominant meaning should show facilitation in ambiguous contexts, followed
in time by the subordinate meaning. Subsequently, the appropriate meaning is
selected based on some post-access process as determined by the meaning of
the sentence. In this experiment, however, there was no differential activation of
target words in the ambiguous context condition at any of the S.OA conditions.

The pattern of data in Experiment 1 favors a multiple access model of
ambiguity resolution account. As stated earlier, a multiple access model predicts
that context will have no effect on meaning access at earlier stages of

processing, with the appropriate meaning being selected later. This was the

2 9

pattern of facilitation found in Experiment 1 where at 0 ms S.O.A. there was no
activation for either the dominant or the subordinate target, with a rise in
facilitation over time so that both targets showed some facilitation at 200 ms
S.O.A Finally, only the dominant target showed facilitation at 600 ms S.OA
The overall pattern supports a modularity effect because of the lack of an effect
of context and only an effect of dominance at 600 ms S.O.A

However, there is an important difference between the procedures used in
this experiment and in previous priming studies of lexical ambiguity. In previous
experiments (Seidenberg, et al. 1979; Tanenhaus et al. 1979; Simpson and
Krueger, 1991), researchers used what was really I.S.I. instead of S.O.A. even
though they referred to the manipulation as S.O.A In their experiments, the
marker used to trigger presentation of the target word was located at the end of
the ambiguous word. Thus, the period of time between the ambiguous word and
the target word was not synchronous with the onset of the ambiguous word. In
other words, 0 ms S.O.A in these experiments meant that the target word was
presented immediately following presentation of the ambiguous word. In
contrast, 0 ms S.OA for the present experiment meant that the ambiguous word
began and the target word appeared on the monitor at the same time.

An analysis of the average duration of the ambiguous words used in this
experiment found that they lasted between 500 and 600 ms. Consequently, the
600 ms S.O.A. manipulation in this experiment was closer to the 0 ms "S.O.A."

used in the Simpson and Krueger experiments, and the earlier S.OA conditions

3 0
were tapping into the processes of ambiguity resolution that take place before
the end of the word.

There may be several reasons for why I failed to replicate Simpson and
Krueger, 1991. As stated, there were no interaction effects between context and
target dominance or target strength, nor was there an interaction between target
relation and target strength at 0 ms or 200 ms S.O.A. However, because the
subject was actually trying to pronounce the target word during the time that the
ambiguous word was being spoken, interference between naming the target
word and listening to the spoken ambiguous word could have occurred, resulting
in the lack of any context effects. There was an interaction between target
relation and target strength at 600 ms S.O.A The ﬁnding that there was an
effect of target dominance only at the 600 ms S.OA condition indicates that it
takes some period of time for the dominant meaning to be accessed but may
indicate a difference in access by the end of the ambiguous word. Thus, it
could be that dominance of the target plays an early role in meaning access with
context effects occuring at some later time.

Finally, in Experiment 1, it was possible that subjects were not paying
close attention to the sentences before pronouncing the target word. In the
experiment, subjects listened to the sentence as it was played through the
speaker. However, there were no checks to make sure that subjects were not
ignoring the sentence, thereby adopting the strategy of just saying the target

word as quickly as they could when it was presented on the monitor. If this

3 1
strategy was adopted by subjects, it becomes difﬁcult to examine the effects of
sentence context.
Experiment 2
Experiment 1 was a failed attempt to replicate Simpson and Krueger's
(1991) experiments. However, it cannot be ruled out that I failed to replicate
Simpson and Krueger because of the difference in the S.O.A manipulation.
Consequently, Experiment 2 examined the effects of context on lexical ambiguity
resolution using I.S.I. instead of S.O.A. In this experiment, markers were placed
at the end of the ambiguous word, and I.S.I.'s of 0, 300 and 700 ms were used
as in Simpson and Krueger's experiments. Additionally, because it cannot be
ruled out that subjects were not paying close attention to the sentences before
pronouncing the target word, an additional task whereby subjects were randomly
prompted to repeat verbatim the previous sentence was used. The design,
materials and procedure were the same as Experiment 1.
Method

Subjects, One hundred and eight introductory psychology students at
Michigan State University participated as subjects in this experiment with the
restriction that they did not participate in the ﬁrst experiment. Subjects received
partial credit in their introductory psychology courses for their participation and
all were native English speakers and had normal or corrected to normal vision
and hearing.

Malling. The apparatus used in this experiment was the same as that

3 2
used in Experiment 1.

W The design and materials used in this experiment
were the same as in Experiment 1, except the tag used to present target words
was placed at the end of the last word in the sentence so that I.S.l. could be
manipulated. Also, a cue was randomly added to 35 of the 108 trials (31%) that
prompted the subjects to repeat a sentence occasionally throughout the block of
experimental trials.

m The procedure was the same as in Experiment 1. As before,
on each trial, subjects listened to a sentence and then named the target word
out loud. After pronouncing the word, subjects were either prompted to press a
button on the button-box to proceed to the next trial, or were prompted to repeat,
to the best of their memory, the sentence they had just heard. The experimenter
recorded whether the subject correctly recalled the sentence. Subjects ran a
practice block of ten trials to familiarize themselves with the task and procedure.
Of the ten practice trials, two trials required that the subject repeat the sentence.
After the practice trials, the experimenter answered any questions about the
procedure before the subject proceeded to the 108 experimental trials. After the
experiment, subjects were debriefed by the experimenter and thanked for their
participation in the experiment. The entire session lasted about 30 minutes.
Results

Mean naming latencies at each I.S.I. for the Omnibus ANOVA for

Experiment 2 were obtained and are shown in Table 2. Errors in which the

3 3
subject incorrectly named the target word did not occur. Subjects who failed to
recall a quarter of the repeated trials were replaced. Trials were eliminated if the
subject made an exbaneous noise before pronouncing the target word which
triggered the voice key, as noted by the experimenter, or if the subject
pronounced the target word in less than 250 ms or failed to pronounce the target

word in less than 1500 ms. The analyses included 98.2% of the data.

 

Insert Table 2 Here

 

Unlike Experiment 1, there was no main effect of I.S.I., p > .05. However,
as in Experiment 1, there was a main effect of context, F (2,105)=3.974,
MSe=2765, p < .05. Naming latencies to targets were equivalent in the
subordinate context condition and the dominant context condition (513 and 516
ms; p>.05), and slower in the neutral context condition (523 ms, p<.05). Thus, it
appears that having some type of meaningful sentence speeds up naming of a
target. There was also a main effect of target strength, F(1,2)=9.0575,
MSe=1622, p < .005. Dominant targets (514 ms) were named faster than
subordinate targets (521 ms), regardless of the context of the sentence or the
relationship of the target to the ambiguous word. While none of the interactions
were significant, the interaction of context X target relation was marginal,
F(2,4)=2.4350, MSe=2392, p = .08. In the neutral context, there was more

priming (unrelated-related target) than in the dominant context, or the

34
subordinate context (12, -2, and 1.22 ms, respectively). There also was a
marginal interaction of context X target strength, F(4,2)=2.6145, MSe=2399, p =
.07. In the neutral and dominant contexts, dominant targets were named faster
than subordinate target (12.54 ms and 9.57 ms, respectively). In the
subordinate context, dominant targets were named slower than subordinate
targets (-1.92 ms). This pattern, however, held for both related and unrelated
targets.

[1' .

Again, the results of Experiment 2 failed to replicate the ﬁndings of
Simpson and Krueger (1991), even when the timing parameters more closely
matched those used in their experiments. Using I.S.l. instead of S.O.A. there
was no evidence of only the appropriate meaning biased by the sentence being
accessed; thus, there was no support for a selective access account of lexical
ambiguity resolution. The main effect of context suggested that when some
information about a particular meaning of the ambiguous word is given,
responses to name a target word are quicker. However, this speeded naming
occurs for both related and unrelated targets, as well as for dominant and
subordinate targets. While dominant targets were named faster than
subordinate targets, this effect did not vary with target relation. Thus unrelated
dominant targets were named just as quickly as related dominant targets.

The lack of priming for the appropriate meaning also carried over across

all I.S.l. conditions, suggesting that the contexts used may not have been

35

sufficient to restrict access to only one meaning of the ambiguous word.

General Discussion

Simpson and Krueger (1991) found support for a purely selective access
model of lexical ambiguity resolution. In their experiments using nonequibiased
words, when the dominant meaning was biased by the sentence, there was only
facilitation for the target word related to the dominant meaning of the ambiguous
word. When the subordinate meaning of the ambiguous word was biased by the
sentence, there was only facilitation for the target word related to the
subordinate meaning of the ambiguous word. What makes these ﬁndings so
interesting is the lack of facilitation of the dominant meaning in the presence of a
context that biases the subordinate meaning of the ambiguous word.

The experiments reported here were an attempt to replicate the ﬁndings
of Simpson and Krueger (1991) using a more controlled cross-modal priming
experimental paradigm which permitted greater control over target word
presentation. In Experiment 1, subjects listened to a sentence being played
through a speaker and pronounced a target word presented to them on a
computer monitor. Target words were either related to the dominant or
subordinate meaning of the ambiguous word in the sentence and appeared at
either 0 ms, 200, or 600 ms S.OA Unlike Simpson and Krueger, when the
dominant meaning of the ambiguous word was biased by the sentence, targets

related to both the dominant and the subordinate meanings were facilitated. In

36

3 7
fact, there was no evidence of context mediating which meaning of the
ambiguous word was accessed across all levels of S.O.A used in this
experiment. However, as indicated by the pattern of results at 600 ms S.O.A.
dominance may play an early role in accessing meaning of the ambiguous word,
followed later by context effects.

Experiment 2 was an additional attempt to replicate the earlier findings of
Simpson and Krueger (1991), using timing parameters closer to those used in
their experiments. Also, because there was a concern that subjects may have
not been paying attention to the sentences, an additional task of having subjects
randomly repeat the sentence was used. This task presumably caused subjects
to process the sentences more carefully. Using I.S.l. instead of S.O.A. I again
found no evidence of context mediating which meaning of an ambiguous word
was accessed. However, the main effect of context suggested that when some
information about a particular meaning of an ambiguous word is given, target
word (both related and unrelated) naming latencies are faster. Nevertheless,
Experiment 2 failed to provide any information about how context may inﬂuence
access of meaning of an ambiguous word.

Replication of Simpson and Krueger (1991) was prompted not only
because of their ﬁndings supporting a selective access theory of lexical
ambiguity resolution, but also because of the experimental paradigm they used.
In their experiments, subjects read a sentence presented to them on a computer

monitor and when they had ﬁnished pronouncing the last word in the sentence

3 8
(which was the ambiguous word), an experimenter pressed a button that
presented the subject with the target word they were to pronounce. This type of
task, I argued, did not provide the same amount of control over timing as cross-
modal priming or eye-movement monitoring tasks because of naming and
pronounciation latencies. In their task, subjects were presented with the target
word as much as one second after initially viewing the ambiguous word. Thus, it
remains unclear whether or not subjects initially accessed both meanings of the
ambiguous word and by the time they were presented with the target word,
context had aided in the selection of the appropriate meaning. Using the same
stimuli and a cross-modal priming paradigm, I failed to replicate their ﬁndings.

A possible reason for this failure to ﬁnd selective access of the
appropriate meaning of an ambiguous word may be due to the timing parameters
used in this experiment. Having more precise control over when the target word
is presented may have uncovered periods of time when context has no effect
over access of meaning. Assuming that the target presentation times used by
Simpson and Krueger would have tapped into the periods of time when there are
differential effects of contexts, their effects may be due to the lack of control over
when the target is actually presented.

An additional reason why I failed to replicate the ﬁndings of Simpson and
Krueger (1991) may be because of the change in modality in the present
experiments. In Simpson and Krueger's work, subjects read aloud the sentence,

and then read aloud the target word. In the present experiments, subjects

3 9

listened to the sentence and then read aloud the target word. While many
experiments examining lexical ambiguity resolution have used such a cross-
modal priming paradigm (Seidenberg, et al, 1983; Tannenhaus, et al 1979;
Tabossi, 1988), it is possible that the discrepancy in the results can be
accounted for by the discrepancy in the tasks. In other words, in the present
experiments, cognitive processes involved in the auditory domain are not the
same as those involved in visual processing used in Simpson and Krueger's
experiments. This differences in cognitive processing may be qualitative in that
during reading, as in the Simpson and Krueger task, subjects are forced to
process the context of the sentence, while listening to a sentence may allow for
less processing of the context of the sentence. Consequently, the lack of
priming of the appropriate meaning of an ambiguous word could be because of
the auditory processing of the sentences.

Finally, an additional difference between the two studies lies in the design
of the studies. Simpson and Krueger (1991) manipulated context and I.S.I. as a
between subjects variable while I manipulated context as a within subjects
variable and S.O.A and I.S.l. as between subjects variables. Because in the
Simpson and Krueger's study subjects saw only one type of sentence context,
they had the greatest chances of developing a strategy for naming the target
word, namely, there is an ambiguous word at the end of each sentence, and the
appropriate meaning of that ambiguous word is occasionally biased by the

sentence. This design is the strongest test of modularity because it allows

40
subjects the opportunity of recognizing a pattern in the trials of the experiment
and if possible, inﬂuence which meaning of the ambiguous word is accessed.
Simpson and Krueger interpreted their ﬁndings as evidence against a modularity
view because they found facilitating for only the appropriate meaning of the
ambiguous word as determined by sentence context (selective access). In the
present experiment, because sentence context was a within subjects
manipulation, it is difﬁcult to determine whether or not me effect of context was
simply diluted in the present case and subjects were not able to take more
advantage of them, or if there is simply no effects of context on the access of the
meanings of an ambiguous word and that context aids in selection of the
appropriate meaning.

Although the data in my experiment did not support it, a reordered-access
account (Duffy, et al., 1988) probably remains the most favorable theory of
lexical ambiguity resolution given the results of other studies. According to this
theory, the dominant meaning of an unequibiased ambiguous word is accessed
quickly, and is instantiated in to the sentence if it is appropriate. Under these
conditions, no trace of the subordinate meaning may exist because of the quick
access of the dominant meaning, and this can look like selective access of the
dominant meaning. The results of the present Experiment 1 are consistent with
the rapid access of the dominant meaning. When the subordinate meaning of
an unequibiased ambiguous word is biased by the sentence, activation levels of

the subordinate meaning increase enough to compete with the activation level of

4 1
the dominant meaning. This competition between the dominant and the
subordinate meanings slows down the selection process of a particular meaning
of the ambiguous word.

For equibiased ambiguous words in a neutral context, again there is
competition between the different meanings of the ambiguous word because of
equal levels of activation. However, when one meaning is biased by the
sentence, the activation level of that meaning increases so that it is access
before the other meaning. This quicker selection of the appropriate meaning
can resemble selective access of that meaning. In conclusion, the present
experiments failed to replicate the ﬁndings of Simpson and Krueger (1991).
Context did not mediate access of meaning of an ambiguous word.
Unfortunately, while the results did not support a selective access account, they
also did not support a reordered access account (Duffy, et al. 1988) of lexical

ambiguity resolution.

Table 1

Mean Reaction Times (RT) in ms for S.O.A X Target Ambigu'ty Strength X Target Relation X

Context
0 ms S.O.A
W W
Belated Males! Belated Unrelam
mm 554 565 562 574
um 572 573 572 585
m 563 561 559 579
M 563 566 564 579
200 ms S.O.A.
Dominantﬁclated MW
5.9m Unrelated Belem! Unrelated
991mm 520 523 513 526
mm 526 539 532 552
W 531 532 527 534
M 523 531 524 537
600 ms S.O.A.
W W
Belated M Related ”Balm
mm 491 497 520 513
mm 494 518 524 525
m 515 516 507 509
M 500 510 517 516

42

Table 2

Mean Reaction Times (RT) in ms for I.S.l. X Target Ambiguity Strength X Target Relation X

Context.

0 ms I.S.l.
W
Belm Unrelated
524 521
529 551
529 530
527 534
300 ms I.S.I.
Qcmlnmtﬁelated
Belated Unrelated
509 510
496 515
513 502
505 509
700 ms I.S.l.
Water:
Belated Madam
502 502
503 506
505 507
503 505

43

W
BEIEIQS! 2031399
538 526
537 558
522 521
532 535

W
Bglm 211919199
512 525
530 520
502 507
515 517
Subordinate Related
W M
518 506
508 523
505 517
510 515

44

Appendix A
W W
1. The princess took a carriage to the ball. BOUNCE
We decided on the size of the ball.
The boy reached up to catch the ball.
She stored her letters in the wooden chest BODY
The yomg man had a very large chest.
The mmwashaving painsinhischest
2. I tied up the papers with a band. MUSIC

We all agreed that we liked the band.

We like to listen to the marching band.

I scratched my hand aga'nst the rough bark.
lwassurprised byilsverystrange bark.

l was scared by his very loud bark.

3. With his gun, Johnny shot the ha'ry bat.
While in the basement, Johnny found a bat
Johnny proudly swung his wooden bat

We went out for dnner on our date.

l'm sue that will be a good date.

A great leader was killed on this date.

4. The man got a cramp in his calf.

The man looked carefdly at my injured calf.
The man fed corn to the hungry calf.
Thisisa brokenand mstyold spring.

DOG

BASEBALL

MONTH

COW

SUMMER

DANCE

TREASURE

RUBBER

TREE

FLY

GIRL

LEG

COIL

45

This really is not a very good spring.
This has been a cold and ra'ny spring.

5. The princess rode home in her silver coach. TEAM
We liked the looks of the new coach.
Wewonendgaveaprizetoourcoach.

The men took their positions on the diamond. GOLD
The two men dscussed the size of the diamond.

The man gave his wife a beautiful diamond.

6. l cheated by dealing from the bottom of the deck. SHIP
There was something wrong with the deck.

The men were told to mop the deck.

They ﬁnally asked me to join the club. STICK
The people were very interested in the club.

His head was hurt by the heavy club.

7. The school children walked outside for the drill. HOLE
I think we need to have a drill.

To help make the cha'r he used a drill.

The grass had grown over a whole foot. TOE
They thought that it would be a foot

I received a bad cuton myfoot

8. I like to watch football games in the fall. DOWN
I hope you don't have a bad fall.

ltrippedonthecurbandhadabadfall.

HORSE

CIRCLE

PACK

GROUP

FIRE

INCH

LEAVES

46

The cowboy thought he had the vw'nring hand. ARM
Thethoughtthe manhadaricehand.
The man bowed and kissed the lady's hand.

9. The quest'ons on that test were too hard. SOFT
I don't like it because it's too hard.

She cooled the metal until it was hard.

lwasstuckinlhe chest byhissharp horn. BLOW
Themanshowedusaveryold horn.

They called us to dinner by soundng a horn.

CARDS

EASY

DEER

47

10. My mother poured water into her steam iron. STEEL
Let's decide what to do with this iron.

The box was made of very thick iron.

I made a mistake by pressing the wrong key. LOCK

I don't think that is the right key.

Icoudnotgetinwiththewrong key.

11. The drl careftlly traced the shape of the letter. NOTE
The girl proudly showed her mother the letter.

Oh histrip myfriend sent me a letter.

I lifted it easily because it was light DARK
Mary thought that the chair was very light

I liked the color because it was light

12. Everyone thinks that man is a real nut SHELL
Thistime I decidedto pickanut

I sat down and ate a toasted nut

The worker built a ten foot wire pen. WRITE
He showed us a very well made pen.

The teacher made a mark with her pen.

SHIRT

TYPE

ALPHABET

HEAVY

CRAZY

PIG

48

13. I won the leading part in the play. GAME
We were all very surprised by the play.
Wescoredsixpointsonapassplay.

The tired woman asked for some more punch. HIT
Iwasverysurprised bythe punch.
Iwaslmockedoutbylisveryhardpunch.

14. I held my ears became of the loud ring. FINGER
Hewantedtog'vethegirl aring.

She loved towear her bearliful new ring.

I opened the letter by breaking the seal. ANIMAL
The woman sad it was a good seal.

My favorite circus act is the Ira'ned seal.

15.l rolled up my sleeve to get the shot GUN
Iwasvery worried aboutthe shot
lwasscaredbecauselheardaloudshot.

I made the rounds on my night watch. TIME
Theygotmad becauselsaton mywatch.

The rich man looked at his gold watch.

ACT

DRINK

BELL

SHUT

DOCTOR

LOOK

49

16. He blewwater athe crowd through his trunk. CAR
Ihoughthewoddcarryitinhistrunk

lputallofmy sdtcasesintohetrmk
Thednnerwasinhonorofhefamousstar. SKY
We all loved tolookathe beautiful star.

We looked up high to see the shiring star.

17. The mediaisabouttoannouncehe

members of his cabinet KITCHEN
Everyone was impressed wih he new cabinet
Jake went to he lumber store to select

the wood for his cabinet
On heir school vacation, he accounting

club towed he mint TASTE
No one was particularly impressed with he mint
Whit liked he drink until she discovered it

wasmadewihrnint

18. Robertwas excited hathe would getto

tour he car plant. FLOWER
Jasonwassuprised byhe slzeofhe plant
I liked he colors in he pot she gave me for

my new plant
I went for a jog along he muddy bank. MONEY
Billy was not happy with he destruction to the bank.

We all hought itwas an odd place to build he new bank.

ELEPHANT

MOVIE

GOVERNMENT

MONEY

BUILDING

WATER

50

Appendix B

Filler Sentences
W

A

masses-9’19

d
P

11.
12.
13.
14.
15.
16.
17.
18.
19.

21.

.Theteacherwenttohe beachwihhersister.

Thosescaedyoung boysatehe poison.

The licensed ﬁshermen were allowed to cahh seven ﬁsh.
The attendant kept all of he passengers calm.

John's house was destroyed by he high winds.

The excited parents decided to name he baby Jack.

. She accepted he offer from her boss.

The decorations were taken down and stored away.

The survivors blamed he company for the lack of lifeboats.

Keih hadhe mostmoneyleftafterourvacation.

The company made sleepwear out of non-certiﬁed fabric.
The tower is not a good place for a restaurant

The best insurance agents will offer you a better rate.
The patients discovered here was a serious emergency.
The cars were parked close to he convention center.
The parents agreed about he beneﬁts of he program.
Many students will be attending he charity dance.

The players were excited about making he team.

The dentist learned new procedures for removing tech.

. The instrument fell and got scratched.

The ballerina was not satisﬁed wih he dance sequence.

. The plant did nothrive in he dry in

. We were elated hat he magazine was accepted for

publication.

1WD.
BROTHER
SERUM
LURE
SERENE
BLOW
BOX

JOB

FAR
WATER
TRIP
CLOTH
FOOD
PRICE
DANGER
BUILDING
COMPUTER
MUSIC
CLUB
WHITE
CUT
ROUTINE
BREATHE

BOOK

24.

27.
28.

(a)
N

51

The princess loved weMng he beautiful jewelry.

. The secretary always stated her day by drinking coffee.

.Therewasanobscenemessagewrittenonhe

blackboard.
The photograph was stored between plates of glass.
None of he textbooks could answer he student's

questions.

. The door withstood he incessant pounding of he man.

.Manyra'ncoatsare madeofmaterialhatisnot

waterproof.

. The highrise on he hill crumbled during he earhquake.

. The group left messages on all around campus.

The gu'tarist hoped shake up he music industry.
The couple's young daughter is a delight
The aspiring auhor had yet to write a word.

The boating trip had been arranged very quickly.

W
1. The busdriverwassickofhe hotdays.

2. His daughter chose to attend a mldwestern college.

3.
4.

The woman moved to Chicago because of her job.

The old man distinctly remembered he war.

. Joan remembered he days when parking on campus

was easy.
The new computer was very powerful and efﬁcient.
Brendathrewa partyhatnooneattended.
Montana is he fourh largest state in his county.

RING
CAFFEINE

CHALK

WINDOW

ANSWER

ADULT

DRY
SHAKE
SCHOOL
BUSINESS
JOY
LETTER

FAST

GROOM

BEACH

HOUSE

BIKE

PAPER
JOLLY

52

9. Jake was working rights md losing sleep.

10.
11.
12.
13.
14.

15.

16.
17.
18.
19.
20.

21,

23.

24.

25.

27.
28.

31.

32.

The woman expected her parents to arrive early.
The man cla'med to have seen he Beatles.
The librarian ordered a second copy of he boolt

The high school football team ﬁnally lost a game.

The morring fog bums off qu'ckly when he sun rises.

Overdrilling in Califomia's geohermal ﬁelds is a
great concern.

The old actress remembered so many stories.

Greenland is he largest island in he world.

The chess club's advertisement got no response.

The man hadn't spoken to his friend in months.

The woman enjoyed bowling on Sundays.

Jack Lemmon won an Oscar for his role.

His wife shot him because he was evil.

Jill ate at he new restaurant on Main street

New manufacturing ideas are now used in
many comparies.

The boy crossed he read and played in he ﬁeld.

. Johnny looked up he phone number in he book.

I did notwantto run for class president

Bill went to school vvih a headache his moming.

Mchael liked to punch he refrigerator.

Football players wear short sleeves so players
can't grab hem.

Susan dreamed of being a movie star.

The young mm cried at he and of he movie.

HAPPY
SUNNY
SPEAK
DOUGHNUT
WINTER

JUMP

WELFARE
JUNK
HOPE
SABLE
KING
BOOM
READ
QUESTION
DISK

SCAN
MELODY
GHOST
HOST
TELEVISION
VEHICLE

LAUGH
BLINK

ROAST

33. The student had to write all weekend.
34. We bought new speakers last right
35. l have to call Jason toright

36. Itra'nedcatsanddogshismorning.

53

CLIP
PROFESSOR
FENCE

DISK

54
AppendixC
Practicesentences
W
1. Eightfeetofsnowfell onhe city.

2. The inauguration of he President is always held in January.

3. The Earth is closest to he Sun in January.
4. Bill and Ted didn't hink time travel was very pleasant

5. Sam's at hissed whenever I came to he house.

W
1. Deja vu is he feeling of having been somewhere before.

2. John F. Kennedy's funeral was modelled after Lincoln's.
3. Elton John wrote a song about Marilyn Monroe.
4. My eyes ached after staring at he stars.

5. Jimmy was afra'd his moher would smell smoke.

IABQEDALQBQ
TOWN

MONTH
CALENDAR
NICE

HOME

AFTER
MUSIC
ANKLE
CAMP

COMMENT

5 5
Bibliography

Duﬁ‘y, S. A; Morris, R. K; & Rayner, K (1988). Lexical ambiguity and ﬁxation
times in reading. MW. 429-446.

Rayner, K; Pacht, J. M.; 8 Duffy, S. A (IN PRESS). Effects of prior encounter
and discourse bias on he processing of lexically ambiguous words:
Evidence from eye ﬁxations.

Serene, S. C.; Pacht, J. M.; & Rayner, K (1992). The effect of meaning
frequency on processing lexically ambiguous words: Evidence from eye

ﬁxations W. 296-300.

Seidenberg, M. 8.; Tanenhaus, M. K, Leiman, J. M.; & Bienkowski, M. (1982).
Automatic access of the meanings of ambiguous words' In context. Some

limitations of knowledge-based processing. WW,
489-537.

Seidenberg, M. 8.; Waters, G. 8.; Sanders, M. 8. Langer, P. (1984). Pre- and
postlexical loci of contextual effects on word recognition. Mam

mum. 12, 315-323.

Simpson, G. 8,; & Burgess, C. (1985). Activation and selection processes in
he recognition of ambiguous words. I l l 0

W1, 28-39

Simpson, G. B.; & Krueger, M. A (1991). Selective access of homograph

meanings in sentence context. WW.
627-643.

vainney, D. A (1979). Lexical access during sentence comprehension.

(Re)Consideration of context effects. Wild
MechaLﬂchamQLlﬁ 645-659

Tabossi, P. (1988). Accessing lexical ambiguity in different types of sentential
contexts. JaymLQLMsmcuransllansuaseZZ. 324-340.

Tabossi, P.; & Zardon, F. (1993). Processing ambiguous words in context.

.leumaLQtMenananguasejz. 359-372.

56

Tanenhaus, M. K; Leiman, J. M.; & Seidenberg, M. 8. (1979). Evidence for
multiplestages in he processing of ambiguous words In syntactic

contexts WWW 427-440

 

"111111111111111111?