– 204
LANGUAGE
L. H. Wurm,
AND SPEECH,
D. A. Vakoch,
2004,and
47 S.
(2),
R.175
Seaman
Authors
175
Recognition of Spoken Words:
Semantic Effects in Lexical Access
Lee H. Wurm,1 Douglas A. Vakoch,2
and Sean R. Seaman 1
1 Wayne
2 SETI
Key words
State University, Detroit
Institute and University of California, Davis
Abstract
Until recently most models of word recognition have assumed that semantic
effects come into play only after the identification of the word in question.
What little evidence exists for early semantic effects in word recognition
lexical decision
has relied primarily on priming manipulations using the lexical decision
task, and has used visual stimulus presentation. The current study uses
semantics
auditory stimulus presentation and multiple experimental tasks, and does
not use priming. Response latencies for 100 common nouns were found to
speech
depend on perceptual dimensions identified by Osgood (1969): Evaluation,
perception
Potency, and Activity. In addition, the two-way interactions between these
dimensions were significant. All effects were above and beyond the effects
word recognition
of concreteness, word length, frequency, onset phoneme characteristics,
stress, and neighborhood density. Results are discussed against evidence from several areas of
research suggesting a role of behaviorally important information in perception.
auditory naming
1 Introduction
When an organism gathers information about its environment, two characteristics of
the objects in that environment are particularly salient for the continued survival of
the organism. One is the threat that the object poses. The other, less often mentioned
but no less important, is the desirability (i.e., usefulness or value) of the object.
Information exchange is beneficial to the extent that the observer learns about both
of these characteristics (Darwin, 1859 / 1968; Davidson, 1992; Schneirla, 1965). The
duality of threat and desirability is consistent with evidence that there are two different
* Acknowledgments: This work was partially supported by a grant to the second author
from The John Templeton Foundation. The opinions expressed in this paper are those
of the authors and do not necessarily reflect the views of The John Templeton Foundation.
We would like to thank Joanna Aycock for help with stimulus preparation and data collection. We are also grateful to the Working group in Human Cognitive Neuroscience at
Wayne State University for making valuable comments about an earlier version of this paper.
Address for correspondence. Lee H. Wurm, Department of Psychology, Wayne State
University, 71 West Warren Avenue, Detroit, MI 48202; tel.: 1-313-577-0771; fax: 1-313577-7636; e-mail: <[email protected] >.
‘Language and Speech’ is ©Kingston Press Ltd. 1958 – 2004
Language and Speech
176
Semantic effects in lexical access
kinds of reactions to affective stimuli: appetitive reactions and defensive reactions
(Bradley, Codispoti, Cuthbert, & Lang, 2001; Lang, Bradley, & Cuthbert, 1998). Such
research has examined physiological responses (e.g., skin conductance, heart rate,
and modulation of the startle reflex) as well as subjective participant ratings, to visually-presented materials with different kinds of affective content. One hypothesis
motivating the current research is that appetitive and defensive responses to stimuli
may correspond to aspects of meaning that are coded in lexical entries.
Researchers in other areas of psychology have also been interested in issues like
these. Osgood (1969) proposed a model in which the meaning of words can be understood in terms of three dimensions, which code behaviorally significant semantic
information. It is adaptive, according to this view, to be able to categorize items quickly
along the three dimensions of Evaluation (“Is something good or bad?”), Activity (“Is
it fast or slow?”), and Potency (“Is it strong or weak?”; Osgood, May, & Miron, 1975;
Osgood, Suci, & Tannenbaum, 1957). These dimensions were derived from participant judgments, for example in studies where the apparent similarity of pairs of a
wide range of objects was rated. Typically, many such similarity ratings are analyzed
using multidimensional scaling techniques, yielding a number of orthogonal vectors
that explain the variance in the judgments. The dimensions that consistently emerge
from these analyses are most readily interpretable as Evaluation, Potency, and Activity.
Some researchers have wondered whether these dimensions are also related to
faster, more automatic cognitive processes than making similarity judgments. Chen
and Bargh (1999) used an interesting methodology to investigate automatic attitude
activation. They had participants either pull a lever toward themselves or push a lever
away as a means of classifying visually-presented words as either good or bad.
Participants responded to “good” words faster by pulling the lever than by pushing
it, and they responded to “bad” words faster by pushing the lever away than by pulling
it. According to the authors, replicating this congruence effect with a task that does
not involve explicit evaluation of stimuli would demonstrate that approach / avoid
responses are linked to automatic evaluative processes. A second experiment was then
run, in which the participants did not make evaluative classifications at all; their task
was simply to pull the lever (or push it away, in another condition) whenever a word
was displayed. Here, too, it was found that “pull” responses were faster for good
words and “push” responses were faster for bad words. The authors interpreted the
results of this study in terms of behavioral predispositions to approach good things
and avoid bad things.
Chen and Bargh’s (1999) discussion of the benefit of an unconscious evaluative
mechanism that responds to all “attitude objects” suggests that one might be able to
observe approach and avoid response tendencies with tasks designed to tap into online language processes such as word recognition. In the current study we look for
evidence of this by defining word meaning on the dimensions of Evaluation, Potency,
and Activity. It would be foolish to argue that the location of a word in a semantic space
defined by these dimensions is by any means sufficient for complete semantic processing.
However, such information can in principle be of tremendous use to an observer even
before a full, detailed semantic analysis is complete, and it is plausible that the mental
representations of words and other objects contain such information.1
Language and Speech
L. H. Wurm, D. A. Vakoch, and S. R. Seaman
177
In some of our earlier work, we have found evidence of such effects in word
recognition. Wurm and Vakoch (1996; Vakoch & Wurm, 1997) found that lexical
decision times were related to Evaluation, Potency, and Activity. The results of these
studies were interpreted within the framework outlined above (Osgood, 1969). In a
study of emotion words (Wurm & Vakoch, 1996), lexical decision times suggested an
“avoid” tendency: The fastest decision times were for words rated low on Evaluation,
high on Potency, and high on Activity (i.e., the combination of “bad” plus “strong”
plus “fast”). Interestingly, in a subsequent study looking at words much more generally (selected at random from a large dictionary), lexical decision times were not fastest
for this combination of maximally threatening connotations; instead, words with
“good” connotations had the fastest decision times. The overall data pattern was
interpreted to indicate that in lexical processing in general (i.e., not specifically for
emotion words), recognition latencies reflect an “approach” response tendency. This
interpretation will be used to guide predictions for the current study, presented below.
The question of whether semantic information can influence lower-level word
recognition is crucial theoretically, as it speaks to whether strictly modular architectures can accommodate the human experimental data, or whether interactive models
are more appropriate. Historically, most models of word recognition have not directly
incorporated semantics. Instead, it has generally been assumed that semantic information becomes available only after word identification. However, some theorists
have demonstrated that semantic information becomes available well before unique
identification of a single word. In Marslen-Wilson’s (1987) cohort model, for example,
the first 150 ms of speech that a listener hears is sufficient for preliminary activation
of limited aspects of semantic information for the subset of words consistent with this
acoustic input. Readers should note that 150 ms of acoustic input is generally not
enough to uniquely identify a spoken word, so some aspects of meaning are believed
to receive activation before the recognition of any single word.
Zwitserlood (1989) reached a conclusion consistent with Marslen-Wilson’s (1987),
based on a series of experiments in Dutch that used cross-modal priming. In these experiments, an auditory prime word preceded a visual probe word, and participants were
required to make a lexical decision about the probe word. She found that presenting
the first three phonemes ( [ k{p ] ) of kapitein (‘captain’) led to faster recognition times
for visually-presented probe words related semantically to both kapitein and kapitaal
(‘capital’). This multiple activation occurred even when the sentence context was
strongly biased toward one or the other word.
More recently, Tyler, Moss, Galpin, and Voice (2002) studied the effects of one
specific semantic variable (concreteness / imageability) in cross-modal priming experiments.2 One goal of the study was to elucidate the time-course of semantic activation,
1
Several metaphors for lexical entries are plausible; when we say that a lexical entry
contains certain kinds of semantic information (or points to it, or is associated with it),
we are not taking a strong theoretical position about the nature of lexical entries. Given
the current state of knowledge in the field, we think it would be inappropriate to do so.
2
Tyler et al. (2002) and Tyler, Voice, and Moss (2000, discussed below) consider concreteness and imageability to be the same concept, and use the terms interchangeably.
Language and Speech
178
Semantic effects in lexical access
which they attempted by using three probe locations: the identification point (located
empirically with gating data), word offset, and 250 ms post-offset. The authors found
that concrete words were primed better than abstract ones, although both showed
significant priming. Importantly, these effects were present at all probe positions,
indicating that this particular kind of semantic information is available very early in
the recognition process.
The authors argue that although no current model is sufficient to account for
these data, one like that of Gaskell and Marslen-Wilson (1997) seems promising. In
this model the speech signal is mapped directly onto a distributed representation that
carries all information associated with a word (i.e., both semantics and phonology,
along with other things). By implication, the connection weights in the model must
encode information about both semantics and phonology, and therefore semantic
information will be intricately involved in the recognition process from the start.
Because there is only the one, all-inclusive representation for a word, lexical (including
semantic) and phonological information are represented and accessed in parallel.
One interesting aspect of this model is the assertion that in the part of the speech
signal prior to the uniqueness point (UP — Marslen-Wilson, 1984; Marslen-Wilson
and Welsh, 1978), all word candidates are activated. This necessarily means that their
associated semantic information is activated, as well. The result, in this distributed
system, is a blend of the semantics of all words that match the partial input. At the
UP, both the phonological and semantic information become disambiguated. As the
authors point out, it is not clear whether this kind of model can handle the experimental literature on multiple activation of meaning, some of which was summarized
above. We will have more to say about this model below.
Many of the studies summarized so far used priming paradigms. There have
also been many attempts to observe unprimed semantic effects in word recognition,
in which words are presented in isolation and with no close associative relationship
to the other items in the list. Nearly all work in this area has used visual presentation
of stimulus materials, which is worth noting because there is evidence that orthographic processing of words is influenced by phonological characteristics of those words,
even though these auditory / phonological processes are not needed for reading.
Most of these attempts have shown no effect or very small effects (Brown & Watson,
1987; de Groot, 1989). However, Rodd, Gaskell, and Marslen-Wilson (2002) were able
to demonstrate that words with multiple senses had faster visual lexical decision times
only if the multiple senses are related. If they are not related, there is a delay associated
with multiple senses. Strain, Patterson, and Seidenberg (1995; Strain & Herdman, 1999)
found a naming-time advantage for concrete over abstract words. However, the effect
held only for low-frequency exception words. Exception words are those in which the
pronunciation of some letter string is not the most frequent pronunciation, or the one
predicted by linguistic rules. Strain et al. (1995) argued that semantic activation occurs
very early for all words, but it is only observable when the translation from orthography
to phonology is slow or noisy, as in the case of low-frequency exception words.
With spoken materials there is no orthography to phonology translation process,
and researchers can therefore see whether the observation of semantic effects depends
on this factor. Tyler, Voice, and Moss (2000) took advantage of this logic, showing in
Language and Speech
L. H. Wurm, D. A. Vakoch, and S. R. Seaman
179
auditory lexical decision and naming experiments that concrete words are identified
faster than abstract words. They further found that only words with large cohorts
showed the concreteness effect, and concluded that a word’s semantic information
can help with recognition when the discrimination process is more difficult.
In the current study we take a similar approach, except that we look beyond the
effects of variables like concreteness, to see if other, more behaviorally relevant
semantic variables also influence the recognition process. Nouns were presented auditorily to participants who were required either to make a speeded lexical decision
about the stimuli (Experiment 1) or to name the stimuli as quickly as possible
(Experiment 2).
The current study extends existing work in this area and differs from it in several
important ways. First, the present study is not a priming study. We are interested in
determining whether the time-course of word recognition is affected by semantic
factors, without any priming or context manipulation. Finding such effects would
allow a more basic conclusion to be drawn about the organization of lexical memory
because it would not be dependent on the possible relationships between pairs of
closely related words.
Second, the current study uses multiple experimental paradigms. The preponderance of evidence summarized above has relied solely on the lexical decision task.
This task has been criticized as susceptible to strategic effects and postperceptual bias
(e.g., Balota, 1990; Balota & Chumbley, 1984; Chumbley & Balota, 1984; Neely, 1991;
Neely & Keefe, 1989). In this study both the lexical decision task and the naming task
were used.
The third major difference between previous studies and the current study is in
the conceptualization of semantic effects. Previous studies have focused on variables
such as concreteness / imageability. Instead, variables such as these will be controlled
for, and a more comprehensive constellation of semantic dimensions will be examined.
These dimensions carry behaviorally significant information and provide a coherent
framework for clarifying subtle nuances of connotative meaning.
Finally, while nearly all of the studies discussed thus far used visual stimulus presentation, stimuli in the current study are presented auditorily. Speech is a signal that
unfolds over time, so auditorily-presented stimuli might allow a better look at the
effects in question. In addition, because speech is an older and more fundamental
cognitive process than is reading, one might expect auditorily-presented language to
be more likely to show a low-level link between semantics and word recognition.
2 Predictions
Following Vakoch and Wurm (1997; Wurm & Vakoch, 2000; Wurm, Vakoch, Aycock,
& Childers, 2003), our general hypothesis is that word recognition times will be characterized by an approach response tendency. This leads to several specific predictions.
First, words with higher ratings on Evaluation should be recognized faster, as they
were in the Vakoch and Wurm (1997) study examining general lexical processing. The
“good” end of this dimension seems quite clearly related to “approach” responding,
whereas the “bad” end seems clearly related to “avoid” responding. Second, the
Language and Speech
180
Semantic effects in lexical access
Activity dimension is characterized as imposing a deadline for response production,
so the higher the Activity rating, the more quickly responses should be made. Third,
lower Potency means that there is minimal risk associated with the hypothesized
tendency to approach, and so responses should proceed more quickly than in the
context of high Potency (Potency represents the overall threat or danger level).
We also expect interactions between ratings on these dimensions. Based on the
results of Vakoch and Wurm (1997), we hypothesize that the interaction patterns
should reflect the same processes captured by the main effects, but sometimes amplified by the combination of facilitative effects, and sometimes modulated by the
ambiguity of certain contexts (e.g., high Potency). Put another way, the main effect
of any particular dimension is expected to be more pronounced for words predicted
to be fast on the basis of a second dimension. As an example, fast response times are
expected for words rated higher on Activity, and for words rated lower on Potency.
In terms of the interaction, then, the expectation is that the effect of Activity will be
more pronounced for words lower on Potency than for words higher on Potency. For
these words, which carry nonthreatening connotations, the behavioral implication of
the Activity dimension is straightforward: higher Activity calls for faster responding.
This relationship is not expected to hold for words with high ratings on Potency. For
these words, the effect of Activity should be weaker, because higher Potency not only
increases the stakes, but introduces substantial ambiguity about how to respond:
Should one flee? Fight? Carefully approach?
Similarly, while lower values of Potency should be associated with faster responding,
this pattern should be seen most clearly with words with high Evaluation ratings. The
relationship between Potency ratings and response times for words low on Evaluation
(i.e., those with bad connotations) is expected to be weaker. This is because low
Evaluation reduces the ambiguity of high Potency, by ruling out any approach response.
Finally, the facilitative effect of higher Activity ratings is expected to be especially pronounced in the context of higher Evaluation ratings. If a word’s referent is
good, then the approach tendency should clearly be seen. If, however, a word’s referent
is bad, the effect should be tempered because of the ambiguity described above.
3 Preliminary rating study
In this study ratings were obtained from participants for the stimulus words on the
three dimensions of interest, and values for each stimulus on a number of potential
control variables were computed. The dimension weights will be used as predictor
variables in an analysis of lexical decision and naming times from the next experiments,
after statistically removing the effects of several control variables described below.
3.1
Method
Participants. Thirty-two undergraduate students from the Wayne State University
psychology subject pool participated. All were native speakers of English who reported
normal hearing and normal or corrected-to-normal vision. Participants received extra
credit in a psychology course for their participation.
Language and Speech
L. H. Wurm, D. A. Vakoch, and S. R. Seaman
181
Materials. One hundred nouns used by Wurm and Vakoch (2000) were used for the
present study. They were selected to roughly fill a 2 × 2 semantic space defined by high
and low danger and usefulness, two dimensions of meaning not being considered in
the current study. Inspection of the Appendix shows that these are common words,
likely to be familiar to undergraduate students. An additional 100 nouns were selected
via the same process, and turned into pseudowords by changing the phoneme at the
UP to a different phoneme from the same broad class. For example, the word ‘poverty’
was changed to the pseudoword ‘pozerty’, and ‘lavender’ was changed to ‘lavelder’.
Procedure. Rating packets were printed that contained the 100 real words (i.e., not
the pseudowords) to be used in the main experiment. Each rating packet was printed
in one of six different random orders. For Evaluation, the instructions printed at the
top of the rating packet said “Please rate how GOOD OR BAD each of the following
things is. Write a number in the blank next to each word, using the following scale.”
The capitalized words were changed to “STRONG OR WEAK” for Potency, and
“FAST OR SLOW” for Activity. In each case, this instruction was followed by a
number line with eight equally-spaced points on it. Extreme points on the scales were
labeled ‘bad’ (1) and ‘good’ (8) for Evaluation; ‘weak’ (1) and ‘strong’ (8) for Potency;
and ‘slow’ (1) and ‘fast’ (8) for Activity. The order in which ratings were given was
randomized.
3.2
Results
A complete list of the stimuli (along with mean ratings) can be found in the Appendix.
The mean ratings on Evaluation, Potency, and Activity were 4.23 (SD = 1.78), 4.70
(SD = 1.74), and 3.85 (SD = 1.84), respectively. Items covered almost the entire range
from one to eight on all dimensions, and the distributions of ratings were similar.
Table 1 shows the intercorrelations for ratings on these dimensions.
TABLE 1
Bivariate intercorrelations for semantic dimensions
Potency
Activity
Evaluation
– .52**
– .29**
Potency
--
.58**
**p < .01.
4 Calculation of control variables
Many variables are known to affect word recognition performance. Given the present
research questions, it was not possible to equate stimuli on all of these variables;
doing so would have left too few stimuli to work with. Instead, these variables were
controlled for by including them as factors in a regression model in the main reaction-time experiment, prior to assessing the effects of Evaluation, Potency, and Activity.
Language and Speech
182
Semantic effects in lexical access
Furthermore, although many readers may be more familiar with ANOVA procedures,
regression is more appropriate for many kinds of cognitive research, and it is also
more powerful (e.g., Cohen & Cohen, 1983; Lorch & Myers, 1990).
Eight classes of control variables were identified. The following paragraphs list
these variables and describe how each was computed.
1.
Frequency and familiarity: Word frequency values were taken from the CELEX
lexical database (Baayen, Piepenbrock, & van Rijn, 1993; Burnage, 1990), and
familiarity ratings from the Oxford Psycholinguistic Database (Quinlan, 1987).
2.
Word length was computed two ways for each word: in milliseconds and in
number of phonemes.
3.
Bigram frequency is a measure of how frequently pairs of letters occur, as a function of letter position and word length. This was calculated for each word using
the tables of Mayzner and Tresselt (1965).3
4.
First-syllable stress was coded as either strong (1) or weak (0). Some models of speech
perception make a distinction between items with strong initial syllables and those
with weak initial syllables (e.g., Cutler & Norris, 1988; Grosjean & Gee, 1987),
and stress has been shown to influence lexical decision times (e.g., Wurm, 1997).
5.
Neighborhood density refers to the number of words that are orthographically
or phonetically similar. There is abundant evidence that processing difficulty
for a given word depends on the number of neighbors it has (e.g., Coltheart,
Davelaar, Jonasson, & Besner, 1977; Goldinger, Luce, & Pisoni, 1989; Luce,
Pisoni, & Goldinger, 1990).
Four different measures of neighborhood density were computed, using the
Carnegie Mellon Pronouncing Dictionary (1995). One of the measures used in
the present study was an auditory analog of Coltheart’s N (Coltheart et al., 1977),
which is the number of words that can be made from a given target word by the
substitution of one letter, preserving letter position. For example, if the target
word is pin, then pen, win, and pit (along with many others) would be included
in the list of neighbors.
The second approach is like Coltheart’s N, but also allows for the addition or
deletion of one letter at the beginning or end of a word. Using the above example,
we would also add to the list of neighbors words such as spin and pins.
For both of these measures, the summed frequency of all of a word’s neighbors
was also computed, providing the third and fourth density measures.
3
Bigram frequency was used as an approximation to its auditory analog. As is the case
with frequency counts, there is often a trade-off in that auditory values would be more
obviously relevant, but visual values are either more readily available or come from larger,
more reliable corpora. Fortunately, the auditory and visual versions of such counts generally correlate very highly.
Language and Speech
L. H. Wurm, D. A. Vakoch, and S. R. Seaman
183
6.
Concreteness, imagery, and meaningfulness were taken from the Oxford
Psycholinguistic Database (Quinlan, 1987). This database includes two different
measures of meaningfulness, both of which were used in the preliminary analyses.
Words that are rated as more concrete, more imageable, or more meaningful
tend to be recognized more quickly (e.g., James, 1975; Whaley, 1978).
7.
Animate or inanimate referent was also included. McRae, de Sa, and Seidenberg
(1997) found that semantic features cluster differently for living things and artifacts, so we decided to include it as a factor.
8.
Experiment 2 used the naming paradigm, and included the control variables
discussed so far plus four more variables that coded characteristics of the initial
phoneme of the word. The initial phoneme can influence the ease (and thus the
speed) of lexical processing, and it affects naming times (e.g., Bates, Devescovi,
Pizzamiglio, & D’Amico, 1995). Typically researchers using the naming paradigm
try to match different groups of stimuli on the initial phoneme. This study does
not have groups of stimuli — each word is essentially its own condition. We
therefore included three different characteristics of the onset phoneme: vowel versus
consonant, place-of-articulation, and voicing. Two of these characteristics (vowel
vs. consonant onset, and voicing) are binary, and so each was simply coded as 0
or 1. Three places of articulation were coded: front (i.e., bilabial and labiodental), middle (i.e., interdental and alveolar), and back (i.e., palatal and velar).
This phoneme characteristic therefore required two (i.e., N-1) dummy-coded
variables. For example, “front” was coded as 0 0, “middle” was coded as 1 0,
and “back” was coded as 0 1 (see Cohen & Cohen, 1983).4
The effects of the control variables were statistically removed before any possible
effects of Evaluation, Potency, and Activity were assessed. In cases where we
had more than one measure of a construct (e.g., there were 4 measures of neighborhood density), only the measure that accounted for the most variance was
included (to avoid collinearity problems).
5 Experiment 1: Auditory lexical decision
This experiment provided a test of the adaptiveness model outlined above, defining
word referents in terms of Evaluation, Potency, and Activity. The major question
addressed by this study is: Controlling for the effects of concreteness, frequency,
length, and other important characteristics of words, will the judged Evaluation,
Potency, and Activity of those words affect response latencies in the ways predicted?
4
Kessler, Treiman, and Mullennix (2002) recommend blocking stimuli according to their
onsets, up to and including the first vowel. This is sometimes feasible, but given the
design of the current study, it is not: We would end up with dozens of different blocks,
most of which contain one or two items. In other recent work (Wurm & Aycock, 2003a,
2003b) we have found equivalent results when comparing Kessler et al.’s (2002) blocking
method to the method we will use in the current study.
Language and Speech
184
Semantic effects in lexical access
5.1
Method
Participants. Participants were 65 undergraduate psychology students. All were native
speakers of English who reported normal hearing. They received extra credit in a
psychology course for their participation. None of them had participated in the
preliminary rating study.
Materials. The critical items presented to subjects were the 100 nouns and 100 pseudowords described above. An additional 230 words were selected for use as filler items,
to be included so that the stimulus list would not be comprised entirely of nouns.
Half of these items were changed to pseudowords by the method described above. Filler
words were common, recognizable words of various parts of speech, and were from
one to five syllables long, as were the critical items. Each stimulus word was digitized
at a sampling rate of 20 kHz (low-pass filtered at 9.8 kHz) and stored in a disc file.
Procedure. Participants were tested in groups of one to three in a sound-attenuating
booth. They listened to the stimuli, played over headphones at a comfortable listening
level, and were instructed to decide as quickly as possible if each stimulus was a
genuine English word or not. Each participant used his / her dominant hand to make
responses on a button board, pushing one button for real words and a different button
for pseudowords. A different random stimulus order was used for each group of
participants.
A practice set of 24 items (12 words and 12 pseudowords) was used prior to the
main experiment to familiarize participants with the procedure.
5.2
Results
Data were discarded for trials on which the word / pseudoword decision was made
incorrectly (9.4% of the data). One item (‘canteen’) was excluded because it was an
extreme outlier in the UP distribution (z = 3.7, p = .0001; see Tabachnick & Fidell,
2001). The dependent variable in all analyses was RT on correct trials.
Lexical decision times were measured from the UP of each word. The first author
and a research assistant made independent measurements of each UP, which was
defined as the middle of the prototypical segment of the particular phone in question (following Radeau, Mousty, & Bertelson, 1989; see also Wurm & Ross, 2001). This
point was located using both visual and auditory criteria, with the help of a commercial waveform editor. Measurements for a given word were generally within a few
msec of each other: the mean difference for the 100 critical items was 4.3 ms. The
value used for the UP of each stimulus was the mean of these two independent measurements. None of the correlations between UP location and the semantic dimensions
approached significance.
A hierarchical multiple regression analysis was conducted in which the main
effects of Evaluation, Potency, and Activity were assessed, along with their interactions. Variables were entered in five steps, with simultaneous entry for all variables within
a given step. This analytic procedure parallels that of earlier studies (Vakoch & Wurm,
Language and Speech
L. H. Wurm, D. A. Vakoch, and S. R. Seaman
185
1997; Wurm & Vakoch, 1996, 2000; Wurm et al., 2003; see also Cohen & Cohen,
1983; Lorch & Myers, 1990).
The results of this analysis are shown in Table 2. At Step 1 of the analysis the
between-subjects variance was partitioned. In repeated-measures regression analyses
this is done by entering a block of N - 1 (i.e., 64 for the current experiment) dummy
variables that represent the participants. At Step 2 of the analysis the effects of the
control variables were statistically removed. Among other things, this step of the
analysis shows a replication of the concreteness effect reported by Tyler et al. (2000)
for auditorily-presented stimuli.
Steps 3 – 5 of the analysis contained the statistical tests of primary interest. At
Step 3 the main effects of Evaluation, Potency, and Activity were assessed. All were
significant. Higher dimension weights on Evaluation were associated with faster
lexical decision times, as were higher dimension weights on Activity. Higher dimension weights on Potency were associated with slower lexical decision times.
TABLE 2
Summary of hierarchical regression analysis for variables predicting lexical decision time
Variable(s)
df
B (SE B)
β
1. Between Subjects
64
...
...
11.06***
1
1
1
1
1
1
1
– 10.7 (4.44)
– 0.1 (0.01)
– 37.5 (6.23)
17.3 (2.91)
2.4 (19.81)
98.1 (7.48)
– 6.6 (7.68)
– .03
– .08
– .09
.09
.00
.25
– .01
5.83*
28.26***
36.22***
35.11***
< 1.0
172.08***
< 1.0
1
1
1
– 3.2 (1.57)
4.0 (1.71)
– 6.7 (1.44)
– .04
.05
– .08
4.02*
5.36*
21.93***
1
1
1
3.4 (0.98)
– 2.1 (1.02)
3.6 (0.78)
.19
– .14
.36
12.33***
4.08*
21.00***
1
– 3.4 (0.45)
– 1.36
58.19***
F
2. Control Variables
Word frequency
Concreteness
Animate / inanimate
Bigram frequency
item duration
Neighborhood density
Stress
3. Main effects
Evaluation
Potency
Activity
4. Two-Way Interactions
Evaluation × Potency
Evaluation × Activity
Potency × Activity
5. Evaluation × Potency × Activity
*p < .05. ***p < .001.
It should be noted that the main effects of the semantic dimensions are significant over and above the effects of the control variables, including concreteness. Thus,
these effects are not due to the semantic variables being confounded with anything at
Language and Speech
186
Semantic effects in lexical access
Step 2 of the analysis. Similarly, the interactions (discussed next) were evaluated only
after the main effects of the semantic dimensions were controlled for.
Step 4 of the analysis assessed the two-way interactions between Evaluation,
Potency, and Activity ratings. Each of these interactions was significant: Activity ×
Potency, shown in Figure 1; Potency × Evaluation, shown in Figure 2; and Activity
× Evaluation, shown in Figure 3. The figures were constructed by plotting the regression equation. The mean value of each variable in the statistical model was multiplied
by its regression coefficient, except for the two variables being contrasted in the plot.
For these two variables (e.g., for Potency and Activity in Fig. 1), the mean plus one
SD was used for “High,” and the mean minus one SD was used for “Low.” Readers
are reminded that the underlying analyses were continuous; this is merely the most
convenient way to show the nature of the interaction.
Figure 1
The slope of the relationship between lexical decision time and Activity, as a function of
Potency. “Low” indicates a value one SD below the mean and “High” indicates a value
one SD above the mean. The dimensions are continuous; Low and High values were used
only for graphical purposes
The significant interaction indicates that the slope of the relationship between
Activity and RT depends significantly on Potency. Inspection of Figure 1 shows this
dependence. For words rated relatively high on Potency, there is no relationship
between Activity and lexical decision times. For words rated relatively low on Potency,
the Activity effect is strongly facilitative.
Language and Speech
L. H. Wurm, D. A. Vakoch, and S. R. Seaman
187
Figure 2 shows the interaction between Potency and Evaluation. For words rated
relatively low on Evaluation, increasing Potency is associated with slightly faster
lexical decision times. For words rated relatively high on Evaluation, the Potency
effect is stronger and in the opposite direction: increasing Potency is associated with
slower decision times.
Figure 2
The slope of the relationship between lexical decision time and Potency, as a function of
Evaluation. “Low” indicates a value one SD below the mean and “High” indicates a
value one SD above the mean. The dimensions are continuous; Low and High values
were used only for graphical purposes
Figure 3 shows the interaction between Activity and Evaluation. As in Figure 2,
the effect is stronger for words high on Evaluation than for words low on Evaluation
(this time both effects are in the same direction).
Step 5 of the analysis assessed the three-way interaction between Evaluation,
Potency, and Activity, which was also significant. As is clear from Figure 4, the overall
facilitative effect of higher Activity ratings does not hold for all subsets of the stimuli.
Specifically, words with relatively high Potency ratings and relatively low Evaluation
ratings (i.e., words with connotations of both “strong” and “bad”) show a weak effect
in the opposite direction.
Language and Speech
188
Semantic effects in lexical access
Figure 3
The slope of the relationship between lexical decision time and Activity, as a function of
Evaluation. “Low” indicates a value one SD below the mean and “High” indicates a
value one SD above the mean. The dimensions are continuous; Low and High values
were used only for graphical purposes
Figure 4
The slope of the relationship between lexical decision time and Activity, as a function of
Potency and Evaluation. “Low” indicates a value one SD below the mean and “High”
indicates a value one SD above the mean. The dimensions are continuous; Low and High
values were used only for graphical purposes
Language and Speech
L. H. Wurm, D. A. Vakoch, and S. R. Seaman
189
5.3
Discussion
The predicted main effects were all found in this experiment. Beyond the effects of the
control variables, words with ratings that were higher on Evaluation were recognized
faster. So were words with lower ratings on Potency, or higher ratings on Activity.
We believe that these effects demonstrate an influence of meaning on word recognition, specifically a predominating approach response tendency.
In addition, the two-way interactions all have the essential characteristic that was
predicted. The strong facilitative effect of Activity, for example, was most pronounced
for items low on Potency or high on Evaluation. Similarly, the inhibitory main effect
of Potency was found to hold only for words rated high on Evaluation.
The significant three-way interaction shown in Figure 4 was not predicted, but
it too can be explained within the current theoretical framework. In general, increasing
Activity ratings are associated with dramatically faster decision times. This holds for
three of the four lines shown in the figure. The excepted items have the combined
connotations of “strong” (i.e., high Potency) and “bad” (i.e., low Evaluation). In this
specific context, increasing Activity is associated with slower decision times, which
suggests the cost (in processing time) in this context of increased stakes and increased
ambiguity. Wurm and Vakoch (1996), too, found that this particular conjunction of
connotations produced a response pattern quite unlike the other conjunctions, using
the lexical decision task.
6 Experiment 2: Naming
The purpose of Experiment 2 is to provide another opportunity to test the theoretical framework developed above. Every experimental paradigm has been criticized
for one reason or another: Lexical decision has been criticized as a measure of
perceptual processing mainly on the grounds that it is susceptible to postperceptual
contamination and bias. While not perfect, the naming paradigm is believed to be
free of these kinds of concerns, and to be a relatively pure measure of perceptual
processing. For this reason, it is important to replicate the results of Experiment 1 using
the naming paradigm.
On each trial of this experiment, a participant heard a word presented over
headphones and repeated the word into a microphone as quickly as possible.
6.1
Method
Participants. Forty-six undergraduate students from the Wayne State University
psychology subject pool participated. None had participated in the preliminary rating
experiment or Experiment 1. All were native speakers of English who reported normal
hearing. Participants received extra credit in a psychology course for their participation.
Materials. The same stimuli used in Experiment 1 were used, except that the pseudowords were not presented.
Language and Speech
190
Semantic effects in lexical access
Procedure. Participants were tested individually in a sound-attenuating booth. They
listened to the stimuli, played over headphones at a comfortable listening level, and
were instructed to repeat back each word they heard as quickly and accurately as
possible. A microphone was positioned approximately 10 cm in front of each participant. A different random stimulus order was used for each participant.
A practice set of 20 words was used prior to the main experiment to familiarize
participants with the procedure.
6.2
Results
Naming times for trials on which the participant pronounced the wrong word were not
included (2% of the trials). Naming times were also excluded from the analyses if they
were due to a noise other than initiation of the verbal response, such as coughing or
shuffling feet (<1% of the trials). One participant was a clear outlier in both mean naming
time (1179 ms) and error rate (24%). That person’s data were not included in the analysis.
Data for the word ‘canteen’ were also excluded, as in Experiment 1.
TABLE 3
Summary of hierarchical regression analysis for variables predicting naming time
Variable(s)
df
B (SE B)
β
1. Between Subjects
44
…
…
92.05***
2
1
1
1
1
1
1
1
1
1
…
25.2 (17.82)
32.4 (8.05)
– 38.2 (5.66)
– 0.1 (0.02)
– 13.2 (9.85)
14.3 (3.35)
125.1 (33.41)
173.9 (13.32)
– 47.9 (12.27)
…
.02
.05
– .07
– .04
– .02
.05
.06
.19
– .05
7.23***
1.99
16.18***
46.58***
14.48***
1.78
18.19***
14.02***
170.54***
15.27***
1
1
1
1.9 (2.65)
9.8 (2.94)
– 10.7 (2.46)
.01
.05
– .06
< 1.0
11.06**
18.76***
1
1
1
3.3 (1.66)
– 4.1 (1.64)
4.2 (1.32)
.08
– .11
.18
3.87*
6.34*
10.04**
1
– 0.5 (0.74)
– .08
F
2. Control Variables
Place of articulation
Vowel onset
Voiced onset
Word frequency
Concreteness
Animate / inanimate
Bigram frequency
Item duration
Neighborhood density
Stress
3. Main effects
Evaluation
Potency
Activity
4. Two-Way Interactions
Evaluation × Potency
Evaluation × Activity
Potency × Activity
5. Evaluation × Potency × Activity
*p < .05. **p < .01. ***p < .001.
Language and Speech
< 1.0
L. H. Wurm, D. A. Vakoch, and S. R. Seaman
191
A regression analysis exactly analogous to that reported in Experiment 1 was
conducted. The results of this analysis are shown in Table 3, and are extremely similar
to those of the earlier experiment. As in Experiment 1, higher dimension weights on
Activity were associated with faster naming times, and higher dimension weights on
Potency were associated with slower naming times. In this experiment, the main effect
of Evaluation was not significant.
Figure 5
The slope of the relationship between naming time and Activity, as a function of Potency.
“Low” indicates a value one SD below the mean and “High” indicates a value one SD
above the mean. The dimensions are continuous; Low and High values were used only
for graphical purposes
Each of the two-way interactions was significant, as in Experiment 1: Activity
× Potency, shown in Figure 5; Potency × Evaluation, shown in Figure 6; and Activity
× Evaluation, shown in Figure 7. Step 5 of the analysis assessed the three-way interaction between Evaluation, Potency, and Activity, which was not significant.
These three figures are very similar to the corresponding figures from Experiment 1
(i.e., Figures 1 – 3). In Figure 5, for example, one again sees a very strong relationship
between Activity and naming times for words rated low on Potency, and a much
weaker relationship for words rated high on Potency. Figure 6 shows a strong inhibitory
effect of high Potency for words rated high on Evaluation, and a weaker (and this time
inhibitory) effect for words rated low on Evaluation.
Language and Speech
192
Semantic effects in lexical access
Figure 6
The slope of the relationship between naming time and Potency, as a function of Evaluation.
“Low” indicates a value one SD below the mean and “High” indicates a value one SD
above the mean. The dimensions are continuous; Low and High values were used only
for graphical purposes.
Finally, Figure 7 shows that the striking facilitative effect of Activity is much
stronger for words rated high on Evaluation, just as was seen in Experiment 1. The
main difference in these last two figures, compared to Figures 2 and 3, is that there
is no overall advantage for words rated high on Evaluation. Therefore the line for
those items is shifted upward a bit, relative to the corresponding lines in the earlier
figures.
6.3
Discussion
The results of Experiment 2 provide strong confirmation of those reported in
Experiment 1. Nearly all of the same effects were significant, and the interactions
looked extremely similar. In contrast to Experiment 1, though, the current experiment failed to show a main effect for Evaluation or a three-way interaction between
Evaluation, Potency, and Activity. There are four main possible explanations for these
differences.
First, there was less residual variance available for the semantic variables to
explain. Between-subject differences accounted for 48% of the variance in Experiment 2,
compared to 15% in Experiment 1. These numbers are typical for the paradigms:
Wurm and Ross (2001), in a study of the processing of morphologically complex
Language and Speech
L. H. Wurm, D. A. Vakoch, and S. R. Seaman
193
Figure 7
The slope of the relationship between naming time and Activity, as a function of Evaluation.
“Low” indicates a value one SD below the mean and “High” indicates a value one SD
above the mean. The dimensions are continuous; Low and High values were used only
for graphical purposes.
words, found 51% and 22% of the variance explained by subjects in a naming and a
lexical decision experiment, respectively.
Second, the correlation between Evaluation and Potency ratings in the current
study of nouns (r = – .52) was substantially stronger than that observed for a sample
of randomly-selected words (r = – .18; these randomly-selected words were taken from
Vakoch & Wurm, 1997). While this level of intercorrelation is well below the level that
would cause concern about collinearity in the analysis (Cohen & Cohen, 1983), this
could have had the effect of masking any possible main effect of Evaluation. Vakoch
and Wurm (1997) did find this main effect; interestingly, though, if we select only the
concrete nouns from the Vakoch and Wurm (1997) stimulus list and recompute the
correlation, its strength increases to about the same value found in the current study
(r = – .57). It is perhaps the case, then, that a moderate negative correlation between
Evaluation and Potency is a characteristic of nouns in general.
Third, although Evaluation is the most extensively studied of the three dimensions, in the great majority of the studies that show this effect, there is no mention of
the other two dimensions. This raises the possibility that these other dimensions are
uncontrolled for in a large number of studies, and that when they are controlled for,
the Evaluation main effect disappears.
Finally, the naming paradigm is less sensitive in general to the effects of several
Language and Speech
194
Semantic effects in lexical access
variables. That makes it a more conservative choice, and is one of the main reasons
for including it in the current study. Given the differences in the two experimental
paradigms, one can have a great deal of confidence in any results that hold across both
of them. Effects that only appear in lexical decision should perhaps be interpreted with
more caution, and followed up with additional research. However, given the pattern
of the two-way interactions, which was quite consistent across the experiments, it is
clear that all three semantic dimensions are playing a significant role in influencing
word recognition times.
7 General Discussion
This study reveals some of the complexity of lexical representations. It was found
that the recognition time for any given word is dependent, in part, on its semantic characteristics (defined fairly coarsely, a point we will return to below). Results showed
that word recognition times, measured using two different experimental tasks, are
related to the three underlying semantic dimensions identified by Osgood and his
colleagues.
The current study also demonstrates that the semantic effects found in earlier
studies (Vakoch & Wurm, 1997; Wurm & Vakoch, 1996, 2000) are not artifacts of
the lexical decision task. Interestingly, researchers’ condemnation of the lexical decision task has been based in part on Chumbley and Balota’s (1984) argument that a
word’s meaning is available before lexical access and lexical decision (see also Hill &
Kemp-Wheeler, 1989). This has been taken to imply that the lexical decision task is
unsuitable as a pure measure of perception, but this logic requires the assumption that
word recognition precedes all semantic processing. A growing experimental literature suggests that this assumption is incorrect (although it may still be that the lexical
decision task is, for other reasons, less than ideal). That literature includes the studies
by Marslen-Wilson (1987), Moss, McCormick, and Tyler (1997), Strain et al. (1995),
Tyler et al. (2002), and Zwitserlood (1989), all of which suggest that some aspects of
semantic information are available before the unique identification of one word (see
also Forster & Hector, 2002).
The present study confirms and extends the evidence for early semantic access
provided by Strain et al. (1995). As mentioned above, they hypothesized that early
semantic activation takes place for all words, but they were only able to observe this
activation for low-frequency exception words. This was to be expected because that
study used the reading-aloud task in which the orthography-to-phonology mapping
is of crucial importance. The current study used two different experimental paradigms and allowed for a demonstration of early semantic effects that do not depend
on very low word frequency.5 Furthermore, the current study extends Strain et al.’s
findings and the others summarized above, in that effects such as concreteness were
5
Post hoc analyses revealed a significant Potency × frequency interaction (for lexical
decision: B = – 19.9 (SE B = 3.63), β = – .39, F = 30.05, p < .001; for naming: B = – 12.8
(SE B = 5.36), β = – .13, F = 5.66, p < .05). The Potency main effect became weaker with
increasing word frequency, but was significant even for the highest frequencies.
Language and Speech
L. H. Wurm, D. A. Vakoch, and S. R. Seaman
195
controlled for prior to the assessment of the semantic variables being examined (variables that can be used in a coherent explanatory framework that has tremendous
behavioral significance).
One possible framework for interpreting the observed semantic effects is a twopart semantic analysis. Both parts of this analysis proceed rapidly and automatically,
but one uses Evaluation, Potency, and Activity or dimensions that code similarly
valuable information. Dimensional values from this part of the analysis would be
integrated with the results of the other, more detailed part of the analysis, because in
comprehension it is necessary to know more detailed information about the object in
question. A rabbit, for example, is not defined simply by the conjunction of “fast” and
“weak,” but by these characteristics in combination with many others. However, under
some circumstances a response can be initiated on the basis of a small number of
these many characteristics, before the more detailed semantic analysis is completed.
Work from several areas, some of which are only loosely connected, is converging
on the idea that a very fast initial semantic analysis takes place. For example, recent
research in the area of morphological processing shows that the way listeners and
readers process prefixed, suffixed, and compound words is affected by the transparency of the semantic relationship between the word constituents (e.g., Libben,
1998; Libben, Derwing, & de Almeida, 1999; Marslen-Wilson, Tyler, Waksler, &
Older, 1994; Schreuder & Baayen, 1995; Wurm, 1997, 2000; Wurm & Ross, 2001). It
is important to note that this is true even though the semantic transparency of the
relationship between constituents cannot be known until well into the recognition
process. This is suggestive of fast retrieval of some information stored with lexical representations that can affect word processing.
In another area of research, van Petten, Coulson, Rubin, Plante, and Parks
(1999) examined the possible relationship between word meanings and event-related
potentials (ERPs) measured from the surface of the scalp. They found that stimulus
words appropriate to and inappropriate to a context manipulation produced N400 ERPs
that were different. For the present discussion, the crucial point of van Petten et al.’s
(1999) study is that these differences were found 200 ms before the UP of the word.
These authors concluded that “ … semantic integration can begin to operate with
only partial, incomplete information about word identity” (p. 394).
Recent psychophysiological research suggests that there is an interaction between
affect, object recognition, and motor responses. Schupp, Junghöfer, Weike, and Hamm
(2003) examined ERPs while participants looked at positive, neutral, and negative
images. They concluded that “ … the affect system not only modulates motor output
(i.e., favoring approach or avoidance dispositions), but already operates at an early
level of sensory encoding” (p. 7). Other studies have led to similar conclusions. For
example, Bargh, Chaiken, Raymond, and Hymes (1996) found priming for words
with similar valences but not for words with dissimilar valences (e.g., the word flowers
primed the word knowledge, because both are judged to be positive — see also Bargh,
1990; Chartrand & Bargh, 1996; Fazio, 1990; Fazio, Sanbonmatsu, Powell, & Kardes,
1986). Results such as these have been interpreted within Murphy and Zajonc’s (1993)
affect primacy hypothesis, which states that a nonconscious mental system processes
affective information (e.g., whether something is seen as positive or negative). This
Language and Speech
196
Semantic effects in lexical access
early evaluative process is believed to be general and widespread, rather than being
tied to narrow or specifically emotional contexts. Furthermore, the words being studied
do not need to refer to objects that elicit a strong evaluative attitude (Bargh, Chaiken,
Govender, & Pratto, 1992; Bargh et al., 1996). Chen and Bargh’s (1999) experiment
demonstrating facilitated responding when lever direction was congruent with an
“approach” or “avoid” word (summarized in the Introduction) links the affect primacy
hypothesis to behavior.
In a model of appraisal proposed by Robinson (1998; see also Scherer, 1984;
Smith & Lazarus, 1990), the valence and urgency of a stimulus or situation are judged
extremely rapidly and preattentively. The preattentive “urgency detection” module
in Robinson’s model determines whether a stimulus or event is personally relevant,
whether it is consistent with or counter to a person’s goals, and whether it can be
dealt with effectively. These aspects of stimuli and events bear similarity to the dimensions of semantic information that may be stored with lexical entries.
As Tyler et al. (2002) noted, current models cannot easily explain early semantic
effects, whether they be effects of concreteness or those demonstrated in the current
study. A wide range of models could in principle accommodate the current findings,
though, with the specifics of implementation depending on the model architecture.
For example, weights on semantic dimensions could be incorporated to alter the
resting activation levels of words or to modify the threshold values needed for word
recognition. It would also be possible to order words according to their dimension
weights, in the same way that some models have bins of words ordered by their frequencies of occurrence. Semantic information could also be incorporated as a higher-level
information source that feeds down to earlier processing levels in a network.
Implementing any of these alternatives would be fairly straightforward.
It is theoretically possible to observe early semantic effects without this semantic
information necessarily being stored with the word’s lexical entry. One interesting
possibility is suggested by a model of semantic categorization proposed by Forster
and Hector (2002), who asked participants to quickly decide whether something was
an animal or not. They found that performance on pseudowords (e.g., turple) was
poorer when it had a neighbor that was an animal name. Their conclusion was that
neighbors had an effect only when they had the proper semantic characteristics, and
that a cascaded system that continuously monitors activation in semantic features
(e.g., “animalness,” or Evaluation) was the best explanation for the results.
This is an attractive possibility in that semantic variables can clearly have effects
prior to the unique identification of a word. However, applied to the current results,
the account seems lacking. This account predicts that the Evaluation effect should be
modeled not as a scale from 1 (‘bad’) to 8 (‘good’), but as the absolute value of the
difference from a true zero point (i.e., 4.5 on this scale). A post hoc analysis of the
data from Experiment 1 showed that the significant main effect of Evaluation disappears if the dimension is rescaled in this way (p = .502). The Forster and Hector
(2002) account may apply in a more straightforward way to Activity (and Potency),
the end points of which can be interpreted as “not active” (‘not potent’) and “active”
(‘potent’). What the account cannot predict is why we should see opposite signs for
the coefficients of these two variables.
Language and Speech
L. H. Wurm, D. A. Vakoch, and S. R. Seaman
197
It is also far from clear how this account would predict the same effects in the
naming paradigm. The crucial question would seem to be why the system is monitoring
activation in semantic features, and the answer appears to be that it is required for the
task Forster and Hector (2002) used, semantic categorization. It is therefore not clear
that this account bears directly on the data at hand, although the cascaded architecture is an interesting possibility for further consideration.
The Gaskell and Marslen-Wilson (1997) model discussed above is a more attractive alternative. We saw in the Introduction that semantics is a part of the single,
distributed representation of a given word, which results in semantics being an integral part of the recognition process from the beginning. The only major modification
that would be required to accommodate the current results is that the model would need
to be specified in more detail, to include specifically Evaluation, Potency, and Activity
(or something like them) as part of the representation associated with each word.
This model appears to make an additional interesting and testable prediction.
As we noted, prior to the UP, the activated semantics in the system is a blend of the
semantics of all words consistent with the input. Within the framework of Evaluation,
Potency, and Activity, this blend will normally be an uninformative mixture with an
“average” overall value on the dimensions. However, it might be possible to identify
a particular cohort of words that are predominantly high on Potency (for example),
and contrast that with other cohorts that have different characteristics. We would
expect to observe a Potency effect that is particularly strong, and evident particularly
early, for words from such a homogeneous cohort.
The methodology used in the current study is relatively simple to implement, insofar
as all words can be located within this connotative meaning space (e.g., Osgood et
al., 1975) or any other (see Wurm & Vakoch, 2000; Wurm et al., 2003). Dimension
weights can be estimated easily through participant judgments. Although other
meaning spaces can be imagined, we believe that the semantic dimensions used in the
current study are good candidates for inclusion in models of word recognition. These
particular dimensions are dominant in several fields of psychology because of their
power to explain variance in a wide range of tasks, including those assumed to tap
into on-line perceptual processing. The significance of the information coded by these
dimensions is important enough that the effects can be seen in word recognition.
Received: April 21, 2003; revision received: September 02, 2003;
accepted: October 15, 2003
References
BAAYEN, R. H., PIEPENBROCK, R., & van RIJN, H. (1993). The CELEX lexical database
[ CD-ROM ].
BALOTA, D. A. (1990). The role of meaning in word recognition. In D. A. Balota, G. B. Flores
d’Arcais, & Rayner, K. (Eds.), Comprehension processes in reading (pp. 9 – 32). Hillsdale,
NJ: Lawrence Erlbaum Associates.
BALOTA, D. A., & CHUMBLEY, J. I. (1984). Are lexical decisions a good measure of lexical
access? The role of word frequency in the neglected decision stage. Journal of Experimental
Psychology: Human Perception and Performance, 10, 340 – 357.
Language and Speech
198
Semantic effects in lexical access
BARGH, J. A. (1990). Auto-motives: Preconscious determinants of social interaction. In E. T.
Higgins & R. M. Sorrentino (Eds.), Handbook of motivation and cognition (Vol. 2,
pp. 93 – 130). New York: Guilford Press.
BARGH, J. A., CHAIKEN, S., GOVENDER, R., & PRATTO, F. (1992). The generality of
the automatic attitude activation effect. Journal of Personality and Social Psychology,
62, 893 – 912.
BARGH, J. A., CHAIKEN, S., RAYMOND, P., & HYMES, C. (1996). The automatic evaluation effect: Unconditional automatic attitude activation with a pronunciation task.
Journal of Experimental Social Psychology, 32, 104 – 128.
BATES, E., DEVESCOVI, A., PIZZAMIGLIO, L., & D’AMICO, S. (1995). Gender and
lexical access in Italian. Perception and Psychophysics, 57, 847 – 862.
BRADLEY, M. M., CODISPOTI, M., CUTHBERT, B. N., & LANG, P. J. (2001). Emotion
and motivation I: Defensive and appetitive reactions in picture processing. Emotion,
1, 276 – 298.
BROWN, G. D., & WATSON, F. L. (1987). First in, first out: Word learning age and spoken
word frequency as predictors of word familiarity and word naming latency. Memory and
Cognition, 15, 208 – 216.
BURNAGE, G. (1990). CELEX: A guide for users. Nijmegen, Netherlands: CELEX.
CARNEGIE MELLON PRONOUNCING DICTIONARY (version 0.4) (1995). Pittsburgh,
PA: Carnegie Mellon University [ computer software ].
CHARTRAND, T. L., & BARGH, J. A. (1996). Automatic activation of impression formation
and memorization goals: Nonconscious goal priming reproduces effects of explicit task
instructions. Journal of Personality and Social Psychology, 71, 464 – 478.
CHEN, M., & BARGH, J. A. (1999). Consequences of automatic evaluation: Immediate behavioral predispositions to approach or avoid the stimulus. Personality and Social Psychology
Bulletin, 25, 215 – 224.
CHUMBLEY, J. I., & BALOTA, D. A. (1984). A word’s meaning affects the decision in lexical
decision. Memory and Cognition, 12, 590 – 606.
COHEN, J., & COHEN, P. (1983). Applied multiple regression / correlation analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum Associates.
COLTHEART, M., DAVELAAR, E., JONASSON, J. T., & BESNER, D. (1977). Access to the
internal lexicon. In S. Dornic (Ed.), Attention and performance (Vol. VI, pp. 535 – 555).
Hillsdale, NJ: Lawrence Erlbaum Associates.
CUTLER, A., & NORRIS, D. (1988). The role of strong syllables in segmentation for lexical
access. Journal of Experimental Psychology: Human Perception and Performance,
14, 113 – 121.
DARWIN, C. (1968). The origin of species (J. W. Burrow, Ed.). New York: Penguin. (Original
work published 1859.)
DAVIDSON, R. J. (1992). Prolegomenon to the structure of emotion: Gleanings from neuropsychology. Cognition and Emotion, 6, 245 – 268.
De GROOT, A. M. B. (1989). Representational aspects of word imageability and word frequency
as assessed through word association. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 15, 824 – 845.
FAZIO, R. H. (1990). Multiple processes by which attitudes guide behavior: The MODE model
as an integrative framework. In M. P. Zanna (Ed.), Advances in experimental psychology
(Vol. 23, pp. 75 – 109). San Diego, CA: Academic Press.
FAZIO, R. H., SANBONMATSU, D. M., POWELL, M. C., & KARDES, F. R. (1986). On
the automatic activation of attitudes. Journal of Personality and Social Psychology,
50, 229 – 238.
FORSTER, K. I., & HECTOR, J. (2002). Cascaded versus noncascaded models of lexical and
semantic processing: The turple effect. Memory & Cognition, 30, 1106 – 1117.
Language and Speech
L. H. Wurm, D. A. Vakoch, and S. R. Seaman
199
GASKELL, M. G., & MARSLEN-WILSON, W. D. (1997). Integrating form and meaning: A
distributed model of speech perception. Language and Cognitive Processes, 12, 613 – 656.
GOLDINGER, S. D., LUCE, P. A., & PISONI, D. B. (1989). Priming lexical neighbors of
spoken words: Effects of competition and inhibition. Journal of Memory and Language,
28, 501 – 518.
GROSJEAN, F., & GEE, J. (1987). Prosodic structure and spoken word recognition. Cognition,
25, 135 – 155.
HILL, A. B., & KEMP-WHEELER, S. M. (1989). The influence of anxiety on lexical and
affective decision time for emotional words. Personality and Individual Differences,
10, 1143 – 1149.
JAMES, C. T. (1975). The role of semantic information in lexical decisions. Journal of Experimental
Psychology: Human Perception and Performance, 1, 130 – 136.
KESSLER, B., TREIMAN, R., & MULLENNIX, J. (2002). Phonetic biases in voice key
response time measurements. Journal of Memory and Language, 47, 145 – 171.
LANG, P. J., BRADLEY, M. M., & CUTHBERT, B. N. (1998). Emotion and attention: Stop,
look, and listen. Current Psychology of Cognition, 17, 997 – 1020.
LIBBEN, G. (1998). Semantic transparency in the processing of compounds: Consequence for
representation, processing, and impairment. Brain and Language, 61, 30 – 44.
LIBBEN, G., DERWING, B. L., & de ALMEIDA, R. G. (1999). Ambiguous novel compounds
and models of morphological parsing. Brain and Language, 68, 378 – 386.
LORCH, R. F., & MYERS, J. L. (1990). Regression analyses of repeated measures data in
cognitive research. Journal of Experimental Psychology: Learning, Memory, and Cognition,
16, 149 – 157.
LUCE, P. A., PISONI, D. B., & GOLDINGER, S. D. (1990). Similarity neighborhoods of
spoken words. In G. T. M. Altmann (Ed.), Cognitive models of speech processing:
Psycholinguistic and computational perspectives. Cambridge, MA: M.I.T. Press.
MARSLEN-WILSON, W. D. (1984). Function and process in spoken word recognition. In H.
Bouma & D. G. Bouwhuis (Eds.), Attention and performance: Control of language processes
(Vol. X, pp. 125 – 150). Hillsdale, NJ: Lawrence Erlbaum Associates.
MARSLEN-WILSON, W. D. (1987). Functional parallelism in spoken word recognition.
Cognition, 25, 71 – 102.
MARSLEN-WILSON, W. D., TYLER, L. H., WAKSLER, R., & OLDER, L. (1994).
Morphology and meaning in the English mental lexicon. Psychological Review, 101, 3 – 33.
MARSLEN-WILSON, W., & WELSH, A. (1978). Processing interactions and lexical access
during word recognition in continuous speech. Cognitive Psychology, 10, 29 – 63.
MAYZNER, M. S., & TRESSELT, M. E. (1965). Tables of single-letter and digram frequency
counts for various word-length and letter-position combinations. Psychonomic Monograph
Supplements, 1, 13 – 32.
McRAE, K., de SA, V. R., & SEIDENBERG, M. S. (1997). On the nature and scope of featural representations of word meaning. Journal of Experimental Psychology: General,
126, 99 – 130.
MOSS, H. E., McCORMICK, S. F., & TYLER, L. K. (1997). The time course of activation
of semantic information during spoken word recognition. Language and Cognitive
Processes, 12, 695 – 731.
MURPHY, S. T., & ZAJONC, R. B. (1993). Affect, cognition, and awareness: Affective priming
with optimal and suboptimal stimulus exposures. Journal of Personality and Social
Psychology, 64, 723 – 739.
NEELY, J. H. (1991). Semantic priming effects in visual word recognition: A selective review
of current findings and theories. In D. Besner & G. Humphreys (Eds.), Basic processes
in reading: Visual word recognition (pp. 264 – 336). Hillsdale, NJ: Lawrence Erlbaum.
Language and Speech
200
Semantic effects in lexical access
NEELY, J. H., & KEEFE, D. E. (1989). Semantic context effects on visual word processing: A
hybrid prospective / retrospective processing theory. In G. H. Bower (Ed.), The psychology
of learning and motivation: Advances in research and theory (Vol. 24, pp. 207 – 248). San
Diego, CA: Academic Press.
OSGOOD, C. E. (1969). On the whys and wherefores of E, P, and A. Journal of Personality and
Social Psychology, 12, 194 – 199.
OSGOOD, C. E., MAY, W. H., & MIRON, M. S. (1975). Cross-cultural universals of affective
meaning. Urbana, IL: University of Illinois Press.
OSGOOD, C. E., SUCI, G. J., & TANNENBAUM, P. H. (1957). The measurement of meaning.
Urbana, IL: University of Illinois Press.
QUINLAN, P. T. (1987). Oxford psycholinguistic database [ Computer Software ]. Oxford,
England: Oxford University Press.
RADEAU, M., MOUSTY, P., & BERTELSON, P. (1989). The effect of the uniqueness point
in spoken-word recognition. Psychological Research, 51, 123 – 128.
ROBINSON, M. D. (1998). Running from William James’ bear: A review of preattentive mechanisms and their contributions to emotional experience. Cognition and Emotion, 12, 667–696.
RODD, J., GASKELL, G., & MARSLEN-WILSON, W. (2002). Making sense of semantic ambiguity: Semantic competition in lexical access. Journal of Memory and Language, 46, 245–266.
SCHERER, K. R. (1984). Emotion as a multicomponent process: A model and some crosscultural data. In P. Shaver (Ed.), Review of personality and social psychology (Vol. 5,
pp. 37 – 63). Beverly Hills, CA: Sage.
SCHNEIRLA, T. C. (1965). Aspects of stimulation and organization in approach / withdrawal
processes underlying vertebrate behavioral development. Advances in the study of Behavior,
1, 1 – 74.
SCHREUDER, R., & BAAYEN, R. H. (1995). Modeling morphological processing. In L. B.
Feldman (Ed.), Morphological aspects of language processing (pp. 131 – 154). Hillsdale, NJ:
Lawrence Erlbaum Associates.
SCHUPP, H. T., JUNGHÖFER, M., WEIKE, A. I., & HAMM, A. O. (2003). Emotional
facilitation of sensory processing in the visual cortex. Psychological Science, 14, 7 – 13.
SMITH, C. A., & LAZARUS, R. S. (1990). Emotion and adaptation. In L. A. Pervin (Ed.),
Handbook of personality: Theory and research (pp. 609 – 637). New York: Guilford Press.
STRAIN, E., & HERDMAN, C. M. (1999). Imageability effects in word naming: An individual differences analysis. Canadian Journal of Experimental Psychology, 53, 347 – 359.
STRAIN, E., PATTERSON, K., & SEIDENBERG, M. S. (1995). Semantic effects in singleword naming. Journal of Experimental Psychology: Learning, Memory, and Cognition,
21, 1140 – 1154.
TABACHNICK, B. G., & FIDELL, L. S. (2001). Using multivariate statistics (4th ed.). Boston:
Allyn and Bacon.
TYLER, L. K., MOSS, H. E., GALPIN, A., & VOICE, J. K. (2002). Activating meaning in
time: The role of imageability and form-class. Language and Cognitive Processes,
17, 471 – 502.
TYLER, L. K., VOICE, J. K., & MOSS, H. E. (2000). The interaction of meaning and sound
in spoken word recognition. Psychonomic Bulletin & Review, 7, 320 – 326.
VAKOCH, D. A., & WURM, L. H. (1997). Emotional connotation in speech perception:
Semantic associations in the general lexicon. Cognition and Emotion, 11, 337 – 349.
Van PETTEN, C., COULSON, S., RUBIN, S., PLANTE, E., & PARKS, M. (1999). Time
course of word identification and semantic integration in spoken language. Journal of
Experimental Psychology: Human Perception and Performance, 25, 394 – 417.
WHALEY, C. P. (1978). Word-nonword classification time. Journal of Verbal Learning and
Verbal Behavior, 17, 143 – 154.
Language and Speech
L. H. Wurm, D. A. Vakoch, and S. R. Seaman
201
WURM, L. H. (1997). Auditory processing of prefixed English words is both continuous and
decompositional. Journal of Memory and Language, 37, 438 – 461.
WURM, L. H. (2000). Auditory processing of polymorphemic pseudowords. Journal of Memory
and Language, 42, 255 – 271.
WURM, L. H., & AYCOCK, J. (2003a). Recognition of low-frequency prefixed words.
Manuscript submitted for publication.
WURM, L. H., & AYCOCK, J. (2003b). Recognition of spoken prefixed words: The role of
early Conditional Root Uniqueness Points. In R. H. Baayen & R. Schreuder (Eds.),
Morphological structure in language processing (pp. 259 – 286). Berlin: Mouton de Gruyter.
WURM, L. H., & ROSS, S. E. (2001). Conditional root uniqueness points: Psychological
validity and perceptual consequences. Journal of Memory and Language, 45, 39 – 57.
WURM, L. H., & VAKOCH, D. A. (1996). Dimensions of speech perception: Semantic associations in the affective lexicon. Cognition and Emotion, 10, 409 – 423.
WURM, L. H., & VAKOCH, D. A. (2000). The adaptive value of lexical connotation in speech
perception. Cognition and Emotion, 14, 177 – 191.
WURM, L. H., VAKOCH, D. A., AYCOCK, J., & CHILDERS, R. R. (2003). Semantic effects
in lexical access: Evidence from single-word naming. Cognition and Emotion, 17, 547 – 565.
ZWITSERLOOD, P. (1989). The locus of the effects of sentential-semantic context in spoken-word processing. Cognition, 32, 25 – 64.
Language and Speech
202
Semantic effects in lexical access
Appendix
Stimuli and mean ratings
Word
Evaluation
Potency
Activity
ant
2.97
2.72
5.47
appendix
3.68
3.25
2.59
apple
6.41
3.13
1.88
arrow
3.56
5.50
4.19
badger
4.50
4.23
4.93
balloon
6.00
2.19
3.06
banana
5.78
2.19
1.44
basket
5.38
2.78
1.41
blanket
6.94
2.63
1.81
bottle
4.41
4.22
1.63
buffalo
5.03
6.63
5.38
burglar
1.19
6.47
6.63
butterfly
7.06
1.66
5.41
cancer
1.31
7.22
6.34
cannon
3.09
6.97
4.81
canteen
5.25
4.23
2.28
canyon
5.84
5.50
1.66
card
5.84
2.19
1.63
carrot
5.16
3.00
1.59
chicken
5.56
3.34
5.69
clothing
6.84
3.50
2.31
club
3.56
5.65
3.72
collision
1.72
6.59
6.66
constellation
6.63
4.31
2.09
cork
4.38
3.09
1.56
corner
3.97
4.81
1.50
crime
1.41
6.56
6.72
crossbow
3.19
6.10
4.03
crow
3.66
4.16
6.06
desert
3.72
5.09
2.59
desk
4.44
5.00
1.34
dust
2.25
1.81
1.81
dynamite
2.78
7.56
6.78
Language and Speech
L. H. Wurm, D. A. Vakoch, and S. R. Seaman
203
Word
Evaluation
Potency
Activity
eagle
6.44
5.75
6.09
echo
5.16
3.09
4.16
egg
5.31
2.66
2.38
electricity
5.69
6.66
6.84
elephant
5.84
6.50
5.63
fire
4.56
7.16
7.03
fish
5.88
3.59
5.91
flag
6.03
3.44
3.69
food
7.00
4.13
2.63
fork
4.75
4.44
2.22
hammer
4.19
6.53
3.72
hook
3.22
5.38
2.13
hurricane
1.91
7.84
7.72
joke
7.25
3.56
5.09
kerosene
3.28
5.56
4.28
knife
3.50
6.78
3.69
ladder
4.50
4.31
1.88
lamp
5.63
3.52
1.94
lava
2.88
6.50
5.63
leaf
5.66
2.09
2.91
lightning
4.03
7.22
7.41
lint
2.41
1.38
1.38
lion
4.34
7.25
6.91
machete
2.50
7.19
4.22
moose
5.38
6.25
5.38
mugger
1.13
6.56
6.59
nail
3.50
5.88
1.97
needle
2.84
4.78
2.81
ointment
4.75
2.88
2.63
opera
5.91
3.13
4.78
parsley
4.31
1.97
1.50
pesticide
1.88
6.16
3.56
philosophy
4.88
3.38
3.13
pin
3.53
4.13
1.84
plague
1.19
7.56
5.25
poetry
5.94
2.69
3.63
Language and Speech
204
Semantic effects in lexical access
Evaluation
Potency
Activity
poison
1.31
7.09
4.16
pollution
1.19
5.94
4.41
potato
5.72
3.34
1.66
quicksand
1.69
5.44
4.94
rabbit
6.13
2.91
6.06
razor
3.19
5.63
3.25
scorpion
1.56
5.44
5.38
shoe
5.63
4.19
3.16
skunk
2.16
3.84
4.84
snake
2.03
6.06
5.72
spear
2.69
6.03
4.47
spoon
4.97
3.50
2.22
stove
5.00
5.13
3.19
strawberry
6.94
1.84
1.69
sunset
7.69
3.22
3.84
syringe
2.22
5.69
3.16
tarantula
1.78
5.00
5.31
telescope
5.19
4.41
1.97
thorn
1.84
5.06
2.06
tiger
3.75
7.31
6.81
toad
3.59
2.69
4.31
Word
tobacco
2.06
5.00
2.53
tonsil
3.91
2.72
2.63
tornado
2.25
7.59
7.59
toxin
1.44
7.00
3.91
tree
6.66
6.56
3.53
twig
4.44
2.53
2.06
waltz
6.38
2.66
5.06
water
7.25
5.03
5.53
wood
5.69
5.63
2.16
wool
5.63
3.19
1.56
Note: All ratings were made on a 1 to 8 scale.
Language and Speech