The study of mental
representations is a central concern in the overall theory of cognition and the
mind. Concepts are mental representations of entities arising from our
experiences, and are fundamental for prediction and communication. Our mental
organization influences how we use our knowledge, and these models of
categorizations have been developed and studied through the past decades. Traditional
categorization models emphasized heavily on feature-based similarity while
recent evidences have shown categorization to be include more than just
comparisons. Contemporary approaches such as Wisniewski & Bassok’s (1999)
and Lin & Murphy’s re-evaluate the role of thematic relations, suggesting
the importance of thematic relations, together with taxonomic relations, in
mental organization. In this paper, we will first discuss the shift of research
looking at thematic relations, and then tie in current research on adult use of
thematic relations with Funnell’s (2001) Levels of Meaning model.

Similarity grounding categorization

underpins many traditional models of the organization in categorization. One
classic model is the prototype approach, in which the categorical membership of
the novel object is determined by comparing the object to the prototype that
represents the category (Rosch & Mervis, 1975). The other is the exemplar
approach, where the comparison is made between the novel object and the actual categorical
exemplars (Brooks, 1978, 1987). While stark differences exist between the two
models, both models hinge on similarity between the novel object and
categories’ representation for categorization purposes.

had been argued against as insufficient for grounding categorization by several
researchers. One main argument comes from Goodman (1972), claiming that
similarity is too flexible to ground categorization. Specifically, Goodman
argued that similarity is not based on solely perceptual input, but involves
other factors. Goodman’s strong claim has been supported by empirical evidences
showing that other factors like expertise (Sjoberg, 1972) and environment
(Hardnad, 1978) influence similarity judgments as well. In response, Goldstone
(1994) suggested that similarity is indeed an integration of multiple sources
of information.  Tversky’s Contrast Model
(1977), for instance, proposes that similarity between two entities increases
as they share more features, and decreases as they possess more distinctive
features. However, Goodman argued that the features entered into comparison can
be very broad and any substantial changes in features can affect similarity
judgment, rendering similarity a slippery slope. Nevertheless, Goldstone’s
(1994) paper attempted to find a mediating ground between similarity and
categorization, suggesting that though similarity may be slippery, it is still applicable
to most categories and is necessary in helping us understand the notion of
categorization, thus meriting a reassessment of its role. 


Need for Dual-Process Model

conventional view of conceptual organization heavily emphasized on the notion
of similarities, with category representing a set of entities sharing an
essential core of similar features, whether perceptual, functional or
biological (Medin & Ortony, 1989). This relation is referred to as the taxonomic relation (e.g. cow-horse; pen-pencil). Taxonomic relation is also
the core of the classic prototype and exemplar models mentioned above. However,
over the recent decades, a growing body of evidence has shown that thematic
relations are as important as taxonomic relations in categorization. Thematic relations are complementary
relations among entities that co-occur together in space and time (Markman,
1981, 1989). These relations can be spatial (e.g. a lid is on top of a
container), functional (e.g. a comb is used to brush hair), temporal (e.g. drying
clothes after washing them) and/or causal (e.g. a lighter produces fire).

Wisniewski & Bassok’s (1999) paper, cognitive research had based similarity
on a single process model involving only feature comparisons. However,
Wisniewski & Bassok’s paper served as a major turning point, when the need
of a dual-process model involving comparison and integration as well as the
effects of stimulus compatibility was highlighted. Their paper was motivated by
unexpected finding in Bassok & Medin’s (1997) study. Their study started
out with the premise that comparison was the only process involved in
similarity judgment. Surprisingly, they found that when actions were taxonomically
unrelated and therefore less alignable, participants tended to thematically
integrate them into a joint theme or scenario. This then led to Wisniewski
& Bassok designing experiments to predict the stimulus compatibility
effects on similarity judgment.

of statements describing actions, object pairs were intentionally designed to
vary orthogonally in terms of alignability and thematic relatedness, resulting
in four types of base-target object pairs. The target was either high (+) or
low (-) on thematic relatedness (T) and on alignability (A), in relation to the
base object in the pair. For example, one quintuplet of base-target objects
used is chair-table (A+T+), chair-bed (A+T-), chair-carpenter (A-T+) and
chair-electrician (A-T-). Participants rated object pairs on a 7-point
similarity scale and were required to write down an explanation for their
ratings. Highly alignable pairs were rated as more similar  than poorly alignable pairs. Object pairs with
preexisting thematic relations were also scored higher similarity ratings than
pairs without thematic relations. Participants’ explanations were divided into
instances of comparison and integration. Participants tended to compare object
pairs that were highly alignable and without preexisting thematic relations;
while participants tended to integrate object pairs that were poorly alignable
and with preexisting thematic relations. The robustness of the stimulus
compatibility effect was further examined in Studies 2 & 3. Despite the
task instructions modified to emphasize the appropriateness of comparison (in
Study 2) and integration (in Study 3), participants’ responses were still
strongly affected by the stimulus compatibility.

old processing model which involves only comparison cannot account for the
above findings. Wisniewski & Bassok’s paper thus demonstrated the need to extend
to a dual-process model and consider not just taxonomic relations (which
invokes comparison), but thematic relations (which invokes integration) which
may be equally important in similarity judgment as well.

Adults’ Use of Thematic Relations

the next decade, an increasing bulk of research were devoted into studying
thematic relations in mental representations. Previous developmental literature
in the 1960s and 1970s generated extensive work on thematic relations but only in
children’s conceptual development, but not in older children or young adults.
Inhelder & Piaget (1964) found that when children were tasked to sort toys,
children under 5 primarily categorized based on objects that “belonged
together” which they then tell a story about. These developmental studies led
to the conclusion of a thematic-to-taxonomic shift as part of children’s
conceptual development. However, recent literature suggested otherwise, arguing
that thematic relations do not fade away as children’s concept develop, but are
still present in adults’ concepts. For example, Smiley & Brown (1979)’s
study found the majority of educated elderly participants preferring to use thematic
over taxonomic relations. Other literatures also found elderly preferring to
use thematic relations as compared to younger adults (Smiley & Brown, 1979;
Annet, 1959; Denney, 1974). Education factor also entered into the study as
Luria (1979) and Sharp et al (1979) found evidence that uneducated adults used
more thematic relations. Motivated by these literatures, Lin & Murphy
(2001) attempted to investigate the use of thematic relations in educated,
adult participants.

& Murphy (2001) presented an intensive study on adult’s use of thematic
relations, extended across ten experiments, with each experiment debunking possible
criticisms in the preceding. Experiment 1 to 8 investigated thematic
categorizations in American adult college students while Experiment 9 & 10
investigated the conceptual content of thematic groupings. Forced choice triad
task were used, in which participants decided which of the two matches (one
taxonomic, one thematic) goes best with the target to form a category. Experiment
1 & 2 found that young, educated adults tended to use thematic
categorization 62% and 49% of the time respectively. The task procedures in
Experiment 3 to 5 differed slightlty (e.g. presenting stimuli in words instead
of word-picture combination) in attempt to replicate Smiley & Brown’s
(1979). Smiley & Brown previously found that 75% of their college student participants
using taxonomic categories 84% of the time. However, up till Experiment 5, Lin
& Murphy still found very different results from Smiley & Brown, which
could be attributed to two-third of Smiley & Brown’s original stimuli not
published. Nevertheless, Lin & Murphy found convincing evidence of adult
readily using meaningful thematic relations in categorization. The second part
of the study provide noteworthy evidence of the utility of thematic relations,
specifically that thematic relations can guide category-based induction, just
like taxonomic relations. Overall, Lin & Murphy’s experiments sum up to
suggest that the importance of thematic relations in adults’ concepts, more than
previous research have typically concluded.


Levels of Meaning Model

demonstrated the importance of thematic relations in addition to taxonomic
relations in categorization, we will now introduce Funnell’s (2001) Level of
Meaning Model to bring the two ideas together. Unlike Tulving’s (1972) classic
distinction between semantic and episodic memory, Funnell’s model represents meaning
on a continuum which connects both semantic and episodic characteristics.

discussed recent evidence from semantic dementia patients whose memory of
objects and words meanings is preserved only for current and personal use. Patient
EP, reported by Funnell (1996), was a woman who had severe semantic memory
deficits. Although she was unable to identify objects in confrontational tasks
(e.g. word-picture matching) when in clinic, her husband reported her using
similar objects appropriately in daily activities back at home. Classic models
of conceptual representation in semantic memory, which focused only on the
encyclopaedic nature of context-free information, could not account for the
observation. Studies of semantic dementia patients thus revealed the
involvement of scripts in semantic memory. 
With this, Funnel developed an extended semantic memory model, where
meaning is represented on a continuum, from being entirely embedded in the
physical context of personal experience, to being entirely stripped of
contextual information. This continuum is continuously sustained and modified
by new learning from our current personal experiences. Contextual information
of current information are stored as a specific physical script of a specific
scene at the level of “Specific Event Knowledge”. This includes information
regarding the specific objects and role players involved, organized around a
specific goal (e.g. preparing a cup of Nescafé coffee this morning). As the
same physical scenes are experienced frequently and centered on same goal,
physical properties specific to the particular scene (e.g. red coffee mug, blue
coffee mug) are replaced by more general properties (e.g. description of a
typical mug). This more abstracted level of the event script is known as “General
Event Knowledge”. Finally, the most abstracted level – “Concepts” – is
represented independently, free from typical contextual information and are
available for use in novel context. Concepts can vary in specificity, from
subordinate (e.g. coffee mug) to basic (e.g. mugs) to superordinate (e.g.
cutleries) levels. This model accounts for the observation in semantic dementia
patients. As semantic memory breaks down, the context-free concepts are
compromised, and patients can no longer recognized objects presented out of
context. However, their less abstract general event knowledge allow them to
still use and recognize objects within the context of a familiar physical

(2001) model is relevant to the research of thematic and taxonomic relations in
categorization as it connects the two into a single model. Concepts relate to
traditional taxonomic theories which organizes information based on similar
properties, while General Event Knowledge relate to thematic theories which
organizes information based on the co-occurrence in event or theme. This model
thus provides further support for people (which includes adults) making use of
both taxonomic and thematic similarity, since thematic relations underlies the
basic learning mechanism of meaning at the event knowledge level, which is then
developed through further abstraction into taxonomic concepts.

Individual Differences in
Taxonomic-Thematic Tendencies

issue that surfaced often in the study of taxonomic-thematic relations was the
individual differences in the strength of taxonomic vs thematic relations. For
instance, in Lin & Murphy’s (2011) study, participants were classified into
predominantly taxonomic or predominantly thematic based on their responses. In
Schwaryz et al.’s (2011) study of picture-naming errors, individuals with
aphasia were found to differ in their tendencies to produce taxonomic and
thematic errors. However, prior to Mirman & Graziano’s (2011) paper, the
extent of such individual differences were neither investigated nor established
for neurologically intact adults. Before the 21st century, most
studies investigating thematic relations have used tasks that explicitly required
assessing the two types of relations (e.g. the “triad” task). Few studies
looked at tasks where thematic relations are not required. Mirman &
Graziano thus investigated the activation of taxonomic and thematic relations
even when task demands do not require them. The inclusion of such tasks is
important in the establishment of real individual differences in the strength
of taxonomic vs thematic relations, as genuine individual differences should be
stable across different types of tasks, whether or not the task demands require
assessing taxonomic or thematic relations.

the first part of the study, participants were presented a four-image display
containing a target object, a semantic competitor (taxonomic or thematic) and
two unrelated distractors. Participants heard a spoken word, and were
instructed to either click or look at the corresponding image. Eye movements were
tracked and analyzed. Participants were found to more likely fixate on the
semantically related competitors than the unrelated distractors in both
taxonomically and thematically related conditions. A relative measure was then
computed to represent each individual’s tendency to activate taxonomic
relations more strongly than thematic relations. This computation was then used
in the second part of the study to predict the individual’s tendency to choose
taxonomic over thematic options in a triad task. The number of taxonomic
choices in the triad tasks were found to positively associate with the
individual participants’ relative taxonomic competition effect size computed
from the spoken-word task. Thus, the cross-task relation suggested that the
tendency to activate taxonomic vs thematic relations was indeed due to
individual differences, rather than being specific to task or stimulus.


Over the years, the study of mental organization
have evolved, from the old single-process model to the contemporary
dual-process model which includes both comparison and integration, as suggested
by Wisniewski & Bassok (1997). The study of thematic relations previously
understood to be present only in children’s conceptual development, is now
thought to also be present in adults’ concepts (Lin & Murphy, 2011). The
interdependence of taxonomic and thematic relations is then tied in with
Funnell’s Levels of Meaning model. Individual differences in tendency to use
taxonomic or thematic are evident, supported by Mirmna & Graziano’s (2011)
cross-task evidence. These recent evidences are critical in the study of mental
representation, because categorization is fundamental to how we use our
knowledge – to identify, predict and communicate. Thus, deepening the understanding
of how our mind organizes information will benefit a wide range of studies in
cognitive research and beyond.