Specificity effect in concrete/abstract semantic categorization task

Current research on concrete and abstract concepts is evolving in the direction of unpacking the widely used notion of conceptual concreteness (Bolognesi and Steen 2018; Langland-Hassan et al. 2021; Borghi 2022; Löhr 2023; Reilly et al. 2025). An early clarification of the underspecified notion of concreteness was introduced by Connell and Lynott, through the differentiation of different perceptibility dimensions (Connell and Lynott 2012). A concept that scores high in concreteness often denotes a referent that can be perceived through the five senses (vision, touch, hearing, olfaction, and taste), such as banana or book. Conversely, concepts that score low in concreteness, namely abstract concepts, typically denote something not available in a spatiotemporal dimension and something we cannot directly perceive through our senses, such as belief or bond. Both concrete and abstract concepts result from abstraction processes, in which “knowledge of a specific category has been abstracted out of the buzzing and blooming confusion of experience” (Barsalou 2005:389). Nevertheless, there is evidence that concrete concepts and abstract concepts are processed in different ways. Typically, words denoting concrete concepts are, on average, acquired earlier (Della Rosa et al. 2010) and are more easily imageable (Paivio 1990) than abstract concepts. Moreover, concrete concepts are more tightly associated with perceptual aspects of meaning and less with linguistic ones, for their acquisition and use (Borghi 2022). This is reflected in a behavioral advantage known as the concreteness effect, namely the processing advantage of concrete words over abstract words. Concrete words are recognized faster and more accurately in a variety of tasks, such as lexical decision, word naming, and recall (Kounios and Holcomb 1994; Jessen et al. 2000; Romani et al. 2008). This effect is considered a cornerstone in cognitive, psychological, and linguistic studies about word processing and it can be explained by two different theories: the dual-coding theory (DCT, Paivio 1991) and the contextual availability theory (CAT, Schwanenflugel and Schoben, 1983; Schwanenflugel et al. 1988). The DCT suggests that word meaning is represented in two interconnected systems: a verbal system for linguistic information and a non-verbal system for visual and sensory information. Concrete concepts would activate both representational systems, while abstract concepts rely more on the verbal system and can only be indirectly represented in the visual system. The DCT argues that the asymmetry in the representation of concrete and abstract concepts explains the concreteness effect. On the other hand, the CAT posits that the difference between concrete and abstract concepts lies in the degree of accessibility of meaning stored in the semantic memory, rather than in the representational code. Abstract concepts, associated with multiple contexts, require more effort to retrieve from memory, while concrete concepts are more easily retrieved because they are strongly linked to a limited number of contexts.

Numerous studies have investigated processing differences between abstract and concrete concepts at both behavioral and neural levels within these theoretical frameworks, yet the findings remain inconsistent. While the empirical evidence tends overall to support the concreteness effect, some psycholinguistic studies have reported challenges in replicating it (Barca et al. 2002; Papagno et al. 2007). In certain cases, abstract concepts have even been found to have an advantage over concrete ones. This pattern has emerged in two studies employing lexical decision tasks, in which the stimuli were controlled for imageability and contextual availability, which are the two psycholinguistic variables associated with dual-coding theory (DCT) and contextual availability theory (CAT), respectively (Kousta et al. 2011; Barber et al. 2013). These results suggest that empirical studies reporting a concreteness effect did not isolate concreteness completely, leaving other variables as potential confounds for the facilitatory effect attributed to concreteness.

The current study focuses on a variable, categorical specificity, which shows a positive but mild/medium correlation with concreteness (Bolognesi et al. 2020; Bolognesi and Caselli 2023). Categorical specificity measures category inclusiveness, or, in classic semantics and logics, the extension (denotation) of applicability of a linguistic label to referents in the world (Ostertag 2020). Categorical specificity is a relational property that establishes taxonomical relationships between concepts and somehow defines the position of a concept within a taxonomical hierarchy: vehicle is a general and concrete concept, which includes more specific concrete concepts like car and at a subordinate level electric car. Quality is a general and abstract concept, encompassing more specific abstract concepts such as beauty and elegance (for a semantic analysis see Villani et al. 2024). Most studies on conceptual taxonomy (e.g., Rosch 1978; Rosch et al. 1976) have primarily focused on concrete concepts (e.g., “animals”, “vehicles”, “objects”), overlooking the taxonomic levels of abstract concepts. Additionally, while the basic-level advantage in object categorization is a robust effect (Murphy and Smith 1982; Jolicoeur et al. 1984; Wisniewski and Murphy 1989), its underlying sensorimotor and linguistic mechanisms remain controversial, and their implications for abstract knowledge have not yet been addressed (see von Hoef et al. 2023; Wingfiel et al. 2024). Recent research has demonstrated that categorical specificity explains a fraction of decision latencies in lexical and semantic tasks (e.g., Iliev and Axelrod 2017; Bolognesi and Caselli 2023; Ravelli et al. 2024), above and beyond conceptual concreteness. In these studies, however, the direction of the effect is unclear. Iliev and Axelrod focus on a dataset of decision latencies collected for 40,481 English words (Balota et al. 2007) and use specificity scores extracted from Wordnet. They report that highly specific words are processed more slowly than highly general words. Conversely, Bolognesi et al., as well as Ravelli et al., focus respectively on the Italian and the English ANEW datasets (Bradley and Lang 1999) and find the opposite pattern: highly specific words are processed faster than highly general words. Given these inconsistent results, the jury is still out about the nature and direction of the specificity effect, and this may be due to the different ways in which specificity is operationalized, on the nature of the task, and on the kind of words used in the studies, as further described in the Theoretical Background section of the current paper.

The present paper aims at clarifying these inconsistent findings, investigating the relationship between concreteness and specificity in two distinct processing tasks: a lexical recognition task and a semantic abstract/concrete categorization task (Pexman et al. 2017). This investigation involves a controlled set of words that were carefully selected and normed, displaying substantial variations in both concreteness and specificity.

The overarching goal of this paper is to provide additional evidence demonstrating that specificity is not merely a variable synonymous with concreteness but is, in fact, a theoretically distinct variable that maintains an intriguing relationship with concreteness. To achieve this objective, participants’ reaction times (RTs) were collected, and processing latencies were analyzed. To the best of our knowledge, no prior research has examined the interaction between word concreteness and word specificity systematically balanced within the same taxonomic relation in behavioral experiments. This is crucial for understanding how categorical specificity forms a distinct yet connected dimension with concreteness in semantic memory. Methodologically, this approach may help to disentangle previous findings on conceptual processing that could be misattributed solely to the concreteness effect. The present paper presents findings from three studies: a rating study, a lexical decision task, and a semantic decision task, all utilizing words that vary in both concreteness and specificity as stimuli. Each study addresses a specific research question, hereby summarized:

According to scientific literature, abstract and concrete concepts differ across multiple dimensions. These distinctions include the quality and quantity of information they convey (Vigliocco et al. 2009), the processes involved in their acquisition (Villani et al. 2021), their dependence on social interaction (Villani et al. 2022; Fini et al. 2023; Mazzuca et al. 2025), and the type and richness of linguistic and perceptual experiences they evoke (Lynott et al. 2020). These differences may help explain why, on average, concrete concepts are acquired earlier (Schwanenflugel 2013; Borghi 2022), are more easily recalled (Romani et al. 2008), and are processed more quickly (Schwanenflugel and Shoben 1983; Schwanenflugel et al. 1988) than abstract concepts.

Recent studies, however, report reversed effects, demonstrating a processing advantage of abstract over concrete words when various psycholinguistic variables are controlled (Kousta et al. 2011; Barber et al. 2013). For instance, the concreteness effect disappears in lexical decision latencies when words are controlled for contextual availability, suggesting that contextual availability, rather than concreteness, should be considered as the factor influencing latencies (Schwanenflugel et al. 1988, 1992).

Kousta et al. (2011) and Vigliocco et al. (2014) reported a reversed effect when concrete and abstract concepts are matched on other psycholinguistic variables: abstract concepts exhibited a processing advantage. This phenomenon may be linked to the stronger affective associations typically found in abstract concepts compared to concrete ones. However, more recent findings suggest that stronger emotion-related content is actually associated with concrete rather than abstract words (Winter 2023).

In general, the concreteness effect seems to be a topic with unresolved aspects, and alternative explanations offer additional insights into this phenomenon. An intriguing approach to exploring this subject involves shifting the focus from a singular trait or variable, like concreteness, to various aspects that may collectively contribute to generating a potential facilitatory effect.

To address this issue, researchers have proposed various hypotheses and developed alternative rating systems to better understand the information underlying concreteness ratings. One prominent approach involves collecting modality-specific norms of perceptual strength, which assess the degree to which a concept is associated with each of the five basic senses: vision, touch, hearing, smell, and taste. Studies have shown that these norms more accurately reflect the perceptual characteristics of words and concepts and offer stronger predictive power for lexical decision latencies than traditional concreteness ratings. Concreteness ratings, in contrast, tend to be biased toward the visual modality and appear to encompass more than just perceptual information (Connell and Lynott 2012).

Another method for unpacking concreteness ratings involves collecting scores on new variables and assessing their ability to explain the variance in concreteness ratings. For example, Troche et al. (2017) argue that various cognitive and perceptual dimensions (e.g., emotion, time, space, color, size, visual form) contribute to forming a conceptual topography that accounts for variability in concreteness. Davis and Yee (2023) demonstrate that approximately 50% of the variance in concreteness ratings is explained by the time and physical space required to perceive a concept. Moreover, Langland-Hassan and Davis (2023) and Langland-Hassan et al. (2021) introduced the concept of Trial Concreteness as an alternative, context-sensitive measure of conceptual abstractness. This approach moves beyond traditional, fixed classifications of abstract and concrete words measured in isolation, by focusing on the degree of concreteness required within specific tasks. In their paradigm, participants complete non-verbal semantic matching trials in which they are asked to identify the item that is most semantically related to a target image choosing from several alternatives. The difficulty of each trial, and thus its level of Trial Concreteness, is determined by two main factors: the visual similarity between the target and the correct choice, and how frequently those items are encountered together in everyday contexts. The approach builds on the notions of low-dimensionality and high-dimensionality (Lupyan & Mirman 2013). A category is low-dimensional if it gathers exemplars that share few salient features, while it is high-dimensional if it gathers exemplars that share many salient features. Similarly, a trial has low Concreteness when “the target and correct choice items were (on average) judged neither to be found together commonly nor to be visually similar, relative to one or more of the other available choices for the trial” (Langland Hassan et al., 2021:8), while a trial has a high Concreteness if the association between the target item and the correct item are judged to be highly visually similar and commonly associable. In the context provided by competitive items, if target and correct choice items share few salient features, it is required more effort to associate them by virtue of semantic information. Conversely, if the target item and the correct item share many salient characteristics, the effort to abstract from the competitive distractors to establish the correct semantic association is smaller. The notion of low/high dimensionality emphasizes the process through which the identification of a concept requires an abstraction from exemplars or instantiation provided.

To the best of our knowledge, three studies (Iliev and Axelrod 2017; Bolognesi and Caselli 2023; Ravelli et al. 2024) have analyzed decision latencies in relation to both specificity and concreteness, finding significant effects. The authors of these studies claimed the theoretical disjunction between these two variables.

Iliev and Axelrod (2017) present a framework in which the degree of specificity within a taxonomic category reflects its informational precision. In this view, more specific categories carry more informational content. Drawing on concreteness ratings from Brysbaert et al. (2014), the authors introduce two measures of lexical precision based on WordNet (Miller 1995): semantic inclusiveness (P-inclusiveness) and semantic depth (P-depth). Semantic inclusiveness captures a word’s generality by measuring how many subordinate nodes it has in a taxonomic tree, with a larger number indicating greater generality and lower precision. Semantic depth, on the other hand, measures how far a word is from the top-level node in the taxonomy, with greater depth suggesting higher specificity.

Using these measures, Iliev and Axelrod examined how concreteness and precision predict word recognition latencies from the Balota et al. (2007) dataset, controlling for word length and frequency. Their results show that concreteness and precision exert opposite effects: more concrete words are processed more quickly, while more precise (i.e., specific) words lead to slower processing times. They interpret this pattern by suggesting that concrete words are easier to identify because they are more readily defined. In contrast, although precise words have well-defined meanings, their lower generality demands more cognitive effort to process.

However, the authors caution that their precision measures are derived from WordNet, a resource built on extensive expert knowledge, which may not reflect the lexical organization or semantic distinctions accessible to average speakers. In contrast, concreteness ratings are based on direct speaker judgments, providing a more realistic approximation of everyday conceptual understanding.

Bolognesi and Caselli (2023) propose a different approach, stressing the importance and the need for human-generated judgement of specificity, to avoid potential biases that may have impacted Iliev and Axelrod’s findings. The authors collected specificity ratings adopting a Best–Worst Scaling method (BWS, Louviere et al. 2015), which considers this variable’s relational nature. Specificity judgments using BWS have been collected for the Italian words contained in ANEW dataset (Montefinese et al. 2014) and used in statistical models to explain the variance in reaction times, available for the same set of words, in a LDT (Vergallito et al. 2020). The authors report that specificity explains a significant fraction of the variance in decision latencies, with a positive effect: the more specific a word is, the more time is required to be processed in LDT, in line with Iliev and Axelrod’s previous findings.

Finally, in a more recent study, Ravelli et al. (2024) collected specificity ratings for the English version of the ANEW dataset, using the same methods adopted by Bolognesi and Caselli (2023), namely the BWS method. Due to the higher availability of datasets of decision latencies in English compared to Italian, Ravelli and colleagues ran various analyses to investigate the role of specificity in lexical and semantic access based on different datasets of English decision latencies. Overall, the authors found that specificity explains a higher portion of variance in responding to semantic decision tasks (using the dataset by Pexman et al. 2017) compared to lexical decision tasks (datasets by Balota et al. 2007; Keuleers et al. 2012; Hutchison et al. 2013; Tucker et al. 2019; Mandera et al. 2020). Interestingly, in semantic decisions, specificity shows a negative effect (more specific words are processed faster than more general ones), while in lexical decisions, specificity does not always play a significant role (nor does concreteness), but when it does, specificity slows down reaction times, in line with previous findings. The differing patterns observed across task types, specifically, lexical decision tasks (LDT) and semantic decision tasks (SDT), suggest that these tasks engage distinct cognitive processes. This highlights the need for further investigation into the nature and behavior of the specificity effect. Additionally, the ANEW dataset used by Ravelli et al. (2024) consists of words selected for their high affective content, which may limit the generalizability of the findings. A more systematic examination of specificity effects on word processing is therefore warranted, using stimuli that are carefully controlled and span a range of conceptual categories across multiple taxonomic levels.

The current study builds on these previous findings and arguments. Since effects and results of Lexical Decision Task and Semantic Decision Task seem to diverge, suggesting that different aspects of language are recognized and gathered with these kinds of tasks (van Hees et al. 2016; Pexman et al. 2017), we decided to collect data both of lexical and semantic access to meaning in words that vary in concreteness and specificity. We based our analyses on a set of stimuli carefully constructed to include words systematically high and low in specificity and high and low in concreteness.

This setup allows us to carefully examine the distinct but integrated aspects involved in the conceptual abstraction process: the extent to which the referent designated by a linguistic term can be perceived through sensory experience (concreteness), and the degree of extension or precision of a conceptual category (specificity) (see Reilly et al. 2025). When speakers engage in lexical or semantic decision tasks involving relatively more abstract or concrete concepts, their performance may be influenced by the cognitive effort required to abstract a referent from a single or multiple category exemplars (e.g., Nosofsky 1986; Daelemans and van den Bosch 2005; Chandler 2017). Consequently, the concreteness effect may interact with word specificity, resulting in a processing advantage for concepts that are both highly concrete and specific, as well as for highly abstract but specific concepts. Conversely, there may be a disadvantage for concepts low in both concreteness and specificity. If there were no interaction between concreteness and specificity, a similar pattern of results would be expected, but this would suggest an additive effect of the two variables on conceptual processing, further reinforcing the idea that concreteness and specificity are independent factors in the abstraction process (Bolognesi et al. 2020).

Study 1 consists of a rating task in which participants evaluate the imageability, concreteness, and specificity of the stimuli. This step serves to verify that the stimuli, as designed by the researchers, are perceived by average speakers as differing along these three dimensions. Study 2 involves a lexical decision task (LDT) conducted using the same set of stimuli. Study 3 is a semantic decision task (SDT) in which participants are asked to judge whether each word refers to a concrete or an abstract entity. This type of semantic decision task is grounded in the theoretical framework outlined by the Calgary Semantic Decision Project (CSDP; Pexman et al. 2017). According to this framework, LDT and SDT afford different processing mechanisms because word processing unfolds through an interactive and cascading sequence of processes, as argued in earlier work (Balota 1990; Yapet al. 2015). Specifically, orthographic input is initially processed at the letter level, followed by processing at the lexical level, which involves structural and formal properties of the word, such as morphosyntactic features and frequency. Semantic processing constitutes an additional level, where conceptual meaning is activated. Information flows bidirectionally between these levels (interactive activation) and does so continuously, without requiring the completion of one level before the next is engaged (cascading activation). Both lexical and semantic information can influence performance in word processing tasks, although their relative contributions may vary depending on task-specific demands.

Data analysis and visualization of all the studies were carried out with R (RCore Team 2023) and RStudio (v. 4.2.3). Ratings scores and Reaction Times (RTs) were modeled with linear mixed models using lmer() function form “lme4” R’s package (Bates et al. 2015). Post-hoc contrasts on rating scores were carried out with “emmeans” R’s package (Lenth 2021) using Tukey’s adjustment for multiple comparisons.

All the experiments reported here were conducted with the Ethical approval of the Ethics Committee of the University of Bologna (Protocol n. 0121435). All participants received guidelines and instructions as well as informed consent. All participants agreed to participate in the experiments voluntarily.

Data and analyses scripts of the three studies are publicly available in the following Open Science Framework repository: https://osf.io/j2smv/

In the present study, we investigated the role of word specificity in language processing and its impact on the widely recognized but still debated phenomenon known as “concreteness effect”, i.e., the processing advantage of concrete over abstract concepts. As anticipated in the introduction of this work, categorical specificity is a variable often neglected and sometimes confused with concreteness. Recent studies have provided evidence of a positive, though mild, correlation between these two variables (Bolognesi and Caselli 2023). Both concrete and abstract concepts can define highly general and inclusive conceptual categories (e.g., animal, feeling) or highly specific and constrained conceptual categories (e.g., pitbull, enthusiasm). However, since existing studies working on differences between concrete and abstract concepts did not control stimuli for specificity, it remains an open question whether the level of specificity may contribute to the concreteness effect, introducing potential confounding evidence reported in the literature. Consistent with this, recent studies reviewed in the present article have shown that word specificity accounts for a significant portion of variance in response latencies during word processing. Nevertheless, those studies reported different directions of such effect, with specific words being processed either faster or slower than general ones. This inconsistency may stem from the diverse criteria used in selecting linguistic stimuli (from WordNet or human-generated data), and the nature of the task performed.

The main aim of our work is to shed light on the direction of specificity effect during language processing and its role in explaining “concreteness effect”. To achieve this goal, we conducted three studies: a norming study and two decision tasks. Specifically, we performed a lexical decision and a semantic categorization task, using a controlled set of words varying in concreteness and specificity (i.e., abstract-specific, abstract-general, concrete-specific, concrete-general) and balanced for word length and frequency. Reaction times for each task were modeled as function of specificity and concreteness ratings elicited by speakers during a norming study on our stimuli.

Our findings support and extend previous research on word specificity and word concreteness. The results of Study 1 confirm that when speakers were asked to judge word specificity both abstract and concrete concepts were characterized by different levels of specificity even though the differences in specificity are less pronounced for abstract than for concrete concepts, in line with what is argued by Borghi (2022). Moreover, norming data show the asymmetry between the knowledge derived by human speakers and that extracted by a dictionary-based resource, such as WordNet. While the latter can be a valuable tool to capture hierarchical taxonomic relations, it does not accurately reflect speakers’ lexical knowledge and organization of the mental lexicon. For this reason, we used specificity and concreteness measures operationalized through human judgments to investigate differences between concepts in lexical and semantic decision tasks.

Results of reaction times analyses differ between the two reported studies. In Study 2, we found that both concreteness and specificity had no effect in explaining RTs during lexical decision. This result can be explained by the relatively shallow process involved in lexical processing task, where the semantic information is less influential than lexical-level elements (Barsalou 2008; Yap et al. 2011), especially when stimuli are balanced for relevant linguistic dimensions, such as word length and word frequency. Similarly, Yap et al. (2011) demonstrated that if controlled for lexical variables, the “semantic variables explained only 2% of additional variance in lexical decision latencies” (Pexman et al. 2017). Nevertheless, a visual and qualitative inspection of the influence of word’s concreteness and specificity on predicted RTs (Fig. 1) suggest that concepts with lower specificity, i.e., general ones, require less effort to be recognized, regardless of their concreteness level. This tendency is in line with previous (Iliev and Axelrod 2017; Bolognesi and Caselli 2023) and current (Ravelli et al. 2024) research showing a positive effect of specificity over response latencies in LDT, where more general words are associated with faster response than specific words. Moreover, our data suggests that the processing disadvantage is more marked for concepts that score high in specificity and low in concreteness exhibit the slowest recognition times (i.e., abstract-specific). The present results are intriguing and merit further investigation. For instance, future studies could replicate the same paradigm with longer word presentation times to ensure a less superficial judgment.

In Study 3, we found a significant main effect of concreteness (i.e., concrete words were recognized faster than abstract ones) and a significant main effect of specificity (specific words were recognized faster than general ones) during abstract/concrete semantic categorization task. Notably, the interaction between concreteness and specificity was not significant. This demonstrates that categorical specificity has an independent significant contribution in predicting responses times in a task implying a deeper semantic processing level. These findings align with Ravelli and colleagues’ study (2024), in which the authors compared SDT’s chronometric data (Pexman et al. 2017) with specificity scores (Bolognesi and Caselli 2023) and found a negative effect of specificity: more specific words are processed faster than more general ones. Similarly, in the present semantic decision task we found that specificity and concreteness have a similar negative effect over RTs: an increase in responses times occurred during processing of words with low specificity scores and low concreteness scores. However, regardless of concreteness level, highly specific words required less effort to be recognized as concrete or abstract respectively, compared to general ones. One plausible interpretation for our findings could be that the categorization of concrete and abstract words may be affected by the cognitive effort needed to abstract a referent from a single or multiple exemplars instantiated. This hypothesis is in line with an exemplar-based theory of conceptual category representations (e.g., Nosofsky 1986; Daelemans and van den Bosch 2005; Chandler 2017), according to which speakers engaging in a semantic decision task or other kinds of semantics-based cognitive decisions (e.g., typicality judgement and exemplar production tasks, see also Heit and Barsalou 1996) are likely to mentally instantiate exemplars or referents. When confronted with a word that is quite general (e.g., substance) speakers might mentally instantiate more diverse exemplars of the category, before deciding whether it denotes a concrete or abstract entity, leading to longer processing times. Conversely, when presented with a more specific word (e.g., Aspirin) they may effortlessly instantiate a single exemplar, resulting in faster processing. This is also consistent with seminal findings conducted on concrete conceptual categories, by Rosch et al. (1976) and Rosch (1978), showing that superordinate categories require more effort to be processed and conceived compared to concept at other taxonomic levels. Interestingly, in our study this effect seems to emerge also among abstract concepts. Traditionally, specific concepts are characterized by high informativeness and low distinctiveness, whereas general concepts tend to be less informative and more distinctive (Murphy 2004). Therefore, specific concepts could activate narrow and dense representations, associated to the instantiation of a small number of precise exemplars and converging on more salient and accessible semantic information. This “high-resolution” semantic content may support categorization in semantic decision tasks by facilitating the retrieval of task-relevant information (e.g., evaluating whether a word is abstract or concrete). Overall, our studies provided pivotal insights into the distinction between concreteness and categorical specificity. Firstly, in line with Bolognesi and colleagues (2020), we confirmed the weak correlation between the two variables (r = 0.29). Secondly, we established their non-collinearity. Thirdly, we found their additive effect over decision latencies in a behavioral experiment, such as the semantic decision task. Lastly, although the results from the lexical decision task did not reach statistical significance, the pattern observed across the two categorization tasks revealed a consistent effect of concreteness, with more concrete concepts requiring less processing time. In contrast, the influence of specificity varied by task. When included in the model, specificity demonstrated opposite effects on reaction times across the two tasks. In the semantic decision task, words with higher specificity were processed more quickly than those with lower specificity. However, in the lexical decision task, the reverse pattern emerged. Taken together, these findings provide further support for the view that concreteness and specificity represent two distinct semantic dimensions, each contributing differently to word processing depending on task demands.

Moreover, the differences observed across categorization tasks support the possibility that the influence of semantic variables may vary depending on the nature of the task (Goh et al. 2016; see also Pexman 2012; Pexman et al. 2017). For instance, studies on numbers of senses (Borowsky and Masson 1996) showed dissociation effect on RTs in LDT and SDT, as pointed out by Yap et al. (2011, p. 724): “words with more senses elicited shorter lexical decision times but less accurate semantic classification responses”. Given that general words tend to have more possible senses and are applicable across a wider range of contexts compared to specific words, it is reasonable to assume that their processing may benefit from broader semantic activation. This could result in faster responses during lexical decision tasks, where deep access to a word’s meaning is not required. In contrast, semantic categorization tasks demand a deeper understanding of word meaning to make accurate classifications. In this context, the greater number of senses and the activation of multiple competing exemplars associated with general words may hinder performance. As a result, general words show a significant disadvantage compared to specific words in terms of semantic decision latencies. Specific words, by constraining the range of possible meanings, offer information that is sufficiently precise and focused to facilitate faster and more accurate semantic classification.

In contrast to the perspective proposed by Iliev and Axelrod (2017), we suggest that informativeness at the subordinate level may actually enhance performance in semantic decision tasks, as it provides more direct access to relevant conceptual information.

Furthermore, the typically low dimensionality of general concepts, as described by Lupyan and Mirman (2013), may help explain the increased cognitive effort required for their processing and the resulting delay in determining whether a concept is concrete or abstract in semantic decision tasks. General concepts often represent broader, more inclusive categories with fewer salient shared features, which may make them more difficult to evaluate quickly. Because low dimensionality is a property commonly associated with abstract concepts, the observed differences in response latencies between highly abstract but general concepts and highly abstract but specific concepts may reflect a more precise and constrained representation for the latter.

Recent findings by Rambelli and Bolognesi (2023) support this interpretation, showing that abstract-specific concepts tend to exhibit low contextual variability (Hoffman 2016)—that is, they are associated with a relatively limited range of accessible contexts. According to Contextual Availability Theory (Schwanenflugel et al. 1988), the ease with which relevant contexts can be retrieved from memory is linked to faster word processing. Therefore, integrating our results with previous research, it is reasonable to propose that the quantity and specificity of information associated with a concept—whether in terms of exemplars, meanings, features, or contexts—can result in either a processing advantage or disadvantage, depending on the task. Further research is needed to clarify the specific contribution of these informational dimensions to semantic processing.

This study has some limitations, including the absence of basic-level concepts and the use of a semantic categorization task focused specifically on the abstract–concrete distinction. However, our primary objective was to conduct a targeted comparison of words positioned at the extremes of the specificity continuum, from highly general to highly specific—an area that remains underexplored in current research. To ensure comparability with existing findings, we employed a well-established task in the study of abstract and concrete concepts (Pexman et al. 2017). Although this task emphasizes judgments of concreteness, it reliably engages in deep semantic processing. In contrast, collecting reaction time data using alternative categorization tasks, such as animacy judgments or direct specificity ratings, would have been considerably more complex due to the diversity of semantic domains in our stimuli and the inherently relational nature of specificity, which depends on comparisons between concepts.

Despite these constraints, the observed specificity effect, along with the improved model fit when specificity was included as a predictor of reaction times, highlights the importance of incorporating word specificity into research on the abstract–concrete distinction.

Furthermore, the ongoing debate on the concreteness effect could benefit from greater attention to taxonomic levels of abstraction when selecting stimuli for behavioral tasks. The identification of a specificity effect, particularly within the context of a semantic categorization task, suggests that some of the inconsistencies observed in previous studies on the concreteness effect may be due to the omission of specificity as a contributing factor to the informativeness of semantic representations. We recommend that future research on the processing of abstract and concrete concepts take specificity into account, whether through the design of new experiments or the reanalysis of existing data.

Specificity is a crucial factor in semantic access and should not be overlooked or confused with concreteness. The results of the present study provide further evidence that specificity should be considered a key variable within the framework of semantic richness. It appears to activate semantic memory automatically and to influence language processing by accounting for independent variance in decision latencies.

The present study demonstrated that concreteness and categorical specificity are distinct dimensions of the abstraction process. In light of the mixed findings surrounding the concreteness effect, our work introduces evidence of a specificity effect in a semantic categorization task: specific concepts are processed more quickly than general ones. Both abstract and concrete specific concepts appear to involve more refined conceptual representations, supported by salient and focused information. In contrast, general concepts are represented through a broader set of features, situations, and category members, which introduces greater competition and increases the cognitive effort required for abstraction, ultimately leading to longer processing times.

These findings call for further investigation into the nature of the information that supports specificity and into the extent to which this specification enhances conceptual representation.