1. Introduction
The way humans mentally represent and organise categories has been a central topic in the study of human cognition (Aristotle, Reference Ackrill4th century BCE; Fodor, Reference Fodor1975; Lakoff, Reference Lakoff1987; Rosch, Reference Rosch, Rosch and Lloyd1978; Wittgenstein, Reference Wittgenstein1953). In a seminal article, Rosch (Reference Rosch1975) departed from traditional feature-based accounts (e.g. Katz & Fodor, Reference Katz and Fodor1963) to a theory that placed the psychological grounding of human experience in how categories are constructed at centre stage. What has become known as prototype theory holds that categories are structured around prototypes rather than defined by strict sets of necessary and sufficient features. Through a systematic set of experiments, Rosch (Reference Rosch1975) established that within a category (e.g. bird or furniture), some members are considered more typical or central (e.g. robin, chair) than others (e.g. penguin, lectern). Participants judged typical members more quickly, listed them first, and used them more often in category-related tasks. Rosch (Reference Rosch1975) argued that people’s mental representations of categories reflect graded structures rather than discrete boundaries, influenced by perceptual and functional similarity among entities, with more typical exemplars cognitively closer to the centre of the category than less typical exemplars.
A few years later, Barsalou (Reference Barsalou1983) introduced the notion of an ad hoc category to refer to a novel concept constructed on-the-go in order to achieve a goal relevant to the situation (e.g. constructing tourist activities to perform in Beijing while planning a vacation, foods to eat on a diet while planning a weight loss programme, ways to stay awake during a long drive for the purposes of such a trip, etc.; Barsalou, Reference Barsalou1983, Reference Barsalou2010). These so-called ad hoc categories are temporary, goal-derived groupings that exhibit many of the same cognitive properties as conventional taxonomic categories. They have prototype effects, graded structure and shared typicality patterns. However, their members are not linked by permanent semantic relations but by their relevance to a particular goal or situation.
Barsalou’s (Reference Barsalou1983) ad hoc categories and Rosch’s (Reference Rosch1975) prototype theory both assume that human categorisation is flexible, graded and psychologically grounded rather than strictly rule based, but they differ in how and why categories are formed. The claim in Rosch (Reference Rosch1975) appears to be that prototypes organise stable, fixed semantic categories, while Barsalou’s (Reference Barsalou1983) theory argued that at least some categories are created spontaneously to meet current goals. This makes them heavily context dependent and transient rather than permanent representations stored in long-term memory as fixed structures. According to Barsalou (Reference Barsalou1983), the key element of their organisation appears to be situational relevance rather than putative perceptual and functional similarity. Despite these key differences between the two approaches, Barsalou (Reference Barsalou1983) appears to treat ad hoc categories as a special class of categories rather than radically departing from Rosch’s (Reference Rosch1975) foundational idea.
Several decades later, Casasanto and Lupyan (Reference Casasanto, Lupyan, Margolis and Laurence2015) proposed that all concepts are ad hoc concepts. Building upon Barsalou’s (Reference Barsalou1983) insight that categorisation is guided by goals and context, they extend it to challenge the traditional distinction between stable (taxonomic) and ad hoc categories. In this re-conceptualisation of ad hoc concepts, even familiar categories like bird or chair are assembled differently depending on the situation, task or linguistic framing. For example, the concept of bird invoked when thinking about pets differs from the one activated when thinking about flying animals or dietary restrictions. In a similar way, the concept of teabag that comes to mind when preparing a relaxing cup of tea at home sharply diverges from the concept activated when hearing its metaphorical use in a mixed martial arts (MMA) sports commentary. This flexibility reflects the influence of language, perception and current goals on conceptual processing. Under this ad hoc concepts framework, cognitive processing is dynamic by default, a constant interplay between an individual’s experiential history and the current context of operation. Casasanto and Lupyan (Reference Casasanto, Lupyan, Margolis and Laurence2015) argued for a usage-based theory of concepts, emphasising that conceptual representations are not fixed entities but processes that emerge as needed. In doing so, they extended Barsalou’s original idea to proposing a general framework in which conceptual thinking itself is inherently ad hoc, fluid and context dependent.
This makes ad hoc concepts an appealing explanatory framework for the idea that language affects cognitive processing of the perceived world, captured in Whorf’s (Reference Whorf1956) linguistic relativity hypothesis (LRH). The LRH continues to generate vigorous debate and often highly contrasted views amongst academics. Much of the contention is rooted in disagreement over what kind of evidence ‘counts’ as an effect of language on cognition, which in itself is rooted in what one takes cognition, or cognitive representation, to be, if indeed ‘representation’ is an adequate conceptualisation of the cognitive resources that are brought to bear in the cognitive processing of the perceived world. For some scholars, the evidentiary hallmark of language shaping cognition is the fast modulation of behaviour in experimental designs where language is up-regulated (e.g. by verbal priming or ‘cueing’, Boutonnet & Lupyan, Reference Boutonnet and Lupyan2015; by manipulating language context of operation, Athanasopoulos et al., Reference Athanasopoulos, Bylund, Montero-Melis, Damjanovic, Schartner, Kibbe, Riches and Thierry2015; or in novel category learning, Maier & Abdel Rahman, Reference Maier and Abdel Rahman2019) or down-regulated (e.g. by a concurrent verbal interference manipulation that disrupts online feedback between stimuli and their corresponding linguistic constructions, Winawer et al., Reference Winawer, Witthoft, Frank, Wu, Wade and Boroditsky2007). For other scholars, however, such modulations of behaviour are taken as evidence that effects of language are transient and superficial, since they do not lead to re-organisation of the underlying conceptual space (e.g. Barner et al. Reference Barner, Inagaki and Li2009; Ünal & Papafragou, Reference Ünal and Papafragou2016). As Pinker (Reference Pinker2007) puts it: ‘speakers of different languages tilt in different directions in a woolly task, rather than having differently structured minds’ (p. 148). But this criticism is only valid if we assume that there is an underlying fixed and static representational level. If one abandons the idea of fixed permanent concepts in the mind of the speaker, one of the key assumptions of the ad hoc concepts framework as articulated by Casasanto and Lupyan (Reference Casasanto, Lupyan, Margolis and Laurence2015), then what may seem like transient effects are actually a natural product of the ad hoc nature of conceptual representation.
The current collection of papers brings together empirical evidence that reshapes our understanding of language–thought interactions, placing the fluid nature of the conceptual–perceptual system at the heart of the epistemological profile of linguistic relativity. This discussion began among the three guest editors at the Cognitive Neuroscience of Language Embodiment and Relativity Conference held at Adam Mickiewicz University (Poznań, Poland) in 2022. It promptly continued and became focused on the theoretical advancement of the linguistic relativity hypothesis by way of a theme session entitled Ad hoc cognition and the Whorfian question: towards an experiential and situated model of language-thought interactions at the 16th International Cognitive Linguistics Conference (ICLC16) at Heinrich-Heine-Universität (Düsseldorf, Germany) in 2023. The papers that constitute this special issue are all by authors that took part at the theme session in ICLC16.
The studies reported offer wide breadth and scope in terms of perceptual domains covered (colour, time, objects, motion, odour, touch), linguistic phenomena (spatio-temporal metaphors, numerical structure, colour and odour terminology, motion event typology, grammatical gender, reading and writing direction), linguistic typologies (Chinese, Greek, English, Ukrainian, Russian, Czech) and methodologies (categorisation, verbal interference, mental calculation, psychophysics).
2. An overview of the papers in the special issue
Kapelle and Flecken (Reference Kapelle and Flecken2025) take recent reviews of language–perception interactions (e.g. Athanasopoulos & Casaponsa, Reference Athanasopoulos and Casaponsa2020; Lupyan et al., Reference Lupyan, Abdel Rahman, Boroditsky and Clark2020) as their pivot to offer a scoping review on which kinds of tasks reveal effects of language specifically on colour perception. This is an area that has been of central interest in empirical investigations of linguistic relativity, and the focus of the article on the cognitive demands of tasks that rely on, or manipulate, working memory and the up-/down-regulation of linguistic resources presents an insightful bird’s eye view of the dynamic role of language in colour perception. Under this prism, the ad hoc nature of findings from different paradigms, which show variable effects of linguistic terminology on colour perception, can be (re)interpreted as the interplay between how much memory is needed for a given task and the likelihood of recruiting language online. Taking the verbal interference technique as an example, the authors conclude that studies with tasks requiring a high memory load are more consistently able to find evidence for verbal interference on colour categorical perception than studies that require little reliance on memory. This is tallied up with a long line of psycholinguistic literature that shows that even in the visual modality, working memory is verbally mediated through verbal recoding. Disrupting this verbal recoding by means of an interference task that recruits the same representational resources recruited for categorisation abolishes or significantly attenuates language effects on cognition. In this way, under an ad hoc cognition framework, one can conceptualise language not as a dichotomous switch set in an off/on position, but rather more like an analogue linear scale where the influence of language can be dialled up or down depending on cognitive demands and ambient context.
Zhuang and Vanek’s (Reference Zhuang and Vanek2025) study on numerical cognition in relation to calendar calculations exemplifies how information structure about temporal sequencing may pose different cognitive demands on speakers. The study asked monolingual Chinese and English speakers as well as Chinese–English bilinguals to compute the target day or month (among other calendric units). Chinese labels calendar terms numerically (Monday = Day 1, October = Month 10), whereas English uses a more opaque system of verbal listing (Monday, October). The study found that participants were faster in calendar calculations that conformed to their native language’s numerical reasoning system. Interestingly, the bilingual group preferred numerical instead of verbal listing, suggesting that it poses fewer cognitive demands due to its transparency. This was also corroborated by the finding that changes from verbal to numerical input speeded up processing time for Chinese speakers but slowed down processing in English speakers, showing how prior experience with language habitually shapes cognitive processing in a context dependent way. Specifically, temporary ad hoc destabilisation of relevant experiential priors (verbal listing) induced slower processing in the English-speaking group, whereas the same contextual change up-regulated an experiential prior in the Chinese-speaking group. This interpretation is further supported by the finding that these contextual changes did not affect processing speed in bilinguals. Developing competence in the reasoning style of the second language means that they can more flexibly adapt their calculations processes when the contextual input changes.
Brand et al. (Reference Brand, Preininger, Kříž and Ceháková2025) present the thesis that the effects of language on cognition do not always manifest to the same degree in all situations but are instead constrained by the type of stimulus perceived by the participant and the experimental context. They use grammatical gender as their locus of investigation to examine across three experiments the degree to which it permeates speakers’ conceptual associations of objects with gender. Experiment 1 was a rating task where written word stimuli were rated on a conceptual gender scale. Experiment 2 repeated the same task, but with pictures instead of words. Experiment 3 utilised a more implicit design, asking participants to choose one of three faces (female-like, male-like and gender-neutral) that they most associate with a target image. Crucially, they systematically varied the type of referents that were used as stimuli, focusing on people, animals and inanimate referents. The findings showed that Czech (a language with grammatical gender) but not English (a language with no grammatical gender system in the nominal domain) speakers were influenced by grammatical gender in their ratings, with the category of animals yielding the strongest effects relative to inanimate entities, where in Experiment 3 no effects were found. The groups did not differ in the people category, since the gender rating could be based on the perceived gender characteristics of the person. Interestingly, a grammatical gender effect was present even when Czech speakers were operating in their L2 English. Since English does not have grammatical gender, there is no competing association to interfere with the pre-potent L1 representation. The study roundly shows that in cases where stimuli can up-regulate grammatical gender (as in the case of animate referents) there follows a predictable effect of language on cognition that is itself task dependent and manifests predominantly in contexts where explicit gender associations are elicited (Experiments 1 and 2) rather than more implicit associations with social stimuli, such as faces (Experiment 3).
Casasanto et al. (Reference Casasanto, Fotakopoulou, Lucero, Pita and Boroditsky2025) take a different approach to highlight the dynamic, ad hoc role of language in shaping human cognition. Training participants to talk about an experiential domain (in this case, time) using the structure of another language seems to bring about short-term changes in the psycho-physical experience of said domain. The authors initially approach the issue using a paradigm that is now classic in empirical treatments of linguistic relativity, namely comparing cognitive behaviour in the same task in populations that differ in how they preferentially encode the domain in question. They focus on Greek and English speakers, showing first that the prevalent metaphors for talking about temporal duration in Greek utilise quantity spatial terms (e.g. the days become ‘bigger’ and the nights become ‘smaller’ when the clocks change to wintertime in October) instead of the linear distance terms used in English (e.g. a long/short night). In a task asking participants to estimate the duration of growing stimuli, English speakers were misled by the length of a growing line, thinking that the longer a line grew the more time has passed (even if the growing time was the same as for a shorter line). Conversely, Greek speakers were misled by the quantity to which a container filled with liquid. The greater the filling volume, the greater the filling time estimated by Greek participants, despite filling time being orthogonal to fill level. Interestingly, this cross-linguistic behavioural difference survived verbal interference, which is certainly at variance with the vast majority of linguistic relativity studies that show a diminished effect of language under verbal interference. Again, in line with the ad hoc concepts framework, the nature of the task may come into play here, as the nature of the task (time estimation) may not readily allow covert verbalisation of the temporal properties of the stimuli. This would then mean that prior experience with language is enough to shape people’s temporal conceptualisation, without the need to recruit language online. In their final experiment, English-speaking participants were exposed to quantity metaphors for duration and behaved like Greek speakers in the duration reproduction task. This shows that ad hoc associations can form as a result of short-term training and have a potent role in (re)shaping cognition, a finding that resonates with other papers in this special issue showing effects of long term exposure to another language (as in the case of bi/multilingualism).
Miller et al. (Reference Miller, Blankenburg and Pulvermüller2025) examine whether facilitative effects of labels on perception are unique to language or can be elicited to a comparable extent by non-verbal auditory stimuli such as musical patterns. They use tactile perception as the test domain. Touch is a particularly well-suited sensory domain to test labelling effects given its fine-grained discrimination demands and limited pre-existing linguistic encoding. In a within-subject design, German participants learned to associate two sets of minimally distinct vibrotactile patterns with either pseudo-word labels or acoustically matched musical tone sequences. Learning occurred over 5 days of training. Tactile discrimination was tested before and after learning through a same-different forced choice task presented without any auditory cues. Discrimination performance improved significantly for the tactile patterns that had been paired with pseudo-words, but not for those paired with musical tones. The discussion interprets this effect as evidence of a causal Whorfian influence, whereby linguistic input actively re-shapes perceptual representations. Within a Hebbian cell assembly framework, consistent verbal labelling binds tactile and phonological representations into multi-modal neural networks, reducing overlap among similar sensory traces and sharpening discrimination. By contrast, the musical condition, lacking linguistic structure and articulatory–phonological grounding, fails to produce comparable neural re-organisation. Importantly, the findings also extend the ad hoc cognition framework (Casasanto & Lupyan, Reference Casasanto, Lupyan, Margolis and Laurence2015) by showing that language can dynamically construct perceptual categories even in a sensory domain like touch, where linguistic encoding of contrasts is typically rather impoverished. The enhancement of tactile discrimination following exposure to arbitrary pseudo-words illustrates that perceptual systems are not fixed but re-weighted by linguistic experience. Through an elegant experimental approach, the study positions language as a flexible system that can momentarily recalibrate perceptual sensitivity according to contextual affordances. This observation emphasises the adaptive nature of linguistic effects on sensory input processing, which aligns with the ad hoc cognition account.
Osypenko et al. (Reference Osypenko, Brandt and Athanasopoulos2025) explore how the co-existence of closely related yet non-identical grammatical gender systems shapes the conceptual associations of bilingual speakers. The study focuses on simultaneous Ukrainian–Russian bilinguals. This participant base forms a rare test case where both languages within an individual share a three-gender system (masculine, feminine, neuter), but diverge in specific noun classifications. Across two experiments, participants completed a series of gender-congruency and adjective–noun pairing tasks in both of their languages. Stimuli-targeted nouns that were gender-congruent (same gender in both languages) and gender-incongruent (different genders in Russian and Ukrainian). Participants rated and matched adjectives with nouns along gender-consistent or -inconsistent lines. The results revealed a striking contrast between conditions where the neuter gender was present (Experiment 1) and where it was absent (Experiment 2). In Experiment 1, both monolinguals and bilinguals showed clear gender-congruent associations. However, for gender-incongruent nouns, that is, those differing in gender assignment across the two languages, bilinguals displayed decreased gender-consistent responses and longer reaction times, indicating competition between co-activated gender representations. In contrast, Experiment 2 excluded neuter nouns, presenting only masculine and feminine genders. Here, the differences between bilinguals and monolinguals largely disappeared. This suggests that the presence of the neuter gender introduces an additional layer of conceptual complexity, amplifying cross-linguistic competition, whereas its absence simplifies the ‘grammarscape’ and promotes context-sensitive alignment. The observed difference demonstrates that simultaneous bilinguals flexibly adjust their grammatical mappings depending on which gender distinctions are active in the task. Rather than forming a stable hybrid system, bilinguals dynamically re-weigh grammatical cues depending on contextual activation. This finding enriches the ad hoc cognition framework (Casasanto & Lupyan, Reference Casasanto, Lupyan, Margolis and Laurence2015) by showing that even deeply grammaticalised features can operate as ad hoc constructs, quickly attuned to the immediate linguistic context.
Cao et al. (Reference Cao, Majid and Vanek2025) study how linguistic labelling may aid the learning of smell categories, introducing yet another sensory test domain. The olfactory domain is particularly interesting in this respect, as smells have low codability (at least in many of the languages in the world), meaning that there is not already a linguistic label available to map the smell onto, and the mapping itself may be untrained. The main novelty of the article lies in the use of different types of linguistic labels that are more or less familiar (native vs. non-native sounding) to their Chinese–English bilingual participants. Participants were trained extensively on different smells, either paired with L1-like linguistic labels, with L2-sounding labels or with completely unfamiliar sounding labels (Georgian, in this case) or without labels at all. The hypothesis predicted superior learning outcomes in the group who received smell training accompanied by native-sounding labels, with a reduction in performance for the L2-label group and the unfamiliar-sound group and lowest performance in the no-label group. During training, participants carried out an ABX odour categorisation task, on four consecutive days. Finally, they took part in an odour categorisation test, in this case without labels. Analyses focused on categorisation accuracy on Day 1 versus performance during the final test phase, that is, on learning gain. Contrary to the authors’ hypotheses, the group trained on the smells with L1 Chinese-like labels did not outperform the control (no-label) group.
The group trained with Georgian (unfamiliar) labels did perform better than the no-label group. However, comparing learning gains across the different labelling groups did not reveal significant differences. The frequentist analyses were further supported by Bayes factor analyses, suggesting that overall, there was inconclusive evidence for learning outcome differences across groups. The authors discuss certain limitations of the design, as well as of the ad hoc cognition framework, suggesting that contextual constraints, such as the long-term experiential histories of the tested groups, may have failed to introduce a strong link between odours and linguistic labels in general.
Gennari (Reference Gennari2025) reviews empirical research on the potential influences of language on cognition, taking into account both cross-linguistic and within-language approaches. Both approaches typically involve contrasting people’s behaviours or processing in a perceptual task, with or without the explicit involvement of language. Language–cognition interactions are reported to different degrees, but the specific conditions or cognitive constraints that determine the degree to which there is linguistic meddling are still elusive. Gennari’s review centres on core principles of goal-directed action planning, that involve semantic memory, starting off with how, in semantic memory, linguistic labels bundle together certain features that frequently co-occur – and that the associations between conceptual features or structures and labels become strengthened through experience, that is, through language use. This does not mean that the associations are necessarily always made in the absence of language use. Further, task goals and cognitive demands may determine the extent to which language is resorted to for task fulfilment. In certain cases, the linguistic route may be the most efficient, fastest or most readily available one. We do not yet know when this happens; according to Bayesian accounts, reliance on linguistic experience may be more likely under conditions of uncertainty, but systematic comparisons of different conditions are scarce. We also do not know whether linguistic interference is strategic or deliberate, or when it may be unconscious. Gennari argues for a systematic and controlled investigation of cognitive context, task manipulation, experimental condition and stimulus types, to shed light on the exact conditions underlying language–cognition interaction.
Crossley and Bylund (Reference Crossley and Bylund2025) investigate the influence of script direction on people’s conceptions of time. Different from earlier research focusing on reading direction, this experiment focuses on influences of writing direction in a controlled experimental context. The reasoning behind this focus is that, as we write, for example, from left to right, we look at the result of past motor activity on the left side of a page, and look ahead at the empty side, anticipating future actions towards the right. The authors draw the analogy between writing and walking, where the past is associated with the space behind oneself, and the future with the space in front, the distance yet to be covered. Using a version of a priming paradigm, in the baseline experiment participants read stories either in left-to-right or right-to-left orthography, after which they were asked to perform a temporal diagram task, placing fictive people/objects in space/time. In three subsequent writing experiments, participants were instructed to write either from left-to-right or right-to-left, after which they performed the same task. There were no differences in whether participants placed the future to the left or to the right, between participants in the two writing direction conditions. These null results were counter to the authors’ hypotheses and are discussed in light of accounts of embodied cognition, suggesting that the body may contribute to temporal cognition in a selective way, rather than to determine it. In particular, findings suggest that the act of writing may be a different kind of experience than reading experience, with a different role for guiding time conceptualisation.
3. Conclusion: Ad hoc relativity
The collection of studies surveyed offers a rich synthesis of how language interacts with perception and cognition across sensory, conceptual and linguistic domains. Read through the lens of Casasanto and Lupyan’s (Reference Casasanto, Lupyan, Margolis and Laurence2015) ad hoc concepts framework, as well as the original postulation of the theory by Barsalou (Reference Barsalou1983, Reference Barsalou2010), these findings converge on the idea that conceptual representations are not static, language-independent entities but dynamically constructed, context-sensitive configurations that depend on the cognitive and linguistic resources available in the moment. Across domains, from colour to touch, time and grammatical gender, language functions as a flexible tool that can be up- or down-regulated depending on task demands, memory load and contextual affordances.
Across the contributions, a coherent picture emerges. Linguistic effects on cognitive processing are not binary but gradient, modulated by how much the cognitive system needs, or prompts, to call on verbal scaffolding. Conceptualisation dynamically re-organises around task-specific constraints rather than reflecting stable linguistic determinism. Even if linguistic features can persistently bias conceptualisation (such as entrenched grammatical features), such bias may surface only when situationally activated. It then follows that language is best understood not as a static code that determines thought, but as a dynamic cognitive resource that flexibly reconfigures conceptual processing to meet situational demands.
This re-conceptualisation of the role of language in human thought is in line with more nuanced debates around the linguistic relativity hypothesis that aim to reformulate the question from one requiring a yes/no response (does language affect thinking?) to one that views such effects as a matter of degree that largely depends on specific experimental conditions that may promote or suppress the implicit use of language online to solve cognitive tasks (for discussions, see Athanasopoulos & Bylund, Reference Athanasopoulos and Bylund2013; Lupyan et al., Reference Lupyan, Abdel Rahman, Boroditsky and Clark2020; Regier & Kay, Reference Regier and Kay2009). The original question is thus reformulated as a series of wh-questions, seeking to shed light on what operational conditions, under which cognitive mechanisms and how much (in terms of degree or extent) language can shape our cognitive processing. Under this new light, any empirical evidence addressing the linguistic relativity hypothesis, whether positive or negative, is interpreted as informing this wider conceptualisation of language–cognition interactions where effects of language can be dialled up or down dynamically, rather than be permanently absent or present.
Open access
