Search

Probabilistic models of language understanding are valuable tools for investigating human language use. However, they need to be hand-designed for a particular domain. In contrast, large language models (LLMs) are trained on text that spans a wide array of domains, but they lack the structure and interpretability of probabilistic models. In this paper, we use chain-of-thought prompts to introduce structures from probabilistic models into LLMs. We explore this approach in the case of metaphor understanding. Our chain-of-thought prompts lead language models to infer latent variables and reason about their relationships in order to choose appropriate paraphrases for metaphors. The latent variables and relationships chosen are informed by theories of metaphor understanding from cognitive psychology. We apply these prompts to the two largest versions of GPT-3 and show that they can improve performance in a paraphrase selection task.

Myriad linguistic cues, subtle and otherwise, have been shown to shape thinking and reasoning. Such linguistic framing effects are attributed variously to irrational biases, affective processes, and analogy. Across two studies, we explored pragmatic reasoning (PR) as a potential unifying mechanism underlying disparate framing effects. In Study 1, we adapted a measure of PR used in attribute framing research, assessing the ability to infer information subtly communicated by a speaker’s choice of frame. Our measure captured individual differences in PR distinct from cognitive reflection and mentalizing ability. In Study 2, PR predicted the effect of identity labels on attitudes towards illegal immigration, but not other framing effects. Two such effects were driven, however, by participants’ explicit recognition of the influence of the framing language. These results suggest that framing effects may indeed be mediated by PR, but that our measure of PR does not fully capture its mediating role.

Games have been integral to the development of cognitive science. As experimental tasks, games can provide an ideal environment for studying questions about mental representation, memory, and strategic decision making. Here, we explore the potential of memory matching games as an experimental task by using a simple online version to conceptually replicate two classic effects from the cognitive scientific literature—the picture superiority effect (Paivio & Csapo, 1973) and the word length effect (Baddeley, Thomson, & Buchanan, 1975). We manipulate the Item Format of the game pieces (pictures vs. words) and their Label Length (short vs. long). As expected, we find a picture superiority effect. We do not find the predicted word length effect. We argue that the results of the study, along with several practical properties of the task, sup- port the use of the game for cognitive scientific research.

Metaphors can both shape and reflect how we think about com- plex issues. Here, we explore the prevalence of WAR language in a large corpus (N = 1.63 million) of Tweets. We compare how different groups of people use the language of WAR to talk about different topics. In Study 1, we find that about 5% of Tweets about Covid-19 from the general public include WAR language, replicating prior work (Wicke & Bolognesi, 2020). In Studies 2 and 3, we find that politicians use WAR language much more often, while Reporters and Medical Experts use WAR language less often. The findings are relevant to current debates about the role of language in democracy, and to theo- ries of metaphor in communication.

Intersectionality refers to the simultaneous and interacting effects of multiple group categorization on individuals with minoritized status, often leading to being perceived in a manner inconsistent with the additive contributions of those categories. For Black women, a number of findings have contributed to the idea that Black women have a unique perceived absence of status, for example, and are perceived as distinct from being Black or a woman. We sought to quantify and visualize the combined effects of race and gender on judgments of persons using data-defined dimensions (the Semantic Differential; Osgood et al. in The measurement of meaning, University of Illinois Press, Champaign, 1957). Our data suggest that gender and race contribute to orthogonal dimensions of difference in the perception of persons. Whereas white males, white females, and Black males all seem to be perceived in accord with additive effects in these two dimensions, Black females seem to be perceived more neutrally, as if neither their gender nor their race is treated as predictive.

Across many cultures, similar sensory metaphors are used for similar kinds of personality traits, including words likesweet and bitter, straight and crooked, warm and cold (Asch, 1958). Although such metaphors seem to make sense,our post hoc intuitions may be tainted by confirmation bias. We measured the strength of alignments between each ofa set of nine sensory pairs (e.g., warm/cold) pictured literally, and a set of eight pairs of literal personality concepts(e.g., friendly/aloof) using dual categorization tasks (IATs), and then extracted principal components from these patternsof alignment between sensory and personality concepts across 72 different pairings. The resulting 2D metaphor spaceseemed to reflect something akin to the stereotype content model (Fiske et al., 2002), with axes representing both warmth(PC1: warm/soft) and competence (PC2: bright/high). When we repeated the experiment, with new images and labels,essentially the same structure captured these nine sensory metaphor pairs.

Behavior in classic framing tasks is unrelated to other cognitive bias measures, but little is known about the relation-ship among different types of framing effects. Across two experiments, participants in the US and India completed aclassic risky choice framing task, a metaphor framing task, and measures of cognitive style, linguistic proficiency, andmetaphor usage. We found no relationship between performance on the framing tasks for either sample, suggesting theytap into different underlying processes. Interestingly, language proficiency predicted risky choice framing behavior innative speakers and metaphor framing in non-native speakers. While there was a positive relationship between metaphorusage and metaphor framing for US participants, the sample from India showed a negative relationship, suggesting thatcurrent measures of metaphor usage may assess different behaviors for native versus non-native speakers. Overall, theresults suggest a heterogenous account of the mechanisms underlying framing effects even as they highlight the importantrole of language.

Word aversion is characterized by visceral disgust in responseto seeing or hearing a word. Unlike taboo words or profanity,aversive words do not seem to have an obvious historicalcontext, referent, or pejorative function that causes people toreact negatively to them. “Moist” is a prototypical example ofan aversive word: roughly 20% of American English speakersequate hearing the word with the sound of fingernailsscratching a chalkboard. Despite widespread aversion to“moist,” the word frequently appears on the packaging ofconsumer products like cake, shampoo, and towelettes. Thepresent study tests whether word aversion affects consumerbehavior. We find that moist-averse participants are less liketo choose hygiene-related, but not food-related, products thathave “moist” on the package. We discuss the implications ofthis finding for theories of language processing and disgust inthe context of consumer behavior.

Metaphors can shape how people reason about complex issues,but most studies of metaphor framing rely exclusively onwritten materials. This is a significant limitation, as peopleregularly encounter linguistic metaphors in a variety ofdifferent communicative settings (e.g., read in the newspaper,heard on the radio, or viewed on television). Because researchfinds that variations in communication modality can influencemessage comprehension, retention, and persuasiveness, weexplored the relative power of metaphor framing in differentcommunication modalities. Across two experiments,participants read, heard, or watched a person describe fourdifferent metaphorically framed issues. They had to answer atarget question about each issue by selecting from two responseoptions, one of which was congruent with the metaphor frame.Results revealed a significant, similarly-sized effect ofmetaphor framing in every communication modality,suggesting that communication modality does not moderate theefficacy of metaphor framing.

People often use spatial language to talk about time, and thisis known to both reflect and shape how they think about it.Despite much research on the spatial grounding of temporallanguage and thought, little attention has been given to howspatial metaphors influence reasoning about real events,especially those in the future. In a large online study(N=2362), we framed a discussion of climate change usingspatial metaphors that varied on reference-frame (ego- vs.time-moving), speed of movement (fast vs. slow), and timehorizon (near, medium, or far future). We found thatdescribing climate change as approaching (time-movingframe) – versus something we approach – made the issueseem more serious, but also more tractable, at least when therate of motion was fast (e.g., “it’s rapidly approaching”).These findings offer novel insights into the relationshipbetween spatial metaphors and temporal reasoning and howwe communicate about uncertain future events.

Skepticism towards science has risen sharply in recent years.Cognitive scientists can help address this issue byilluminating how people conceptualize the scientific process,paving the way for improved communication with the public.We recruited a large sample of lay Americans, as well asacademics in the sciences and humanities, to answer a seriesof questions assessing their views about science. Becausemetaphors have been identified as useful tools forcommunicating about complex domains, we askedparticipants to choose which of two metaphors––working on apuzzle or scaling a mountain––best captured their beliefsabout the scientific process. Results revealed substantialvariation in perceptions of science across groups, and wehighlight the ways in which scientists seem to conceptualizescience differently from non-scientists. Importantly, metaphorpreference was associated with particular patterns of thinking,though not always in our originally hypothesized direction.We discuss the implications of these findings.

Metaphors have been shown to be effective explanatory andcommunicative tools, shaping how people think and reasonabout complex domains. To date, however, most studies haveaddressed only coarse-grained effects of metaphor framing,leaving many questions unanswered about the relative powerof metaphor compared to more literal linguistic framingdevices. We addressed this issue in a large, pre-registeredframing study, comparing the effects of describing the role ofpolice officers as (a) metaphorical guardians of a community(b) literal protectors of a community, and (c) a no-labelcontrol. We found no main effect of framing condition,suggesting that positively valenced metaphors may exert littleinfluence on their own in this domain. However, we didobserve an interaction between condition and politicalideology, such that the guardian metaphor was especiallyeffective at improving attitudes towards police officers forliberals, whose initial approval ratings were relatively low.

When faced with a moral dilemma, following your headversus your heart can result in very different decisions. Earlierwork has argued that people who “self-locate” in the headtend to make more rational and less emotional decisions tomoral dilemmas than those who “self-locate” in the heart. Wereplicate this finding, suggest an alternative interpretation ofthe result, and then extend it with a novel experiment. In ametaphor framing task, we manipulated the salience of thehead/heart metaphors—by using them (a) in a single sentence,(b) a more elaborate paragraph, or (c) by emphasizing one incontrast to the other. We found that people who received thehead metaphor made more rational decisions than those whoreceived the heart metaphor, but only in the high saliencecondition that contrasted the two metaphors. This findingillustrates the communicative value of metaphor, which canbe enhanced through comparison.

Popular approaches to modeling analogical reasoning havecaptured a wide range of developmental and cognitivephenomena, but the use of structured symbolicrepresentations makes it difficult to account for the dynamicand context sensitive nature of similarity judgments. Here, theresults of a novel behavioral task are offered as an additionalchallenge for these approaches. Participants were presentedwith a familiar analogy problem (A:B::C:?), but with a twist.Each of the possible completions (D1, D2, D3), could beconsidered valid: There was no unambiguously “correct”answer, but an array of equally good candidates. We find thatparticipants’ recent experience categorizing objects (i.e.,manipulating the salience of the features), systematicallyaffected performance in the ambiguous analogy task. Theresults are consistent with a dynamic, context sensitiveapproach to modeling analogy that continuously updatesfeature weights over the course of experience

Metaphors pervade discussions of critical issues and influ-ence how people reason about these domains. For instance,when crime is a beast, people suggest enforcement-orientedapproaches to crime-reduction (e.g., by augmenting the po-lice force); when crime is a virus, on the other hand, peoplesuggest systemic reforms for the affected community. In thecurrent study, we find that extending metaphoric language intothe descriptions of policy interventions bolsters the persuasiveinfluence of metaphoric frames for an array of important is-sues. When crime is a beast, people are even more likely toendorse “attacking” the problem with harsh enforcement tac-tics; when crime is a virus people are even more likely to toendorse “treating” the problem through social reform.

Metaphors suffuse language and affect how people think. Ameta-analysis of metaphor framing studies conducted between1983 and 2000 concluded that metaphors are about 6% morepersuasive than literal language (Sopory & Dillard, 2002).However, each of these studies was conducted in English withsamples drawn from populations of native English speakers.Here, we test whether and how language proficiency moderatesthe influence of metaphor frames. Sampling from a populationof non-native, but generally proficient English speakers fromIndia, we found that metaphor frames systematically affectedpeople who reported using English primarily in informal con-texts (i.e., among friends and family and through the media)but not those who reported using English primarily in formalcontexts (i.e., for school or work). We discuss the implica-tions of this finding for countries like the US, where Englishis increasingly a non-native language for its residents, and fortheories of language processing more generally.

What makes some metaphors easier to understand than oth-ers? Theoretical accounts of metaphor processing appeal todimensions like conventionality and aptness to explain vari-ability in metaphor comprehensibility. In a typical experiment,one group of naive participants rates a set of metaphoric sen-tences along these dimensions, while another is timed readingthe same sentences. Then, the ratings are used to predict re-sponse times in order to identify the most relevant linguistic di-mension for metaphor comprehension. However, surprisinglyhigh correlations between ratings of theoretically orthogonalconstructs and the results of an experiment in which a con-text manipulation affected ratings of metaphor conventionalityand aptness suggest that these measures should be treated asdependent, rather than explanatory, variables. We discuss theimplications of this perspective for theories of language pro-cessing.

Spatial metaphors for affective valence are common inEnglish, where up in space=happy/positive and down inspace=sad/negative. Past research suggests that thesemetaphors have some measure of psychological reality:people are faster to respond to valenced words and faceswhen they are presented in metaphor-congruent regions ofspace. Here we explore whether the orientation of a stimulus– rather than its position – is sufficient to elicit such spatial-valence congruency effects, and, if so, which spatial referenceframe(s) people use to represent this orientation. InExperiment 1, participants viewed images of happy and sadprofile faces in different orientations and had to identify theemotion depicted in each face. In Experiment 2, participantscompleted this task while lying down on their sides, therebydisassociating environmental and egocentric reference frames.Experiment 1 revealed a metaphor-congruent interactionbetween emotion and orientation, while Experiment 2revealed that this spatial-valence congruency effect was onlyreliable in the environmental frame of reference.

Metaphoric frames are prominently featured in public discourse. They highlight certain aspects of the target issuesthey are used to describe, thereby encouraging specific patterns of inference. Our goal was to test whether they would influencememory as well. Building off prior work, we contrasted two metaphors for crime: virus and beast. In a pilot study, we identifiedspecific causes, examples, and solutions to crime that were congruent with each frame (one but not the other; e.g., people thought“drug use” better exemplified a crime virus, whereas “murder” better exemplified a crime beast). Participants (n = 469) read orlistened to a short metaphorically-framed crime report, completed a filler task, and were prompted for the information they hadseen/heard. Results indicated the virus metaphor facilitated memory, overall, but not the specific frame-congruent information,suggesting a more general influence of the frame than predicted.

People demonstrate systematic logical failures when reasoning about conditional statements. In the Wason selectiontask, a test typically interpreted as a measure of abstract deductive reasoning, only about 10% of participants choose the cardsprescribed by deductive logic. One possibility is that people are simply bad at hypothesis testing – biased toward confirmingrather than falsifying abstract conditional rules. A second possibility, however, is that performance on the task is stronglyinfluenced by pragmatic effects of linguistic interpretation. In three experiments, we find that manipulating the instructions toemphasize falsification and that changing the formulation of the rule to increase the pragmatic salience of the correct choicesimproves performance. These results arise because people do not merely decode the logical content of linguistic expressions.Rather they attempt to understand the communicative intentions of the individual who produced the expression even in abstractreasoning tasks.

Clinical disorders are multidimensional phenomena that areimportant to both clinicians and the lay public, as well as tocognitive scientists interested in understanding how peoplethink and reason about complex domains. To date, however,little work has examined the factors that influence the folkpsychiatry of addiction. Participants in the present study reada brief paragraph about addiction pitched at either an abstractor personal level, followed by a series of questions about thecauses and treatment of drug addiction. We furthermanipulated whether addiction was described using a medicalor psychological label. Results revealed that liberals andconservatives varied dramatically with respect to their folkpsychiatric reasoning, with liberals preferring a morebiological/medical view, which is associated increasedsupport for medical interventions, reduced feelings ofpersonal responsibility, and elevated feelings of stigma.Framing addiction using medical labels and at an abstractlevel pushed people towards this biological view, suggestingthat media reports and messaging campaigns may influencehow people conceptualize addiction.

Metaphors pervade discussions of sociopolitical issues andinfluence the way we think. One challenge facing researchers,however, is that it can be difficult to make principledpredictions about exactly how metaphors will influencethought. Here, we use an explicit comparison task to quantifythe persuasive capacity of metaphors. In Experiment 1, peoplewere given two metaphors and two policy responses. Theywere asked to match one policy to each metaphor. InExperiment 2, people read metaphorically framed issues andchose between policy responses. We found that data from theexplicit comparison task predicted behavior from anothergroup of participants on the metaphor framing task; a measureof linguistic association from LSA did not predict behavior onthe framing task. These results suggest a relationship betweenexplicit analogical comparison and more implicit naturallanguage metaphor processing. It also provides a method formeasuring the conceptual entailments of metaphors

Metaphors pervade discussions of critical issues, making upas much as 10-20% of natural discourse. Recent work hassuggested that these conventional and systematic metaphorsinfluence the way people reason about the issues theydescribe. For instance, Thibodeau & Boroditsky (2011, 2013)found that people were more likely to want to fight backagainst a crime beast by increasing the police force but morelikely to want to diagnose and treat a crime virus throughsocial reform. Here, we report two norming studies and twoexperiments that reveal a shift in the overall landscape ofopinion on the topic of crime. Importantly, we find that themetaphors continue to have an influence on people’sreasoning about crime. Our results and analyses highlight theimportance of up-to-date opinion norms and carefullycontrolled materials in metaphor research.

Is income inequality more of a blemish or a failing organ inour economy? Both metaphors capture something aboutwealth disparities, but only failing organ seems to emphasizethe fact that our economy is a complex system where activityin one region may lead to a cascade of problems in other partsof the system. In the present study, we introduce a novelmethod for classifying such “systemic” metaphors, whichreveals that people can reliably identify the extent to which ametaphor highlights the complex causal structure of a targetdomain. In a second experiment, we asked whether exposingpeople to more systemic metaphors would induce a systemsthinking mindset and influence reasoning on a seeminglyunrelated task. We found that participants who were primedwith systemic metaphors scored higher on subsequent tasksthat measured relational and holistic thinking, supporting theview that these metaphors can promote systems thinking.

Recent work suggests that people think more systematically when using a second language because second languagesare less automatic or emotionally valenced. Here, we use a different population (from India) to further investigate thispossibility. We also test whether nuanced factors like language proficiency, usage context, and age of acquisition affect thedegree to which people show a foreign language effect. We do not find a strong difference between native and second languagespeakers. However, we do find a more nuanced effect of language proficiency: people who are more proficient in the targetlanguage show more loss aversion. We also find that proficient English speakers are more willing to take on risk in both experiments,suggesting that English, itself, may lead people to think differently – possibly because it is a highly agentive languageor because it is associated with individualistic cultural values.

Despite an urgent need to address the issue of obesity, little research has examined the psychological factors that influence support for obesity-related policy initiatives, which represent an important tool for addressing this complex health issue. In the present study, we measured the degree to which people supported obesity-related policy interventions and empathized with a person struggling with obesity after reading a personal account of his or her situation. The narrative described an obese individual who was portrayed as either successfully losing weight or not, and as attributing his or her weight-loss outcome to personal or environmental factors. We found that protagonists who successfully lost weight and/or took personal responsibility for their situation elicited more empathy from participants, which was associated with support for societal policy interventions for obesity. These findings suggest that specific features of personal narratives influence support for obesity-related policies and highlight empathy as a mechanism through which such narratives affect obesity-related attitudes.

Three experiments explored whether conceptual mappings in conventional metaphors are productive, by testing whether the comprehension of novel metaphors was facilitated by first reading conceptually related conventional metaphors. The first experiment, a replication and extension of Keysar et al. [Keysar, B., Shen, Y., Glucksberg, S., Horton, W. (2000). Conventional language: How metaphorical is it "Journal of Memory and Language" 43, 576-593] (Experiment 2), found no such facilitation; however, in the second experiment, upon re-designing and improving the stimulus materials, facilitation was demonstrated. In a final experiment, this facilitation was shown to be specific to the conceptual mappings involved. The authors argue that metaphor productivity provides a communicative advantage and that this may be sufficient to explain the clustering of metaphors into families noted by Lakoff and Johnson [Lakoff & Johnson, M. (1980a). The metaphorical structure of the human conceptual system. "Cognitive Science" 4, 195-208].