Change detection
The effects of linguistic focus, hierarchical word level
and proficiency
Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
Wayne State University
A version of the change-detection paradigm was used to examine Good-Enough
Representation (Ferreira, Bailey, & Ferraro, 2002). Participants read sentence
pairs where a subject noun (e.g., flower) could change to a Superordinate (e.g.,
plant), Subordinate (e.g., rose), or an Unrelated (e.g., prince) noun. The task was
completed cross-linguistically for bilinguals, where the first sentence appeared in
English (L1) and the second in French (L2). Linguistic focus was also manipulated. Change detection was extremely high in all conditions in the monolingual
sample. In the bilingual sample, focused changes were detected more often, as
were changes to unrelated words. Proficiency was related to change detection for
monolinguals and bilinguals. The relationships between these and other participant and stimulus variables are also explored.
Keywords: change detection, bilingual, monolingual, linguistic focus,
superordinate, subordinate, language proficiency
Multiple definitions have been proposed for the term bilingual (as discussed in
Schreuder & Weltens, 2005). The most widely accepted definition classifies someone as bilingual if he or she has the ability to communicate in more than one
language (Francis, 1999). This is the definition that is retained for the purpose of
this paper.
Similarly, there is no standard way to measure a bilingual’s proficiency. In the
present paper we follow Francis’ (1999) guidelines about the types of demographic
information to collect and report about bilinguals so that results can be compared
across studies.
The Mental Lexicon 5:1 (2010), 47–86. doi 10.1075/ml.5.1.03ken
issn 1871–1340 / e-issn 1871–1375 © John Benjamins Publishing Company
48 Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
The Bilingual Lexicon
There has been much debate about the lexico-semantic organization of the bilingual brain and numerous models have been proposed. The interdependence
view, which proposes a shared conceptual store but separate lexical stores for each
language (Basnight-Brown & Altarriba, 2007; Francis, 1999; Sharifian, 2002), has
amassed the most corroborating evidence. For example, Kroll and Stewart’s (1994)
Revised Hierarchical Model (RHM) proposes such a system. Further, this model
proposes that the L1 and L2 lexical stores have different strengths of connections
between each other as well as individually with the shared conceptual stores; the
strengths of these connections are dependent on language proficiency (Sharifian,
2002), which is a more accurate predictor of actual performance than is age of
acquisition (Ferré, Sanchez-Casas, & Guasch, 2006; Heredia, 1997). A more recent
revision of the RHM (Heredia, 1997) suggests that the labels L1 and L2 for the
lexicons might be deceiving and that they should be replaced by more dominant
and less dominant language.
Good-Enough Representation and Depth of Processing
There is much research to support the idea that language (whether written or auditory) is not always remembered verbatim (Sachs, 1967) or even fully processed
(Ferreira et al., 2002; Sanford, 2002; Sanford & Sturt, 2002). Under many circumstances the gist of a message is retained but the surface details may be forgotten.
Consequently, the stored representation may not be an accurate depiction of the
stimulus (Ferreira et al., 2002; Ferreira & Patson, 2007). Despite this, people are
usually able to correctly interpret and understand a sentence (Sanford & Sturt,
2002).
Ferreira et al. (2002) have proposed the notion of Good-Enough Representation (GER) to account for this. GER suggests that people only include as much
detail in their mental representations as is necessary for the task at hand (Ferreira
et al., 2002). A shallow level of processing for language (or GER) has been demonstrated in numerous laboratory settings; even when explicitly informed that comprehension would be important, participants seem to construct a GER (Sanford
& Sturt, 2002).
Support for GER comes from the Moses illusion (Bredart & Docquier, 1989;
Bredart & Modolo, 1988) and the Survivors anomaly (Barton & Sanford, 1993).
In these examples, participants fail to notice the untrue component in the sentences: How many animals of each sort did Moses put on the Ark? and Suppose
there was an aircrash right on the border of France and Spain. Where should the
survivors be buried? (Barton & Sanford, 1993). This effect is attributed to the fact
Change detection 49
that representation is simply “good enough” to answer the question and that each
constituent is not deeply processed (Sturt, Sanford, Stewart, & Dawydiak, 2004).
However, when the erroneous terms (Moses and survivors, respectively) are focused in the sentence, detection rates increase significantly (Bredart & Modolo,
1988). We will return to the issue of focus below.
Word Taxonomies
A hierarchical relationship exists between some words, whereby the characteristics of a given level are shared by all those below it (Aberra, 2006). These hierarchical levels are typically termed superordinate, basic, and subordinate (e.g., Aberra,
2006; Rosch, 1978).
Words that belong to the basic level (hyponym in linguistics) are the usual
names for objects (Aberra, 2006). They are generally short, concrete words (Brown,
1958), whose referents are all similar in shape and function (Mervis, 1987; Rosch,
Mervis, Gray, Johnson, & Boyes-Braem, 1976). Basic words are learned at an early
age, before superordinate or subordinate words (Liu, Golinkoff, & Sak, 2001; Rosch
et al., 1976), and there is a preference for them in speech (Rogers & Patterson,
2007). Examples of basic-level words are: milk, dog, flower, and chair (Brown, 1958).
A subordinate word is a more specific instance of the basic level (Rogers &
Patterson, 2007), and represents a subset of it (McGregor & Waxman, 1998). For
example, rose is a subset of flower (Mervis, Johnson, & Mervis, 1994). Superordinate words refer to broader, more inclusive categories than the basic level (Liu
et al., 2001; McGregor & Waxman, 1998). Examples of superordinate words (or
categories) are: plant, fruit, and furniture (Brown, 1958). Superordinates’ referents
vary more, both perceptually and functionally (Liu et al., 2001; Rosch et al., 1976).
A basic-level word necessarily includes all of the features of a corresponding
subordinate-level category. Consequently, for the purpose of the present paper,
we can think of conceptual inclusivity in two ways. First, a basic-level term will
include the features of the more specific subordinate terms below it in the hierarchy. Second, a basic-level term is itself included within the representation of the
broader superordinate term above it in the hierarchy. Recently, Aberra (2006) has
argued that, because a superordinate word (e.g., fruit) must include features from
many different basic-level terms (e.g., apple and banana), superordinate terms
have more features and consequently provide the most retrieval cues. Although
this view is logical, it is counter to what is typically argued in psychology and linguistics- that basic-level terms have the most features (Aberra, 2006). Further, given that superordinates and subordinates are learned after basic-level labels, their
processing may rely more heavily on cognitive factors such as language proficiency
(Malabonga & Pasnak, 2002).
50
Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
Linguistic Focus
Linguistic focus refers to the way in which a constituent in the sentence is emphasized (Birch & Garnsey, 1995), bringing one’s attention to it so that it is processed
in greater depth (Langford & Holmes, 1979). Linguistic focus can occur in a variety of manners. In written language, lexical markers and syntactic structures can
be employed to focus attention on a target word. For example, the indefinite this
lexically marks the noun which is focused (Birch, Albrecht, & Myers, 2000). Itclefts (e.g., It was the…) and there-insertions (e.g. There was this…) are common
syntactic structures that bring focus to a noun (Birch et al., 2000). A variety of
languages, including French and English, use similar syntactic focusing structures
such as clefting (Féry, 2001). More recently, Sanford, Sanford, Molle, and Emmott
(2006) have also used italicized text to produce focusing effects.
Linguistic focus has been shown to have important effects on perception,
including better and more detailed memory (McKoon, Ratcliff, Ward, & Sproat,
1993; Sanford, Sanford, Filik, & Molle, 2005; Singer, 1976) and faster retrieval
(Birch et al., 2000; Birch & Garnsey, 1995) for focused words. Sturt et al. (2004)
suggested that when a word that is not in focus is changed to another, closely related word, those two words might appear to be semantically indistinguishable (e.g.,
cap and hat); but at a finer level of representation (when focused), the distinction
is clear. This more detailed representation corresponds to representation at a finer
level of granularity by Hobbs’ (1985) account.
Textual Change-Detection Paradigm
The visual perception literature has long used the change detection paradigm to
demonstrate change blindness. In this paradigm, Rensink, O’Regan, and Clark
(1997) alternated pictures of an airplane and that same picture with an engine
missing. Simons and Rensink (2005) later proposed that if an object is focused in
the scene, it is more likely to be encoded (and thus a change to that object is more
likely to be noticed).
Sturt et al. (2004) used sentences as stimuli (rather than pictures) to examine the effect of linguistic focus on change detection. Sturt et al. (2004) hypothesized that if a word was not in focus, then the GER would suffice and the meaning
encoded would be that of a superordinate category. For example, the word cider
might be encoded more broadly as beverage. Consequently, a change to another
example from that same superordinate category (e.g., beer) in a subsequent sentence should be more difficult to detect than a change to an unrelated word (e.g.,
book). They presented participants with a short passage, which they were asked
to read for meaning. Then, after a 500 ms grey screen, the passage was presented
Change detection
again, but with one word changed. Participants were asked whether there had been
a change and, if so, to identify the word that changed. Target sentences were manipulated for focus and semantic proximity (whether the new word was related to
the original word). They found it was easier to detect a related change when the
word was focused, but that focus made no difference for unrelated word changes,
which participants rarely failed to detect. This was found using both it-cleft constructions (Experiment 1) and context (Experiment 2) focus manipulations.
The Present Study
In the present study, we examined the relationship between linguistic characteristics (linguistic focus and word hierarchy) and representation of meaning. In order
to extend Sturt et al.’s (2004) findings, their paradigm was adapted using new sentence stimuli, with the addition of a hierarchical change manipulation (Pre-test).
Note that in the monolingual version of the change detection paradigm, physical form and meaning are coupled: If the physical form changed, the meaning
changed. The task can thus be performed entirely based on physical form, which
poses a problem because we are primarily interested in meaning. In our main experiment we therefore extended the Pre-test cross-linguistically using a bilingual
adaptation of the task. The physical form of the stimulus changes on all trials, even
“no change” trials, increasing the chance that the task will be performed on the
basis of word meaning.
As with all studies using a change detection paradigm, our analyses will focus
on missed changes. In addition, we will determine whether additional information
can be gained by analyses of reading times (RTs). Several hypotheses are proposed.
First, it is expected that Sturt et al.’s (2004) focus effect will be replicated and extended cross-linguistically. That is, changes to focused items should be more easily detected (irrespective of change direction- superordinate or subordinate) than
when that same constituent is not focused. In line with Hobbs’ (1985) granularity
account, this coarser grain of representation (not in focus) would mean that nuances that would usually allow a differentiation between similar words may no
longer provide this distinction (Sanford et al., 2005).
Second, it is expected that changes to a subordinate category, compared to
a superordinate category, will be missed more often because the features of the
subordinate referent are subsumed by the representation of the basic-level word
(which is also semantically similar). Superordinate terms are also semantically
similar to their corresponding basic level terms, but only some of their features
are shared.
Third, a semantic relatedness effect is expected. That is, change detection is
expected to be very high for word changes that alter the meaning of an utterance
51
52
Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
(Sanford & Sturt, 2002), and less high for word changes that are semantically
related.
Finally, proficiency is expected to have an impact on representation. For proficient bilinguals, processing is likely to be more automatic (Chee, Hon, Lee, &
Soon, 2001), much as it is for monolinguals. It therefore seems reasonable to expect that highly proficient bilinguals would have more cognitive resources available to engage in the encoding task, and consequently, they would have the potential to encode the stimuli with more detail. Speakers who are less proficient,
however, would be required to engage in more effortful processing of the linguistic
stimuli, which would require more cognitive resources (Chee et al., 2001). These
individuals may not have the necessary cognitive resources available to be able to
process the stimuli more deeply.
RTs have never before been measured in this paradigm, and for reasons discussed below, no predictions are made a priori about them.
Pre-test
A pre-test was deemed necessary prior to undertaking the bilingual change-detection paradigm proposed here. Since this is a fairly new paradigm (used by only one
other researcher at the present time, to the best of our knowledge), we felt it necessary to familiarize ourselves with this methodology prior to its administration to
our target participants (English-French bilinguals). Our rationale for using English monolingual speakers for this pre-test was two-fold. First, it would provide us
with an experiment that was more analogous to the Sturt et al. (2004) study, which
would allow for better comparison between studies. Second, we had access to only
a limited number of bilingual participants, which we did not want to waste on our
first attempt with a new paradigm.
Method
Participants
Monolingual English-speaking participants were recruited from the Psychology
subject pool and received extra credit in exchange for their participation. All participants (N = 53) were undergraduate students. Their ages ranged from 18–44
(M = 22.53).
Change detection
Materials
Word Selection. Through the help of published works (e.g., Brown, 1958; Malabonga & Pasnak, 2002; Ruts et al., 2004; Van Overschelde, Rawson, & Dunlosky, 2004)
and colleagues, words were generated and grouped into potential SuperordinateBasic-Subordinate triads, which reflected a “common sense” of what the Superordinate, Basic and Subordinate levels should be. These triads were then separated
into individual words and then presented to undergraduate students (N = 34; these
participants did not take part in any change detection tasks) to categorize as Superordinate, Basic or Subordinate. Participants were also given the option to categorize the word in a fourth category if they were unable to categorize the word, or
if they were unfamiliar with that word. Definitions for each hierarchical level were
provided to participants in the instructions. To better illustrate this hierarchical
relationship, the following example was provided to participants along with the
definitions: plant (Superordinate); flower (Basic); rose (Subordinate). The words
that conformed to the original classifications were then reconstituted into triads.
For example, the percent of participants who categorized the word dog at the Superordinate level was 11.8%, at the Basic level, it was 67.6% and at the Subordinate
level it was 20.6%. Thus, the word dog was retained as a Basic-level term.
Sentence Creation. For each triad, eight sentences were formulated: in half, the
Basic-level noun phrase (NP) was focused, and in the remaining half, it was unfocused. For example, for the triad plant-flower-rose, the focused (and unfocused)
sentences for each level were: It was the flower that was left on the table (The flower
was left on the table); It was the plant that was left on the table (The plant was left on
the table); It was the rose that was left on the table (The rose was left on the table).
In total, this created 8 sentences for each of the 32 triads (focused and unfocused
Superordinate; focused and unfocused Subordinate; focused and unfocused Unrelated; and focused and unfocused with no change), resulting in a total of 256
target sentences. In addition to the no-change sentences for the critical stimuli, 96
no-change filler sentences were created. Filler sentences were similar to the other
sentences used in the experiment in length and complexity. Half of the sentences
were focused and half were unfocused and the subject NP was always a basic-level
word, as was the case with every initial-presentation sentence used in the experiment. Two example filler sentences are The telephone rang and It was the pedestrian
who crossed the road.
To minimize repetition, two lists were created, each with half of the stimuli.
For each triad, participants saw 2 different focused and 2 different unfocused sentences from among the possibilities. For example, for the triad plant-flower-rose, a
participant might have seen the two focused sentences It was the flower that was
left on the table and It was the plant that was left on the table and the two unfocused
53
54
Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
sentences The rose was left on the table and The book was left on the table. Consequently, each participant saw a total of 224 sentence pairs (128 target sentences
and 96 fillers).1
The second column in Appendix A shows the full list of stimulus sentences
(note that TN refers to the target noun for a given trial). The English TNs used are
shown in Appendix B.
Language proficiency. Language proficiency was assessed using the online sample
items from the Wisconsin Language Placement Test (Foreign Language Placement
Tests, 2008). The makers of this test caution against using this as a proficiency measure as its designed purpose is for placement into language classes. However, given
the lack of alternative resources at the present time (Flege, MacKay, & Piske, 2002),
we felt that this placement test could be used as an objective indicator of grammatical proficiency. In addition, self-rating measures asked participants to rate their
English proficiency on Likert-type scales ranging from 1 (Not very fluent/You can
just get by) to 5 (Native-like fluency/You get along extremely well) for the following
aspects of language: vocabulary, pronunciation, writing and comprehension. The
distributions of self-ratings had very bad negative skew but indicated overall high
English proficiency (median response = 5). Proficiency scores on the grammatical
measure revealed somewhat lower aptitude (range = 14–86%; M = 59%).
Procedure
Participants were instructed to read the first sentence (in which the noun phrase
(NP) was either focused or unfocused) for meaning and to press a button indicating that they had finished. After a 500 ms grey screen, a second sentence appeared
Figure 1. Experimental procedure used for pretest. Note that each stimulus sentence appeared as a single line of text on the screen.
Change detection
and participants were again asked to indicate they had finished reading it by pressing a button. Reading times (RTs) for both sentences were recorded by the computer. The target NP in the second sentence on each trial contained one of four
possibilities: a Superordinate change, a Subordinate change, a same-level (Basic)
Unrelated change or the same word (no change). Participants indicated whether
the meaning of any one word changed from the first to the second sentence, and if
so, typed the word that changed in the response box. Figure 1 illustrates a sample
trial for the pre-test.
Data Analysis
Correct detection of a change was recorded if the participant identified the correct
word (including misspellings, pluralizations or recognizable typos of the target).
Detections and RTs were analyzed using linear mixed-effects analyses of covariance with participants and items as crossed random effects (Baayen, Davidson, &
Bates, 2008; Bates, 2005; Bates & Sarkar, 2005; Faraway, 2006; Pinheiro & Bates,
2000; Quené & van den Bergh, 2008). Our variables of primary interest are Focus,
Change condition (i.e. Unrelated, Subordinate, or Superordinate), and Proficiency,
as measured by scores on the objective proficiency measure. We also included log
word length, log word frequency, and participant age and gender. Participant ages
had significant positive skew, so we applied an inverse transform to improve normality. We then multiplied these values by −1 so that the age coefficients could be
interpreted naturally (i.e. older participants have larger scores on the transformed
age variable).
Results and Discussion
Change Detection
Mean accuracy was 99%. There was one significant main effect: Higher proficiency led to fewer missed changes (B = −0.05, SE B = 0.02, z = −2.54, p < .05). This is in
line with our expectation that highly proficient bilinguals would have the cognitive resources available to engage in more detailed processing, thus supporting a
granularity account (Hobbs, 1985). No two- or three-way interactions reached significance, but these results need to be interpreted with caution. Because of the very
small number of missed changes, the model predicted a zero likelihood of an error
in some cases. This nearly perfect performance was not expected, because in Sturt et
al. (2004, Experiment 1), the experiment on which ours was based, mean change detection was only about 90%. We will return to this issue in the General Discussion.
55
56
Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
Reading Times
Reading times for the original sentence (prior to the change manipulation) showed
no main effects or interactions. The analysis of interest was thus with regard to
the second sentence, where the manipulation occurred. We analyzed log reading
times for sentences in which there was a change that was correctly detected. The
results of this analysis are shown in Table 1.
As mentioned earlier, RTs have never before been measured in this paradigm,
but we believed it might be useful to include them, as they may shed some light on
cognitive processing. However, we need to be cautious about making predictions.
First, we must consider that although instructed to press the button after they had
completed reading the sentence, participants were free to press the button at any
time, which could be well before they had actually finished reading the sentence
(e.g., as soon as they noticed a changed word). Second, the focused structure is less
common than the unfocused structure. Thus opposing predictions can be made
with respect to RTs: the focused structure contains more words and is less common, which would suggest longer reading times, but because the focused structure
brings attention to the target noun, reading times may actually be shorter.
There were significant main effects of word Length, Frequency (CELEX English database, 1993), Focus and Change. Sentences with longer words had longer reading times as did those with lower frequency words. Focused sentences
took significantly longer to read than Unfocused sentences. Both types of related changes produced significantly faster reading times than Unrelated changes
(B = −0.08, SE B = .03, t = −2.38, p < .05 for Subordinate changes; B = −0.08, SE
B = .04, t = −2.29, p < .05 for Superordinate changes). RTs for Subordinate and Superordinate changes did not differ from each other. The Change effect might be
due to a kind of semantic priming, whereby residual activation from the noun in
the first sentence results in faster processing of the changed noun in the second
sentence when it is related; this is the case for the Subordinate and Superordinate
change conditions, but not the case in the Unrelated change condition.
In addition to these main effects, there were several interactions. Figure 2
shows the interaction between Word Length and Proficiency. It shows the results
of the statistical model with all variables held constant at their medians except for
the variables involved in the interaction. “Low” and “High” here mean one standard deviation below and above the median on Proficiency, respectively. Other figures in the manuscript were made in an analogous way. As can be seen in Figure 2,
sentences with longer words were associated with longer RTs, and people with
higher proficiencies were protected from this effect to some degree.
Readers should note that proficiency was not dichotomized for the analysis.
Because Proficiency is a continuous variable, error bars or confidence intervals on
Change detection
Table 1. Summary of Multilevel Analysis of Covariance for Variables Predicting Log
Reaction Time (msec) in Monolingual Participants
Variable
B
Age
−8.62
8.25
0.14
0.04
Frequency
−0.02
0.01
−2.89**
Proficiency
−0.00
0.00
−1.03
Gender (male)
−0.04
0.13
−0.33
Focus (unfocused)
−0.08
0.01
−5.88***
0.33
3.53
0.09
3.03**
Length
Age x Length
Age x Frequency
SE (B)
t
−1.04
3.31***
1.92
0.63
16.83
32.83
0.51
0.40
0.51
0.78
Age x Focus (unfocused)
−0.74
2.60
−0.29
Length x Frequency
−0.01
0.00
−0.43
Length x Gender (male)
−0.05
0.06
−0.91
Length x Proficiency
−0.00
0.00
−2.58*
Length x Focus (unfocused)
−0.02
0.04
−0.40
Frequency x Gender (male)
0.02
0.01
1.92
Frequency x Proficiency
0.00
0.00
1.33
Frequency x Focus (unfocused)
0.00
0.01
0.62
Gender (male) x Proficiency
−0.02
0.01
−1.63
Gender (male) x Focus (unfocused)
Age x Gender (male)
Age x Proficiency
−0.01
0.04
−0.40
Proficiency x Focus (unfocused)
0.00
0.00
0.07
Age x Length x Frequency
0.77
2.51
0.31
12.07
14.43
0.84
Age x Length x Gender (male)
Age x Length x Proficiency
0.30
0.23
1.30
Age x Length x Focus (unfocused)
−9.27
8.02
−1.16
Age x Frequency x Gender (male)
0.09
2.59
0.03
Age x Frequency x Proficiency
−0.02
0.04
−0.44
Age x Frequency x Focus (unfocused)
−0.52
1.44
−0.36
Age x Gender (male) x Proficiency
−3.35
2.26
−1.48
6.43
10.07
0.64
Age x Proficiency x Focus (unfocused)
−0.04
0.19
−0.21
Length x Frequency x Gender (male)
−0.01
0.04
−0.32
0.00
0.00
0.69
Age x Gender (male) x Focus (unfocused)
Length x Frequency x Proficiency
57
58
Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
Table 1. (continued)
Variable
Length x Frequency x Focus (unfocused)
B
SE (B)
t
0.02
0.03
0.55
Length x Gender (male) x Proficiency
−0.01
0.00
−1.22
Length x Gender (male) x Focus (unfocused)
−0.15
0.11
−1.35
0.00
0.00
1.28
Length x Proficiency x Focus (unfocused)
Frequency x Gender (male) x Proficiency
0.00
0.00
0.16
−0.04
0.02
−1.80
Frequency x Proficiency x Focus (unfocused)
0.00
0.00
1.02
Gender (male) x Proficiency x Focus (unfocused)
0.00
0.00
0.00
Frequency x Gender (male) x Focus (unfocused)
Significant Effects Involving Change
Main effect: F(2,4497) = 3.90, p = .020
Change x Gender x Focus: F(2,4420) = 3.55, p = .029
Change x Proficiency x Focus: F(2,4420) = 4.59, p = .010
Change x Length x Focus: F(2,4420) = 4.26, p = .014
*p < .05 **p < .01 ***p < .001
the two lines would be meaningless, as these are not specific “lines” that the statistical model sought to explain or predict. This is merely the most convenient way to
show the nature of an interaction involving a continuous variable.
We also observed an Age x Frequency interaction, which is shown in Figure 3.
The frequency effect on RT is steeper for younger participants. This effect could
Figure 2. Pre-test participants’ log reading times as a function of word length and proficiency.
Change detection
Figure 3. Pre-test participants’ log reading times as a function of word frequency and age.
result from more exposure to reading. Although the age range of our participants
was not dramatic, the older participants in our study would have had an additional
7 years of reading experience on average, which may result in some of the lowerfrequency words not being quite as rare to them.
Focus and Change were involved in three significant three-way interactions.
Figure 4 shows the Focus x Change x Proficiency interaction. The top panel shows
RTs for participants above the median on proficiency, and the bottom panel shows
RTs for participant below the median on proficiency. This median split allows us
to see how the Focus x Change interaction depends on proficiency, but proficiency
was not dichotomized in our analysis; as noted above, error bars would therefore
not be meaningful. In the Superordinate condition there is a very small advantage
for focused sentences for low-proficiency participants (bottom panel). As proficiency increases (top panel), the effect changes direction (faster RTs in unfocused
sentences) and becomes markedly stronger. The other two conditions show just
the opposite pattern of outcomes with respect to Proficiency.
There seem to be two possibilities for why Superordinate changes should behave differently: First, Superordinate words were longer than words in the other
two conditions (mean length = 7.8 letters, vs. 6.5 for Subordinate and 5.8 for Unrelated; F(2,86) = 5.38, p = .006). Second, as outlined in the Introduction, we can
think of conceptual inclusivity in different directions. Any given word includes all
features of more specific terms below it in the hierarchy. Thus, an initially-encoded
basic word in sentence 1 includes all features of the more specific Subordinate term
that could appear in sentence 2. But the basic-level features are themselves included
59
60 Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
by the representation of the broader Superordinate term that could appear in sentence 2. These different “kinds” of inclusivity depending on whether the change is
to a Subordinate or Superordinate term could drive different data patterns.
For highly proficient participants, when the focusing structure calls attention
to a Superordinate change, RTs are elevated (far left bar in the top panel). We know
from Figure 2 that high-proficiency participants suffer less from word length than
Figure 4a. High-proficiency pre-test participants’ log reading times as a function of focus
and change condition.
Figure 4b. Low-proficiency pre-test participants’ log reading times as a function of focus
and change condition.
Change detection
low-proficiency participants, so the elevated RTs are not due to reading difficulty.
It could instead be that with a better understanding of these richer conceptual representations, proficient participants require more time to decide if a Superordinate
change should count as a change.
Figure 5 shows the Focus x Change x Length interaction (top panel shows
sentences with shorter words; bottom panel shows sentences with longer words).
Figure 5a. Pre-test participants’ log reading times for short words as a function of focus
and change condition.
Figure 5b. Pre-test participants’ log reading time for long words as a function of focus
and change condition.
61
62
Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
The Superordinate condition shows a larger focus effect for sentences with shorter
words, while the Unrelated condition shows a larger focus effect for sentences with
longer words. The Subordinate condition does not show a focus effect regardless
of word Length. We did not have clear-cut predictions about ways in which word
length might interact with our variables of theoretical interest. As noted above,
Superordinate words had the greatest mean length, Unrelated words had the
Figure 6a. Female pre-test participants’ log reading times as a function of focus and
change condition.
Figure 6b. Male pre-test participants’ log reading times as a function of focus and change
condition.
Change detection
smallest, and Subordinate words were intermediate, but it is not clear why this
pattern should emerge in the reading times.
Finally, Figure 6 shows the Focus x Change x Gender interaction (women in
top panel; men in bottom panel). Note that because Gender is a two-level factor,
it is appropriate to put error bars (+/− 1 SEM) on this plot. Women had a tendency to show very small RT advantages for Unfocused sentences. Men showed
larger differences overall, including RT disadvantages for Unfocused sentences in
the Subordinate and Unrelated conditions. As in the case of Length, we did not
make predictions about how participant gender might interact with our variables
of theoretical interest. We will note, though, that the data patterns shown in Figure 6 (especially for the men; bottom panel) are similar to what was observed for
higher-proficiency participants in Figure 4, and to what will be observed in the
data of bilingual women (to be discussed below).
Main Experiment
Method
Participants
Participants (N = 32) were recruited from three sources: the Psychology subject
pool, the Classical and Modern Languages, Literature and Culture department,
and the English Language Institute. Participants were compensated either with
extra credit toward a psychology class or a gift card.
Participants’ language proficiency was assessed using the measures described
below, in the Materials section. Participants consisted mostly of moderately fluent
English-French bilingual students, though there was a lot of variability in both
objective and subjective measures of fluency. For English proficiency, bilingual
participants’ self-ratings suggested they were all highly proficient (all median responses were again equal to 5 on the 5-point Likert scales, with badly skewed distributions as in the Pre-test). However their actual performance on the grammatical measure revealed slightly weaker language skills (range = 29–100%; M = 60%).
These data were similar to those obtained from the monolingual participants. For
French, participants rated themselves as less proficient (the median response was
either 2 or 3 on each Likert subscale) and the objective grammatical measure also
showed lower proficiency (range = 9–100%; M = 48%). Demographic data are displayed in Table 2 and proficiency data are displayed in Table 3.
63
64 Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
Table 2. Bilingual Participants’ Demographic Data
Variable
Frequency (N = 32)
Gender
Males
Females
13
19
Country of Origin
USA
Canada
Africa
23
4
3
Number of Years Using
English
1 year
or less
2–4 years 5–10 years
More than
10 years
0
0
30
Number of Years Using
French
Other
2
2
1 year or less 2–4 years 5–10 years
More than
10 years
7
5
11
Age of Acquisition- English Before
5 years
25
Age of Acquisition- French Before
5 years
4
9
5–9 years 10–13 years 14–18
years
After 18 years
4
0
3
0
5–9 years 10–13 years 14–18
years
After 18 years
5
5
6
12
Table 3. Bilinguals’ Proficiency Data
Variable
English- Likert
French- Likert
Mean
SD
Vocabulary
4.69
0.59
Pronunciation
4.78
0.55
Writing
4.69
0.69
Comprehension
4.81
0.59
Vocabulary
2.44
1.13
Pronunciation
2.90
1.22
Writing
2.50
1.19
Comprehension
2.84
1.14
English- Objective
Grammar (%)
60.32
15.09
French- Objective
Grammar (%)
48.13
25.81
Materials
Stimuli. The words and sentences used in the Pre-test were used as the basis for
the stimuli for the bilingual experiment. All sentences were translated into French
by two proficient bilinguals (consulting with l’Office Québécois de la Language
Change detection
Française, 2007 when necessary). For example, the basic focused sentence It was
the flower that was left on the table was translated as ‘C’était la fleur qui avait été
laissée sur la table’. This again created eight sentences per triad- 2 superordinate,
2 subordinate, 2 basic (unrelated), and 2 no-change- where one sentence was focused and one was unfocused for each of these pairs. Filler sentences were also
translated. In addition, 10 practice trials with feedback were created to give participants a chance to learn that direct translations were not considered changes.
Appendix A lists the stimulus sentences (note that TN refers to the target noun
for a given trial).
Language proficiency. English proficiency was assessed using the same sample
items as the Pre-test. In addition, the French version of the same survey was also
administered in order to obtain a quantitative and objective measure in both languages. Again, participants were asked to rate their proficiency on Likert-type
scales ranging from 1 (Not very fluent/You can just get by) to 5 (Native-like fluency/
You get along extremely well) for the following aspects of language: vocabulary,
pronunciation, writing and comprehension.
Procedures
This experiment followed the same procedure as the pre-test. Participants read
the first sentence in English, then were shown the second sentence in French
and asked to indicate whether a word had changed in meaning. In each case the
participant pressed a button to indicate that they had finished reading the sentence, and the computer recorded RTs. Figure 7 illustrates a sample trial for the
bilingual task.
Figure 7. Experimental procedure used for the main experiment. Note that each stimulus
sentence appeared as a single line of text on the screen.
65
66 Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
Results and Discussion
In the analyses reported below, proficiency refers to the participants’ scores on the
objective measure of French proficiency. Predictor variables were transformed in
the same way as in the Pre-test.
Change Detection
The bilingual experiment has more to tell us than the Pre-test because of a much
higher proportion of missed changes. The mean accuracy here was 75%. Table 4
summarizes the findings of our analysis of the change detection data.
Table 4. Summary of Multilevel Analysis of Covariance for Variables Predicting Likelihood of a Missed Change by Bilingual Participants
Variable
B
Age
SE (B)
z
−31.95
19.55
−1.63
Length
−0.19
0.32
−0.59
Frequency
−0.10
0.07
−1.51
Proficiency
−0.02
0.01
−2.65**
Gender (male)
0.04
0.33
0.11
Focus (unfocused)
0.47
0.11
4.18***
29.86
29.19
1.02
4.32
5.78
0.75
−36.06
46.10
−0.78
Age x Length
Age x Frequency
Age x Gender (male)
Age x Proficiency
Age x Focus (unfocused)
Length x Frequency
Length x Gender (male)
0.05
0.85
0.06
10.93
18.18
0.60
0.18
0.22
0.82
−0.23
0.40
−0.58
Length x Proficiency
0.01
0.01
1.52
Length x Focus (unfocused)
0.23
0.42
0.54
Frequency x Gender (male)
0.05
0.08
0.61
Frequency x Proficiency
0.00
0.00
1.46
Frequency x Focus (unfocused)
0.01
0.08
0.06
Gender (male) x Proficiency
−0.01
0.01
−1.28
Gender (male) x Focus (unfocused)
−0.02
0.27
−0.07
0.00
0.01
0.75
Age x Length x Frequency
−2.90
21.62
−0.13
Age x Length x Gender (male)
44.17
75.92
0.58
Proficiency x Focus (unfocused)
Change detection
Table 4. (continued)
Variable
B
SE (B)
z
Age x Length x Proficiency
−2.40
1.49
−1.61
Age x Length x Focus (unfocused)
11.72
68.97
0.17
Age x Frequency x Gender (male)
−16.15
14.78
−1.09
−0.08
.29
−0.26
−14.08
13.85
−1.02
6.38
2.14
Age x Gender (male) x Focus (unfocused)
69.51
55.22
1.26
Age x Proficiency x Focus (unfocused)
−2.34
1.10
−2.13*
0.31
0.29
1.06
Age x Frequency x Proficiency
Age x Frequency x Focus (unfocused)
Age x Gender (male) x Proficiency
Length x Frequency x Gender (male)
2.98**
Length x Frequency x Proficiency
0.01
0.01
1.56
Length x Frequency x Focus (unfocused)
0.16
0.30
0.54
Length x Gender (male) x Proficiency
−0.01
0.02
−0.52
Length x Gender (male) x Focus (unfocused)
−0.4
1.05
−0.43
Length x Proficiency x Focus (unfocused)
0.00
0.00
0.17
Frequency x Gender (male) x Proficiency
−0.00
0.00
−0.34
Frequency x Gender (male) x Focus (unfocused)
0.01
0.21
0.05
Frequency x Proficiency x Focus (unfocused)
0.00
0.00
0.24
Gender (male) x Proficiency x Focus (unfocused)
0.02
0.02
1.24
Significant Effects Involving Change (see Footnote 2)
Main effect: χ2(2) = 17.53, df = 2, p < .001
Change x Frequency: χ2(2) = 9.00, df = 2, p = .01
Change x Gender x Focus: χ2(2) = 6.69, df = 2, p = .04
*p < .05 **p < .01 ***p < .001
There were three significant main effects. First, changes were detected more
often when they were in focus, as predicted. This is in line with Hobbs’ (1985)
Granularity account (with the focused structure resulting in fewer missed changes
because of a more detailed initial representation) and Ferreira et al.’s (2002) Good
Enough Representation account, suggesting that, unless one’s attention is focused,
the gist of a sentence is more likely to be encoded.
Second, participants with higher proficiency scores were less likely to miss a
change, also as predicted. More proficient bilinguals can and do engage in more
detailed processing during the initial sentence reading, as they have more cognitive resources available with which to engage in such processing (Chee et al.,
2001).
The three-level factor “Change” also had a significant main effect (χ2(2) = 17.53,
p < .001).2 As predicted, changes to either a Subordinate or Superordinate word
67
68 Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
were significantly more likely to be missed, compared to the Unrelated baseline.
Change detection in the Subordinate and Superordinate conditions did not differ
from each other, which appears to be counter to Aberra’s (2006) argument that
Subordinate changes would be missed more often because their features are included in the Basic-level representation encoded in the first sentence.
Change also interacted significantly with word Frequency (χ2(2) = 9.00,
p = .01). Higher word frequency was associated with better change detection in the
Superordinate condition (B = −0.33, SE B = 0.14, z = −2.32, p = .02), and marginally
associated with better detection in the Unrelated condition (B = −0.15, SE B = 0.09,
z = −1.65, p < .10). There was no relationship between frequency and change detection in the Subordinate category (B = 0.15, SE B = 0.11, z = 1.40, p = .16). Words in
the Subordinate condition had significantly lower frequencies than those in the
Superordinate and Unrelated conditions (mean = 26.6, vs. 74.9 for Superordinate
and 81.2 for Unrelated words; F(2,89) = 12.71, p < .001). This could thus be a case
of range restriction in the Subordinate condition.
There were also three significant three-way interactions. The three-way interaction between Change, Gender, and Focus is shown in Figure 8 (women are
shown in the top panel; men in the bottom panel). For both genders, the Subordinate and Unrelated conditions show sizeable differences in the expected direction
(i.e. the focusing structure improves accuracy). Performance was different in the
Superordinate condition, though. In the Pre-test we noted that words in the Superordinate condition were significantly longer, and that is true for these French
words as well (mean length = 8.1 letters, vs. 6.9 letters for Subordinate and 6.3 letters for Unrelated words; F(2,89) = 4.70, p = .011).
We hypothesized in connection with Figure 4 that the combination of longer
words plus the less common focused structure caused highly proficient readers
to require extra time to decide whether a Superordinate change should count as a
change. Here in Figure 8 we are seeing additional evidence of uncertainty in this
particular condition, especially among women. It is not the case that women had
higher proficiency scores than men — in fact they were slightly lower although the
difference did not approach significance (t(30) = −1.11, p = .28). What might make
women more likely than men to show performance similar to highly proficient
participants could be compliance (or willingness to try one’s best). Recent literature points to the possibility that male research participants may be less responsible or conscientious than female participants, being both more likely than women
to rush through an online rating task, and also less likely to agree to participate in a
multi-part study (Barenboym, Wurm, & Cano, 2010). This is a speculative account
but we will see additional intriguing hints of it below.
There were two three-way interactions involving participant Age and Proficiency. First, Age and Proficiency interacted with Focus (Figure 9). The Focus and
Change detection 69
Figure 8a. Female participants’ proportion of missed changes as a function of focus and
change condition.
Figure 8b. Male participants’ proportion of missed changes as a function of focus and
change condition.
Proficiency main effects are clear in the figure. In general, changes were missed
more often in the Unfocused condition, and more often by less proficient participants, which is in line with our predictions. Examining the performance of
high-proficiency participants across panels of the figure, we see that age does not
influence performance (i.e. both “H” lines are flat) and the results amount to a
70 Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
simple main effect of focus. This result is consistent with the reasoning of Chee et
al. (2001) that proficient participants have a processing advantage due to greater
availability of cognitive resources. The less proficient participants also gain some
benefit from the focusing structure, but for them the effect is related to age.
Why the focus effect should be larger for older participants than for younger
participants is not clear. It should be noted that the “older participants” were mostly
in their mid-twenties; but there is no obvious reason to expect that an 18-year-old’s
performance would differ in any significant way from a 26-year-old’s performance.
Recent findings by Carrier, Cheever, Rosen, Benitez, and Chang (2009) suggest a
difference across generations in the ease of multitasking. These researchers found
that younger generations (i.e., the Net Generation, those born after 1980) find it
easier to multitask, and the present cross-linguistic memory task may be less cognitively taxing for them. Whether 26-year olds and 18-year olds should be considered to represent different generations is of course debatable. Potential explanations based on a hypothesized relationship between age and proficiency can be
ruled out, as this correlation is small and non-significant (r(30) = .13, p = .54).
Finally, Age and Proficiency also interacted with participant Gender (Figure 10). This interaction is very clearly driven by less proficient men, whose relationship between age and change detection was very different than everyone else’s.
Proficient participants of either gender showed little effect of age, which is in line
with the notion that they have additional cognitive resources available for processing (e.g., Chee et al., 2001). However, men who were younger and less proficient
Figure 9. Proportion of missed changes as a function of focus, proficiency and age.
Change detection
Figure 10. Proportion of missed changes as a function of gender, proficiency and age.
had very poor change detection performance. As discussed above, there may be
sex differences in conscientiousness with respect to cognitive tasks, and there is
also some literature to suggest that compliance increases with age, at least with
children (Zupančič, Gril, & Kavčič, 2006). It seems then that there could be some
relationship between age and the likelihood of making one’s best effort for men
but not women. We should guard against overinterpretation, though, especially in
light of the small number of men in the current experiment (n = 10).
Reading Times
As in the Pre-test, reading times for the first sentence showed no main effects or
interactions, as expected. We analyzed log RTs for second sentences with changes
that were correctly detected. The results of this analysis are shown in Table 5. The
only significant main effect was Gender, with men being faster than women, but
there were several interactions. The Change x Focus interaction will be explored
below, as it was part of a significant 3-way interaction.
Focus interacted with word Length. Follow-up comparisons showed that sentences with longer words took marginally longer to read in the Focused condition
(B = 0.16, SE B = 0.09, t = 1.77, p < .10), but there was no relationship in the Unfocused condition (B = −0.05, SE B = 0.08, t = −0.57, p > .57). It seems logical that
sentences with longer words ought to produce longer RTs, but participants may
have occasionally pressed the button before finishing the entire sentences, once
71
72
Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
Table 5. Summary of Multilevel Analysis of Covariance for Variables Predicting Log
Reaction Time (msec) in Bilingual Participants
Variable
B
SE (B)
t
Age
3.59
7.88
0.46
Length
0.07
0.07
0.95
−0.02
0.02
−1.48
Frequency
Proficiency
−0.00
0.00
−1.32
Gender (male)
−0.32
0.13
−2.40*
Focus (unfocused)
−0.02
0.03
−0.66
Age x Length
−1.69
6.42
−0.26
Age x Frequency
−0.14
1.39
−0.10
Age x Gender (male)
30.20
19.19
1.57
Age x Proficiency
−0.55
0.35
−1.56
Age x Focus (unfocused)
−2.59
4.80
−0.54
Length x Frequency
−0.00
0.05
−0.06
0.16
0.11
1.41
Length x Proficiency
−0.00
0.00
−0.14
Length x Focus (unfocused)
−0.22
0.11
−2.10*
Length x Gender (male)
0.03
0.02
1.16
Frequency x Proficiency
Frequency x Gender (male)
−0.00
0.00
−0.69
Frequency x Focus (unfocused)
−0.04
0.02
−1.65
0.01
0.00
Gender (male) x Proficiency
Gender (male) x Focus (unfocused)
1.97*
0.07
0.08
0.87
Proficiency x Focus (unfocused)
−0.00
0.00
−0.62
Age x Length x Frequency
−2.06
4.84
−0.43
Age x Length x Gender (male)
−31.72
17.05
−1.86
0.17
0.30
0.56
Age x Length x Focus (unfocused)
Age x Length x Proficiency
33.34
15.87
2.10*
Age x Frequency x Gender (male)
2.32
3.72
0.62
−0.05
0.06
−0.85
Age x Frequency x Proficiency
Age x Frequency x Focus (unfocused)
1.50
3.48
0.43
Age x Gender (male) x Proficiency
0.59
0.92
0.64
Age x Gender (male) x Focus (unfocused)
4.79
15.78
0.30
−0.08
0.29
−0.31
Age x Proficiency x Focus (unfocused)
Length x Frequency x Gender (male)
Length x Frequency x Proficiency
0.05
0.08
0.54
−0.00
0.00
−1.52
Change detection
Table 5. (continued)
Variable
B
SE (B)
t
Length x Frequency x Focus (unfocused)
0.13
0.08
1.66
Length x Gender (male) x Proficiency
0.00
0.00
0.37
Length x Gender (male) x Focus (unfocused)
0.04
0.26
0.14
−0.01
0.00
−1.24
Length x Proficiency x Focus (unfocused)
Frequency x Gender (male) x Proficiency
0.00
0.00
0.56
Frequency x Gender (male) x Focus (unfocused)
0.00
0.06
0.10
−0.00
0.01
−0.13
0.00
0.01
0.19
Frequency x Proficiency x Focus (unfocused)
Gender (male) x Proficiency x Focus (unfocused)
Significant Effects Involving Change
Change x Focus: F(2,1820) = 2.93, p = .050
Change x Proficiency: F(2,1820) = 6.39, p = .002
Change x Gender x Focus: F(2,1770) = 5.46, p = .004
*p < .05 **p < .01 ***p < .001
Note. The Age x Length x Focus interaction is only marginally significant if stimulus repetitions are taken
into account: B = 28.42, SE B = 16.10, t = 1.77, p = .077.
a changed word was noticed. In these critical Unfocused sentences, the changed
word always occurred quite early. If this was done systematically it would be expected to create a main effect of Focus, which this analysis did not show. If done
less systematically, it would serve primarily to increase variance, and as we will see
below, the variances in RTs were considerable.
Gender also interacted with Proficiency. Follow-up comparisons showed that
higher proficiency was associated with faster RTs, but only for women (B = −0.01,
SE B = 0.00, t = −4.93, p < .001; for men, B = 0.00, SE B = 0.00, t = 0.08, p = .93). The
literature does provide some evidence that girls consistently (and cross-culturally)
outperform boys on measures of linguistic ability, especially on reading and literacy (OECD/UNESCO, 2003). The utter lack of a proficiency effect on RTs for men
may also be partly explained by the conjectured lower conscientiousness of male
subjects, which was discussed previously. Men were perhaps not finishing the entire sentence before pressing the button (as instructed), so they produced faster
RTs, but as we saw in Figure 10, at least some of them missed a lot of changes.
Proficiency also interacted with Change. Higher proficiency was significantly associated with faster RT in the Unrelated condition (B = −0.01, SE B = 0.00,
t = −2.34, p < .05), but this relationship did not hold for either of the related conditions (for Subordinate changes, B = −0.00, SE B = 0.00, t = −1.02, p > .30; for Superordinate changes, B = −0.00, SE B = 0.00, t = −0.68, p > .49). Put another way, people
with more cognitive resources available (Chee et al., 2001) can perform the task
faster than those with fewer cognitive resources in the Unrelated condition, but
73
74
Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
Figure 11. Log reading times as a function of focus, age and word length.
if there is a semantic relationship between the original and the changed words,
people with less cognitive resources are able to perform just as quickly.
There were two significant three-way interactions. The Age x Length x Focus
interaction is shown in Figure 11. Older participants showed the expected Length
effect (i.e. longer RTs for sentences with longer words), but only in the Unfocused
condition. Younger participants showed the effect only in the Focused condition,
and in fact showed a reverse length effect in the Unfocused condition. We had no
reason to expect this kind of interaction, and it was not significant in any of the
other analyses. We can think of no reason why the Length x Focus interaction
should itself depend significantly on the age of the participant, particularly when
the age range of the participants is fairly restricted.
Finally, the Gender x Focus x Change interaction was significant (see Figure 12). For men (bottom panel), both the Subordinate and the Unrelated conditions produced faster RT for Unfocused sentences, while the Focused structure
produced faster RT in the Superordinate condition. This looks like the pattern observed for low-proficiency participants in the Pre-test (bottom panel of Figure 4),
but with much larger effects. Women showed smaller effects overall, and a pattern that was more reminiscent of the high-proficiency performance in Figure 4.
Because there was no proficiency difference between men and women, we have
offered the conjecture above that women’s similarity to high-proficiency performance could stem from higher conscientiousness or desire to comply.
Change detection
Figure 12a. Female participants’ log reading times as a function of focus and change
condition.
Figure 12b. Male participants’ log reading times as a function of focus and change
­condition.
75
76
Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
General Discussion
Effects of the variables under study here have not previously been examined with
a bilingual sample. This was important in terms of experimental design because it
allowed us to decouple form changes from meaning changes. It is also important
from a theoretical perspective because it allowed a more detailed look at the role
of proficiency, alone and in interaction with other variables.
Our Pre-test was used to pilot-test the paradigm, and while we concluded that
it worked well enough to proceed with the main experiment, it produced a surprisingly low number of missed changes (about 1%, compared to nearly 10% in Sturt
et al., 2004). This might be due to the fact that Sturt et al.’s related changes seem to
have remained within the Basic level. The example stimuli they provided are beer/
cider and hat/cap. Our related changes always involved either a Superordinate or
Subordinate term, and in the native language of monolinguals, these changes seem
to be as noticeable as changes to completely unrelated words. Analysis of the Pretest RTs did reveal additional information about processing, though. Although
change detection performance was not worse in the Unfocused sentences as Sturt
et al. found, these sentences were read faster (perhaps suggesting less detailed,
or “good enough” processing). This finding is in line with Langford and Holmes
(1979) who suggested that unfocused items are attended to less than focused items.
When change detection performance was brought down from the ceiling (i.e.
in the bilingual main experiment), interesting differences did emerge and many of
our hypotheses were supported. First, it was expected that Sturt et al.’s (2004) focus
effect would be replicated cross-linguistically. Indeed, Focused changes were more
likely to be noticed than Unfocused changes. This is in line with Ferreira et al.’s
(2002) account of GER, which supposes that deeper processing occurs when attention is focused. In the present project, like that of Sturt et al., focus affected the level
of granularity (or detail) that was encoded with respect to the detection of change.
Second, as expected, Unrelated changes were detected more often than either
type of related change. We only found partial support for our prediction that Subordinate changes would be missed more than Superordinate changes. Overall, the
miss rates for Subordinate and Superordinate changes were 31.5% and 25.5%, respectively. As we saw above, both of these were significantly worse than the Unrelated rate (16.5%), but they did not differ from each other as predicted. Figure 8
shows that the predicted effect (Subordinate producing more misses than Superordinate) was there for women, in unfocused sentences. The failure to find a difference more generally could be due to the low-to-moderate level of proficiency of
our participants — perhaps they have not fully associated the featural details with
the relevant terms, and so the “semantics” of Superordinate terms are still works in
progress in their mental lexicons.
Change detection
Third, it was expected that Proficiency would impact the granularity of the
representation that is formed. Indeed, highly proficient bilinguals missed fewer
changes than less proficient bilinguals, both in the pre-test and in the main experiment. In the main experiment there was also a significant RT advantage for
women with higher proficiency. Proficiency also seemed to temper some of the
slowing effects of age, focus and gender on RTs (Figures 9 and 10), and in the pretest, proficiency protected participants from an inhibitory effect of word Length
on RTs.
Our most consistent finding was an interaction between Gender, Focus, and
Change condition. Changes in focused sentences in the Unrelated condition were
almost always noticed (Figure 8), usually with moderate or fast RTs (Figure 12).
These words are unexpected and the focusing structure highlights them. With one
exception, the tendency was for changes in the focused condition to be detected
more often than those in the unfocused condition, as predicted. Developing a coherent theoretical explanation of the RTs is much more difficult though. As we
noted above, interpretation of the RTs is speculative because of the contradictory
predictions that can reasonably be made about them.
The same caution should be repeated about the age effects we have found in
the current study. While it is possible to speculate about their underlying causes,
all but two of our participants were between the ages of 18 and 26 (with the additional two participants being 40 and 41 years old). This research is the first of its
kind and our results should be replicated.
Comparing the results from this experiment to those of Sturt et al. (2004), we
found some support for the notion that both bilingual and monolingual processing follow similar principles. Like Sturt et al. (2004) report in their monolingual
data, we also found that focused constructions led to fewer missed changes in
bilinguals. An additional parallel involves related versus unrelated changes, where
related changes were consistently missed more. Thus, regardless of language status
(monolingual or bilingual), there is evidence that linguistic focus results in deeper
processing and that related changes, by virtue of their semantic overlap with the
original noun, are more often missed.
Our bilingual data also offer support for Ferreira et al.’s (2002) notion of GER.
In many cases, only the gist of the message is encoded and the level of detail in
the representation of linguistic stimuli depends, in part, on linguistic focus. It is
also important to address the practical importance of both GER and more detailed processing for the human cognitive system. GER allows for quick and efficient processing and is thus adequate in many situations. However, focusing
structures can be used to help ensure that any particularly important information
benefits from deeper processing. This finding could be used in real-world applications like creation of forms that people need to fill out, signs that direct or caution
77
78
Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
pedestrians or drivers, and so on. Making such messages longer (by adding words)
is probably not a good idea, but focusing can also be done in other ways (e.g. bold
type, italics, capital letters).
Conclusions and Future Directions
The current study provides a valuable jumping-off point for future work. A number of potential explanations are hinted at by the structure of our item and participant samples, and future research could be aimed at pulling these effects apart.
For example, a better factorial approach to participant recruitment, making sure
that there are equal numbers of men and women across a larger age range, could
replicate some of our findings and help place the age findings on a more secure
footing. Future studies should also use larger samples than we were able to recruit.
We observed considerable variance in performance, especially for the RTs, which
kept some potentially interesting effects from reaching statistical significance.
We used Aberra’s (2006) arguments to derive the prediction that change detection in the Subordinate condition should be worse than in the Superordinate
condition. Contrary to expectation, we found that performance in these two conditions was not significantly different. Future research should attempt to explore
Aberra’s prediction further, and also to separate possible effects of word frequency
and length from those of hierarchical level (i.e. Change). This will be difficult given
that there do seem to be natural associations between hierarchical level and these
variables, but it could perhaps be done with subsets if a large enough pool of potential items were identified.
Bilinguals process information more deeply than monolinguals, but only given a high enough level of proficiency in both languages (Bialystok, 1999; Malabonga & Pasnak, 2002). Interactions with proficiency in the current study suggest
that higher levels of proficiency produce performance more like the monolinguals.
Additional research using participants who are extremely proficient would thus be
interesting.
Research with people proficient in two very different languages would be interesting as well. Lemhöfer et al. (2008) have shown that results of bilingual studies
using various tasks generalize to speakers of different L1s (at least for European
languages). This suggests that our findings would generalize beyond EnglishFrench bilinguals, and indeed, we have no reason to expect that they would not. It
might be informative, though, to replicate this experiment using bilingual subjects
whose languages include one alphabetic and one non-alphabetic language.
Finally, more work needs to be done with respect to the RT measure. It is a
potentially rich source of information, but in the current paradigm the predic-
Change detection
tions to be made about RT are debatable. As one example, researchers could use a
moving text window or a self-paced reading paradigm to make sure that RTs apply only to the crucial word, rather than to an entire sentence. This would greatly
reduce variances, and also allow researchers to be more certain about the proper
interpretation of these RTs.
The current findings have two important implications. First, it suggests that
subject compliance or conscientiousness might be an issue, supporting the findings of Barenboym et al. (2010) regarding gender differences. In neither of our
experiments was there a gender difference on Proficiency, so in cases in which
one gender or the other showed the “high Proficiency” pattern, we resorted to
a speculative account based on conscientiousness or compliance. It was usually
women who showed this pattern, which sometimes included a paradoxical damaging effect on performance (i.e., in the Superordinate condition in Figure 8). We
did not expect to uncover effects like these in the current study, and thus we did
not include any measures of conscientiousness or compliance. Our data suggest
that it might be worth doing so, in order to determine whether our speculative
account of the data is plausible.
Second, our data partially support the interdependence view of bilingual language processing, such as proposed in Kroll and Stewart’s (1994) RHM. The RHM
suggests that meaning (rather than surface details) is encoded into a central conceptual storage, to which both languages have differential access based on relative
proficiency. Highly proficient bilinguals retrieve information conceptually, while
less proficient bilinguals have access by way of a more lexical, or form-based, route
such as translation equivalents. Furthermore, Ferré et al. (2006) have shown that
semantic variables (such as word taxonomy in our experiment) only affect bilinguals who have attained a certain proficiency (cf. Cummins’ [1979] “threshold”),
perhaps, because they are processing conceptually, rather than lexically.
Author Note
The first author is presently pursuing a Doctoral degree at Wayne State University, Detroit,
Michigan, USA.
This research was conducted in partial fulfilment of the primary author’s M.A. in Cognitive
Psychology at Wayne State University and was presented as a poster at the Sixth International
Conference on the Mental Lexicon, in October, 2008. Special thanks to the first author’s other
Master’s committee members — Dr. Patricia Siple and Dr. Katherine Paesani — for their valuable input.
79
80 Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
Notes
1. Our study design raises the question of repetition effects, as the critical words were presented
to each participant multiple times. While repeating an item four times may seem like a lot, because of the inclusion of a large set of filler items most of the words a participant saw (75%) were
not repeated. Furthermore, there were an average of 40 intervening trials between repetitions
of any given word. Subjectively the repetition may have been even less salient than this, because
half of the repetitions used the focused structure and half used the unfocused structure. We also
performed analyses supplemental to those presented in this paper, in which we added “Trial
Number” (and interactions with Trial number) as predictors. Twenty-two of our 23 significant
effects remain significant in these supplemental analyses and the 23rd goes from significant to
marginal (p < .08; see note to Table 5). We conclude that the repetition of items did not affect the
data in any major way.
2. The current version of the software (Baayen, 2009; Bates & Maechler, 2009) does not support
omnibus tests for multi-level factors if the dependent variable is dichotomous (as is the case in
Table 4). What we have done instead is to run two statistical models that are identical except
that one contains the “Change” factor, and one does not. The model containing the factor will
have two df more than the one without it. We then compare the predictive power of the two
models by making use of a χ2 test, the significance of which indicates whether the more complex model (i.e. the one with two more df) is warranted by improved predictive power (Baayen,
2008). Note that this approach is only needed for effects involving Change, and only when the
DV is dichotomous.
References
Aberra, D. A. (2006). The hierarchical relationship of words: Superordinate, hyponym and
subordinate. University of Alberta- Working Papers in Linguistics, 1. Retrieved from http://
www.uofaweb.ualberta.ca/linguistics/UofAWPL.cfm
Baayen, RH. (2008). Analyzing linguistic data: A practical introduction to statistics using R. Cambridge, UK: Cambridge University Press.
Baayen, RH. (2009). The languageR library. R package version 1.0. http://cran.r-project.org/
web/packages/languageR/index.html.
Baayen, R. H., Davidson, D. J., & Bates, D. M. (2008). Mixed-effects modeling with crossed random effects for subjects and items. Journal of Memory and Language, 59, 390–412.
Barenboym, D. A., Wurm, L. H., & Cano, A. (2010). A comparison of stimulus ratings made online and in person: Gender and method effects. Behavior Research Methods 42(1), 273–285.
Barton, S. B., & Sanford, A. J. (1993). A case study of anomaly detection: Shallow semantic processing and cohesion establishment. Memory & Cognition, 21(4), 477–487.
Basnight-Brown, D. M., & Altarriba, J. (2007). Differences in semantic and translation priming across languages: The role of language direction and language dominance. Memory &
Cognition, 35(5), 953–965.
Bates, D. M. (2005). Fitting linear mixed models in R. R News, 5, 27–30.
Bates, D. M., & Maechler, M. (2009). lme4: Linear mixed-effects models using S4 classes. R
package version 0.999375-32. http://CRAN.R-project.org/package=lme4
Change detection
Bates, D. M., & Sarkar, D. (2005). The lme4 library, from http://www.lib.stat.cmu.edu/CRAN
Bialystok, E. (1999). Cognitive complexity and cognitive control in the bilingual mind. Child
Development, 70(3), 636–644.
Birch, S. L., Albrecht, J. E., & Myers, J. L. (2000). Syntactic focusing structures influence discourse processing. Discourse Processes, 30(3), 285–304.
Birch, S. L., & Garnsey, S. M. (1995). The effect of focus on memory for words in sentences.
Journal of Memory and Language, 34(2), 232–267.
Bredart, S., & Docquier, M. (1989). The Moses illusion: A follow-up on the focalization effect.
Cahiers de Psychologie Cognitive/Current Psychology of Cognition, 9(3), 357–362.
Bredart, S., & Modolo, K. (1988). Moses strikes again: Focalization effect on a semantic illusion.
Acta Psychologica, 67(2), 135–144.
Brown, R. (1958). How shall a thing be called? Psychological Review, 65(1), 14–21.
Carrier, L. M., Cheever, N. A., Rosen, L. D., Benitez, S., & Chang, J. (2009). Multitasking across
generations: Multitasking choices and difficulty ratings in three generations of Americans.
Computers in Human Behavior, 25, 483–489.
CELEX English database (1993). from Dutch Centre for Lexical Information: http://www.mpi.
nl/world/celex
Chee, M. W. L., Hon, N., Lee, H. L., & Soon, C. S. (2001). Relative language proficiency modulates BOLD signal change when bilinguals perform semantic judgments. NeuroImage, 13,
1155–1163.
Faraway, J. J. (2006). Extending linear models with R: Generalized linear, mixed effects, and nonparametric regression models. Boca Raton, FL: CRC Press.
Ferré, P., Sanchez-Casas, R., & Guasch, M. (2006). Can a horse be a donkey? Semantic and form
interference effects in translation recognition in early and late proficient and nonproficient
Spanish-Catalan bilinguals. Language Learning, 56(4), 571–608.
Ferreira, F., Bailey, K. G. D., & Ferraro, V. (2002). Good-enough representations in language
comprehension. Current Directions in Psychological Science, 11(1), 11–15.
Ferreira, F., & Patson, N. D. (2007). The ‘good enough’ approach to language comprehension.
Language and Linguistics Compass, 1(1–2), 71–83.
Féry, C. (2001). Focus and Phrasing in French. In C. Féry & W. Sternefeld (Eds.), Audiatur Vox
Sapientiae: A Festschrift for A. v. Stechow (pp. 153–181). Berlin: Akademie Verlag.
Flege, J. E., MacKay, I. R. A., & Piske, T. (2002). Assessing bilingual dominance. Applied Psycholinguistics, 23(4), 567–598.
Foreign Language Placement Tests (2008). Madison, Wisconsin: University of Wisconsin Center for Placement Testing.
Francis, W. S. (1999). Cognitive integration of language and memory in bilinguals: Semantic
representation. Psychological Bulletin, 125(2), 193–222.
Heredia, R. R. (1997). Bilingual memory and hierarchical models: A case for language dominance. Current Directions in Psychological Science, 6(2), 34–39.
Hobbs, J. (1985). Granularity. Paper presented at the International Joint Conference on Articial
Intelligence, Los Angeles.
Kroll, J. F., & Stewart, E. (1994). Category interference in translation and picture naming: Evidence for asymmetric connection between bilingual memory representations. Journal of
Memory and Language, 33(2), 149–174.
Langford, J., & Holmes, V. M. (1979). Syntactic presupposition in sentence comprehension. Cognition, 7(4), 363–383.
81
82
Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
Lemhöfer, K., Dijkstra, T., Schriefers, H., Baayen, R. H., Grainger, J., & Zwitserlood, P. (2008).
Native language influences on word recognition in a second language: A megastudy. Journal of Experimental Psychology: Learning, Memory, and Cognition, 34(1), 12–31.
Liu, J., Golinkoff, M., & Sak, K. (2001). One cow does not an animal make: Young children can
extend novel words at the superordinate level. Child Development, 72(6), 1674–1694.
Malabonga, V., & Pasnak, R. (2002). Hierarchical categorization by bilingual Latino children:
Does a basic-level bias exist. Genetic, Social, and General Psychology Monographs, 128(4),
409–441.
McGregor, K. K., & Waxman, S. R. (1998). Object naming at multiple hierarchical levels: A comparison of preschoolers with and without word-finding deficits. Journal of Child Language,
25, 419–430.
McKoon, G., Ratcliff, R., Ward, G., & Sproat, R. (1993). Syntactic prominence effects on discourse processes. Journal of Memory and Language, 32(5), 593–607.
Mervis, C. B. (1987). Child-basic object categories and early lexical development. New York, NY,
US: Cambridge University Press.
Mervis, C. B., Johnson, K. E., & Mervis, C. A. (1994). Acquisition of subordinate categories by
3-year-olds: The roles of attribute salience, linguistic input, and child characteristics. Cognitive Development, 9(2), 211–234.
OECD/UNESCO (2003). Gender differences and similarities in achievement Literacy Skills for
the World of Tomorrow: Further Results From PISA 2000 (pp. 141–162). Paris: OECD Publishing.
Office Québécois de la Language Française (2007). Le Grand Dictionnaire Terminologique. Retrieved October 20: http://www.granddictionnaire.com/
Pinheiro, J. C., & Bates, D. M. (2000). Mixed-effects models in S and S-PLUS. New York: Springer.
Quené, H., & van den Bergh, H. (2008). Examples of mixed-effects modeling with crossed and
random effects and with binomial data. Journal of Memory and Language, 59, 413–425.
Rensink, R. A., O’Regan, J. K., & Clark, J. J. (1997). To see or not to see: The need for attention
to perceive changes in scenes. Psychological Science, 8(5), 368–373.
Rogers, T. T., & Patterson, K. (2007). Object categorization: Reversals and explanations of the
basic-level advantage. Journal of Experimental Psychology: General, 136(3), 451–469.
Rosch, E. (1978). Principles of categorization. In E. Rosch & B. Lloyd (Eds.), Cognition and Categorization (pp. 27–48). Hillsdale: Lawrence Erlbaum.
Rosch, E., Mervis, C. B., Gray, W. D., Johnson, D. M., & Boyes-Braem, P. (1976). Basic objects in
natural categories. Cognitive Psychology, 8(3), 382–439.
Ruts, W., De Deyne, S., Ameel, E., Vanpaemel, W., Verbeemen, T., & Storms, G. (2004). Dutch
norm data for 13 semantic categories and 338 exemplars. Behavior Research Methods, Instruments & Computers. Special Issue: Web-based archive of norms, stimuli, and data: Part 1,
36(3), 506–515.
Sachs, J. S. (1967). Recognition of semantic, syntactic and lexical changes in sentences. Psychonomic Bulletin, 1(2), 17–18.
Sanford, A. J. (2002). Context, attention and depth of processing during interpretation. Mind &
Language. Special Issue: Pragmatics and cognitive science, 17(1–2), 188–206.
Sanford, A. J., & Sturt, P. (2002). Depth of processing in language comprehension: Not noticing
the evidence. Trends in Cognitive Sciences, 6(9), 382–386.
Sanford, A. J. S., Sanford, A. J., Filik, R., & Molle, J. (2005). Depth of lexical-semantic processing
and sentential load. Journal of Memory and Language, 53(3), 378–396.
Change detection
Sanford, A. J. S., Sanford, A. J., Molle, J., & Emmott, C. (2006). Shallow processing and attention
capture in written and spoken discourse. Discourse Processes, 42(2), 109–130.
Schreuder, R., & Weltens, B. (2005). The bilingual lexicon: An overview. In J. F. Kroll & A. M.
B. De Groot (Eds.), Handbook of bilingualism: Psycholinguistic approaches (pp. 1–10). New
York: Oxford University Press.
Sharifian, F. (2002). Processing hyponymy in L1 and L2. Journal of Psycholinguistic Research,
31(4), 421–436.
Simons, D. J., & Rensink, R. A. (2005). Change blindness: Past, present, and future. Trends in
Cognitive Sciences, 9(1), 16–20.
Singer, M. (1976). Thematic structure and the integration of linguistic information. Journal of
Verbal Learning & Verbal Behavior, 15(5), 549–558.
Sturt, P., Sanford, A. J., Stewart, A., & Dawydiak, E. (2004). Linguistic focus and good-enough
representations: An application of the change-detection paradigm. Psychonomic Bulletin &
Review, 11(5), 882–888.
Van Overschelde, J. P., Rawson, K. A., & Dunlosky, J. (2004). Category norms: An updated and
expanded version of the Battig and Montague (1969) norms. Journal of Memory and Language, 50(3), 289–335.
Zupančič, M., Gril, A., & Kavčič, T. (2006). Child and early adolescent personality: Its structure,
age trends and gender differences. Studia Psychologica, 48(4), 311–332.
Appendix A: Stimulus Sentences (bilingual)
Not Focused
Le/la TN a sauté la clôture.
The TN leaped over the gate.
Le/la TN chantait de la cage.
The TN chirped from the cage.
Le/la TN fut importé(e) de la Floride.
The TN was imported from Florida.
Le/la TN le protégea de la précipitation.
The TN protected him from the precipitation.
Le/la TN était préféré(e).
The TN one was preferred.
Le/la TN a été ajouté(e) à la sandwich.
The TN was added to the sandwich.
Le/la TN était au sommet de la colline.
The TN was on top of the hill.
Le/la TN a été acheté(e) à la vente.
The TN was purchased at the sale.
Le/la TN a duré toute la soirée.
The TN lasted all night.
Le/la TN a brisé(e) lorsqu’il fut échappé.
The TN shattered when it was dropped.
Le/la TN grinçait lorsqu’elle marchait.
The TN squeaked as she walked.
Le/la TN était sucré(e).
The TN was sweet.
Le/la TN était vieux/vieille et déchiré(e).
The TN was old and ripped.
Le/la TN fut livré(e) ce matin.
The TN was delivered this morning.
Le/la TN était très coûteux(se).
The TN was quite expensive.
Le/la TN fut laissé(e) sur l’estrade.
The TN was left on the stage.
Le/la TN s’est fait mouillé dans la piscine.
The TN got wet in the pool.
83
84
Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
Le/la TN remuait sa queue.
The TN wagged its tail.
Le/la TN fut utilisé(é) pour fabriquer une
chemise.
The TN was used to make a shirt.
Le/la TN fut importé(e) du Canada.
The TN was imported from Canada.
Le/la TN a grondé l’enfant.
The TN scolded the child.
Le/la TN a été ajouté(e) au jardin.
The TN was added to the garden.
Le/la TN a grandi pendant l’été.
The TN grew big over the summer.
Le/la TN était sur le seuil de la fenêtre.
The TN was on the window sill.
Le/la TN était aimé(e) par la communauté.
The TN was liked by the community.
Le/la TN regardait à travers la vitre de l’aqua- The TN looked through the glass of the
rium.
aquarium.
Le/la TN a tombé dans l’escalier.
The TN fell down the stairs.
Le/la TN a fait beaucoup de bruit.
The TN made a lot of noise.
Le/la TN était utilisé(e) pour servir la nourriture.
The TN was used to serve the food.
Le/la TN remplissait l’assiette.
The TN filled the plate.
Le/la TN s’est précipité dans la rue.
The TN sped down the street.
Le/la TN était pris(e) dans le filet.
The TN was caught in the net.
Focused
C’était le/la TN qui il a sauté la clôture.
It was the TN that leaped over the gate.
C’était le/la TN qui chantait de la cage.
It was the TN that chirped from the cage.
C’était le/la TN qui fut importé(e) de la
Floride.
It was the TN that was imported from Florida.
C’était le/la TN qui le protégea de la précipitation.
It was the TN that protected him from the
precipitation.
C’était le/la TN qui était préféré(e).
It was the TN one that was preferred.
C’était le/la TN qui a été ajouté(e) à la sandwich.
It was the TN that was added to the sandwich.
C’était le/la TN qui était au sommet de la
colline.
It was the TN that was on top of the hill.
C’était le/la TN qui a été acheté(e) à la vente.
It was the TN that was purchased at the sale.
C’était le/la TN qui a duré toute la soirée.
It was the TN that lasted all night.
C’était le/la TN qui a brisé(e) lorsqu’il fut
échappé.
It was the TN that shattered when it was
dropped.
C’était le/la TN qui grinçait lorsqu’elle
marchait.
It was the TN that squeaked as she walked.
C’était le/la TN qui était sucré(e).
It was the TN that was sweet.
Change detection
C’était le/la TN qui était vieux/vieille et
déchiré(e).
It was the TN that was old and ripped.
C’était le/la TN qui fut livré(e) ce matin.
It was the TN that was delivered this morning.
C’était le/la TN qui était très coûteux(se).
It was the TN that was quite expensive.
C’était le/la TN qui fut laissé(e) sur l’estrade.
It was the TN that was left on the stage.
C’était le/la TN qui s’est fait mouillé dans la
piscine.
It was the TN that got wet in the pool.
C’était le/la TN qui remuait sa queue.
It was the TN that wagged its tail.
C’était le/la TN qui fut utilisé(é) pour fabriquer une chemise.
It was the TN that was used to make a shirt.
C’était le/la TN qui fut importé(e) du Canada.
It was the TN that was imported from Canada.
C’était le/la TN qui TN a grondé l’enfant.
It was the TN that scolded the child.
C’était le/la TN qui a été ajouté(e) au jardin.
It was the TN that was added to the garden.
C’était le/la TN qui a grandi pendant l’été.
It was the TN that grew big over the summer.
C’était le/la TN qui était sur le seuil de la
fenêtre.
It was the TN that was on the window sill.
C’était le/la TN qui était aimé(e) par la communauté.
It was the TN that was liked by the community.
C’était le/la TN qui regardait à travers la vitre It was the TN that looked through the glass of
de l’aquarium.
the aquarium.
C’était le/la TN qui a tombé dans l’escalier.
It was the TN that fell down the stairs.
C’était le/la TN qui a fait beaucoup de bruit.
It was the TN that made a lot of noise.
C’était le/la TN qui était utilisé(e) pour servir It was the TN that was used to serve the food.
la nourriture.
C’était le/la TN qui remplissait l’assiette.
It was the TN that filled the plate.
C’était le/la TN qui s’est précipité dans la rue. It was the TN that sped down the street.
C’était le/la TN qui était pris(e) dans le filet.
It was the TN that was caught in the net.
Appendix B: English TN Triads
BASIC
SUPERORDINATE
SUBORDINATE
UNRELATED
horse
animal
stallion
secretary
bird
animal
pigeon
whistle
orange
fruit
clementine
shrub
coat
clothing
*raincoat
block
yellow
color
amber
heavy
85
86 Lynne N. Kennette, Lee H. Wurm and Lisa R. Van Havermaet
mustard
condiment
dijon
toothpick
house
dwelling
cottage
detective
camera
electronics
digital
armoire
game
entertainment
darts
class
glasses
eyewear
bifocals
lamp
shoe
footwear
heels
medal
melon
fruit
cantaloupe
nurse
bed
furniture
crib
curtain
chair
furniture
recliner
film
piano
instrument
keyboard
book
flute
instrument
clarinet
candy
watch
jewelry
*stopwatch
friend
dog
mammals
dalmatian
turtle
fabric
material
silk
pattern
wood
materials
oak
frame
parent
person
father
television
herb
plant
mint
bench
tree
plant
pine
monkey
flower
plant
rose
book
doctor
professional
pediatrician
park
lizard
reptile
chameleon
lawyer
ball
toy
*ball
knife
airplane
transportation
jet
cow
spoon
utensil
ladle
volunteer
lettuce
vegetable
romaine
money
car
vehicle
sedan
fly
fish
animal
salmon
ring
*In French, this word does not include the basic level term.
Author’s address
Lynne N. Kennette
Department of Psychology
Wayne State University
5057 Woodward Ave, 7th Floor
Detroit, MI, 48202
Phone: (313) 577-2800
Fax: (313) 577-7636
[email protected]
Copyright of Mental Lexicon is the property of John Benjamins Publishing Co. and its content may not be
copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written
permission. However, users may print, download, or email articles for individual use.