Professional Documents
Culture Documents
DOI 10.1007/s10936-012-9206-6
Abstract Extensive evidence has shown that presentation of a word (target) following a
related word (prime) results in faster reaction times compared to unrelated words. Two primes
preceding a target have been used to examine the effects of multiple influences on a target.
Several studies have observed greater, or additive, priming effects of multiple related primes
compared to single related primes. The present study aims to eliminate attentional factors that
may have contributed to findings in previous studies that used explicitly presented primes
and targets. Thus, a continuous priming paradigm where targets are unknown to participants
is used with noun-noun-verb triads filling agent, patient, and action roles in situation schemas
(tourist, car, rent). Results replicate priming of single nouns preceding related verbs but do
not suggest an additive effect for two nouns versus one. The absence of additive priming
suggests that attentional processes may have been a factor in previous research.
Keywords
Introduction
In an extensive body of research using a single-word semantic priming paradigm, presentation
of a word following a semantically related word results in faster reaction times (RTs) compared to unrelated or less related primes (Ferretti et al. 2001; Hare et al. 2009; McRae et al.
2005; Neely 1991; Perea and Rosa 2002). Designs of semantic priming tasks are manipulated
to investigate factors that influence processing of target words or language structures, by contributing to either faster or slower reactions. This line of research uses behavioral responses to
controlled language tasks to illuminate aspects of semantic and syntactic processing that cannot ordinarily be separated from the complex interactions of meaning, concepts and structure
in natural language production.
123
Semantic priming studies have investigated various hypotheses about the neural
organization and processing of language, including types of semantic relatedness, factors
in language processing, and the time course of activation and decay of various types of
information that may be made available when lexical entries are processed (Balota et al.
1992; McRae and Boisvert 1998; Moss et al. 1995; Neely 1991; Perea and Rosa 2002).
This line of research has offered evidentiary support for many types of semantic relatedness
including synonyms and antonyms; category names and exemplars (i.e. bird, robin); prototypical and atypical category exemplars (i.e. robin, penguin); closely associated words that
are either semantically related (i.e. doctor, nurse) or unrelated (i.e. bed, pan); verbs sharing
qualities related to motor planning and execution (i.e. drop, toss); as well as mediated associates that are related through a third concept (i.e. lion-(tiger)-stripes) (Jones 2010; Moss
et al. 1995; Neely 1991; Perea and Rosa 2002; Raczaszek-Leonardi et al. 2008; Setola and
Reilly 2005). It has also been hypothesized that the degree of relatedness between words
has an influence on word processing. Vigliocco et al. (2004) compared concrete noun-noun
related pairs across very close, close, medium and far (e.g. dagger-sword; dagger-hammer)
semantic distances (as defined by the number of shared features) to test a hypothesis that
conceptual feature representations are part of the stored meaning of words, and found that
reaction times increase as semantic distance increases.
Much of the early research on semantic priming of words focused on the relationships
between concrete animate or inanimate nouns, which can be easily and intuitively grouped
into superordinate categories (e.g., robin and bluejay share the superordinate category birds)
(Neely 1991). Expanding from that basis, similar semantic priming tasks have been used to
extend our understanding of relatedness. For example, Moss et al. (1995, experiment 1)
included prime-target pairs with different relation types (category coordinates, i.e. pig-horse,
and functionally related pairs, i.e. restaurant-wine) in a single experimental task and found
priming with no significant effect of relation type, providing evidence that the contextually
related pairs were related similar to the previously established category relationships.
Thematic roles are components of lexical entries that unite aspects of semantic and syntactic information. Both semantic priming and sentence processing studies have provided
support for an account of language processing in which the stored meaning of verbs contains
information about the entities that are central to the representation of action concepts and
that interact with their stored syntactic argument structures (Ferretti et al. 2001; Shapiro et al.
1991). Thus, conceptual features of words interact with each other in a structured way, with
the verb serving as the linking component of the relationship. Verbs can be described in terms
of their required or permissible thematic structure in addition to information such as syntactic
category and argument structure (Shapiro et al. 1991). Thematic roles include the initiator of
action (agent), recipient of action (patient), location, instrument, and event. For example, the
meaning of the verb fix cannot be represented without reference to a person who fixes (agent),
and a thing that is fixed (patient). The basic syntactic argument for fix requires subject and
object slots that are filled by thematic role nouns to construct a meaningful message.
Extending the theory of semantic relatedness to include thematic roles necessitates the
inclusion of an event memory account, in which real-world knowledge and experience about
common situations and events form part of the stored meaning of words (Ferretti et al. 2001).
While a thematic role can be imagined as an empty slot that can be filled by any number
of words to convey a unique message, specific filler words may be prototypical or atypical
features of an action concept. An event memory account of the representation of the verb
fix, for example, would include experiential knowledge about the types or features of people
who fix or things that are fixed in the stored concept of the verb. Both electrician and
plumber are common agents of the verb fix, yet electrician commonly occurs with the
123
patient lights, while the combination of the thematic role fillers plumber-fix-lights forms a
meaningful but atypical message. Thematic role fillers that appear together in a common or
prototypical group have been referred to as a situation schema (McRae et al. 2005). Research
investigating ongoing sentence processing supports the theory that thematic information is
part of the lexical entry for verbs (Ahrens and Swinney 1995; Shapiro et al. 1991).
Thematic role investigations offer insights into the interactions between the semantic and
syntactic information that is made available during activation of words. Priming research
using both word and sentence contexts suggest that the thematic role information contributing to language processing is not limited to thematic role structure, but includes conceptual information about prototypical or common fillers for those roles. Ferretti et al. (2001,
experiments 1 and 4) used a sentence fragment condition, with a cross-modal naming task
designed to cue specific thematic roles, to compare agents and patients that had previously
shown priming effects on related verbs in a single word semantic decision task. Presenting
target agents and patients in their congruent and incongruent roles (i.e. He was arrested by
the [cop], [crook]), Ferretti et al. (2001) observed a significant priming effect for both agents
and patients in their congruent but not in their incongruent thematic role assignments.
Single word semantic priming paradigms stripped of syntactic context that could establish expectations have also been used to explore relationships between thematic role nouns
and related verbs. Ferretti et al. (2001) observed priming effects of verbs on related agents,
patients, and instruments, (but not locations) on an animacy decision task. Using a pronunciation task, McRae et al. (2005) observed priming for the reverse presentation of stimuli where
the thematic roles of agent, patient, instrument, and location primed related verb targets.
Edmonds and Mizrahi (2011) replicated the findings of bidirectional priming with agents
and patients in younger adults using a lexical decision task (LDT), a behavioral task in which
participants are asked to read letter strings and make rapid yes or no decisions about
whether or not they are real English words. Manipulations of task complexity can be used
to differentially promote attentional or automatic processing by affecting cognitive demands
and the amount of semantic information that must be accessed to complete a task (Balota
and Paul 1996; Ferretti et al. 2001; Grondin et al. 2009).
Effects of multiple primes on targets have been explored to evaluate the possibility of additive priming for two related noun primes as compared to single related primes. In multiple
priming tasks, targets follow exposure to two primes that are either both related, both unrelated, or mixed with one related prime in either the first or second position, resulting in four
experimental conditions compared to two, related and unrelated, that are typical of single
prime designs (Balota and Paul 1996; Khader et al. 2003). In this paper, the term additive
priming is used to refer to priming effects of multiple related primes that are greater than the
effects of single related primes combined with unrelated primes.
Balota and Paul (1996) investigated effects of two noun primes on noun targets with
several manipulations of relatedness type and task complexity. Taking advantage of the fact
that English allows lexical ambiguities in which entries can have multiple conceptual representations (i.e. organ =< type of musical instrument> or <type of body part>), Balota
and Paul (1996) manipulated the relationship between multiple primes and targets by using
related noun primes that were not directly related to each other but converged on the conceptual features of unambiguous noun targets (i.e. lion-stripes-tiger) or ambiguous targets (i.e.
kidney-piano-organ), or that converged on a category label (i.e. copper-bronze-metal). Using
both standard LDT and a relatedness judgment task, additive priming was seen for targets
following two related primes compared to a single related prime in either the first or the
second prime position, regardless of the type of relatedness. However, the task did affect the
magnitude of the priming effect. Mean reaction time differences for two versus one related
123
prime ranged from 10 to 33 ms for all three relatedness types in the LDT condition. In the
relatedness judgment task condition, Balota and Paul (1996, experiment 6) observed similar
additive priming for ambiguous targets, while for unambiguous targets additive priming was
far greater, with mean reaction times more than 120 ms faster for two converging primes compared to single related primes. In the LDT, the relatedness manipulation did not affect the size
(in milliseconds) of the additive priming effect, whereas the relatedness judgment task with
unambiguous targets received a much larger boost in additive priming from a second related
prime, perhaps because judgment tasks demand more explicit semantic processing than LDT.
Balota and Paul (1996) also used different stimulus onset asynchrony (SOA) times (299 ms
total for both primes and intervals in the LDT task and 533 ms in the relatedness judgment
task), giving participants more time for conscious processing of primes in the relatedness
task.
Khader et al. (2003) also used a relatedness judgment task to investigate priming effects
of multiple primes, using noun and verb stimuli. The overall SOA for each trial, from presentation of the first prime to onset of the target, was 1,300 ms. Prime-target relationships were
constructed using a pilot questionnaire in which participants filled noun or verb gaps in triads
with subject-object-verb relationships, where two related words were provided. The stimuli
used in Khader et al. (2003) had more explicit relationships between primes and targets than
the stimuli in Balota and Paul (1996), and they were also verified by native speaker responses
in a pilot questionnaire, so that the participants in the priming task were more likely to recognize relationships within trials. Participants also had a greater potential of recognizing
patterns of relationship across trials. Using these noun-noun-verb stimuli, a long SOA, and a
task that demanded conscious semantic processing and explicit attention to the prime-target
relationships, Khader et al. (2003) reported an additive priming effect that reduced reaction
times significantly. A mean reaction time difference of 145 ms was observed for two nouns
on a related verb target (e.g. cheese-mouse-eat), with a priming effect of 154 ms, compared
to a single related noun in the second position (e.g. wood-mouse-eats) or in the first position
(e.g. cheese-carpenter-eats), which had priming effects of 9 and 6 ms, respectively. Khader
et al. (2003) observed a similar pattern of priming for verb and noun primes on related noun
targets (verb-noun-noun presentation, e.g. cuts-carpenter-wood). The much larger additive
priming seen in relatedness judgment tasks compared to the LDT experiments in Balota
and Paul (1996), suggests attentional processing benefits rather than automatic processing
(Balota and Paul 1996; Khader et al. 2003).
The amount of information available to participants is a potential explanation for dramatic differences in the amount of additive priming observed in previous studies, with
factors including the length of processing time and amount of conscious attention to the
prime-target relationships. Because the relationships between multiple primes and targets in
Balota and Paul (1996) ambiguous condition did not converge on single conceptual representations, the prime-prime-target relationship was less likely to be apparent to participants
completing the experiment. In contrast, the multiple primes-target relationships in Balota
and Paul (1996) unambiguous condition and in Khader et al. (2003) would have made the
prime-target relationships more likely to be recognized by participants. By converging on
a single conceptual representation in Balota and Paul (1996) unambiguous condition, and
filling a situational context with targets in Khader et al. (2003), multiple primes would
have increased the information explicitly available to participants. The long SOAs in multiple prime trials and slower overall reaction times in relatedness judgment tasks compared
to lexical decisions would also have increased the processing time, thereby increasing the
potential for strategic processing such as expectancy generation to contribute to the priming
effect.
123
123
continuous LDT, although the processing time allowed by the SOA or RSI in experimental
designs is a factor. McNamara and Altarriba (1988, experiments 1 and 2) used the same
group of mediated (i.e. lion-stripes), directly related (i.e. lion-tiger), and unrelated pairs in
separate standard and continuous LDT experiments. Task complexity was manipulated by
having equal numbers of participants respond to either mediated-only or both mediatedand directly-related pairs in both the standard and continuous experiments (McNamara and
Altarriba 1988). The mediated-only groups demonstrated priming in both experiments, but
mediated pairs that were mixed with directly-related pairs only demonstrated priming in
the continuous LDT experiment (McNamara and Altarriba 1988). McNamara and Altarriba
(1988) attributed the difference to the elimination of a post-retrieval, task specific strategy
described as relatedness checking in the continuous LDT task. Thus, continuous list designs
may reduce the likelihood that participants will be able to recognize some or all prime-target
relationships and use attentional or strategic processing to anticipate targets.
Table 1 Mean RTs with sample stimuli for participant (part.) and item results for each experimental condition
Priming condition
Sample stimuli
Prime 1
Prime 2
Part. mean
Target
(N = 40)
SD
Item mean
SD
(N = 30)
Car
Tourist
Renting
557.51
66.94
555.88
Box
Tourist
Renting
560.70
70.81
564.61
44.45
Car
Dentist
Renting
573.84
82.60
573.39
47.44
Box
Dentist
Renting
574.24
80.37
572.75
51.17
32.71
123
Table 2 Planned comparisons for participant and item results to address all research questions (RQ)
Comparisons
Participants (N = 40)
Items (N = 30)
Mean
difference
SE
p values
Mean
difference
SE
p values
RR v UU (RQ1)
16.73
6.96
.021
16.86
6.16
.011
RR v UR (RQ2)
3.20
5.57
.570
8.73
5.25
.107
RR v RU (RQ3)
16.34
6.19
.012
17.51
5.94
.006
Based on previous findings (Khader et al. 2003), priming was expected for both participants and items for related agent and patient nouns compared to two unrelated noun primes
preceding the same target verbs, using a continuous LDT task (RQ1). If observed, this will
demonstrate a priming effect on verbs of multiple thematic role nouns related via familiar situation schemas. No specific predictions were made for RQ2 or RQ3, since previous findings
indicating an additive effect for two related primes (RR) compared to single related primes
were the result of different priming paradigms. However, if the RR condition produced reaction times greater than single primes in both the first and second position paired with unrelated
noun primes (UR and RU) (both RQ2 and RQ3), then it would indicate additive priming for
the RR condition. Previous findings regarding multiple primes strongly suggest that priming
of single related primes are likely to be greater in the second than in the first position (where
the unrelated prime appears between the related prime and the target), so that additive priming can only be inferred if the priming effect in the RR condition is significant greater than
both the UR (RQ2) and RU (RQ3) conditions (Balota and Paul 1996; Khader et al. 2003).
See Table 2 for a list of planned post hoc mean comparisons used to test the three research
questions.
Method
Participants
To judge the relatedness of stimuli, 15 young adults were recruited to complete a survey
using a rating scale, and 45 participants completed the priming experiment. All participants
were young adult, right-handed, monolingual English speakers, age 1830, with no history
of learning disabilities or neurological disorders. Participants completed informed consent
procedures, were recruited from the University of Florida and Gainesville, FL communities,
and received compensation in the form of either course credit from their instructors or a $5
gift card.
Stimuli
Stimuli consisted of prime-prime-target triads in which all primes were nouns and all targets were verbs in the present progressive (-ing) form. Relatedness of the primes to their
targets was manipulated, and in the related condition one prime was a typical agent, and the
other a typical patient, of the target verb. The presentation order of agents and patients was
evenly mixed in the priming experiment. Primes and targets were related via familiar situation
123
schemas in which they commonly occur together. Target verbs were also paired with unrelated primes. The effect of animacy was controlled by matching related and unrelated primes
for animacy on a one-to-one basis. Related and unrelated primes were matched as closely as
possible on the psycholinguistic variables of written frequency, length (in letters, phonemes,
and syllables), concreteness, printed familiarity, imageability and age of acquisition ratings.
Psycholinguistic data as well as unrelated nouns were collected using the MRC Psycholinguistic Database (Clark 1997). Frequency (KFFRQ) was drawn from Kucera and Francis
(1967) Computational analysis of present-day American English. Printed familiarity (FAM),
concreteness (CNC), and imageability (IMG) values were derived by merging normative
data from three sources, the Paivio Norms (an unpublished expansion of Paivio et al 1968),
Toglia and Battig (1978), and Gilhooly and Logie (1980), by a method described by Coltheart
(1981). Age of acquisition (AOA) values were drawn from Gilhooly and Logie (1980) normative data, mathematically transformed to match the 100700 point range used for the other
rated variables. Average values for these variables were collected for both related and unrelated primes, as well as the verb targets and the verbs in their infinitive forms, for which ratings
data are more often available. Attempts were made to avoid the inclusion of strongly classambiguous words using Francis and Kucera (1982) Frequency analysis of English usage,
which gives frequency counts of English words by part of speech (e.g. noun, verb, adjective,
etc.). All words used as nouns in this study were more frequent than verb forms by a ratio of 6
or better. All words used as verbs, which are as a class used less frequently than nouns, were
dominant over noun forms by a ratio of 2 or better, with one conflict disregarded because it
is not consistent with modern usage (healing, produced with equal frequency as a noun and
a verb in the text corpus, from Francis and Kucera (1982).
Because it is possible for thematically-related words to also be strongly associated,
attempts were made to minimize association as a potential factor in this priming experiment. Word pairs with strong associations were avoided by collecting forward associations
in prime-prime and prime-target pairings (three pairings per triad) where available using the
University of South Florida free association norms (Nelson et al. 1998). Association norms
were available for 83 % of the pairings, and all forward pairings had a lower than 0.10 association ratio except for one direct pair (boss-employee, 0.147) and two prime-target pairs that
were presented with an intervening prime (dice-[gambler]-rolling, 0.124 and thief-[money]steal, 0.388).
Potential related stimuli were previously developed in a thematic-role completion questionnaire, in which participants responded to a series of related questions and statements asking for single word responses fitting a thematic role, and by combining semantically-related
agent-verb and patient-verb dyads identified in earlier research (Edmonds and Mizrahi 2011;
McRae et al. 2005). The questionnaire consisted of a series of related questions and statements asking participants to enter single word responses fitting one of five thematic roles;
verb, agent, patient, instrument, or location, with one to five response spaces each.
Potential stimuli triads created in the stimuli questionnaire phase were tested for semantic
relatedness using a rating survey. The goal of this survey was to guide final selection of stimuli
for the priming experiment and to demonstrate that the selected role fillers accurately match
participant judgments about the commonality of the scenarios formed by the triads. Related
agents and patients and their psycholinguistically matched unrelated nouns were mixed in
four relatedness conditions. Unrelated items were matched individually to agent and patient
nouns based on animacy and on frequency range (low: 110; medium: 1160; high >60 after
Merritt et al. 2006), and as closely as possible on length in letters, phonemes and syllables.
Related and unrelated agents and patients were matched by group across the psycholinguistic variables of age of acquisition, frequency, concreteness, familiarity, and imageability as
123
described above for stimuli development. To verify that related and unrelated primes were
not significantly different on any of these psycholinguistic variables, a series of one-way
ANOVA comparisons of related and unrelated prime 1 s and related and unrelated prime 2 s
was performed. These analyses did not find a difference between these stimuli groups on any
of the psycholinguistic factors at a 0.05 significance level.
Participants rated the triads on a survey form for each verb in all four relatedness conditions.
Participants were asked to rate how common it is for the scenarios described by each triad to
occur. Similar to previous investigations (Edmonds and Mizrahi 2011; Ferretti et al. 2001;
McRae et al. 1997), triads were rated by circling a number on a 7-point ordinal scale with 1 =
not common at all, and 7 = highly common. Mauchlys test was significant ( 2 (5) = 26.81,
p < .001) indicating that the assumption of sphericity in the condition variances was not
met, as expected, and adjustment to degrees of freedom was required (Field 2009). Repeatedmeasures ANOVA was performed with a Huynh-Feldt correction, which is a conservative
procedure to adjust for the unequal variance between conditions (Huynh-Feldt Epsilon =
.716) (Field 2009). This measure showed a significant effect of relatedness (F(2.15, 62.32)
= 491.98, p < .001) across conditions. A Sidak post hoc analysis with Holms sequential
correction (Holm 1979) revealed significant differences between the related agent + patient
condition, unrelated condition, and the two mixed related + unrelated conditions ( p < .001
for all comparisons), and there was no significant difference between the two mixed (agent +
unrelated and unrelated + patient) conditions ( p = .877). From the results of the relatedness
survey, a final list of 30 verbs with related and unrelated primes was selected for the priming
experiment. The experimental stimuli list is included in Appendix.
Nonexperimental filler words were included in the task to separate the experimental triads,
reduce the likelihood of recognizing the pattern of relationships in experimental triads, and
to keep the related-unrelated proportion low. The relatedness proportion is the ratio of the
trials in which targets are preceded by a related word to the total number of trials (Ferretti
et al. 2001; McRae et al. 2005; Neely 1991). A low relatedness proportion reduces the likelihood that participants will recognize or anticipate the pattern of relatedness used in the
experimental trials. A total of 140 filler words were drawn from the MRC Psycholinguistic
Database (Clark 1997), similar to the process used to select unrelated primes. The relatedness
proportion cannot be calculated in the same way for continuous LDT as it is for standard LDT
because primes are not obviously paired with targets (McRae and Boisvert 1998). However,
an attempt was made to estimate the relatedness proportion by calculating the ratio of trials
that were deliberately preceded by a related word to the total number of trials, yielding an
estimated relatedness proportion of 0.18, which can be considered low for comparison to
studies examining similar thematic role relationships (McRae et al. 2005).
Nonwords were individually matched to words for length in letters and syllables to
avoid unintended effects of length between word and nonword items during the experimental task. Nonwords were either drawn from the ARC nonword database (Rastle et al.
2002) or by changing one or more letters in a real word (e.g. g d and n m changing
engineer endimeer).
Apparatus
Participants were seated at a standard desk before a Dell Latitude D820 laptop with a 15 inch
monitor. Two keys were marked for responses on the lexical decision task; the < key was
labeled with a Y for Yes responses, and the > key with an N for No responses. This
experiment was conducted using DirectRT v2006 software (Jarvis 2006).
123
10
Procedures
Experimental triads were embedded in a presentation list in random order. Participants were
instructed to read and respond to every letter string presented on a screen by pressing a yes
or no key to indicate whether it was a real English word. Letter strings appeared in the center
of the monitor screen in Times New Roman size 32 black font with a white background. Letter
strings remained on the screen until a response was given, followed by a response-stimulus
interval [RSI] of 100 ms consisting of a blank screen before the next item in the list appeared.
Participants were given no information about which items were of experimental interest. See
Fig. 1 for a schematic of the task presentation.
Participants completed an instruction and practice section with an experimenter. During
the practice section, items advanced automatically following correct responses, and feedback
was given following incorrect responses, or if they responded faster than 100 ms or slower
than 1,200 ms. Following the practice section participants were given an opportunity to ask
questions about the task and conveyed understanding before starting the experiment. Participants were informed that they would no longer receive any feedback before they began the
experiment.
During the experiment participants saw 1,000 letter strings consisting of 120 experimental triads (360 word presentations), 140 filler words, and 500 nonwords. The ratio of words
to non-words was 0.5, such that yes and no responses were equally likely for each letter string. The experimental triads were divided into four lists such that each target verb
appeared only once in each list and the experimental conditions were balanced across the
lists. The order of presentation of the lists, and of the experimental and nonexperimental letter
Fig. 1 Continuous list lexical decision task schematic. Participants were instructed to respond to every letter
string in the priming experiment (1,000 items total) by pressing keys marked Y for yes, for real English
words, or N for no, for pronounceable nonwords. Experimental triads (e.g., thief-money-steal) were embedded in the list along with filler words (e.g., blanket) and nonwords (e.g., falping) so that items of interest were
disguised from participants
123
11
strings within each list, were randomized for each participant so that any effects of learning
or fatigue were distributed across all targets and conditions. The lists were organized with
a semi-random number of letter strings appearing between each of the experimental triads,
with a minimum of 3 and an average of 5.3 nonexperimental letter strings between each
triad. To reduce fatigue, participants were given two 5 min breaks, and the experiment was
completed in less than 30 min. To clarify each component of the testing session, an asterisk
appeared for 500 ms as a warning signal before the practice section, before the first letter
string at the beginning of the experimental task and after each break.
Design and Analyses
A repeated-measures design was used, with four relatedness conditions. Each participant
completed all conditions, distributed throughout the experimental task, so that participants
served as their own controls. Participant results were initially analyzed for overall accuracy of
lexical decisions. A minimum of 95 % accuracy was required for inclusion in further analysis.
Lexical decisions for both primes and the target in each experimental triad were examined,
and RTs to targets were excluded from further analysis for any triad that had one or more
lexical decision errors. It is standard to exclude trials containing decision errors from further
analysis (Ferretti et al. 2001; Jones 2010; Khader et al. 2003; Perea and Rosa 2002; Vigliocco
et al. 2004). Trials including errors on either the prime or target were excluded from reaction time (RT) analysis, similar to previous continuous LDT experiments (McNamara and
Altarriba 1988; Moss et al. 1995).
Response latencies (RTs) were analyzed by condition for participants and items. RTs that
fell greater than two standard deviations above or below the mean for each participant or
item by condition were excluded from further analysis. Effects of relatedness were examined
using repeated measures ANOVA. Then, the planned comparisons (RR v UU [RQ1], RR v
UR [RQ2] and RR v RU [RQ3]; refer to Table 2) were conducted with paired t tests. The
mean error rate for each participant or item by condition was determined. Mean error rates
were compared for effects of relatedness using repeated measures ANOVA. Based on the
results of similar research, no significant effect of relatedness on error rates was anticipated
(Ferretti et al. 2001; Hare et al. 2009; Jones 2010).
Results
The requirement of 95 % accuracy on the lexical decision task was not met by 5 participants,
and their data were excluded from further analysis. The remaining 40 participants had a mean
age of 20.55 (SD 2.09) years, and 18 % were male. The overall accuracy of lexical decisions
was 98 %, which falls in a range typical of lexical decision tasks (Ferretti et al. 2001; Jones
2010; McRae and Boisvert 1998). An error on lexical decisions for either prime or the target
occurred on 6 % of experimental triads. The average SOA from presentation of prime 1 to
onset of the target for experimental triads was 1,432 ms, consisting of the mean RT to primes
(616.15 ms) plus the 100 ms RSI, for two primes per triad.
Participant Analysis
Target responses that were greater than 2 SDs above or below the condition mean
for each participant were excluded from participant analysis, consisting of 6 % of data.
123
12
In combination with responses excluded due to lexical decision errors in a triad (not a
sum because some responses qualified for exclusion for both reasons), 10 % of total data
were excluded from participant analysis. An unplanned series of one-way ANOVA analyses
comparing the condition means (as used in further analysis, with trials that have any lexical
decision error excluded) with the condition means when only errors to the targets are excluded
found no significant difference for any of the four experimental conditions ( p > .05 for all
comparisons).
A repeated-measures ANOVA showed a significant effect of relatedness across participants (F(3,39) = 4.62, p = .004). Mauchlys test was not significant ( 2 (5)=8.34,
p = .139) indicating that the assumption of sphericity in the condition variances was met
and no adjustment to degrees of freedom was required (Field 2009). Participants had a
mean RT of 557.57 ms (SD 66.94) to targets in the RR condition, a mean of 560.70 ms
(SD 70.81) to targets in the UR condition, a mean of 573.84 ms (SD 82.60) to targets in
the RU condition, and a mean of 574.24 ms (SD 80.37) to targets in the UU condition (see
Table 1).
There was a difference of 16.73 ms between RR and UU means for participants (RQ1).
A paired t-test analysis of the RR-UU mean comparison showed a significant difference
(t (39) = 2.404, p = .021). Participants showed a mean difference of 3.20 ms between
RR and UR means (RQ2). This difference did not reach significance (t (39) = .574, p =
.570). Participants had a mean difference of 16.34 ms between RR and RU means (RQ3),
which was a significant difference (t (39) = 2.640, p = .012). Table 2 summarizes the
RT mean differences for participants for each of the three research questions posed in this
study.
The mean error rate for each participant by condition was calculated. The results of a
repeated measures ANOVA showed that the effects of relatedness on error rates was not
significant for participants (F(3,39) = 1.64, p =.183). This result was expected based on
findings in previous research as well as the low overall error rate of 2 % (Ferretti et al. 2001;
Hare et al. 2009; Jones 2010).
Item Analysis
Target responses that were greater than 2 SDs above or below the condition mean for each item
were excluded from item analysis, consisting of 5 % of data. In combination with responses
excluded due to lexical decision errors in a triad (not a sum because some responses qualified
for exclusion for both reasons), 10 % of total data were excluded from item analysis. An
unplanned series of one-way ANOVA analyses comparing the condition means (as used in
further analysis, with trials that have any lexical decision error excluded) with the condition
means when only errors to the targets are excluded found no significant difference for any
of the four experimental conditions ( p >.05 for all comparisons).
A Repeated-measures ANOVA showed a significant effect of relatedness across items
(F(3,29) = 3.67, p = .015). Mauchlys test was not significant ( 2 (5) = 6.39, p = .271)
indicating that the assumption of sphericity in the condition variances was met and no adjustment to degrees of freedom was required (Field 2009). RTs to items had a mean of 555.88 ms
(SD 32.71) in the RR condition, a mean of 564.61 ms (SD 44.45) to targets in the UR condition, a mean of 573.39 ms (SD 47.44) to targets in the RU condition, and a mean of 572.75 ms
(SD 51.17) to targets in the UU condition (see Table 1).
There was a mean difference of 16.86 ms between RR and UU means for items (RQ1).
A paired t test showed a significant difference (t (29) = 2.736, p = .011). Items showed
123
13
a mean difference of 8.73 ms between RR and UR means (RQ2). This difference was not
statistically significant (t (29) = 1.664, p = .107). Participants had a mean difference of
17.51 ms between RR and RU means (RQ3), which was a statistically significant difference
(t (29) = 2.945, p = .006). See Table 2 for details.
The mean error rate for each item by condition was calculated. The results of a repeated
measures ANOVA showed that the effects of relatedness on error rates was not significant
for items (F(3,29) = 1.43, p = .240). This result was expected based on findings in previous
research as well as the low overall error rate of 2 % (Ferretti et al. 2001; Hare et al. 2009;
Jones 2010).
Discussion
The purpose of the current study was to evaluate whether additive priming effects occur
with two related thematic role primes preceding verb targets within a continuous priming paradigm. This study was motivated by previous findings that have shown priming of
verbs following related agents (Edmonds and Mizrahi 2011; McRae et al. 2005) and patients
(Edmonds and Mizrahi 2011; McRae et al. 2005) in young adults as well as one study that
found additive priming with two thematic role primes related to the verb (Khader et al. 2003).
These studies evaluated priming where the prime(s) and target were presented in explicit pairs
or triads, and the participants only responded to the targets, thus providing the opportunity
for the pattern of presentation and multiple presentation of targets to activate attentional
processes. In the current study, we aimed to reduce attentional processes to potential relationships between primes and targets with a continuous list paradigm. Since the fundamental
goal of this study was to understand whether two related thematic role fillers result in additive priming, we compared the RR condition to the three other conditions (UU, UR, and
RU). We recognize that repeated presentations of the verb target could result in a repetition
effect. In order to reduce any potential repetition effects, we (1) used the continuous priming paradigm, which eliminates explicit presentation of primes and targets, so participants
responded to each word, thereby reducing the chance of recognition of specific words in
specific word order, (2) had a low estimated relatedness proportion (0.18) within our stimuli, which is consistent with previous semantic priming studies (Ferretti et al. 2001; McRae
et al. 2005; Neely 1991), and (3) distributed experimental verbs across all four relatedness
conditions by participants and items with semi-random presentation of the lists so that each
target-condition combination appeared first to last an equal number of times throughout the
experiment, while the order of the experimental and nonexperimental letter strings within
each list were simply randomized. Thus, any chance of a repetition effect was randomized
across conditions.
Statistically significant mean differences between the RR and UU conditions, compared
within participants and items for research question 1 (RQ1), showed a priming effect of
approximately 17 ms for two related agent and patient noun primes preceding related verbs.
This is an expected result based on previous studies that observed priming effects of single thematically-related noun primes to verbs as well as of two related primes to targets
(Balota and Paul 1996; Edmonds and Mizrahi 2011; Ferretti et al. 2001; Khader et al. 2003;
McRae et al. 2005). These effects have been seen using explicitly presented primes and
targets; however, results of continuous and standard LDT using the same stimuli have also
been found to produce similar patterns of results (McNamara and Altarriba 1988; Perea and
Rosa 2002). While explicit attentional processing likely contributed to more robust priming
observed in previous studies, the continuous list paradigm used in this study supports the
123
14
123
15
thematic role order can be eliminated as a factor in the different results seen in RQ2 and
RQ3.
The RR versus RU comparison did not directly address whether there was priming in
the RU condition. However, it is apparent by looking at the averages for the RU and UU
conditions that there was no priming, since the reaction times across conditions were the
same for participants and items. Taken together, the RU and UR findings suggest that a single
related noun prime is sufficient to account for the observed priming in healthy young adults
when it directly precedes the target (UR), whereas preactivation decays too rapidly in the
presence of an intervening unrelated noun (RU) to facilitate a related verb. Further, there is no
added benefit of two related primes as compared to one related thematic role prime directly
preceding the verb (UR).
Khader et al. (2003) also failed to observe a priming effect of a single related prime in
the first position of a multiple prime task. However, Balota and Paul (1996) did observe
priming effects of single related nouns on noun targets in all of their relatedness and task
conditions, suggesting that the influence of the first prime in a multiple priming paradigm
cannot be disregarded. In Balota and Paul (1996) converging prime paradigm with nouns,
each prime activated different conceptual aspects of its prime (e.g. lion and stripes activate
distinct category and visual conceptual features of the target tiger). In contrast, the related
primes in Khader et al. (2003) and in the present study are related via common situation
schemas. It is possible that intervening, unrelated thematic-role fillers interfered with the
activation of situation schemas, while presentation of intervening unrelated nouns that did
not share features or category membership with the targets in Balota and Paul (1996) did
not interfere with activation of conceptual features of concrete nouns. Further, Ferretti et al.
(2001) previous finding that agents and patients did not prime related verbs when used in
an incongruous sentence context suggests the possibility that exposure to semantic or syntactic information that contradicts real-world knowledge about common situation schemas
may have an interference effect on activation of concepts related in this way. The absence
of syntactic context in the present study supports a semantic basis for such an effect. In
the future, a neutral condition using generic nouns (i.e. someone or something) in addition
to the related and unrelated prime conditions included in this study would test the proposed hypothesis that failure to observe a priming effect of single related primes in the first
position is due to intervening incongruous role fillers that interfere with activation of situation
schemas.
The continuous list design of the present study attempted to reduce or eliminate attentional processes to potential relationships between words, and the results showed priming
in the UR and RR conditions, but no additive effect for the RR condition. With the primeprime-target relationships of experimental interest sufficiently disguised by the continuous
list, there was a greatly reduced likelihood of conscious expectancy generation over the
related primes to the target in the observed priming. Thus, the priming effect of the thematic
role nouns on related verbs is believed to be the result of automatic semantic processing,
suggesting a direct relationship between conceptual representations of agents and patients
to verbs that is not dependent on syntactic context or conscious memory of common situations. In previous reports, where primes were explicitly paired with targets and conscious
attention to the relationship between them was demanded, the addition of a second related
prime resulted in large increases in the priming effect compared to single related primes
(Balota and Paul 1996; Khader et al. 2003). These additive effects may have been boosted
when the nature of the relationship between primes and targets was clear enough to allow for
activation of attentional processes and/or task-specific strategic processing. Longer SOAs
and explicit semantic processing demands such as those observed in previous studies have
123
16
been associated with greater attentional processing (Neely 1991), and thus may have provided an added boost in activation to result in additive priming. But longer SOAs alone
are not likely the reason for additive priming, because the current study had fairly long
average SOAs (1,432 ms) similar to Khader et al. (2003) (1,300 ms), but additive priming
was not observed in this study, suggesting that the continuous list paradigm was effective at reducing the potential of strategic processing sufficiently to observe relatively isolated effects of automatic processing. Therefore, it may be prudent to evaluate questions of
additive effects in future studies with more conservative paradigms such as continuous list
presentation.
Future Directions
The current study used carefully controlled stimuli and a conservative priming paradigm to
evaluate whether additive priming occurred with a verb target following two related agent
and patient primes. The results showed no priming advantage for two related thematic roles
as compared to one that directly precedes the related verb. However, the absence of additive priming does not mean that additive priming cannot occur within situation schemas, as
there are myriad variables which could be manipulated to understand better the co-activation
effects among thematic roles and related verbs. Thus, more research is needed to understand
better the implications of the current findings with respect to the semantic relationships of
verbs and their thematic roles.
The selection of target verbs in this study was limited to 2-place thematic structures,
which require two thematic role fillers to fill their simplest permitted syntactic frames (NP
V NP). These findings cannot be applied to all verbs, since the thematic role properties used
to restrict selection of verbs in this study are not shared by all verb classes. Given previous
sentence processing studies which have found differences in reaction times to 2-place and
3-place verbs, it is difficult to predict the possible effects of multiple primes on activation of
verbs with more complex stored lexical representations in semantic priming where there is
an absence of syntactic context (Ahrens 2003; Shapiro et al. 1991).
The current findings also do not extend to fillers of thematic roles that are optional in
these verbs, such as instruments and locations. It is conceivable that different thematic role
combinations could activate more and different features in the conceptual representation of
verbs which could result in additive priming. Instruments have been shown to prime verbs,
and it is possible that they may be inherent in the meaning of a verb even if not required
syntactically, which could potentially result in a greater boost in activation (McRae et al.
2005). Thus, for example, the combination of a patient (tree) and instrument (ax) may have
an additive effect on the verb chopping where the agent (lumberjack) and patient (tree) do not.
Attempting to understand semantic processing of multiple concepts is a complex task
that will require much research with different experimental methods. Additionally, we must
extend this work to other populations in order to improve our understanding of the effects of
aging and injury on semantic processing. Such understanding is crucial to improving the sensitivity and effectiveness of assessment and treatment of neurogenic language disorders that
can affect the semantic and/or lexical semantic systems (e.g., aphasia, semantic dementia).
Appendix
See Table 3
123
17
Related
prime 2
Target
Unrelated
prime 1
Unrelated
prime 2
Role
order
Defendant
Prosecutor
Accusing
Spectator
Vegetarian
PA
Traitor
Country
Betraying
Builder
Evening
AP
Couple
Anniversary
Celebrating
Master
Examination
AP
Dog
Bone
Chewing
Boy
Coal
AP
Wood
Lumberjack
Chopping
Dirt
Cartoonist
PA
Evidence
Detective
Discovering
Division
Architect
PA
Ambulance
Paramedic
Driving
Appliance
Biologist
PA
Tenant
Landlord
Evicting
Broker
Guardian
PA
Electrician
Lights
Fixing
Participant
Sphere
AP
Shuttle
Astronaut
Flying
Luggage
Machinist
PA
Doctor
Patient
Healing
Father
Teacher
AP
Host
Guest
Inviting
Ward
Clerk
AP
Grandma
Sweater
Knitting
Rancher
Cottage
AP
Math
Student
Learning
Mall
Captain
PA
Boss
Employee
Managing
Aunt
Composer
AP
Baker
Dough
Mixing
Widow
Sheet
AP
Toilet
Plumber
Plunging
Closet
Athlete
PA
Priest
Sermon
Preaching
Knight
Locker
AP
Car
Tourist
Renting
Box
Dentist
PA
Cat
Fireman
Rescuing
Cow
Nominee
PA
Dice
Gambler
Rolling
Lime
Florist
PA
Maid
Floor
Scrubbing
Fool
Scale
AP
Waitress
Customer
Serving
Princess
Engineer
AP
Soldier
Enemy
Shooting
Pianist
Member
AP
AP
Artist
Portrait
Sketching
Bishop
Hardware
Pig
Butcher
Slaughtering
Rat
Actress
PA
Thief
Money
Stealing
Clown
Water
AP
Ball
Pitcher
Throwing
Clay
Scholar
PA
Movie
Audience
Watching
Arrow
Chairman
PA
Novel
Author
Writing
Motor
Cowboy
PA
Primes 1 and 2 varied between related and unrelated by experimental condition. The order of presentation of
agents and patients in experimental triads was balanced across patient-agent (PA) and agent-patient (AP). The
role order of primes was the same across all conditions for each target verb.
References
Ahrens, K. (2003). Verbal integration: The interaction of participant roles and sentential argument
structure. Journal of Psycholinguistic Research, 32, 497516.
Ahrens, K., & Swinney, D. (1995). Participant roles and the processing of verbs during sentence
comprehension. Journal of Psycholinguistic Research, 24, 533547.
Balota, D. A., Black, S. R., & Cheney, M. (1992). Automatic and attentional priming in young and
older adults: Reevaluation of the two-process model. Journal of Experimental Psychology: Human
Perception and Performance, 18, 485502.
123
18
Balota, D. A., & Paul, S. T. (1996). Summation of activation: Evidence from multiple primes that converge
and diverge within semantic memory. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 22, 827845.
Clark, C. (1997). MRC psycholinguistic database (Version 2.00). [Machine usable dictionary]. Available
from University of Western Australia Web site, http://www.psy.uwa.edu.au/mrcdatabase/uwa_mrc.
htm.
Coltheart, M. (1981). MRC psycholinguistic database user manual: Version 1. Retrieved February 9, 2011,
from http://www.psych.rl.ac.uk/User_Manual_v1_0.html.
Edmonds, L. A., & Mizrahi, S. (2011). Online priming of agent and patient thematic roles and related verbs
in younger and older adults. Aphasiology, doi:10.1080/02687038.2011.599527. Retrieved December 1,
2011, from http://dx.doi.org/10.1080/02687038.2011.599527.
Ferretti, T. R., McRae, K., & Hatherell, A. (2001). Integrating verbs, situation schemas, and thematic role
concepts. Journal of Memory and Language, 44, 516547.
Field, A. (2009). Discovering statistics using SPSS: Introducing statistical methods (3rd Ed.). London: SAGE
Publications., Ltd..
Francis, W. N., & Kucera, H. (1982). Frequency analysis of English usage: Lexicon and grammar.
Boston: Houghton Mifflin.
Gilhooly, K. J., & Logie, R. H. (1980). Age of acquisition, imagery, concreteness, familiarity and ambiguity
measures for 1944 words. Behaviour Research Methods and Instrumentation, 12, 395427.
Grondin, R., Lupker, S. J., & McRae, K. (2009). Shared features dominate semantic richness effects for
concrete concepts. Journal of Memory and Language, 60, 119.
Hare, M., Jones, M., Thomson, C., Kelly, S., & McRae, K. (2009). Activating event knowledge. Cognition, 111, 151167.
Holm, S. (1979). A simple sequentially rejective multiple test procedure. Scandinavian Journal of
Statistics, 6, 6570.
Jarvis, B. G. (2006). MediaLab (Version 2006.2.40) [Computer Software]. New York, NY: Empirisoft
Corporation.
Jones, L. L. (2010). Pure mediated priming: A retrospective semantic matching model. Journal of
Experimental Psychology: Learning, Memory and Cognition, 36, 135146.
Khader, P., Scherag, A., Streb, J., & Rsler, F. (2003). Differences between noun and verb processing in
a minimal phrase context: A semantic priming study using event-related brain potentials. Cognitive
Brain Research, 17, 293313.
Kucera, H., & Francis, W. N. (1967). Computational analysis of present-day American English. Providence: Brown University Press.
McRae, K., & Boisvert, S. (1998). Automatic semantic similarity priming. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 24, 558572.
McRae, K., Ferretti, T. R., & Amyote, L. (1997). Thematic roles as verb-specific concepts. Language and
Cognitive Processes, 12, 137176.
McRae, K., Hare, M., Elman, J. L., & Ferretti, T. (2005). A basis for generating expectancies for verbs
from nouns. Memory and Cognition, 33, 11741184.
McNamara, T. P., & Altarriba, J. (1988). Depth of spreading activation revisited: Semantic mediated
priming occurs in lexical decisions. Journal of Memory and Language, 27, 545559.
Merritt, P., DeLosh, E. L., & McDaniel, M. A. (2006). Effects of word frequency on individual-item and
serial-order retention: Tests of the order-encoding view. Memory and Cognition, 34, 16151627.
Moss, H. E., Ostrin, R. K., Tyler, L. K., & Marslen-Wilson, W. D. (1995). Accessing different types of
lexical semantic information: Evidence from priming. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 21, 863883.
Neely, J. H. (1991). Semantic priming effects in visual word recognition: A selective review of current
findings and theories. In D. Besner & G. W. Humphreys (Eds.), Basic processes in reading (pp. 264
336). Hillsdale NJ: Lawrence Erlbaum Associates.
Nelson, D. L., McEvoy, C. L., & Schreiber, T. A. (1998). The University of South Florida word association, rhyme, and word fragment norms. Retrieved January 21, 2011, from http://web.usf.edu/
FreeAssociation/.
Nickels, L. (2002). Therapy for naming disorders: Revisiting, revising, and reviewing. Aphasiology, 16, 935
979.
Paivio, A., Yuille, J. C., & Madigan, S. A. (1968). Concreteness, imagery and meaningfulness values for
925 words. Journal of Experimental Psychology Monograph Supplement, 76 (3, part 2).
Perea, M., & Rosa, E. (2002). The effects of associative and semantic priming in the lexical decision
task. Psychological Research, 66, 180194.
123
19
Raczaszek-Leonardi, J., Shapiro, L. P., Tuller, B., & Kelso, J. A. S. (2008). Activating basic category
exemplars in sentence contexts: A dynamic account. Journal of Psycholinguistic Research, 37, 87113.
Rastle, K., Harrington, J., & Coltheart, M. (2002). 358,534 nonwords: The ARC nonword database. Quarterly
Journal of Experimental Psychology, 55, 13391362.
Sailor, K., Brooks, P. J., Bruening, P. R., Seiger-Gardner, L., & Guterman, M. (2008). Exploring the time
course of semantic interference and associative priming in the picture-word interference task. The
Quarterly Journal of Experimental Psychology, 62, 789801.
Schnur, T. T., Schwartz, M. F., Brecher, A., & Hodgson, C. (2006). Semantic interference during
blocked-cyclic naming: Evidence from aphasia. Journal of Memory and Language, 54, 199227.
Setola, P., & Reilly, R. G. (2005). Words in the brains language: An experimental investigation. Brain
and Language, 94, 251259.
Shapiro, L. P., Brookins, B., Gordon, B., & Nagel, N. (1991). Verb effects during sentence processing. Journal
of Experimental Psychology: Learning, Memory, and Cognition, 17, 983996.
Thompson, C. K., Shapiro, L. P., Kiran, S., & Sobecks, J. (2003). The role of syntactic complexity in
treatment of sentence deficits in agrammatic aphasia: The complexity account of treatment efficacy
(CATE). Journal of Speech, Language, and Hearing Research, 46, 591607.
Toglia, M. P., & Battig, W. R. (1978). Handbook of semantic word norms. New York: Erlbaum.
Vigliocco, G., Vinson, D. P., Lewis, W., & Garrett, M. F. (2004). Representing the meanings of object and
action words: The featural and unitary semantic space hypothesis. Cognitive Psychology, 48, 422488.
123