The EX Factor: Evidence for ExperienceDriven Comprehension

1
The EX Factor Evidence for Experience-Driven
Comprehension Thomas A. Farmer1, Morten H.
Christiansen1, and Karen A. Kemtes2 1Cornell
University, 2University of NevadaLas Vegas
Purpose
Supporting Correlational Data and Regression
Analyses
As such, it appears that vWM scores, at least on
the WC span task, are more grounded in language
experience than in memory capacity. This is not
to say that there is no capacity component of the
task. The association between vWM and the
experience measures is, however,
stronger. Factor 1 was subsequently interpreted
as a measure of language experience, which we
have termed the EX-Factor. Because only one
measure loaded on Factor 2, it was deemed
unstable and not interpreted.
Traditional accounts of individual differences in
sentence processing have focused on individual
differences in verbal working memory as a key
explanatory variable (Just Carpenter, 1992, for
example). MacDonald and Christiansen (2002),
arguing for an experience-driven comprehension
system, however, proposed that individual
differences in language comprehension are, in
part, a product of differences in language
experience. Crucially, the authors proposed that
reading span tasks, traditionally thought to
measure verbal working memory capacity, actually
measure differences in language experience, at
least more so than a capacity-based construct
given the highly linguistic nature of these
tasks, people with more language experience
exhibit superior performance. Currently, the
experience-based view of individual differences
is supported only by indirect behavioral data and
connectionist simulation results. In order to
test experience effects on on-line comprehension
more directly, we derived a measure of language
experience, the EX-Factor, and tested its ability
to predict individual differences on garden-path
relative clause sentences.
Table 2. Original Correlation Matrix

The Validity of the EX-Factor
In order to assess the validity of the EX-Factor,
we also administered the same participants the
Main Verb / Reduced Relative Clause (MV/RRC)
ambiguous sentences from MacDonald, Just, and
Carpenter (1992) in a self-paced word-by-word
moving window task.
Although reflected within the factor loadings,
analyses of the correlations among scores on the
five individual difference tasks help to
illuminate the nature of the relationships among
the variables, and revealed that scores on 1)
vWM task correlated significantly with ART
(r.29, p.014) and with 2) VOCAB
(r.30, p.011), the two most direct measures of
experience. 3) ART and VOCAB also
significantly correlated, r.44, plt.0005. Of
interest, scores on BDS, the measure of memory
capacity, did not significantly correlate with
any of the three reading experience measures, or
with vWM (all ps gt .28). Subsequently, the RT
differences associated with the RC sentences were
predicted from all possible linear combinations
of all individual difference variables. The
regression equation that accounted, significantly
(F(2, 68)5.31, p.007, R2.135), for the most
variance in RC resolution was the one with only
vWM (t(71)2.73 ,p.008) and BDS (t(71)-2.02,
p.037) as predictors. IMPORTANTLY Verbal WM
accounted for 9.43 of the variance in the
magnitude of the garden-path effect (squared
semi-partial correlation .307), and BDS
accounted for 5.76 of the variance (squared
semi-partial correlation -.24). That is, the
processing-skill-heavy WC span task was the
superior predictor of the garden-path effect.
MV-Amb The experienced soldiers warned about
the dangers before the midnight
raid. RRC-Amb The experienced soldiers warned
about the dangers conducted the
midnight raid. MV-UnAmb The experienced
soldiers spoke about the dangers before the
midnight raid. RC-UnAmb The
experienced soldiers who were warned about the
dangers conducted the midnight
raid. These materials have been shown,
consistently, to elicit a garden-path effect on
the relative-clause ambiguous sentences at
disambiguation. We also created a score on the
EX-Factor by converting vWM, ART, VOCAB, and
COGNEED to z-scores and then summing them (summed
z-score method). Prediction EX-Factor scores
should predict the magnitude of the garden-path
effect on ambiguous sentences resolved in
accordance with the relative clause
interpretation. Before examining the predictive
utility of EX-Factor scores, it is important to
note that the results of MacDonald et al. (1992)
did replicate, yielding a significant 2 (MV vs.
RC) x 2 (Amb vs. UnAmb) X 2 (ambiguity vs.
disambiguation) ANOVA, F(1, 70)18.6, plt.0005.
Measuring Reading Experience
Seventy-one participants completed 5 tasks
hypothesized to measure either experience with
language or verbal working memory. Experience
Measures Author Recognition Task (ART) (West,
Stanovich, Mitchell, 1993) measures print
exposure. M33, SD11 (possible
range0-100). Vocabulary (VOCAB) (Shipley,
1940) argued to be an indicator of reading
experience (see MacDonald and Christiansen, 2002,
for a discussion). M31.32, SD3.14 (possible
range 1-50). Need for Cognition (COGNEED)
(Cacioppo, Petty, Kao, 1984)a personality
variable, referring to the degree to which one
desires to be engaged in cognitively taxing
activities. We assumed that people high in
COGNEED would be more frequent readers. M10.68,
SD22.76 (possible range -72 to 72) Verbal
Working Memory (vWM) (Waters Caplan, 1996)
modified version of the Daneman and Carpenter
(1980) reading span task. Instead of reading
sentences out-loud, participants are asked to
make a semantic acceptability judgment of each
sentence. Notably, given that participants must
process lexical, syntactic, and semantic
information, the task demands are very similar to
on-line reading tasks (M4.43, SD1.09). This
task was chosen over the more traditional DC
reading span task in light of findings by Waters
and Caplan (1996) that the DC span task tends to
be unreliable and probably does not include a
substantial processing component. In light of
the added processing component (see above), it is
more processing-skill-heavy than other vWM
measures, and thus, more likely to provide a
measure of experience. Verbal Working Memory /
Memory Measure Backwards Digit Span (BDS)
(Wechsler, 1981) STM capacity measure.
Importantly, although there is a minor linguistic
component, this task is much less grounded in
language comprehension processes than vWM.
Conclusion
Through these analyses, we demonstrated that 1)
a measure of processing skill, EX-Factor scores,
significantly predicted processing difficulty
associated with reduced relatives 2) the high
experience individuals were more sensitive to the
frequency-based bias against RC-resolution 3)
the original correlations indicate a significant
association between vWM and the experience
measures, but not with the capacity measure 4)
although there is a capacity component to the WC
span task (cross-loading), it appears to measure
language experience over and above capacity 5)
vWM is the superior predictor of individual
differences in the garden-path effect. To the
degree that the WC span task taps processing
skill over and above capacity, experience (or
whatever seems to be measured by the
processing-skill-heavy WC span task) appears to
be a better predictor of complex syntactic
processing. Although the factor scores (and
loadings), their association with the garden-path
effect, the correlational analyses, and the
regression analyses, taken together, do not
rule-out a capacity-based argument, they DO
provide support for a role of experience in
explaining individual differences in complex
syntactic processing. Once again, we do not
argue that processing skill is the only process
measured by the span task. Clearly, processes
such as attention (Swets, Desmet, Hambrick,
Ferreira, 2004), and capacity also influence task
performance. Many other factors, including
general intelligence, cognitive aging, and so
forth may also exert influences on task
performance as well. What does seem clear from
this work is that span tasks do measure, to a
large degree, language processing skill, and as
such, do serve as a proxy measure of language
experience.
Figure 1 reveals that participants were
garden-pathed on the RC-resolved sentences,
t(70)6.72, plt.0005.
Importantly, EX-Factor scores significantly
predicted the difference between RC-Amb and
RC-UnAmb at disambiguation, t(70)3.03, p.003,
whereas BDS scores did not (p gt .05).
Exploratory Factor Analysis
Figure 1. Participants are garden-pathed when the
sentence is resolved in accordance with the RC
interpretation of the ambiguity. Importantly,
EX-Factor scores significantly predict the RT
difference between ambiguous and unambiguous RC
sentences at disambiguation.
Notably, this trend is highly compatible with an
experience-based approach to individual
differences in sentence processing. Tabossi et
al. (1994), for example, found that the RC
resolution of the MV/RC ambiguity is much less
frequent than the MV interpretation. As such,
high experience individuals should be more
sensitive to RC-Resolution than low experience
individuals such a trend is evident in Figure 2.
Extraction Method Principal Axis
Factoring Rotation Method Varimax (although the
same factor structure, and very similar factor
loadings, occurred with non-orthogonal
rotation) Factor Evaluation Two factors
extracted via Eigenvalues-greater-than-1 rule,
scree plot analysis, and percent of variance
explained Two factors were extracted, accounting
for 35.17 of the variance (an amount consistent
with other published social science studies)
References
Cacioppo, J., Petty, R., Kao, C. (1984). The
efficient assessment of need for cognition.
Journal of Personality Assessment, 48,
306-307. Farmer, T. A., Christiansen, M. C.,
Kemtes, K. A. (2005). The impact of experience on
individual differences. Proceedings of the
Twenty-Seventh Annual Conference of the Cognitive
Science Society. Mahwah, NJ Erlbaum. MacDonald,
M. C., Just, M., A., Carpenter, P. A. (1992).
Working memory constraints on the processing of
syntactic ambiguity. Cognition, 24,
56-98. MacDonald, M. C., Christiansen, M. H.
(2002). Reassessing working memory Comment on
Just and Carpenter (1992) and Waters and Caplan
(1996). Psychological Review, 109(1),
35-54. Shipley, W. C. (1940). A self-administered
scale for measuring intellectual impairment and
deterioration. Journal of Psychology, 9,
371- 377. Swets, B., Desmet, T., Hambrick, D.
.Z., Ferreira, F. (2004). The role of working
memory in attachment preferences. Poster
presented at the 45th Annual Meeting of the
Psychonomic Society, Minneapolis, MN. Tabossi,
P., Spivey-Knowlton, M., McRae, K., Tanenhaus,
M. (1994). Semantic effects in syntactic
ambiguity resolution Evidence for a
constraint-based resolution process. In C. Umilta
M. Moscovitch (Eds.) Attention and Performance
XV Unconscious processes. (pp. 589-616).
Cambridge, MA MIT Press. Waters, G., Caplan,
D. (1996). The measurement of verbal working
memory capacity and its relation to reading
comprehension. Quarterly Journal of Experimental
Psychology Hum Exp, 49, 51-79. Wechsler, D.
(1981). The Wechsler Adult Intelligence Scale
Revised. New York Psychological
Corporation. West, R. F., Stanovich, K. E.,
Mitchell, H. R. (1993). Reading in the real world
and its correlates. Reading Research Quarterly,
28(1), 35-50.
Noteworthy Points The three measures identified
as indicators of reading experience, ART, VOCAB,
and COGNEED, all loaded together on the first
factor. In line with an experience-based view of
the WC span task, vWM (traditionally seen as a
measure of verbal working memory capacity)
loaded, along with the three hypothesized
language experience measures, on Factor 1
(although there is some evidence of a
cross-loading, with BDS, onto Factor 2 hence, a
small capacity component as well). BDS did not
load with the experience measures (or the span
task hypothesized to measure verbal working
memory capacity), but instead loaded by itself on
Factor 2.
Figure 2. The magnitude of the garden-path effect
on the RC-Resolved sentences (RC-Amb - RC-UnAmb
at disambiguation).
These data are consistent with other studies
demonstrating that participants who score well on
the WC span task are much more sensitive to
distributional biases in comprehension that their
low span counterparts (Farmer, Christiansen,
Kemtes, 2005).