Concepts featuring in eminent nutritional recommendations (e.g., USDA’s five food categories, namely fruits, vegetables, grains, proteins, and dairy) are generally referred to as “taxonomic” in the literature (see for instance Mirman et al. 2017; Markman and Hutchinson 1984; Markman 1981). The hierarchical structure displayed by taxonomic concepts has led scholars to draw a distinction among basic level categories (e.g., apple, dog, guitar…), characterized as those that subjects spontaneously refer to in a variety of tasks, (see Murphy 2002); superordinate categories (e.g., fruit, animal, musical instrument…), which are higher up in the hierarchy with respect to the basic level; and subordinate categories (e.g., granny smith, labrador, classical guitar…) which are below the basic level and typically require some expertise to be mastered. However, this is a subject-dependent partition to some extent. As shown by Tanaka and Taylor (1991), the basic level of categorization varies according to the subject’s expertise in a domain: when asked to categorise the image of a dog, the first name that comes to the mind of dog experts is more specific (usually breed-level, e.g., “German shepherd”) than that given by non-experts (“dog”).

Several studies investigated children’s understanding of taxonomic hierarchies (e.g., Blewitt 1994; Johnson et al. 1997) and discovered that two-year-olds already prefer basic level categories in a variety of conceptual tasks, that three-year-olds are able to learn subordinate categories (Mervis et al. 1994) and sort items according to basic level categories (Rosch et al. 1976), and that five-year-olds can generalize the properties of a category to a subordinate one.

Taxonomic categories distinctively display other features too. One is the so-called strong typicality effects, namely, the fact that typically shaped and colored apples are usually considered good examples of fruit compared to tomatoes or grapes, which are known to be fruits, although in a somewhat different way (Murphy 2002: 22); furthermore, taxonomic categories display correlational structures, meaning that if a member of a category has a feature that is typical of said category, it is likely to have other typical properties. These structures are apparent in taxonomic food concepts: for instance, there are strong color-taste correlations among vegetables, such as greenish-bitter or reddish-sweet (Foroni et al. 2016; Foroni and Rumiati 2017). Correlational structures explain why members of the same taxonomic category often share many features and motivate the widespread assumption that taxonomic categories are similarity-based (see Nguyen and Murphy 2003; Mirman et al. 2017; Hampton 2006; Mervis and Rosch 1981)Footnote 1. It is worth noting that scholars seem to understand food taxonomies as somewhat objective and grounded in the very nature of the items grouped in taxonomic categories (much like biological taxonomies), although there might in fact be countless different criteria for building culinary taxonomies (see for instance Borghini and Gandolini 2020; who build an authoriality-based taxonomy of recipes). Strong typicality effects and correlational structures jointly make taxonomic categories stable in users’ minds, easy to access and to store in long-lasting memory (Barsalou 1991).

Researchers have devoted great attention to taxonomic food concepts in children, since taxonomic knowledge has been shown to support a variety of cognitive abilities (e.g., categorization and category-based induction). Three-year-olds already display some proficiency in using taxonomic food knowledge to tell apart vegetables from fruits (Rioux et al. 2016 https://doi.org/10.1016/j.cogdev.2016.09.003): when presented with a picture of a food item (for instance, a red apple), they are able to pick a taxonomic match (for instance, a green apple) instead of a more perceptually similar one (for instance, a red tomato) when asked to put the same kind of items in a box. Children’s understanding of taxonomic knowledge undergoes significant development until the seventh year of life, as demonstrated by their performance in taxonomic matching and induction tasks (Nguyen and Murphy 2003). Nguyen (2012; Experiment 1) found that three-year-olds can use selectively taxonomic food categories for drawing biochemical inferences (e.g., inferences about the substance food items are made of). Perturbations in this developmental trajectory during preschool years are in fact predictive of food rejection dispositions in children (Rioux et al. 2018a, b).

Although prominent in the literature, information about taxonomies is not the only one that is relevant to children’s representation of food: in everyday decision-making, it is hard to detach it from other kinds of knowledge that we are going to illustrate in the following sections.

Food knowledge often involves knowledge of external relations between food items, events, and concepts. For instance, in a grocery store one might find a section devoted to breakfast foods, a shelf dedicated to bakery ingredients and tools, maybe a corner for Halloween or Christmas treats.

Although many authors have considered the very concept of food to be eminently taxonomic (see for instance Estes et al. 2011Footnote 2), relational thinking is indeed crucial in the food domain since food acceptability depends on certain relations holding between food items (e.g., food pairings consisting of otherwise acceptable foods like chocolate and chicken might appear disgusting or inherently wrong to somebody) and eaters (e.g., some foods might be rejected due to someone’s food intolerances or religious beliefs).

Sometimes relational knowledge can be used to group items and form categories (Gentner 2005). To an extent, this is unsurprising: after all, taxonomic knowledge itself revolves around set-inclusion relations. Nonetheless, relational knowledge stresses the importance of external relations in conceptual representations, understood as relations holding between items that are represented as distinguished or not belonging to the same superordinate category (Markman and Stilwell 2001).

Recent empirical literature in food cognition focuses mostly on two kinds of relational knowledge: one hinges on complementarity or spatio-temporal co-occurrence between instances of two concepts (“thematic knowledge”, see Estes et al. 2011), the other on the roles that two concepts play within a script, defined as a schema for a routine event (Nguyen and Murphy 2003). Both kinds of knowledge are crucial in determining food acceptance, since most human communities develop strong expectations about what counts as an appropriate way of serving and consume a given dish. Consider for instance fries, which are usually highly palatable among children. Fries for breakfast might trigger rejection among kids that are accustomed to eating sweet food in the morning and presenting a dish in which fries are paired with an unusual food associate (e.g., strawberry ice cream) might cause the same effect.

Thematic knowledge has been mainly framed as knowledge about the links between concepts that perform complementary roles in the same event or scenario (Golonka and Estes 2009; Wisniewski and Bassok 1999; Mirman et al. 2017). More specifically, the nature of said links can be either functional (e.g., fork and knife, spoon and soup…), causal (knife and watermelon, grill and meat), co-occurrence-based (established food pairings, e.g., strawberries and whipped cream, bread and butter…), or a combination of the former (e.g., buns and patty or ice cream and cone, which are arguably both co-occurrence- and function-based). Those links are often referred to as “themes” (for instance, see Estes et al. 2011) and are more culturally bound than the similarity and inclusion links underlying taxonomical knowledge (Markman 1989, 1994).

According to Estes and colleagues (2011), one distinctive feature of thematic relata is that they perform different roles in the context of their theme. For this reason, unlike cognate taxonomic concepts and items, thematic relata (e.g., bread and butter or soups and spoons) do not usually share many intrinsic features. Nonetheless, the authors have found that thematic knowledge is especially suited to support analogical reasoning and that sometimes it even intrudes in similarity-based judgements. For instance, it has been shown that subjects tend to judge milk and coffee as more similar than milk and lemonade, although lemonade and milk share more key perceptual features (namely, the color white)Footnote 3 (Wisniewski and Bassok 1999; Golonka and Estes 2009).

The centrality of role-based thinking in developing themes suggests a strong connection between thematic knowledge and role-governed food categories. According to Markman and Stilwell (2001), role-governed categories refer to roles within the relational structures underlying relational categories. Although it is not always clear what the theoretical purpose of categories is according to the authors (they alternatively state that categories either “name”, or “identify”, or “represent” ontologically heterogeneous entities such as objects, verbs, events, and relations), their distinction can prove useful when applied to food. Relational food knowledge in general may underlie the understanding of multiple configurations of food items and concepts (e.g., main course-side dish couplings, Russian- and French-style service, mereotopological structuresFootnote 4), whereas thematic knowledge involves more precisely our ability to fill up the (conceptual) slots of complementary relational structures with (concepts of) food items.

It is important to point out that thematic and taxonomic food concepts can be acquired by different means: while familiarity with certain items has a positive effect on children’s ability to detect both taxonomic and relational matches, the recognition of thematic fit depends specifically on the capacity to rely on contextual cues in real-world scenarios (see McRae et al. 1998; Markman 1989). Due to the conventional nature of many themes, children need to get accustomed with a food item in several different contexts to grasp the relevant thematic connections that it bears to other entities. For instance, the shape of the cone lends itself to contain several kinds of foods equally well (e.g., ice cream, cheese, sauces…), and ice cream is in fact often served in different containers or recipes (e.g., cones, cups, glasses, cakes, and so on). Since the thematic association between ice cream and cones is not grounded in any principled reason, it is necessary to engage regularly with scenarios in which the two items are co-occurrent in order to grasp the relevant piece of thematic knowledge. Although there are indeed physical constraints involved in thematic relations (for instance, Estes and colleagues (2011) say that thematic fit is based on affordancesFootnote 5), the nature of themes is conventional and arbitrary to a certain degree.

Empirical evidence shows that humans usually acquire thematic thinking early during childhood (Lucariello et al. 1992; Waxman and Namy 1997; Blanchet et al. 2001) although significant interindividual differences may occur due to factors such as education, cultural norms, and language learning. Pickard and colleagues (2021) devised food-based analogy tasks to detect possible gaps in preschoolers’ thematic knowledge: after being exposed to two pictures of food items exemplifying either taxonomically- or thematically-related couples, test subjects were asked to match a target stimulus with one of two options in analogy to the kind of relationship instantiated by the previous couple of pictures. Each participant was tested on sixteen trials, half of which assessed their ability to understand taxonomy-based analogies (e.g., apple: banana::burger patty: chicken leg), whereas the other half tested thematic analogies (e.g., ice cream ball: cone::burger patty: burger bun). The sixteen trials were presented in pseudo-randomized order. The study revealed that increased levels of food rejection were predictive of poorer thematic understanding. Since it conveys information about the cultural and conventional features of eating settings, poor thematic knowledge may in fact be a crucial factor triggering uncertainty among young eaters. In a series of three studies run on preschoolers, Pickard and colleagues (2023) investigated the links between food neophobia and different pieces of relational food knowledge. In study 1, they used a forced-choice trial task to assess the developmental differences in the acquisition of different kinds of relational food knowledge. Among the four associations investigated (which have been adapted from Lucariello et al. 1992), children showed proficiency in mastering food-utensils (“functional”) associations earliest, then food-food (“conventional”) associations, and event- or meal-based script relations later on (more about scripts in the next section). These results suggest that a more fine-grained distinction between kinds of thematic relations might be useful to better understand how children conceptualize the food domain and why some thematic relations are mastered earlier than others.

Another relevant kind of relational food knowledge that empirical research has focused on is the so-called script knowledge. We classify script food knowledge as relational for it conveys information about the items, processes, and relations that occur within a given customary scenario or situation (“script”) (Ross and Murphy 1999; Nguyen and Murphy 2003; Estes et al. 2011). It is thus easy to see the relevance of this kind of knowledge in the food domain, both for grouping items and for forming concepts: breakfast food, birthday cakes, and ritual food are all perspicuous notions which rest upon script knowledge. Empirical research confirms its importance as well. In a category generation task with American college-aged participants, Ross and Murphy (1999; Experiment 1) found that all the subjects spontaneously formed at least one script categoryFootnote 6 to group the food items they were presented with.

Unfortunately, the literature displays a high degree of ambiguity about the exact characterization of the notion of script knowledge, especially when it comes to its definition and relation with other kinds of knowledge. Some authors (see Lucariello and Nelson 1985; Nelson 1983; Lucariello et al. 1992) seem to conceive script knowledge as a sub-kind of taxonomic knowledge, which they call “slot-filler”. According to these accounts, slot-filler categories are composed of items which share the same function within a given event, where said function “creates” slots that can either be filled simultaneously or alternatively by the items in the category (the authors exemplify slot-filler categories by mentioning bologna, peanuts, and cheese as members the category “items that can be eaten for lunch”).

Others treat script knowledge as a subcategory of thematic knowledge. Estes and colleagues (2011) define scripts as “generalized sequence(s) of actions and instruments associated with the execution of some common event”, in which the “various objects, concepts, people, and actions (…) are externally related by the event itself, and they perform complementary roles in the execution of the script” (p. 254). Scripts are thus an event-based subset of themes, where the relevant thematic associations are links between actions and instruments that are typical of a given common event. It is not clear whether this frame can accommodate available empirical data: for instance, some authors (Lucariello et al. 1992; Berger and Donnadieu 2006) suggest that thematic and script knowledge should be kept separated, for the second is mastered later in development due to his more cognitively demanding nature, but other studies have not found different developmental trajectories for the two kinds of knowledge (Nguyen and Murphy 2003).

Third, some authors (Nguyen and Murphy 2003; Nguyen 2012) propose a distinction between thematic and script associates in which the former play complementary roles within a scenario, whereas the latter play the very same role within the relevant script. Thematic associates are thus co-occurrent, while script associates are mutually substitutable. A principled distinction between complementarity and substitutability relations seems plausible because understanding these two different relations requires different kinds of knowledge and cognitive abilities, but it seems hard to draw clear-cut boundaries between the two kinds of relations in the food domain. Consider “breakfast food”, a textbook example of script food categories (Bian and Markman 2020a, b): it seems uncontentious that both milk and cereals are typical breakfast foods to some people, although they cannot be strictly speaking substituted with each other; in fact, they are linked by a strong thematic association because they play complementary roles in breakfasts. Within this framework, more fine-grained role-based distinctions are called for to tell the two types of relation apart: for instance, some items within the breakfast food category can be linked by substitutability relations relative to the role “beverage” (e.g., milk and orange juice), others by complementarity relations according to the “ingredients-to-the-same-recipe” role (e.g., peanut butter and jelly), while others by both (e.g., milk and coffee can be both co-occurrent or alternatives to the role of “beverages” within the breakfast script).

Finally, some studies endorse notions of script knowledge that do not comply with any of the aforementioned approaches. For instance, Thibaut and colleagues (2016; Experiment 1) tested children’s abilities to extend psychological and biological properties on taxonomic or script food associates, but the stimuli used to exemplify script relations do not appear to be event-specific (fish-rice, carrot-fish, strawberry-whipped cream…), rather resembling thematic associates. In Ross and Murphy’s (1999) studies, script associates were introduced to participants as “foods that are eaten at the same time or in the same situation” (p. 510), thus conflating script, thematic, and other kinds of food groupings under a single label (e.g. “junk food” counted as a script category in their study, while more recent research classifies it as an evaluative category, see Nguyen and Murphy 2003 and Sect. 2.3.1 of this paper).

Despite these differences in approaches and the lack of a commonly accepted definition, the intuition that script knowledge plays a central role in food cognition and that it hinges on the recognition of external relations among items occurring in a given scenario is shared by most researchers in the field and supported by empirical findings. Ross and Murphy’s (1999) studies on food categories showed that script categories are almost as relevant as taxonomic ones in category generation and sorting tasksFootnote 7, and that both kinds of knowledge support the inference of various properties from a target stimulus onto another relatum with comparable strength (Experiment 6).

Some previously covered studies (Estes et al. 2011) studied thematic and script food knowledge alongside each other. This literature shows that children attend both kinds of knowledge since an early age (usually between the third and fourth year, see Pickard et al. 2023, Nguyen 2012Footnote 8) and that knowledge gaps in both fields are positively correlated with increased levels of food rejection tendencies (Pickard et al. 2021).

Both scripts and thematic associations are difficult to investigate empirically, as they are highly sensitive to culture and to interindividual differences. For instance, Bian and Markman (2020a) have run a comparative study to assess the differences between the “breakfast food” notion endorsed by children in China and in America (four- and five-year-olds). They found that the former had more permissive categorization criteria and more graded category boundaries than the latter.

Another study by Bian and Markman (2020b) explored a further complexity of script food knowledge that might account for the interindividual differences in flexibility of script categories’ boundaries in cognitive rather than cultural terms. Due to a phenomenon known as inherence heuristic bias (Cimpian and Salomon 2014), some people are inclined to misrepresent script categories as being objectively given and grounded in the “intrinsic” or “inherent” properties of the food items they refer to, rather than being conventional and historically determined. This bias has been detected in young children (see Hussak and Cimpian 2015; Tworek and Cimpian 2016) and might be a precursor of essentialist dispositions, defined as the tendency to represent categories as having non-obvious essences accounting for the typical features of their members (Newman and Knobe 2019; Neufeld 2022). Going back to Bian and Markman’s study (2020b), the authors found that a substantial number of adults provided “inherent” explanations to their categorization of breakfast food items (e.g., some items are typical breakfast foods because they are highly energetic or easy to digest) and that such explanatory strategy negatively correlates to the willingness of trying new food items for breakfast. One can expect that the distinction between the biological (inherent) and cultural (conventional) dimensions of food in script-based tasks is especially hard to attend for children, for preschoolers are prone to essentialize spontaneously many kinds of categories (e.g., artifacts, see Gelman 2013).

In conclusion, the research on food script knowledge displays some shortcomings. Most importantly, the lack of a shared notion of script knowledge makes it hard to integrate empirical findings across different studies; then, many relevant food scripts, e.g., course-level (e.g., appetizer), meal-level (e.g., breakfast), or special event (e.g., birthday) scripts, are yet to be studied. Finally, due to the cultural nature of food scripts, further research is needed to assess the impact of factors such as linguistic cues and meal structures on children’s food choices.Footnote 9

To be able to make certain food choices an agent requires knowledge about the most appropriate means to a certain end as well as the nature of the goal, be it health (Nguyen 2007), moral values (Lakritz et al. 2022), or practical aims (e.g., make the best recipe out of an almost empty fridge).In this section, we first discuss food knowledge concerning somewhat stable and general value assessments, then that involving provisional and impromptu means-to-an-end reasoning.