Professional Documents
Culture Documents
Abstract
The standard approach to analogical reasoning posits that the
mechanism that people employ to ensure the soundness of
analogical inferences consists in copying unmapped individual
explicit base relations, substituting corresponding source entities
with target entities, and generating slots for base entities that
were unmapped. Alternatively, we contend that when the gist of
the information to be transferred is better captured by relational
categories than by explicit individual relations, people resort to
searching for target exemplars of the base relational categories,
disregarding similarity between relations. Experiment 1 revealed
that for this kind of analogy, inferences that did not resemble
the base analog in terms of explicit individual relations but were
built on exemplars of the base relational category were judged
as sounder than inferences that matched the base analog in terms
of relations but not in terms of a common category. Within the
framework of the proposed approach, we postulated that inference
evaluation also depends on the similarity between the base and
target exemplars on relevant aspects. Experiment 2 revealed that
inferences were judged as sounder when the exemplars upon
which the inferences were built matched the base exemplars
along salient dimensions of the relational category they shared.
1. Introduction
Base analog
ADMIRE (George, car, Peter)
LEND (Peter, car, George)
CAUSE [ADMIRE (George, car, Peter), LEND (Peter, car, George)]
TALL (George)
LIVE (George, Buenos Aires)
Target analog
ADMIRE (Bruno, shoes, Marcos)
LEND (Marcos, shoes, Bruno)
CAUSE [ADMIRE (Bruno, shoes, Marcos), LEND (Marcos, shoes, Bruno)]
TALL (Bruno)
LIVE (Bruno, New York)
404 Ricardo A. Minervino, Adrián Margni, Lucía Micaela Tavernini and Máximo Trench
which they belong. The pairings established during mapping will serve as
an input for the formulation of coherent inferences.
There is a broad consensus about the basic mechanism that people try to
follow to ensure the soundness of analogical inferences. This mechanism
involves constructing target equivalents for source propositions that did
not find counterparts in the target, but that are part of the base structure
that was partially matched with the target. To formulate these inferences,
the cognitive system applies some variant of a copy with substitution and
generation mechanism (CWSG; e.g., Falkenhainer et al., 1989; Hofstadter
& Mitchell, 1994; Holyoak et al., 1994; Hummel & Holyoak, 2003; Keane
et al., 1994; Kokinov, 1994), which consists of copying base relations (or
replacing them by similar ones), substituting base entities for target entities
as dictated by the mapping, and generating slots for unmapped base entities.
The level of soundness of an inference will depend on how large and deep
the shared system that supports it is, as well as on how similar the relations
matched during mapping are.
Suppose that the base analog now includes the information that in order
to reciprocate Peter’s generosity, George washed Peter’s car. In such a case,
the system would generate the template WASH [or a similar relation] (Bruno,
shoes, Marcos). The selection of the specific action to be postulated in the
target domain will depend on the specificities of the target. In this case,
since the target is about shoes, wash could be replaced by shine.
While in the previous example entity substitutions are entirely dictated
by the mapping process, in other occasions the analogizer must identify
which entities from the target domain are capable of instantiating the slots
that were generated out of entities that did not find correspondences during
mapping. Suppose that the base analog now includes the information
that the car wash was carried out employing a hose. Given that no
correspondence was established for hose during mapping, the template
generated by CWSG for this situation would be WASH [or a similar
406 Ricardo A. Minervino, Adrián Margni, Lucía Micaela Tavernini and Máximo Trench
1
Since the 1980s, Cognitive Science research has been interested in determining
to what concept of analogy people adhere, something that is evidenced, for
example, by the type of inferences people make when asked to suppose that a
current situation is analogous to a previous event. It is assumed that, faced with
the request to make an analogy, the cognitive system responds by shifting into an
analogical comparison mode. What we seek to establish is whether analogy making
of this kind focuses on properties of objects, relationships, relational categories,
etc. It is not an objective of the present project to establish the logical foundation
of this type of inferences or their validity. The purpose is to determine what people
understand by analogy and what operations they activate when they are asked to
make an analogy.
Category Membership as a Criterion to Evaluate the Soundness of Analogical Inferences 407
Base analog: John loves Mary and Mary likes roses. This made John send
roses to Mary.
Target analog: Peter loves Susan and Susan likes red peppers.
The mechanism of CWSG would yield the inference that Peter sent (or a
similar relation) red peppers to Susan. As the system inherits the roses-red
peppers matching, it is only able to make adjustments on send, looking for
similar relations that are appropriate for red peppers (e.g., Peter could take
the red peppers to Susan’s instead of sending them). Via this mechanism,
the system is incapable of instantiating the template in a way that results in
a sound inference. What differentiates these analogies from those discussed
in the previous section is that unmapped relations (e.g., send) do not capture
the nucleus of the information to be transferred from base to target. Rather,
this gist is captured by a more global schema (e.g., expression of love) that
gets activated in response to the combination of send with roses. Let us now
consider a case where the mapping does not provide the replacement entity,
and therefore the object role has to be filled in the target domain:
Base analog: John loves Mary. This made John send flowers to Mary.
Target analog: Peter loves Susan.
is not just dealing with a sending relation, but rather with a sending action
that constitutes a component of an expression of love. Thus, the system
must provide a subtler characterization of the object role that flowers were
filling in the base situation. While features such as those described can
capture the semantic content of the object role in a sending relation, in this
last example it would be necessary to invite other (or different) features,
such as surprising, romantic or original: while perfume or some mariachis
could be adequate fillers, a bottle of water will not.
If on the one hand CWSG seems too liberal (e.g., it is not capable of
imposing sufficient restrictions to the entities filling the role of object), on
other occasions it can be unnecessarily restrictive. Many combinations of
non-similar verbs and non-similar entities could result in manifestations
of love, such as spending quality time together, giving a back rub or
saying words of caring or appreciation. In principle, the system could
break free from the restrictions of repeating the base relation or replacing
it with a similar one, or of conforming to the formal structure of the base
proposition. Going back to our red peppers example, our creative mind
could take the pairing between flowers and red peppers and, taking into
account the facts that Susan likes red peppers and Peter wants to express
his love to Susan, it could postulate that Peter might prepare a surprise meal
for Susan employing red peppers. In this expression of love, neither send
nor a similar relation are involved (the action of preparing is used instead),
flowers are replaced by meal (leaving aside the pairing flowers-peppers) and
red peppers become a complement of prepare, a role that was not present
in the original unmapped fact. What really matters is that the inference
constitutes an instance of expression of love.
The concept of expression of love constitutes an example of a type of
relational category known as schema-governed categories (SGC). A SGC
describes the semantic structure of a type of event, in terms of a network
that links the typical components of those events (Gentner & Kurtz, 2005;
410 Ricardo A. Minervino, Adrián Margni, Lucía Micaela Tavernini and Máximo Trench
Goldwater et al., 2011; Markman & Stilwell, 2001). Instead of sharing a set
of probabilistic features and feature correlations, members of SGCs (e.g.,
reparation, betrayal, assassination or help) share a structure that can be
instantiated by exemplars that can be very different in the relations and
entities that they include, and also in their formal structure.
Our proposal aligns with prior research exploring the influence of
semantic dependencies between entities and relations when assessing
analogical relatedness (e.g., Bassok & Medin, 1997). In their study on
factors determining similarity judgments, Bassok and Medin discovered
that explicit relations and entities in the base and target could be
overshadowed by abstract interpretations resulting from meaningful
combinations of relations and entities. For instance, a scenario where a
carpenter fixed a chair was considered more similar to an electrician fixing
a radio than to a carpenter fixing a radio. This judgment was based on the
abstract interpretation of “a professional performing a job-related activity”
in the first pair of facts. Bassok and Medin argued that similarity between
pairable elements no longer matters when relations and entities combine
to form more abstract meanings. Extending this idea, we can envision
that a target scenario where a carpenter installs a door frame might be
considered similar to a carpenter fixing a chair, despite the dissimilar verbs.
By examining sentence pairs that involved dissimilar verbs but the same
relational category, Minervino et al. (2008, 2013) demonstrated that events
can be seen seen as analogous despite lacking local-level similarity.
Transitioning from analogical relatedness to re-representation, evidence
suggests that this process operates both at the level of single explicit
relations and on a broader category scale. At a local scale, “John pushed the
box of books” might be deemed analogous to “Mary guided the wheelchair”
if push and guide are re-represented as instances of “causing directed
change to an object’s position” (Gentner & Kurtz, 2006; Silliman & Kurtz,
2019). Alternatively, propositions can undergo re-representation even when
Category Membership as a Criterion to Evaluate the Soundness of Analogical Inferences 411
analyzing their single explicit relations fails to unveil a hidden identity. For
instance, a dubious case of romantic courtship like “Sammy played a joke
on a girl in his class” could be categorized as an instance of such category
through analogical comparison with a more typical exemplar like “Sammy
bought a perfume for a girl in his class” (Oberholzer et al., 2018).
The primary aim of our research was to extend this perspective to
analogical inference, exploring whether, in analogies where crucial
information is better captured by SGCs than by explicit base relations,
people prefer the proposed category mechanism to ensure the soundness of
the generated inferences.
As t he mere use of one or the other mechanism cannot be
straightforwardly taken to prove that people regard it as the preferred
criterion, we thought that the best way to determine which mechanism
represents the favorite strategy for ensuring the soundness of analogical
inferences would consist in having participants choose between inferences
derived by CWSG and inferences produced by the category mechanism
here proposed (for the mechanisms actually employed by people when
generating analogical inferences see, Minervino et al., 2023).
As the kind of analogical inference in which the present paper
concentrates involves searching for new exemplars of the base SGCs in the
target domain, it seems reasonable to hypothesize that the soundness of the
inferences will somewhat depend on the particular exemplar upon which
the inference was built. To maximize soundness, it is likely that people tend
to select new exemplars of the base SGC that resemble the base case in
many relevant aspects.
The values of exemplars along relevant dimensions of their overarching
SGC have proven relevant during other subprocesses of analogical thinking.
For example, Tavernini et al. (2017) presented participant with triplets of
images displaying a base situation plus two target situations. For example,
in the base of one of the sets a man was passing a computer mouse to a
412 Ricardo A. Minervino, Adrián Margni, Lucía Micaela Tavernini and Máximo Trench
woman. While in one of the targets a man was passing a laptop to a woman
(an object bearing taxonomic similarity with the mouse), in the other target
a man was passing a pair of socks to a woman (an object lacking taxonomic
similarity with the mouse). After being told that the situations were cases
of a particular relation (i.e., “These are three cases of passing”) or cases
of a SGC (i.e., “These are three cases of awarding”), participants had to
rate the analogical relatedness between the base and each of the targets.
Participants’ ratings revealed that general object similarities (membership
to same category) positively affected the evaluations of analogical quality,
but only when the compared situations were framed by their shared
relation. When framed by SGCs, other kinds of similarities between objects
seemed to count, determining opposite elections. More specifically, the
analogical relatedness was affected by the closeness of the exemplars in
critical dimensions of the SGC to which they belonged (in this example the
importance of awarding), something that, in this example, was determined
by the value of the entities that fulfilled the role of prize. In this way, under
the SGC condition, people judged the situation of the pair of socks to be
more analogous to the base than that of the laptop.
Category dimensions also seem to play a role during analogical retrieval.
Olguín et al. (2017) assessed whether the closeness of the analogs on their
dimensional value along the SGC to which they belong plays a role during
retrieval. Target situations consisted in “me too” analogies (see Hofstadter
& Sander, 2013) where a fictional character tells a story and the participant
is asked to report an analogous episode. While in half of the occasions
participants received instances of SGCs displaying high values along a
relevant dimension of these SGCs, in the other half they received instances
displaying low levels along relevant dimensions of these SGCs. Base
analogs recalled in response to target analogs with high dimensional values
scored higher along said dimension than those recalled in response to target
analogs showing low values, thus demonstrating an effect of dimensional
Category Membership as a Criterion to Evaluate the Soundness of Analogical Inferences 413
2. Experiment 1
The purpose of the first experiment was to determine if, for analogies
in which the critical information to be transferred lies at the level of a
relational category instead of explicit individual relations, the inferences
representing exemplars of the base SGC are judged as being sounder than
those that maintain similarities at the level of explicit individual relations
with the base information. Participants received base analogs comprising
a base cause and its effect, followed by a target analog that consisted in a
situation in which a very an action that was very similar to the base cause
was carried out by another individual or group. Participants were then
provided with two possible target effects, having to choose the one that
followed more reasonably from the analogy. One of such consequences
was an exemplar of the SGC to which the base effect belonged, but did
not involve an explicit individual relation that resembled its counterpart in
the base effect. Conversely, the other consequence was not an exemplar of
the relational category to which the base effect belonged, but its explicit
individual relation was similar to that of the base effect.
In order to assess whether an eventual preference of participants towards
one of the two competing alternatives was due to the analogical comparison
and not to its higher intrinsic plausibility as a consequence of the target
cause, participants’ choices were compared to those of a second group of
participants who were asked to choose the most reasonable consequence of
the target cause, but without having previously received any base analog.
414 Ricardo A. Minervino, Adrián Margni, Lucía Micaela Tavernini and Máximo Trench
2.1 Method
Participants
Fifty students at the University of Comahue (M = 22.42 years, SD = 2.46)
volunteered to participate in the experiment. After signing an informed
consent, they were randomly assigned in equal number to the analogy or
the target-only condition.
Materials
Ten sets of materials were employed, each one comprising a base analog
and a target cause that was similar to the base cause (see Table 1). Our
intention was to show that when the critical information to be transferred
does not reside in the explicit individual relation of the base effect, people
prefer analogical inferences that are created by selecting further exemplars
of the SGC to which the base effect belongs. Therefore, we intentionally
selected facts whose explicit individual relations were poorly informative,
but which clearly belonged to SGCs. For example, in the base situation “A
student questioned the theory to his professor, with the consequence that
the professor raised the student’s grade”, it would be extremely difficult to
guess what may have happened had the sentence ended with “raise” (had he
raised his hand? or his voice? or the quality of his arguments?). However,
this action becomes perfectly comprehensible as an instance of reward once
we know that the professor raised the student’s grade. While base analogs
comprised a cause and its effect, target analogs consisted of a further
situation in which the action referred to in the base cause was carried out by
another individual or group (e.g., “Another student questioned the theory
to his professor”). The target causes were coupled with two alternative
target effects which were formally equivalent to their corresponding base
effects. In the “similar relation” target effect, the explicit relations were
semantically similar to those of the base effect, but were chosen in such
a way that the base effect and the target effect could not be encompassed
Category Membership as a Criterion to Evaluate the Soundness of Analogical Inferences 415
by the same relational category. Following with the example in which the
base effect was “The professor raised the student’s grade” (i.e., a case of the
SGC reward), the similar relation target effect stated that “The professor
increased the suspension to the student”. Conversely, in the “same SGC”
target effect, the verbs were not semantically similar to those of the base
effect, but were chosen in such a way that the base and the target effects
could indeed be encompassed by the same SGC (e.g., “The professor
offered a scholarship to the student”, another exemplar of the SGC reward).
To prevent participants in the analogical condition from inducing an
association between the similarity of the base and target relations and the
absence of an encompassing relational category, ten filler sets were built.
Their structure was similar to that of the critical sets, except for the fact that
the alternative consequences for the target cause either maintained similar
relations as well as a shared relational category with the base effect, or did
not maintain any of these characteristics.
In order to gather an independent measure of the degree of similarity
between base and target verbs, we asked an independent group of 30
students taken from the same population to rate the similarity between
the verb of the base effects and the corresponding verbs of the two
alternative target effects, using a 5-point Likert scale ranging from 1 to
5 (1 = non similar, 2 = hardly similar, 3 = moderately similar, 4 = very
similar, 5 = extremely similar). The order of presentation of the pairs was
counterbalanced, and in all of the cases the pairs extracted from the same
set of materials were separated from each other by at least three interleaving
pairs taken from different sets. The verbs included in the effects with
similar relations were judged to be more similar to the base verbs than the
verbs included in the target effects sharing the SGC with the base, Mdn
= 4.3 (Range = 0.6) vs. Mdn = 1.5 (Range = 0.5), Z = -4.798, p < .001. To
determine the degree of representativeness of the base and target effects as
exemplars of the critical SGCs, another group of 30 students who belonged
416 Ricardo A. Minervino, Adrián Margni, Lucía Micaela Tavernini and Máximo Trench
to the same population received the base effect of each set of materials (e.g.,
“The teacher raised the student's grade”), preceded by the corresponding
base cause (for example, “The student questioned his teacher’s theory”) and
followed by the critical SGC applicable to the base effect (e.g., reward).
Participants also received each of the inferences included in the choice task
(e.g., “The teacher increased the student's suspension” and “The teacher
offered the student a scholarship”), both preceded by the same target
cause (e.g., “A student questioned his teacher's theory”) and followed by
the critical SGC applicable to the target effect of each set (e.g., reward).
Participants had to rate the extent to which the base and target effects
pertained to the SGCs presented using a 5-point Likert scale (1 = not an
example, 2 = hardly an example, 3 = moderately an example, 4 = a very
good example, 5 = an extremely good example). The order of presentation
of the items was counterbalanced, and in all cases the items extracted from
the same set of materials were separated from each other by at least three
items from another set of materials. On the one hand, result showed that
the base effects clearly constituted instances of the critical SGC, Mdn = 4.3
(Range = 0.6). Furthermore, results confirmed that the "same SGC" target
effects were judged to be better examples of that SGC compared to the
"similar relations" target effects, Mdn = 4.3 (Range = 0.5) vs. Mdn = 1.4
(Range = 0.5), Z = -4.806, p < .001.
Procedure
All tasks and instructions were presented on computers, with participants
working in individual stations at the laboratory. Participants of the analogy
group read a brief explanation about the potential of analogical comparisons
to infer new information about a target situation. They were then told that
they would receive a first episode involving a cause and an effect, followed
by a second episode which only included a cause, their task being to
postulate an effect for this second episode based on what had happened in
the first episode. In each experimental trial (be it a critical or a filler one),
they were asked to choose which of the two alternative effects follows
418 Ricardo A. Minervino, Adrián Margni, Lucía Micaela Tavernini and Máximo Trench
more reasonably from the analogy (Figure 1 shows how the tasks appeared
to participants). They were informed that there was no time limit and that
they could not go back to previous tasks to review them. Filler and critical
materials were presented in random order, and the order in which the two
alternative target options were displayed was counterbalanced. Prior to the
actual experimental tasks, participants performed two practical trials, during
which they could ask any questions regarding how to carry out the activity.
the same as those of the analogy condition, except that the target cause
was not preceded by a base analog. Critical and filler items were presented
in random order, and the order in which the two alternative target options
were displayed was counterbalanced.
In the analogical
2.2 Results and Discussion condition, the inferences that belonged to the same
SGCInofthethe base condition,
analogical effect were chosen
the inferences that in 83%toofthethe
belonged samecases
SGC of(see Figure
the base effect 2).
This preference in the analogical condition to choose the category-based
were chosen in 83% of the cases (see Figure 2). This preference in the analogical condition to
options cannot be attributed to a higher intrinsic plausibility of such options
choose the category-based options cannot be attributed to a higher intrinsic plausibility of such
as reasonable consequences of the target causes, since participants of the
options as reasonable consequences of the target causes, since participants of the analogy condition
analogy condition (Mdn = .8, Range = 0.5) chose the category-based option
(Mdn = .8, Range = 0.5) chose the category-based option in a higher proportion of the trials than
in a higher proportion of the trials than participants in the target-only
participants in the target-only condition (Mdn = .4, Range = 0.6), U = 31, p < .001, who chose the
condition (Mdn = .4, Range = 0.6), U = 31, p < .001, who chose the similar
similar relations option in a higher proportion of the cases (54.4% vs. 45.6%; see Figure 3).
relations option in a higher proportion of the cases (54.4% vs. 45.6%; see
Figure 3).
100
90 Same SGC
Similar relation
Chosen Effects (%)
80
70
60
50
40
30
20
10
0
Analogy Group Target-only Group
WeWehave
have documented that for those analogies in which the gist of the information to be
documented that for those analogies in which the gist of the
transferred is better captured by relational categories than by explicit individual relations, people
information to be transferred is better captured by relational categories than
judge as sounder those inferences that are exemplars of the relational category that was readily
applicable to the base effect. A sensible research question concerns whether belonging to such a
420 Ricardo A. Minervino, Adrián Margni, Lucía Micaela Tavernini and Máximo Trench
3. Experiment 2
The objective of the present experiment was to test the hypothesis that
the perceived soundness of analogical inferences depends on the extent to
which the SGC exemplar on which the target inference is based matches
the base exemplar along critical dimensions of the SGC. As in Experiment
1, participants in the analogical group received base analogs comprising
a base cause and its effect, followed by a target analog that consisted
in a situation in which an action that was similar to the base cause was
carried out by another individual or group. The main difference with
Experiment 1 was that instead of having to choose between a “similar
relation” and a “same category” effect, participants had to choose between
two inferences that involved exemplars of the relational category to which
the base effect belonged, with one of them being closer to base effect
Category Membership as a Criterion to Evaluate the Soundness of Analogical Inferences 421
3.1 Method
Participants and Design
Fifty students at the University of Comahue (M = 22.06 years, SD = 3.23)
volunteered to participate in the experiment. After signing an informed
consent, they were randomly assigned in equal number to the analogical
and the target-only conditions.
.01. The experimental procedures for both conditions were identical to their
corresponding groups of Experiment 1 (see Figure 3).
(see Figure 4). This preference cannot be attributed to a higher intrinsic preferability of the matching
options as consequences of the target situations, since participants in the analogical condition chose
Mdn = 1, range = 0.5 vs. Mdn = .3, range = 0.5, U = 10.500, p < .001, where the opposite alternatives
= 10.500,
were chosen < .001,
morep often where
(66.8% vs. the opposite
33.2%; alternatives
see Figure were
4). Results thuschosen
showedmore often
that inferences
(66.8% vs. 33.2%; see Figure 4). Results thus showed that inferences based
based on exemplars that match the base case along relevant dimensions of the SGCs to which they
on exemplars that match the base case along relevant dimensions of the
belongSGCs
are perceived
to whichas they
sounder than are
belong inferences basedasonsounder
perceived exemplars
thanwhose values along
inferences basedsaid
on exemplars
dimensions whose
do not match those values along
of the base said dimensions do not match those of
analog.
the base analog.
100
90 Same VED
Chosen Effects (%)
80 Different VED
70
60
50
40
30
20
10
0
Analogy Group Target-only Group
4. General Discussion
4. General Discussion
A key
A prediction of the of
key prediction CWSG approach approach
the CWSG to analogical inference is inference
to analogical that people iswill
thatgrant
people
soundness will grant
to inferences soundness
whose relations aretosimilar
inferences whose relations
to those contained are similar
in the propositions to
from which
those contained in the propositions from which the inferences were derived.
In the present study we distinguished between analogies in which the
core of the information to be transferred resides in the explicit individual
relations of unmapped propositions and analogies in which this nucleus is
better captured by the SGCs to which these propositions belong. For this
second type of analogies, in Experiment 1 we obtained that soundness did
not depend on the similarity between base and target relations, but instead
on the fact that the target exemplar pertains to the same SGC as the base
426 Ricardo A. Minervino, Adrián Margni, Lucía Micaela Tavernini and Máximo Trench
Base analog: John loves Mary. This made John send flowers to Mary.
Target analog: Peter loves Susan.
Suppose that the analogizer interprets that the act of sending flowers is an
expression of love and that CWSG then operates on that representation. We
believe this is the more reasonable way to represent this interpretation of
the base analog:
problems that we have indicated that CWSG suffers when the nucleus of
the information does not reside in the relational predicate will reappear at
this new level of representation. Indeed, the generated inference template
would be:
Given these representations, the system could then generate the following
inference: LOVE (Peter, Susan), EXPRESSION OF LOVE (Peter, Susan).
As we understand it, the proposal that relational predicates can include
all kinds of elements (e.g., entities and properties of entities) would devoid
430 Ricardo A. Minervino, Adrián Margni, Lucía Micaela Tavernini and Máximo Trench
category love expression includes the relation send, the agent’s intention
should be to awake certain emotions in a person, the patient has to be a
candidate for being emotionally affected by the agent at stake, and the
object should be pleasant to the patient. The complex interdependency of
the constituents of a fact that makes it pertain to a SGC makes it proper
to talk about these categories as “relational”, but the sense of the term
is broader than the one employed by dominant theories of analogy (i.e.,
systems of multiplace predicates).
A second problem with the standard approach´s response to our proposal
concerns its assumption that the analogical machinery works on a single
representational level (i.e., the one that captures the abstract meaning
of the facts by applying, for example, a SGC). Such assumption entails
that the system discards as irrelevant the specific content of the base fact
from which the inference is derived (e.g., the fact that sending flowers
was the specific instance of expression of love). As we have shown in this
study, however, this second level could be necessary for selecting target
exemplars that match the base exemplar along relevant dimensions of
the shared SGC. Again, the standard approach could reply by stating that
this commonsensical fact would be naturally explained by the CWSG
approach by assuming that the cognitive system begins its operation with
ideal representations in which all the relevant aspects of the analogs were
included, so as to facilitate the production of sound inferences. Thus, if
sending flowers was judged as a romantic expression of love, the system
could provide this input to the analogical machinery:
It seems clear that the post-inference subprocesses would not know how
to assign specific content to this inference. Indeed, what the system should
know is that the base fact has been assigned to a SGC, and that it has
been qualified along a certain dimension in a particular way, so that what
becomes necessary is to search the target domain for a comparable instance
of that category. This is what the categorical perspective defended here
proposes, and what CWSG is not capable of including as a guideline in the
template it generates.
As we have seen, the categorial account of analogical inference herein
proposed seems relevant for other subprocesses of analogical thinking such
as retrieval (Olguin et al., 2017), re-representation (Oberholzer et al., 2018)
or analogy evaluation (Tavernini et al., 2017). Its central idea would be that
whenever two facts belong to a common SGC, the comparison process is
framed by the semantic structure of said category, and often follows the
criteria that govern how category exemplars are compared at large. The
study of concepts has recently witnessed a growing interest in relational
categories, long overshadowed by the study of entity or artifact categories
(Goldwater & Schalk, 2016; Popov et al., 2020). The development of these
investigations will be crucial for theories connecting the study of concepts
with the field of analogical thinking. It is possible that this disconnection
stemmed from an artificial separation between the activity of representing
analogs and that of interpreting the analogy, long criticized precisely by
authors who have sought to promote a view of analogy-making as a case of
categorization (e.g., Hofstadter & FARG, 1995; Hofstadter & Sander, 2013).
Category Membership as a Criterion to Evaluate the Soundness of Analogical Inferences 433
Funding This work was supported by Grants PICT 23522019-02542 and PICT 2019-
32682650 from the National Agency for Scientific and Technological Research of
Argentina (ANPCyT), and by Grant PIP018 from the National Council for Scientific and
Technical Research (CONICET).
Declarations
Consent to Participate All participants signed an informed consent for participation.
References
Bassok, M., & Medin, D. L. (1997). Birds of a feather flock together: Similarity
judgments with semantically-rich stimuli. Journal of Memory and
Language, 36(3), 311-336. https://doi.org/10.1006/jmla.1996.2492
Blanchette, I., & Dunbar, K. (2002). Representational change and analogy: How
analogical inferences alter target representations. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 28(4), 672–685. https://doi.
org/10.1037/0278-7393.28.4.672
Clement, C. A., & Gentner, D. (1991). Systematicity as a selection constraint
in analogical mapping. Cognitive Science, 15(1), 89-132. https://doi.
org/10.1016/0364-0213(91)80014-V
Falkenhainer, B., Forbus, K. D., & Gentner, D. (1989). The structure-mapping
engine: Algorithm and examples. Artificial Intelligence, 41(1), 1-63. https://
doi.org/10.1016/0004-3702(89)90077-5
Forbus, K. D., Ferguson, R. W., Lovett, A., & Gentner, D. (2017). Extending SME
to handle large-scale cognitive modeling. Cognitive Science, 41(5), 1152-
1201. https://doi.org/10.1111/cogs.12377
Gentner, D. (1983). Structure-mapping: A theoretical f ramework for
analogy. Cognitive Science, 7(2), 155–170. https://doi.org/10.1207/
s15516709cog0702_3
Gentner, D. (1989). The mechanisms of analogical transfer. In S. Vosniadou &
434 Ricardo A. Minervino, Adrián Margni, Lucía Micaela Tavernini and Máximo Trench