Swinney, D.A. (1979). Lexical access during sentence comprehension: re(consideration) of context effects. Journal of Verbal Learning and Verbal Behavior, 18, 645-659.

PANEL 1650 DAVID A. SWINNEY

TABLE 1

SCHEMATIZED SAMPLE OF EXPERIMENTAL MATERIALS

Each list contained only one of the three visualwords which were created in conjunction witheach sentence pair. The three visual wordconditions were equally represented on eachlist. Half of the materials on each list werewords (36 experimental materials, and 2words which were paired with filler sentences)and the other half (32) were nonwords (pairedwith filler sentences). For six of the fillersentences, no visual word appeared on thescreen.

Thus, there were 12 presentation condi-tions : each of 3 lists paired with each of the4 tape conditions. A 1000-Hz signal wasplaced on a separate- channel of the tapeexactly coincident with the offset of eachambiguous or control word in the experiment-al sentential materials, and with the offset ofa pseudorandomly chosen word in the fillersentences. These signals, inaudible to the sub-jects, signaled a PDP8/e computer to presentthe appropriate visual word and to start thetiming mechanism which measured the la-tencies for the subject's lexical decisions. (SeeOnifer, Hirshkowitz, & Swinney (1978) fordiscussion of hardware and software involvedin this procedure.)

Subjects. Eighty-four undergraduates from

Tufts University participated in partial ful-fillment of a course requirement. Seven sub-

jects were randomly, assigned to each of the12 experimental conditions. Data for sixadditional subjects were omitted from analy-sis for failure to achieve a score of at least85 % correct on the comprehension test.

Procedure. The subjects were seated infront of a CRT screen and listened throughheadphones to the 82 binaurally presentedsentence pairs. Subjects were tested in groupsof up to 3 at a time: each subject was in abooth isolating him/her from other subjectsin a group. Subjects were instructed to listencarefully to each sentence and to understandit. They were told that they would be tested ontheir comprehension during the experiment,and that the result of this test was crucial totheir successful participation in this experi-ment.

In addition, subjects were told that theyhad a second task. It was explained that astring of letters would appear on the screenduring some of the sentences they listened to,and that they were to decide as quickly aspossible whether each letter string formed aword or not. No hint was given that wordsand sentences might be related and, in thefive practice trials, no such relationshipexisted.

At both the midpoint and the end of theexperimental session, subjects were given a

No context

Biasing context

Rumor had it that, for years, the governmentbuilding had been plagued with problems.The man was not surprised when he foundseveral bugsΔ in the corner of his, room.Rumor had it that, for years, the governmentbuilding had been plagued with problems.The man was not surprised when he foundseveral spiders, roaches, and other bugsΔ inthe corner of his room.

Rumor had it that, for years, the governmentbuilding had been plagued with problems.The man was not surprised when he foundseveral insectsΔ in the corner of his room.Rumor had it that, for years, the governmentbuilding had been plagued with problems.The man was not surprised when he foundseveral spiders, roaches, and other insectsΔ inthe corner of his room.

Visual wordsDisplayed at "Δ"

ANTSPYSEW

(contextually related)(contextually inappropriate)(unrelated)

Context condition Ambiguous Unambiguous

Ambiguity condition

PANEL 2

CONTEXT EFFECTS ON LEXICAL ACCESS 651

sheet of paper containing 2, sentence pairs.They were required to decide whether eachof these was either identical or similar tosentences they had heard, or whether -thesentence, had not occurred at all in the,experi-ment. These materials were scored on a per-centage correct basis. At the end of the experi-mental session, subjects were questionedabout whether they had noticed ambiguitiesin the sentence material and about whetherthey thought the words on the screen relatedin any specific fashion to the sentences theyhad heard.

ResultsThe mean reaction times for the 12 experi-

mental conditions, calculated across all ma-terials and subjects, are presented in Table 2.It is apparent that lexical decisions for wordsrelated to both readings of the ambiguity arefacilitated (relative to decisions for an un-related control word) in conditions containinga lexical ambiguity and no biasing context.Similarly, and of greatest interest, this sameeffect holds for the condition in which thereis a strongly biasing semantic context present;lexical decisions for words related to boththe contextually relevant and the contextuallyinappropriate meanings of the ambiguityappear to be facilitated compared to decisionsfor unrelated control words. The effects forboth of the unambiguous conditions alsoappear quite straightforward : Lexical deci-sions for the "related" word appear to befacilitated, but those for the other two words

are not. Thus, by inspection, the results appearto support the Postdecision Hypothesis; evena very strong semantic context apparentlydoes not direct lexical access. Statisticalanalysis supports this contention.

An analysis of variance revealed that maineffects for Context, Ambiguity, and VisualWord Type were each significant for analysesemploying both subjects and materials asrandom factors, min F'(1, 79) =7.01, p < .01;Min F'(1, 86) = 6.32, p <.025; Min F'(2, 188)=52.6, p<.001, respectively. Both the Con-text x Visual Word Type and the Context xAmbiguity interactions failed to reach signi-ficance, Min F'(1, l19)=0.42; Min F'(1, 74)=.01, respectively. Most revealing for thepresent purposes, however, was the fact thatAmbiguity interacted significantly with VisualWord Type, Min F'(2,157)=4.71, p<.01,but that the Context x Ambiguity x VisualWord Type interaction was not significant,Min F(2, 161)=0.04.

In order to examine the predicted effects,planned multiple comparisons were made onthe relevant Visual Word Type categories foreach of the Ambiguity x Context conditions.For the condition containing a biasing con-text and an ambiguity, reaction times to visualwords in both the contextually related andcontextually inappropriate categories weresignificantly faster than latencies for unre-lated words, t(83) = -6.1, p <.0009; t(83) =-5.04, p <.0009, respectively. The contextu-ally related and contextually inappropriatecategories, however, did not differ from each

TABLE 2MEAN REACTION TIMES, IN MILLISECONDS, FOR CONDITIONS OF THE AMBIGUITY X CONTEXT X VISUAL

WORD INTERACTION: EXPERIMENT I

Visually presented words

Ambiguitycondition

Contextcondition

Contextuallyrelated

Contextuallyinappropriate Unrelated

Ambiguous

Unambiguous

Biasing contextNo context

Biasing contextNo context

890916

88794

910925

958967

960974

963972

PANEL 3

6 Under Bonferroni t analysis (for α=.05, d=28.4), allsignificant effects reported for the multiple t tests are alsosignificant under the more conservative Bonferroni tanalysis.

656

DAVID A. SWINNEY

but the latter two categories did not differsignificantly, t(71)=-0.6. Finally, in thesentential conditions containing an unambi-guous control word and no biasing context,lexical decisions to words in the contextuallyrelated condition were responded to fasterthan those in the contextually inappropriate,t(71) = -3.4, p<.001, and unrelated, t(71) =-3.2, p<.002, conditions, but, again, thelatter two categories did not differ significant-ly, t(71) = 0.02.

6

DiscussionThe results of the second experiment repli-

cate those of the first, demonstrating thatlexical decisions for words related to both therelevant and the contextually inappropriatemeanings of an ambiguous word are facili-tated, even in the presence of a strong, prior,biasing context, when these decisions aremade immediately following occurrence ofthe ambiguity in a sentence. In addition, theexperiment shows that when this test is appliedthree syllables following occurrence of theambiguous word, only lexical decisions forwords related to the contextually relevantmeaning of the ambiguity are facilitated; atthis point lexical decisions for words relatedto contextually inappropriate meanings of anambiguity no longer show facilitation.

It might be noted that in the extensionportion of Experiment 2 (where the test pointwas three syllables following the ambiguity)only the "relevant" meanings of the ambi-guities were found to be facilitated in thematerials containing no biasing context.Because half of the materials had originallybeen chosen to have a priori biases towardthe interpretation picked as the "related"meaning, and the other half chosen to have apriori biases toward the other (inappropriate)meaning, one might have expected that eachof these interpretations would have shownsome facilitation in the unbiased contextcondition. However, although all ambiguitieswere originally chosen to be approximatelyequibiased, with a balanced representation ofwhatever a priori biases they contained, in-terpretations of words change over time. Thebias ratings used in Experiment 2 were basedon those used for Experiment 1. Thus, at least2 years separated the gathering of bias ratingsand performance on those materials in Experi-ment 2. In order to discover whether changesin these, biases had taken place over thisperiod of time, a post-test was performed (forthe unbiased context sentences) using 35Tufts University undergraduates. It was foundthat, of the 36 experimental ambiguities, 29actually had biases favoring the reading whichhad been chosen as the "related" meaning inthe experiment (although some of these werevery small). Adding this information to thefact that all biases, whatever their direction,were relatively small, leads to the conclusion

TABLE 4

MEAN REACTION TIMES, IN MILLISECONDS, FOR CONDITIONS OF THE AMBIGUITY X CONTEXT X VISUALWORD INTERACTION: EXPERIMENT 2 (EXTENSION)

Visual word condition

Ambiguitycondition

Contextcondition

Contextuallyrelated

Contextuallyinappropriate unrelated

Ambiguous

Unambiguous

Biasing contextNo context

Biasing contextNo context

795800

808811

849846

843847

848845

849846

LING257 - 10/22/2013 - Ambiguity, Context and Selection

1

Chen, L. & Boland, J.E. (2008). Dominance and context effects on activation of alternative homophone meanings. Memory & Cognition, 36, 1306-1323.

LING257 - 10/22/2013 - Ambiguity, Context and Selection

2

PA

NE

L 4

PA

NE

L 5

LING257 - 10/22/2013 - Ambiguity, Context and Selection

3

PA

NE

L 6

PA

NE

L 7

LING257 - 10/22/2013 - Ambiguity, Context and Selection

4

PA

NE

L 8

1316 CHEN AND BOLAND

text condition was fixated longer than that in the neutral context. This effect confirms that the subordinate-biasing context boosted activation of the subordinate meaning. There was, however, no context effect on shape competi-tors [t1(28) 1.10, p .10; t2(30) 1.66, p .10].

Visual Bias of Shape CompetitorsAs another test of whether the dominant homophone

meaning was activated in the subordinate-biased context, we compared the observed proportion of looks with the proportion that would be expected on the basis of chance alone. To evaluate this activation, we used log gaze ratios as a measure of visual bias to the shape competitors, as compared with the filler objects. Log gaze ratios provide a measure of bias toward the dominant meaning, as compared with filler objects, that is independent from the level of acti-vation of the subordinate meaning. If shape competitors are activated more than filler objects, this would be evidence for higher activation than would be expected by chance. Figure 5 plots the time course of log gaze ratios for each context condition. Due to missing values, statistics could be computed starting only from the 200-msec interval.

One-way ANOVAS compared the log gaze ratios with 0 across participants and items for each 100-msec interval from 200 msec to 1,000 msec after target onset. In the neu-tral context, there was a visual bias toward shape competi-tors, as compared with filler objects, starting at 400 msec, significant only by participants, not by items [F1(1,26) 16.17, p .001; F2(1,30) 3.23, p .10], and con-tinuing from 500 to 1,000 msec by both participants and items [500 msec, F1(1,26) 40.26, p .001, and F2(1,30) 14.24, p .01; 600 msec, F1(1,26) 50.45, p .001, and F2(1,26) 38.15, p .001; 700 msec, F1(1,26) 23.79, p .001, and F2(1,28) 8.48, p .01; 800 msec, F1(1,24) 10.82, p .01, and F2(1,28) 24.80, p .001; 900 msec, F1(1,24) 5.57, p .05, and F2(1,30) 17.18, p .001]. More important, in the subordinate- biased context, there was still a visual bias toward the shape competitors, as compared with the filler

and at 600 msec, there were fewer both by participants and by items [500 msec, t1(1,28) 2.56, p .05, and t2(1,30) 1.43, p .10; 600 msec, t1(1,28) 4.34, p .001, and t2(1,30) 2.20, p .05]. Together, these findings indicate that at early time intervals, the subordinate- biased context both increased activation of the subordinate meaning and decreased activation of the dominant meaning, relative to a neutral context.

First-Run Dwell TimeFirst-run dwell time was analyzed in order to evaluate

initial processing time for each of the fixated objects. The mean first-run dwell times (in milliseconds) for the criti-cal objects in the neutral context condition (with standard errors) were as follows: actual referent, 590 (13); shape competitor, 246 (6); filler objects, 205 (3). The means for the subordinate-biased context condition were as follows: actual referent, 706 (15); shape competitor, 228 (9); filler objects, 192 (3).

In order to determine whether the dominant meaning of the homophone was activated at above-chance levels, first-run dwell times for the shape competitors were compared with mean first-run dwell times for the unrelated filler objects. For each contextual bias condition, independent t tests were performed. In the neutral context condition, shape competitors had longer first-run dwell times than did fillers, as was expected [t1(43) 6.26, p .001; t2(46) 6.49, p .001]. Interestingly, this effect also occurred in the subordinate-biased context condition: Shape competi-tors had longer first-run dwell times than did filler objects [t1(43) 5.29, p .001; t2(46) 2.93, p .01]. These findings suggest that the dominant meaning, represented by the shape competitor, was activated at levels higher than chance, even in the subordinate-biased context.

Contextual bias effects on the first-run dwell times to actual referents and shape competitors were analyzed using t tests. There was an effect of context type on actual referents [t1(28) 3.20, p .01; t2(30) 4.14, p .001], so that the actual referent in the subordinate-biased con-

–1.5

–1

–0.5

0

0.5

1

1.5

Time Intervals From Target Onset (msec)

ln(P

[Sha

pe C

ompe

titor

]/P[

Fille

r Obj

ect]

)

1000 300 400 500 600 700 800 900200

Neutral context

Biased context

Figure 5. Average log gaze ratios for each context condition in 100-msec time intervals. The first time interval is 0–99 msec after homophone onset.

PA

NE

L 9

Folk, J.R. & Morris, R.K. (2003). Effects of syntactic category assignment on lexical ambiguity resolution in reading: an eye movement analysis. Memory & Cognition, 31, 87-99.

PANEL 9

PANEL 10

LING257 - 10/22/2013 - Ambiguity, Context and Selection

5

NN

dominantThe police arrested the crook (felon) who robbed the gas station.

subordinateMary went to the fireplace and picked up the poker (bucket) from the ledge.

NV

dominantThe farm hand had to clean the stall (barn) and put down fresh hay.

subordinateThe game warden knew that the injured duck (sheep) was not going to live.

PANEL 11

PANEL 12

LING257 - 10/22/2013 - Ambiguity, Context and Selection

6

NVLaurie took the prune (grape) out of the fruit bowl and ate it.

NNApril used the straw (spoon) to stir her glass of iced tea.