How Synonymic Taste Words Alter Perceived Taste in American Consumers

Abstract

Investigations into crispy and crunchy in American English have demonstrated that these synonymic taste words have differing effects on perceived taste depending on association. To test the generalizability of these findings, category fluency tasks were used to elicit foods and beverages more and less associated with several pairs of synonymic taste words. Next, taste tests were conducted using synonymic taste words and some of their more and less associated products as stimuli. The results showed that more associated taste words have a marginally significant, positive effect on taste ratings, with significant interaction effects for certain products resulting in lower taste ratings. This study confirms that synonymic taste words beyond crispy and crunchy can alter perceived taste in American consumers. Moreover, it demonstrates that the underlying mechanisms are complex and, in addition to association, depend on the particular food or beverage as well as further factors.

1. Introduction

Synonymy or “the existence of more than one word with the same meaning (Croft, 2003, p. 358)” is, at best, an exceedingly rare phenomenon if taken to mean that two or more words must constitute perfect substitutes for one another in all contexts and without even the subtlest change in meaning (e.g., Cruse, 2017; Murphy, 2013). Therefore, in this article, we talk about synonyms in the sense of near synonyms, i.e., words with similar semantic meanings distinguishable by differences in their collocates (e.g., Hanks, 2012). For instance, corpus-based research reveals that in American English, crispy tends to describe foods that have been “dehydrated after the application of heat (Diederich, 2015, p. 194)”, such as chips, while crunchy more often describes “wet foods as vegetables and fruit … in their natural state (Diederich, 2015, p. 195)”, such as apples.

Using consumer research methods, Johnson and Pfenninger (2021) showed that American consumers perceive crunchy chips and crispy apples as better-tasting than crispy chips and crunchy apples. It would appear, therefore, that less associated labels produce better perceived taste. However, given the narrow focus of the taste tests, it remains unclear “whether and to what extent these findings can be generalized (Johnson & Pfenninger, 2021, p. 79)”. Similarly, an otherwise stellar review of the aforementioned corpus-based research into crispy and crunchy (Diederich, 2015) concluded that ideally, a greater number of adjectives should have been taken into consideration (Gerhardt, 2018, p. 376).

We address both of these issues: Firstly, we propose a fast method of identifying subtle semantic differences between synonymic taste words and of determining their more and less associated foods and beverages. Secondly, we replicate Johnson and Pfenninger’s (2021) taste tests with several pairs of synonymic taste words and a variety of products. We therefore expand the hitherto primarily descriptive literature on English taste words (e.g., Bagli, 2021), offering more comprehensive insights into the effects of taste words on taste perception.

1.1. Taste Words

We define taste words very broadly as words used to describe a food or beverage as it is experienced in the mouth. This definition includes taste words that refer to the basic tastes, i.e., sweet, salty, sour, bitter, and umami (e.g., Bagli, 2018). Additionally, it encompasses taste words that refer to pain sensations (Cheok et al., 2023), e.g., hot, to ingredients, e.g., peachy (Yingst et al., 2017), to evaluative terms, e.g., fresh (Koukova et al., 2023), and to mouthfeel or texture, e.g., creamy (Fong et al., 2020). We use this broad definition because it corresponds to neurophysiological findings that taste largely depends on the perception of stimuli as originating from the oral cavity rather than on specific sensory phenomena (Small & Prescott, 2005, p. 354). Moreover, it captures the wide variety of words used to talk about taste, e.g., depending on level of expertise in a certain area or on the context in which a taste word is uttered.

In everyday English, speakers’ descriptions of taste commonly include vague terms, e.g., hearty, terms that refer to ingredients, e.g., gingery, and terms that refer to properties beyond taste, e.g., burnt (Ankerstein & Pereira, 2013). In contrast, experts use highly specialized vocabularies. Food scientists develop specific flavor lexicons to describe and compare particular types of food such as beef (Maughan et al., 2012) or Sichuan peppers (Yang et al., 2021). Similarly, expert reviews might contain words such as tobacco or even leather to describe the taste of whiskey (Hamilton et al., 2023), while wine connoisseurs might refer to a particular bottle as unbalanced or even brawny (Lehrer & Lehrer, 2016). Differences in taste vocabulary have even been reported in medical contexts, e.g., how cancer patients and oncologists describe taste issues (Boltong et al., 2011).

Taste words also vary according to context. Purely evaluative terms such as indulgent, irresistible, or simply great, which have no actual taste-related meaning even in the broad sense defined in this study, are especially common in food advertising (e.g., Ke & Wang, 2013; Strauss, 2005). Mackeson Beer, for instance, ran a successful commercial using the purely evaluative taste word good multiple times: It looks good, tastes good, and by golly, it does you good. (Ke & Wang, 2013). This strategy can also be observed in restaurants, which frequently label menu items as, e.g., mouth-watering or zesty, implying that these dishes will please, without actually describing their taste (Jurafsky, 2015). The culinary arts provide another specific context which influences the way speakers talk about taste. Consequently, several corpus studies have looked at, e.g., recipes (e.g., Yamakata et al., 2020), cooking shows (e.g., Fiedler, 2022), and menus (e.g., Jurafsky et al., 2016) as distinct genres. Finally, taste words frequently feature in contexts that have nothing to do with consuming, preparing, or promoting foods and beverages, i.e., they are employed metaphorically. For instance, landscapes can be delicious (Jaworska, 2017) and unimaginative teaching styles can be vanilla (Harland & Wald, 2018). Even concepts, e.g., feelings and moods, can be described using taste words such as bitter (e.g., Torres Soler, 2021) or sour (Zawisławska & Falkowska, 2018, p. 6).

Considering how versatile taste words appear to be, we will attempt to include synonymic taste words of different types, e.g., basic and ingredient-based, and which exhibit domain- and context-specific variation.

1.2. Associated Taste Words and Foods

Words can be said to be associated if they frequently co-occur or collocate in natural speech (Vilkaitė-Lozdienė & Conklin, 2021, p. 362), i.e., they are derivable from corpora (Rapp, 2014, p. 1). However, corpus- or usage-based approaches to semantics require extensive amounts of—ideally—representative and balanced data (Gries, 2009, p. 1231) to properly take the many possible contexts in which words occur into account. Additionally, the “meticulous and painstaking analysis of corpus data” (Gerhardt, 2018, p. 376) can be very time consuming as one cannot assume that collocational frequencies are directly translatable to strength of association (e.g., Mollin, 2009) and must instead rely on additional, qualitative analyses of contexts. This is less than ideal as we want to generate multiple more and less associated pairings of taste words and products in order to validate and generalize previous findings pertaining to crispy and crunchy. Fortunately, given speakers’ “ability to activate words by association with a category or topic” (Barbeiro et al., 2011, p. 312), associated words can also be elicited, e.g., with category fluency tasks (CFTs). CFTs involve determining a semantic category, e.g., animals, as a stimulus and requiring participants to list as many corresponding exemplars, e.g., bird, as they can in a limited timeframe (Prescott et al., 2006, p. 3), thereby providing insight into what is most readily accessible (Fernández-Fontecha, 2021, p. 2).

Several studies have successfully employed CFTs and similar experimental methods to study taste words and food. For instance, Bieler and Runte (2010) discovered that the central taste vocabulary of German is dominated by the taste words süss (sweet), sauer (sour), bitter (bitter), salzig (salty), and scharf (spicy). Furthermore, positive evaluative taste words such as lecker (tasty) are more central than negative evaluative taste words such as eklig (disgusting) or taste words relating to texture such as cremig (creamy). Going in the opposite direction, i.e., using taste words as categories, Ghemulet et al. (2014) found that in Greek, different taste words elicit different numbers of foods. Specifically, sweet elicited the most exemplars, followed by salty, sour, and bitter. Furthermore, bitter elicited significantly more errors than any of the other taste words. In a recent study involving several food categories across languages, Mazzuca and Majid (2023) showed how exemplars reflect a food’s importance in a particular culture: In the category pasta, Italian consumers produced more exemplars than British consumers. Furthermore, Italian consumers’ exemplars revealed in-depth knowledge of ingredients and cooking methods while British consumers’ exemplars reflected a more superficial categorization, e.g., based on visual features.

Data elicited through CFTs reveals what comes to mind the fastest and most often. However, it is not as semantically rich as corpus data. For example, CFTs with speakers of American English show that apple is more and mango is less strongly associated with the category fruit (Kiang & Kutas, 2006). A plausible explanation might be that apples are grown in all fifty states and represent the most frequently consumed fruit nationwide (Industry at a Glance, n.d.) while mangoes require a tropical climate and are only produced domestically in parts of Florida, Hawaii, Texas, and California (Evans et al., 2017, p. 5). In contrast, a corpus-based study of the occurrences of apple and mango in American English would likely have brought to light that their respective frequencies vary by region and/or topic.

Given the lack of surrounding information in data elicited through CFTs, we expect some nuances in meaning, that might only surface in very particular contexts, to be missing. For instance, corpus-based research shows that while generally positively connoted, crispy “may have a negative value judgment in health contexts (Diederich, 2015, p. 150)” owing to its associations with foods high in fat and oil. Therefore, while we may not be able to explain some peripheral exemplars, we should be able to extract and intuitively understand the core foods associated with a taste word, enabling us to establish more and less associated combinations.

1.3. Aim and Scope

As associative patterns can be gleaned both from natural and experimental data, we investigate whether CFTs constitute a valid alternative to corpus research in terms of synonymic taste words with the power to alter perceived taste. Specifically, we ask the following research questions:

RQ1a: Can CFTs uncover distinctions between the foods and beverages associated with synonymic taste words?

RQ1b: If so, does the perceived taste of identical foods and beverages vary depending on food labels containing more and less associated, synonymic taste words?

To answer our research questions, we report on the results of two experiments. Experiment 1 is a CFT with six pairs of synonymic taste words. Experiment 2 comprises taste tests with a subset of the taste words from Experiment 1.

2. Experiment 1: Category Fluency Tasks

2.1. Participants

Sixty-five participants were recruited from Amazon Mechanical Turk, which requires that its “Workers” be at least 18 years old. To ensure that all participants possessed a comparatively large vocabulary (Twenge et al., 2019), we set the screener to require a bachelor’s degree or higher. In order to obtain American rather than regional exemplars, we attempted to recruit native speakers of American English from all regions of the USA, i.e., the West (n = 14), the Southwest (n = 10), the Midwest (n = 9), the Southeast (n = 12), and the Northeast (n = 20). Since we used a within-subjects design, participants served as their own controls. Therefore, we did not require that participants fall into a certain age range, nor did we ask about their gender or any other personal information. Each participant received USD 5 for their participation.

2.2. Procedure

As is often done with verbal fluency tasks (e.g., Vogel et al., 2020), we administered our CFTs as multiple rounds, each lasting one minute. Participants completed the tests online via the platform VideoAsk in October and November of 2023. Participants saw a video recording of a researcher who, after briefly introducing herself, gave participants the following instructions:

Once you push the button to go to the next step, I will say a word, more precisely, a food descriptor such as “hot”, and your task will to be to name as many foods or drinks as possible that you can think of in one minute that match “hot”, for example, “cocoa”, or “tamales”, or “sauce”. In total, there will be twelve food descriptors, so twelve rounds lasting one minute each. Please remember that there are no right or wrong answers, we are simply interested in what people typically think of when we mention a certain food descriptor. So please answer freely and spontaneously by talking into your device. We will transcribe your answers and then delete the recordings immediately.

Additionally, each round began with a short video recording of the researcher reminding the participants of the task at hand and naming the prospective taste word:

In the next sixty seconds, please name as many foods or drinks as you can that match the food descriptor “…”.

Since VideoAsk does not offer randomization, we created 5 versions of the CFT in 5 different orders using random.org’s List Randomizer tool.

To ensure that the instructions were sufficient to conduct the CFTs online, i.e., without a researcher present, we pilot-tested this procedure with 5 non-native speakers of English. These participants indicated that the instructions were clear. However, they expressed a desire to track their progress. Consequently, we added a countdown of the rounds of the CFTs as overlay text.

2.3. Stimuli

The Contemporary Corpus of American English (COCA, Davies, 2020) lists crispy and crunchy as synonyms. Starting with an extensive list of words to describe menu items (73 Words & Adjectives to Describe Food, 2023), we searched the COCA for further pairs of synonymic taste words, making sure to include a wide variety of taste words with shared topics or collocates and relating to multiple senses and properties (details including frequency rankings available on the Open Science Framework OSF). With crispy and crunchy, this resulted in six pairs of synonymic taste words as stimuli for the CFTs (

Table 1. Stimuli used in CFTs.

Pair 1	Pair 2	Pair 3	Pair 4	Pair 5	Pair 6
crispy	foamy	juicy	sour	sugary	piquant
crunchy	frothy	succulent	tart	sweet	spicy

).

2.4. Data Analysis

Since 5 out of 65 participants were unfamiliar with the term piquant and another 4 participants produced fewer than 3 exemplars, piquant and its counterpart spicy were not considered for Experiment 2 and were thus not further analyzed.

For all other taste words, we followed Vogel et al. (2020, p. 117), accepting responses if they constituted food categories (e.g., cookies; fruit), types (e.g., chocolate chip cookies; apples), and subtypes or brands (e.g., Chips Ahoy!; Granny Smith). As we were only interested in the most associated foods given a taste word and not, e.g., in clustering and switching strategies (e.g., Fong et al., 2020), we only coded participants’ initial responses. Importantly, we did not assume any categories a priori, especially not across pairs. Instead, we studied and grouped responses for each pair of synonymic taste words individually. Naturally, this resulted in codes on different levels. For instance, based on the responses obtained for sugary and sweet, we selected the more general code “fruit”. In contrast, the more specific “apple” seemed appropriate given the responses for tart and sour. Despite best efforts to code as objectively as possible, coding necessarily “involves choices and interpretation (Perdue et al., 2017, p. 273)”, e.g., crab Rangoon could just as plausibly be coded as either “dumplings” or “seafood”. Therefore, in order to guarantee transparency and replicability, we uploaded spreadsheets containing all coding decisions as well as the resulting lists for each pair of synonymic taste words to the OSF.

We inputted, organized, and inspected our coded data using R (R Core Team, 2019) and the packages readxl (Wickham & Bryan, 2023), tidyr (Wickham et al., 2024), dplyr (Wickham et al., 2023), and ggplot2 (Wickham, 2016). Importantly, we filtered our data to include only foods named at least 7 out of 65 times, i.e., with more than a 10% chance of being the initial response across a pair (

Table 2. Foods with at least a 10% chance of being the initial response across a pair.

Pair	Foods
Crispy and crunchy	Chips, bread, cereal, chicken, potatoes, produce
Foamy and frothy	Coffee, beer, cream, milk
Juicy and succulent	Steak, apple, chicken, peach, succulents, watermelon
Sour and tart	Lemons, apple, candy, cherries, grapefruit, tarts
Sugary and sweet	Soda, cake, candy, cookies, ice cream, sweeteners, tea

). This was to ensure that the focus of the analysis lay on the core foods associated with a taste word rather than on more peripheral exemplars.

We used nnet (Venables et al., 2002) and car (Fox & Weisberg, 2019) to fit multinomial logistic regression models with the taste words as the independent variable and their respective foods as the dependent variable (R scripts and plots available on the OSF).

2.5. Results

Multinomial logistic regression models (

Table 3. Responses to crispy versus to crunchy.

Variable	Coefficients		Standard Errors		z		p		Odds Ratio
	(Intercept)	crunchy	(Intercept)	crunchy	(Intercept)	crunchy	(Intercept)	crunchy	(Intercept)	crunchy
Bread	−0.8872507	−1.149643	0.4490929	0.7605888	−1.9756508	−1.511518	0.048	0.131	4.117863e-01	3.167497e-01
Cereal	−2.8331441	1.894840	1.0289865	1.1015464	−2.7533345	1.720164	0.006	0.085	5.882761e-02	6.651486e+00
Chicken	0.3023341	−2.339214	0.3198518	0.6921788	0.9452320	−3.379495	0.345	0.001 *	1.353013e+00	9.640334e-02
Potatoes	−0.8872096	−2.248171	0.4490863	1.1158089	−1.9755882	−2.014835	0.048	0.044 *	4.118032e-01	1.055922e-01
Produce	−10.0345402	9.201591	36.6243461	36.6263049	−0.2739855	0.251229	0.784	0.802	4.385858e-05	9.912891e+03

* significant at α = 0.05. Reference level = chips.

,

Table 4. Responses to foamy versus to frothy.

Variable	Coefficients		Standard Errors		z		p		Odds Ratio
	(Intercept)	frothy	(Intercept)	frothy	(Intercept)	frothy	(Intercept)	frothy	(Intercept)	frothy
Beer	−0.6506158	−0.3187888	0.3561085	0.5022005	−1.827015	−0.6347839	0.068	0.526	0.5217244	0.7270291
Cream	−0.7375887	−1.0202978	0.3665866	0.6073499	−2.012045	−1.6799177	0.044	0.093	0.4782658	0.3604876
Milk	−2.0368901	1.1545102	0.6138505	0.7032976	−3.318219	1.6415670	0.001	0.101	0.1304337	3.1724691

Reference level = coffee.

,

Table 5. Responses to juicy versus to succulent.

Variable	Coefficients		Standard Errors		z		p		Odds Ratio
	(Intercept)	succulent	(Intercept)	succulent	(Intercept)	succulent	(Intercept)	succulent	(Intercept)	succulent
Apple	0.1539458	−2.86215105	0.5563313	1.1731619	0.2767161	−2.43968982	0.782	0.015 *	1.1664276606	5.714570e-02
Chicken	−0.6934319	−0.06869529	0.7071184	0.8423390	−0.9806447	−0.08155302	0.327	0.935	0.4998576530	9.336111e-01
Peach	−0.1825416	−0.91607407	0.6055189	0.7958116	−0.3014631	−1.15111928	0.763	0.250	0.8331499675	4.000867e-01
Succulents	−10.4134969	9.90260565	74.5004544	74.5016476	−0.1397776	0.13291794	0.889	0.894	0.0000300245	1.998237e+04
Watermelon	0.9161278	−13.22887336	0.4830192	121.7962563	1.8966695	−0.10861478	0.058	0.914	2.4995926733	1.797932e-06

* significant at α = 0.05. Reference level = steak.

,

Table 6. Responses to sour versus to tart.

Variable	Coefficients		Standard Errors		z		p		Odds Ratio
	(Intercept)	tart	(Intercept)	tart	(Intercept)	tart	(Intercept)	tart	(Intercept)	tart
Apple	−1.3437448	−0.3301623	0.4584340	0.7784519	−2.9311629	−0.42412677	0.003	0.671	2.608669e-01	7.188071e-01
Candy	−0.6504981	−0.5126180	0.3561085	0.6239524	−1.8266850	−0.82156595	0.068	0.411	5.217858e-01	5.989255e-01
Cherries	−3.1335093	2.5581976	1.0205736	1.1023532	−3.0703414	2.32066968	0.002	0.020 *	4.356465e-02	1.291252e+01
Grapefruit	−1.7492017	0.0752561	0.5417563	0.8302640	−3.2287612	0.09064117	0.001	0.928	1.739127e-01	1.078160e+00
Tarts	−11.1699927	10.5947186	55.5504905	55.5520531	−0.2010782	0.19071696	0.841	0.849	1.409074e-05	3.992343e+04

* significant at α = 0.05. Reference level = lemons.

and

Table 7. Responses to sugary versus to sweet.

Variable	Coefficients		Standard Errors		z		p		Odds Ratio
	(Intercept)	sweet	(Intercept)	sweet	(Intercept)	sweet	(Intercept)	sweet	(Intercept)	sweet
Cake	−2.1400564	3.8448341	0.7475393	1.0722622	−2.862801	3.5857220	0.004	0.000 *	0.11764820	46.750928
Candy	−0.6360057	1.7346854	0.4122322	0.9146686	−1.542833	1.8965179	0.123	0.058	0.52940279	5.667145
Cookies	−0.8873120	0.8873239	0.4490887	1.0962283	−1.975806	0.8094335	0.048	0.418	0.41176107	2.428622
Ice cream	−1.2237777	2.4765987	0.5087458	0.9495778	−2.405480	2.6081050	0.016	0.009 *	0.29411699	11.900717
Sweeteners	−1.7346242	3.2387528	0.6262284	1.0016452	−2.769954	3.2334333	0.006	0.001 *	0.17646650	25.501897
Tea	−2.8332220	3.9318639	1.0289924	1.3135886	−2.753394	2.9932232	0.006	0.003 *	0.05882302	51.001951

* significant at α = 0.05. Reference level = soda.

) and chi-square model fit tests for each pair of synonymic taste words showed significant relationships between the particular taste words and their most frequently named foods.

All fitted models showed a significant improvement over the null models. For crispy versus crunchy, residual deviance went from 330.0758 down to 284.141 (p < 0.001) and nearly 14% of the variability in the elicited foods was explained by the particular taste word (pseudo r2 = 0.1391636, df = 10). For foamy versus frothy, residual deviance went from 263.5227 to 255.3014 (p = 0.042) with approximately 3% of the variability in the foods explained by the taste word (pseudo r2 = 0.03119764, df = 6). For juicy versus succulent, residual deviance went from 252.3711 to 208.4217 (p < 0.001) with approximately 17% of the variability in the foods explained by the taste word (pseudo r2 = 0.174146, df = 10). For sour versus tart, residual deviance went from 286.4948 to 261.2727 (p < 0.001) with nearly 9% of the variability in the foods explained by the taste word (pseudo r2 = 0.08803707, df = 10). For sugary versus sweet, residual deviance went from 331.1808 to 299.7973 (p < 0.001) with approximately 9% of the variability in the foods explained by the taste word (pseudo r2 = 0.09476244, df = 12). Therefore, within each pair, the particular taste word had a significant effect on the elicited foods.

To confirm previous, corpus-based findings that crispy is more associated with foods that have been “dehydrated after the application of heat (Diederich, 2015, p. 194)”, such as chips, while crunchy is more associated with “wet foods as vegetables and fruit … in their natural state (Diederich, 2015, p. 195)”, such as apples, we additionally conducted frequency analyses of all 65 initial responses for crispy and crunchy (Figure 1).

The frequency analyses showed that chips are indeed strongly associated with crispy. However, they are even more strongly associated with crunchy. Furthermore, produce, to which apples belong, is not only less associated with crispy, it is entirely absent in this category.

2.6. Discussion of Experiment 1

Each taste word elicited specific foods, distinguishable from those elicited by a synonymic taste word, thereby indicating a positive answer to RQ1a. Importantly, this was shown even if identical foods emerged within a pair (e.g., chips for crispy and crunchy). These results also suggest that synonymic taste words generally possess subtle semantic differences, which is in line with previous, corpus-based research into crispy and crunchy (Diederich, 2015).

The frequency analyses somewhat support previous findings that crispy is more associated with foods that have been “dehydrated after the application of heat (Diederich, 2015, p. 194)”. For instance, chicken, which cannot be eaten raw and must necessarily undergo some sort of cooking process, i.e., a heat application such as frying or baking, was shown to be the food most strongly associated with crispy. However, apples (and produce in general) were not simply less associated with crispy but entirely absent in this category. Furthermore, contrary to the finding that crunchy is more associated with “wet foods as vegetables and fruit … in their natural state (Diederich, 2015, p. 195)”, chips, which are neither wet nor natural, were found to be the food most strongly associated with crunchy. These discrepancies are in line with research comparing corpus-based patterns to speakers’ psychological reality (e.g., Arppe & Järvikivi, 2007; Dąbrowska, 2014).

Importantly, the frequency analyses force us to reinterpret Johnson and Pfenninger’s (2021) findings that American consumers perceive crunchy chips and crispy apples as better tasting than crispy chips and crunchy apples. As we showed that chips are even more strongly associated with crunchy than with crispy, it seems that Johnson and Pfenninger’s (2021) tasters actually preferred the more (rather than less) associated combination, in line with research into the positive effects of association on perceived taste and food choice (e.g., Okamoto et al., 2009; Peng-Li et al., 2020). In contrast, as the frequency analyses in the present study revealed that crispy is entirely unassociated with produce, crispy apples likely surprised Johnson and Pfenninger’s (2021) tasters, which can result in a positive experience “in situations where consumers are expecting to be surprised” (Johnson & Pfenninger, 2021, p. 89).

3. Experiment 2: Taste Tests

3.1. Procedure

Following Anderson and Barrett (2016, Study 1), each participant tasted and rated pairs of identical samples with differing labels, cleansing their palate with water before sampling each item. The labels consisted solely of the taste word and were displayed both on tabletop sign holders at each station as well as on the individual sample containers (Figure 2). Unlike Anderson and Barrett (2016, Study 1), each participant tasted and rated 8 (rather than 2) samples. Following Johnson and Pfenninger (2021, Experiment 2), the questionnaires read “This product tastes good”. and could be answered on a 7-point Likert scale from strongly disagree (−3) to strongly agree (3). Unlike Johnson and Pfenninger (2021, Experiment 2), each participant first rated the suitability of the taste word by responding to the statement “sour/tart/sugary/sweet/juicy/succulent/foamy/frothy accurately describes this product” on a 7-point Likert scale from strongly disagree (−3) to strongly agree (3). This was carried out to ensure thorough exposure to each taste word, as initially, some participants may not have paid close attention to the labels displayed on the tabletop sign holders or on the samples.

To ensure that participants remained unaware of the linguistic (rather than food-related) manipulation and that participants rated pairs of identical foods against one another, the 8 foods were divided into 2 groups of 4 foods per day. Participants worked at their own pace, going from station to station, tasting and rating samples in the order specified on their questionnaires (5 random orders generated with Word Shuffler Tool, written instructions adapted from Goldberg, 2001).

3.2. Participants and Stimuli

Ninety-three native speakers of American English were recruited from the Food Innovation Center in Portland, Oregon, and assigned to one of two test days (

Table 8. Participants and stimuli on day 1.

Day 1 (N = 52)
Sampled Product	More Associated Label	Less Associated Label
Cherries	tart	sour
Cake	sweet	sugary
Steak	succulent	juicy
Milk	frothy	foamy

and

Table 9. Participants and stimuli on day 2.

Day 2 (N = 41)
Sampled Product	More Associated Label	Less Associated Label
Candy	sour	tart
Soda	sugary	sweet
Watermelon	juicy	succulent
Whipped cream	foamy	frothy

). Per day, we used a within-subjects design, i.e., participants served as their own controls. Therefore, we purposely marked year of birth and gender as optional. All participants and their coded responses can be viewed on the OSF. Each participant received USD 20 for their participation.

The products were all among those with more than a 10% chance of being the initial response across a pair (Table 2) in Experiment 1. Where possible, we combined each taste word with their most frequently named product (see “Category fluency tests” folder on the OSF) to create the more associated condition. For instance, the label sugary was combined with soda, and the label sweet was combined with cake. To create the less associated condition, these labels were reversed, i.e., sugary cake and sweet soda. If a product was the most frequent across a pair of synonymic taste words, which was the case for lemons in the categories sour and tart, the second most frequently named product was used to create the more associated condition. Finally, if the second most frequently named product was also identical across a pair, the third most frequently named product was used, which was the case for foamy and frothy.

We considered additionally retesting crispy and crunchy with more appropriate foods given the results of the frequency analyses, e.g., bread and cereal. However, we decided against this as we wanted to obtain data for as many additional taste words as possible and 8 samples is already assumed to be the limit for any one taster to evaluate at a time (e.g., Heintz & Kader, 1983, p. 21; Mason & Koch, 1953, p. 43) with many studies today limiting samples per person and session to three or four to avoid sensory fatigue (e.g., Singh & Seo, 2020).

3.3. Data Analysis

We analyzed the taste ratings in R (R Core Team, 2019). We inputted, organized, and inspected our data using the packages readxl (Wickham & Bryan, 2023), tidyr (Wickham et al., 2024), and dplyr (Wickham et al., 2023). The descriptive statistics for taste can be seen in

Table 10. Descriptive statistics for taste.

Product	Average Rating with More Associated Label	Average Rating with Less Associated Label
Cherries	0.692	0.25
Cake	1.88	1.40
Steak	1.37	1.31
Milk	0.538	0.769
Candy	1.20	1.63
Soda	−1.05	−1.02
Watermelon	1.85	2.15
Whipped cream	1.94	1.88

. We used ordinal (Christensen, 2023), lmerTest (Kuznetsova et al., 2017), lattice (Sarkar, 2008), and car (Fox & Weisberg, 2019) to fit and diagnose a mixed-effects model using association, product, and interaction term as fixed effects, random effects to account for variability in participants due to the repeated-measures design, and taste ratings as the dependent variable (R script and plots available on the OSF).

3.4. Results

The mixed-effects model showed that more associated taste words have only a marginally significant, positive effect on taste ratings (estimate = 0.48, p = 0.051). Furthermore, there were significant interaction effects for candy (estimate = −0.92, p = 0.013), milk (estimate = −0.71, p = 0.041), and watermelon (estimate = −0.77, p = 0.037). Interestingly, a likelihood ratio test showed that adding association and its interactions did not significantly improve model fit over the null model (p = 0.157), suggesting that there are other factors at play beyond association and specific types of products. As most participants specified their age and gender, we additionally tried adding these variables to the model. However, they did not significantly improve model fit.

3.5. Discussion of Experiment 2

The results showed only a marginally significant, positive effect of more associated taste words on perceived taste. Furthermore, candy, milk, and watermelon tasted significantly worse in the more associated condition. These findings indicate a positive—if not straightforward—answer to RQ1b. They further suggest that by and large, more associated taste words are preferable to less associated taste words, but not for all products. Importantly, it is difficult to establish common denominators for the products that taste better (cherries, cake, steak, soda, whipped cream) versus worse (candy, milk, watermelon) labeled with a more associated taste word. For instance, both groups contain fruits, beverages, and sweets.

Therefore, the unknown variables likely have to do with the tasters. This is in line with food choice research, indicating that there are a myriad of factors from genetic makeup (e.g., Risso et al., 2017) and cultural background (e.g., Jeong & Lee, 2021) to environmental concerns (Rombach et al., 2023) that can influence consumer preferences. But how might consumer preferences cause some products to taste better with a less associated label? A plausible explanation is that the processing of less associated combinations increases cognitive load, which studies show reduces consumers’ attentional resources and thus taste perception (e.g., Van Der Wal & Van Dillen, 2013; Van Meer et al., 2023). Therefore, if our participants were, e.g., mostly crunchers and not chewers (Jeltema et al., 2015), their perception of the soft candy we used may have been dulled by increased cognitive load given a less associated label, thereby improving an undesirable texture.

Alternatively, the frequent, predominantly negative metaphorical uses of the taste words sour, frothy, and juicy could be at play. According to the Contemporary Corpus of American English (COCA, Davies, 2020), sour frequently collocates with nouns such as mood or economy. Furthermore, several studies (e.g., Zawisławska & Falkowska, 2018; Zhang et al., 2025) attest to the negative valence of sour in English metaphors. In contrast, while the synonymic taste word tart can be used metaphorically (e.g., tart reply), the COCA shows that collocations with non-food words are rare, i.e., it is mostly used in a literal sense. Similarly, while the common term frothy market indicates a dangerous disconnect between the prices and the intrinsic value of assets (Frothy Market, n.d.), foamy is used literally, e.g., with nouns such as egg or surf. Finally, the COCA reveals that juicy frequently collocates with nouns such as detail or gossip—with potentially negative connotations related to sensationalism and scandals. In contrast, collocates for succulent (e.g., lamb or garden) indicate that it is used literally to convey high levels of moisture. Therefore, it is plausible that taste words with widespread, predominantly negative metaphorical senses worsen taste experiences under more associated conditions. This is particularly convincing given that sweet, which is often employed metaphorically but positively (e.g., Zhou & Tse, 2020), and sugary, which the COCA shows is predominantly used literally, both adhere to the general preference for more associated combinations.

4. General Discussion

Given Johnson and Pfenninger’s (2021) discovery that crispy and crunchy enhance consumer liking of apples and chips, i.e., foods with which they were presumed less associated, we investigated whether this finding can be generalized. To this end, we asked whether CFTs can uncover distinctions between the foods and beverages associated with synonymic taste words (RQ1a) and if so, whether the perceived taste of identical foods and beverages varies depending on food labels containing more and less associated, synonymic taste words (RQ1b). Our results suggest positive answers to both research questions. In addition, they shed light on previous research.

Experiment 1 showed that how foods are accessed and retrieved given a taste word such as crispy or crunchy deviates from how such taste words are used (Diederich, 2015) and, by extension, that there are semantic differences between synonymic taste words that do not surface in usage. Consequently, the study of synonymic taste words and perhaps even taste words in general should not be limited to usage, as this type of data may not capture every semantic nuance. For instance, it would undoubtedly enrich our understanding of taste words as “affectively loaded part[s] of the English lexicon (Winter, 2016, p. 975)” if studies were to investigate, e.g., whether taste words are processed more quickly in more versus less emotional contexts. Similarly, Experiment 1 revealed that chips are more associated with crunchy than with crispy and that apples are entirely unassociated with crispy. Therefore, Johnson and Pfenninger’s (2021) results should likely not be interpreted as a general preference for foods labeled with less associated taste words. On the contrary, the preference for crunchy chips was likely due to stronger association, whereas the preference for crispy apples was likely due to an expected and thus pleasant surprise given the experimental setting. This is corroborated by Experiment 2, which found a general preference for more associated combinations.

4.1. Limitations and Future Directions

Assuming that less associated taste words increase cognitive load, the main limitation of this study was the laboratory setting. In real-life consumption contexts such as restaurants or grocery-store sampling booths, consumers are confronted with countless distractions, all of which could potentially increase cognitive load. Consumers’ attention may be directed towards product-specific cues such as the shape and color of packaging (e.g., Wang & Chou, 2010) or task-specific cues such as ordering food with an app (e.g., Xu & Huang, 2019). Consumers could also be distracted by their surroundings, e.g., the mélange of different odors or sounds including background music (e.g., Meilleur, 2012, p. 23; Shengze et al., 2022). We nonetheless chose a laboratory setting as it would have been very difficult and time consuming to obtain data on multiple taste words in a real-life setting such as at a restaurant. Firstly, we could only have asked diners to evaluate the labels and tastes of the items they selected, perhaps three to four per person. Secondly, we could not have used diners as their own controls without arousing suspicions as this would have entailed serving two of each of their orders, e.g., one frothy cappuccino and one foamy cappuccino. Consequently, while our results represent new and significant findings for cognitive linguistics, they do not provide directly actionable advice, e.g., for menu or product design.

Another limitation of this study was its inability to include crispy and crunchy in the taste tests in order to avoid sensory fatigue. However, given the large number of participants in Johnson and Pfenninger’s (2021) taste tests showing a preference for crunchy chips as well as the association effect we observed with two separate groups of consumers, our study design seems valid.

Given that certain foods and beverages with seemingly little in common do not adhere to the general preference for more associated combinations of taste words and products, future studies should repeat our CFTs and taste tests with particular segments, e.g., consumers with varying degrees of nutritional knowledge (e.g., Melekoglu, 2023). We also recommend that future studies include more labels with negative metaphorical senses in English, e.g., bitter or salty (Ting et al., 2023; Torres Soler, 2021). Understanding how individual differences and consumer preferences interact with association while additionally controlling for negative metaphorical associations could lead to the discovery of fruitful methods to generate food labels that improve perceived taste, at least for certain types of consumers. Additionally, we recommend that future studies include corpus-based investigations into words for products that are perceived as tastier with less associated food labels. Akin to crispy’s aforementioned “negative value judgment[s] in health contexts (Diederich, 2015, p. 150)”, it is plausible that some products have negative connotations, the potential negative effects of which on perceived taste might be dulled given a less associated label and thus increased cognitive load.

Future studies should also repeat our CFTs and taste tests in other countries and in other languages, as results have been shown to vary even between countries that speak the same language (e.g., Banks & Connell, 2023). In other words, outside the USA, advertising a cake as sweet cannot be assumed to be preferable to advertising it as sugary. Furthermore, not all taste words need to necessarily have meaningful translations in other languages (e.g., Kumar & Chambers, 2019). Even those that do will likely exhibit culture-dependent nuances, both in literal (e.g., Mazzuca & Majid, 2023) and figurative contexts (e.g., Depierre, 2009; López Arroyo & Roberts, 2016). In addition to creating food labels with the potential to increase consumer liking, replication studies in other countries and other languages could provide valuable insights, e.g., into food brand awareness (e.g., Karagiannis et al., 2022) in different markets. Furthermore, such studies could confirm or disprove language-agnostic effects of association on taste perception.

Finally, as corpus data show that the language of sensory perception undergoes semantic shifts over time (e.g., Paccosi et al., 2023; Pettersson-Traba, 2021; Poortvliet, 2017), periodic replication studies would be highly valuable. Supplementing diachronic patterns found in natural data with experimental data, it would become possible, for instance, to map how a fairly new taste word such as umami (e.g., Bieler & Runte, 2010; Ninomiya, 2015) gradually establishes itself.

4.2. Conclusions

While several studies have investigated the influence of descriptions on perceived taste (e.g., Gomez & Spielmann, 2019; Okamoto et al., 2009), ours is the first to analyze and compare the effects of multiple sets of synonymic taste words on taste perception. We succeeded in quickly identifying subtle semantic differences between synonymic taste words using CFTs and in replicating Johnson and Pfenninger’s (2021) taste tests with several pairs of synonymic taste words, as were the goals of this study. Our results suggest that foods and beverages are generally perceived as better tasting when labeled with more associated taste words. However, less associated taste words, the processing of which likely increases cognitive load and thus decreases consumers’ attention to actual taste, may be preferred for disliked products or in the event that more associated taste words commonly feature in negative metaphors. More broadly, our results provide further proof that most synonyms should be regarded as near synonyms and that the study of synonyms should include both corpus-based and experimental data in order to obtain a more complete picture of their associative differences.