Abstract

How do people think about things they can never see or touch? The ability to invent and reason about domains such as time, ideas, or mathematics is uniquely human, and is arguably the hallmark of human sophistication. Yet, how people mentally represent these abstract domains has remained one of the great mysteries of the mind. This chapter explores a potential solution: perhaps the mind recruits old structures for new uses. Perhaps sensory and motor representations that result from physical interactions with the world (e.g., representations of physical space) are recycled to support our thinking about abstract phenomena. This hypothesis is motivated, in part, by patterns observed in language: in order to talk about abstract things, speakers often recruit metaphors from more concrete or perceptually rich domains. For example, English speakers often talk about time using spatial language (e.g., a long vacation; a short meeting). Cognitive linguists have argued such expressions reveal that people conceptualize abstract domains like time metaphorically, in terms of space. Although linguistic evidence for this Conceptual Metaphor Theory is abundant, the necessary nonlinguistic evidence has been elusive; people may talk about time in terms of space, but how can we know whether people really think about time in terms of space? This chapter describes a series of experiments that evaluate Conceptual Metaphor Theory as an account of the evolution and structure of abstract concepts and explore relations between language and nonlinguistic thought, using the abstract domain of time and the relatively concrete domain of space as a testbed. Hypotheses about the way people mentally represent space and time were based on patterns in metaphorical language, but were tested using simple psychophysical tasks with nonlinguistic stimuli and responses. Results of the first set of experiments showed that English speakers incorporate irrelevant spatial information into their estimates of time (but not vice versa), suggesting that people not only talk about time using spatial language, but also think about time using spatial representations. The second set of experiments showed that (a) speakers of different languages rely on different spatial metaphors for duration, (b) the dominant metaphor in participants' first languages strongly predicts their performance on nonlinguistic time estimation tasks, and (c) training participants to use new spatiotemporal metaphors in language changes the way they estimate time. A final set of experiments extends the experimental techniques developed to explore time and space to the domain of musical pitch. Together, these studies demonstrate that the metaphorical language people use to describe abstract phenomena provides a window on their underlying mental representations, and also shapes those representations. The structure of abstract domains such as time appears to depend, in part, on both linguistic experience and on physical experience in perception and motor action. How long will it take you to read this chapter? The objective time, as measured by the clock, might depend on whether you're scrutinizing every detail, or just skimming to get the main ideas. The subjective time might vary according to physiological factors like your pulse and body temperature Time as an abstract domain Although subjective duration is among the earliest topics investigated by experimental psychologists (Mach, 1886), the cognitive sciences have yet to produce a comprehensive theory of how people track the passage of time, or even to agree on a set of principles that consistently govern people's duration estimates. A excerpt from a review by People may estimate filled durations as being longer than empty durations, but sometimes the reverse is found. Duration judgments tend to be shorter if a more difficult task is performed than if an easier task is performed, but again the opposite has also been reported. People usually make longer duration estimates for complex than for simple stimuli, although some researchers have found the opposite. (pg. 12) What makes time perception so difficult to understand? One major reason for the continuing scattering of [researchers'] effort has been that time is treated as if it were a sensory process. If time were a sensory process like vision…we would have an 'organ' of time experience such as the eye. (pg. 34) Although time is not something we can see or touch, we often talk about it as if it were They moved the truck forward two meters. ii. They moved the meeting forward two hours. The truck in sentence i. is a physical object which can move forward through space, and whose motion we might see, hear, or feel, from the staring point to the ending point. By contrast, there is no literal motion described in sentence ii. The meeting is not translated through space, and there is no way to experience its 'movement' through time via the senses. Events that occur in time are more abstract than objects that exist in space insomuch as we typically have richer perceptual evidence for the spatial than for the temporal. 1 In this chapter, I will argue that (a) the language people typically use to talk about duration reveals important links between the abstract domain of time and the relatively concrete domain of space, (b) temporal representation must be understood, in part, in terms of spatial representations, and (c) the domains of space and time provide a testbed for hypotheses about the evolution and structure of abstract concepts. 1.2 Metaphor and the problem of abstract thought The mystery of how people come to mentally represent abstract domains such as time, ideas, or mathematics has engaged scholars for centuries, sometimes leading to proposals that seem unscientific by modern standards. Plato (Meno, ca. 380 B.C.E.) argued that we cannot acquire abstract concepts like virtue through instruction, and since babies are not born knowing them, it must be that we recover such concepts from previous incarnations of our souls. Charles Darwin contended that evolution can explain the emergence of abstract thought without recourse to reincarnation, yet it is not immediately obvious how mental capacities that would have been superfluous for our Pleistocene forebears could have been selected for. What selection pressures could have 4 resulted in our ability to compose symphonies, invent calculus, or imagine time travel? How did foragers become physicists in an eyeblink of evolutionary time? The human capacity for abstract thought seems to far exceed what could have benefited our predecessors, yet natural selection can only effect changes that are immediately useful. The apparent evolutionary uselessness of human intelligence drove Alfred Wallace, Darwin's co-founder of the theory of evolution by natural selection, to abandon their theory and invoke creationism to explain our capacity for abstract thought (Darwin, 1859 2 Darwin's own formulation of evolutionary theory points toward an elegant potential solution to Wallace's dilemma: sometimes organisms recycle old structures for new uses. An organ built via selection for a specific role may be fortuitously suited to perform other unselected roles, as well. For example, the fossil record suggests that feathers were not originally 'designed' for flying. Rather, they evolved to regulate body temperature in small running dinosaurs, and were only later co-opted for flight Can exaptation account for mental abilities in humans that could not have been selected for directly? If so, how might this have happened? Which adapted capacities might abstract domains be exapted from? Steven Pinker (1997) sketched the following proposal: Suppose ancestral circuits for reasoning about space and force were copied, the copies' connections to the eyes and muscles were severed, and references to the physical world were bleached out. The circuits could serve as a scaffolding whose slots are filled with symbols for more abstract concerns like states, possessions, ideas, and desires. (pg. 355) As evidence that abstract domains arose from circuits designed for reasoning about the physical world, Pinker appeals to patterns observed in language. Many linguists have noted that when people talk about states, possessions, ideas, and desires, they do so by co-opting the language of intuitive physics 1l a high shelf 1m a high price 2l a big building 2m a big debate 3l forcing the door 3m forcing the issue 4l pushing the button 4m pushing the limit 5l keeping the roof up 5m keeping appearances up 5 For each pair above, sentence l illustrates a literal use and sentence m a metaphorical use of the italicized words. The concrete objects described in the literal sentences (e.g., shelf, building, door, button, roof) belong to a different ontological category than the abstract entities in the metaphorical examples, according a test of what physical relations they can sensibly be said to enter into. For example, it is sensible to say "the cat sat on the shelf / building / door / button / roof", but it may not be sensible to say that "the cat sat on the price / debate / issue / limit / appearance." This test is similar to a test of sensible predicates for concrete vs. abstract entities devised by Fred Based on examples like these, linguists have argued that people create abstract domains by importing structure from concepts grounded in physical experience. Although anticipated by others (e.g., The psychological claim behind [Gruber's linguistic discovery] is that the mind does not manufacture abstract concepts out of thin air…it adapts machinery that is already there, both in the development of the individual organism and in the evolutionary development of the species. (1983, pg. 188-9) Not all theorists agree on the significance of metaphorical language for theories of mental representation. Gregory 1.3 Evidence for Conceptual Metaphor The idea that conventionalized metaphors in language reveal the structure of abstract concepts is often associated with linguist George Lakoff and philosopher Mark Johnson, who described Conceptual Metaphor theory as one of "three major findings of cognitive science" (1999, pg. 3). Yet, this claim was supported almost entirely by linguistic evidence, supplemented by a computational model providing an in principle demonstration of how the meanings of some linguistic metaphors might be learned and represented Boroditsky If people use spatial schemas to think about time, as suggested by metaphors in language, then do people who use different kinds of spatiotemporal metaphors in their native tongues think about time differently? To find out, Boroditsky (2001) compared performance on space-time priming tasks in speakers of English, a language which typically describes time as horizontal, and speakers of Mandarin Chinese, which also commonly uses vertical spatiotemporal metaphors. English speakers were faster to judge sentences about temporal succession (e.g., March comes earlier than April) when primed with a horizontal spatial event, but Mandarin speakers were faster to judge the same sentences when primed with a vertical spatial stimulus. This was true despite the fact that all of the sentences were presented in English. In a follow-up study, Boroditsky (2001) trained English speakers to use vertical metaphors for temporal succession (e.g., March is above April). After training, their priming results resembled those of the native Mandarin speakers. Together, Boroditsky's studies provide some of the first evidence that (a) people not only talk about time in terms of space, they also think about it that way, (b) people who use different spatiotemporal metaphors also think about time differently, and (c) learning new spatial metaphors can change the way you mentally represent time. Yet, these conclusions are subject to a skeptical interpretation. Boroditsky's participants made judgments about sentences containing spatial or temporal language. Perhaps their judgments showed relations between spatial and temporal thinking that were consistent with linguistic metaphors only because they were required to process space or time in language. Would the same relations between representations of space and time be found if participants were tested on nonlinguistic tasks? The fact that people communicate via language replete with anaphora, ambiguity, metonymy, sarcasm, and deixis seems proof that what we say provides only a thumbnail sketch of what we think. Most theorists posit at least some independence between semantic representations and underlying conceptual representations Any utterance is a selective schematization of a concept -a schematization that is in some ways dependent on the grammaticized meanings of the 7 speaker's particular language, recruited for the purposes of verbal expression. (pg. 75-76) Slobin argues that when people are "thinking for speaking" (and presumably for reading or listening to speech), their thoughts are structured, in part, according to their language and its peculiarities. Consequently, speakers of different languages may think differently when they are using language. But how about when people are not thinking for speaking? Eve Clark [When people are] thinking for remembering, thinking for categorizing, or one of the many other tasks in which we may call on the representations we have of objects or events -then their representations may well include a lot of material not customarily encoded in their language. It seems plausible to assume that such conceptual representations are nearer to being universal than the representations we draw on for speaking. (pg. 21) Clark predicts that results may differ dramatically between tests of language-thought relations that use language and those that do not: …we should find that in tasks that require reference to representations in memory that don't make use of any linguistic expression, people who speak different languages will respond in similar, or even identical, ways. That is, representations for nonlinguistic purposes may differ very little across cultures or languages. (2003, pg. 22) Clark adds: Of course, finding the appropriate tasks to check on this without any appeal to language may prove difficult. (2003, pg. 22) Clark's skepticism echoes concerns raised by Papafrougou, Massey, and Gleitman (2002) regarding the difficulty of studying the language-thought interface: …domains within which language might interestingly influence thought are higher-level cognitive representations and processes, for instance, the linguistic encoding of time […] A severe difficulty in investigating how language interfaces thought at these more "significant" and "abstract" levels has been their intractability to assessment. As so often, the deeper and more culturally resonant the cognitive or social function, the harder it is to capture it with the measurement and categorization tools available to psychologists. (pg. 191-192) For the studies reported here, new experimental tools were developed in order to (a) evaluate Conceptual Metaphor Theory as an account of the structure and evolution of abstract concepts, and (b) to investigate relations between language and nonlinguistic thought, using the domains of space and time as a testbed. Specifically, the goal of the first set of experiments was to determine whether English speakers' nonlinguistic mental representations of space and time are related in ways predicted by linguistic metaphors. The second set tested whether nonlinguistic mental representations of time differ among speakers of different languages, in ways consistent with their languageparticular metaphors. The third set of experiments investigated whether language can cause differences in the nonlinguistic mental representations of time to arise among speakers of different languages. A final set of experiments sought to generalize these findings beyond the domain of time. These experiments used novel psychophysical tasks with nonlinguistic stimuli and responses in order to distinguish two theoretical positions, one which posits shallow and the other deep relations between language and nonlinguistic thought (see Do people think about time in terms of space? Do people use mental representations of space in order to mentally represent time, as metaphors in language suggest they do --even when they're not using language? The first six experiments reported here test the hypothesis that temporal thinking depends, in part, on spatial thinking. In each task, participants viewed simple nonlinguistic, nonsymbolic stimuli (i.e., lines or dots) on a computer screen, and estimated either their duration or their spatial displacement. Durations and displacements were fully crossed, so there was no correlation between the spatial and temporal components of the stimuli. As such, one stimulus dimension served as a distractor for the other: an irrelevant piece of information that could potentially interfere with task performance. Patterns of crossdimensional interference were analyzed to reveal relationships between spatial and temporal representations. Broadly speaking, there are three possible relationships between people's mental representations of space and time. First, the two domains could be symmetrically dependent. John Locke (1689/1995) argued that space and time are mutually inextricable in our minds, concluding that, "expansion and duration do mutually embrace and comprehend each other; every part of space being in every part of duration, and every part of duration in every part of expansion" (p. 140). Alternatively, our ideas of space and time could be independent. Any apparent relatedness could be due to structural similarities between essentially unrelated domains These three possible relations between space and time predict three distinct patterns of cross-dimensional interference. If spatial and temporal representations are symmetrically dependent on one another, then any cross-dimensional interference should be approximately symmetric: line displacement should modulate estimates of line duration, and vice versa. Alternatively, if spatial and temporal representations are independent, there should be no significant cross-dimensional interference. However, if mental representations of time are asymmetrically dependent on mental representations of space, as suggested by spatiotemporal metaphors in language, then any crossdimensional interference should be asymmetric: line displacement should affect estimates of line duration more than line duration affects estimates of line displacement. For Experiment 1, native English speaking participants viewed 162 lines of varying lengths (200-800 pixels, in 50 pixel increments), presented on a computer monitor for varying durations (1-5 seconds, in 500 ms increments). Lines 'grew' horizontally from left to right, one pixel at a time, along the vertical midline. Each line remained on the screen until it reached its maximum displacement, and then disappeared. Immediately after each line was shown, a prompt appeared indicating that the participant should reproduce either the line's displacement (if an 'X' icon appeared) or its duration (if an 'hourglass' icon appeared), by clicking the mouse to indicate the endpoints of each temporal or spatial interval. Space trials and time trials were randomly intermixed. Results of Experiment 1 showed that spatial displacement affected estimates of duration, but duration did not affect estimates of spatial displacement (figure 1a). For stimuli of the same average duration, lines that travelled a shorter distance were judged to take a shorter time, and lines that travelled a longer distance were judged to take a longer time. Subjects incorporated irrelevant spatial information into their temporal estimates, but not vice versa. Estimates of duration and displacement were highly 9 accurate, and were equally accurate in the two domains. The asymmetric crossdimensional interference we observe cannot be attributed to a difference in the accuracy of duration and displacement estimations, as no significant difference in was found. Experiments were conducted to assess the generality of these results, and to evaluate potential explanations. In Experiment 1, participants did not know until after each line was presented whether they would need to estimate displacement or duration. They had to attend to both the spatial and temporal dimensions of the stimulus. Experiment 2 addressed the possibility that cross-dimensional interference would diminish if participants were given the opportunity to attend selectively to the trialrelevant stimulus dimension, and to ignore the trial-irrelevant dimension. Materials and procedures were identical to those used in Experiment 1, with one exception. A cue preceded each growing line, indicating which stimulus dimension participants would need to reproduce. Results of Experiment 2 (figure 1b) replicated those of Experiment 1. Participants were able to disregard line duration when estimating displacement. By contrast, they were unable to ignore line displacement, even when they were encouraged to selectively attend to duration. The cross-dimensional effect of space on time estimation in Experiment 1 was not caused by a task-specific demand for subjects to encode spatial and temporal information simultaneously. Experiments 3-5 addressed concerns that spatial information in the stimulus may have been more stable or more salient than temporal information, and that differences in stability or salience produced the asymmetrical cross-dimensional interference observed in Experiments 1 and 2. One concern was that participants may have relied on spatial information to make temporal estimates because stimuli were situated in a constant spatial frame of reference (i.e., the computer monitor). For Experiment 3, stimuli were also situated in a constant temporal frame of reference. Temporal delay periods were introduced preceding and following line presentations, which were proportional to the spatial gaps between the ends of the stimulus lines and the edges of the monitor. Results (figure 1c) replicated those of Experiments 1 and 2. Experiment 4 addressed the possibility that space would no longer influence participants' time estimates if stimulus duration were indexed by something non-spatial. For this experiment, a constant tone (260 Hz) accompanied each growing line. Materials and procedures were otherwise identical to those used in Experiment 2. The tone began sounding when the line started to grow across the screen, and stopped sounding when the line disappeared. Thus, stimulus duration was made available to the participant in both the visual and auditory modalities, but stimulus displacement was only available visually. Results (figure 1d) replicated those of the previous experiments. Displacement strongly influenced participants' duration estimates, even when temporal information was provided via a different sensory modality from the spatial information. Experiment 5 was designed to equate the mnemonic demands of the spatial and temporal dimensions of the stimulus. Materials and procedures were identical to those used in Experiment 2, with one exception. Rather than viewing a growing line, subjects viewed a dot (10x10 pixels) that moved horizontally across the midline of the screen. In the previous experiments, just before each growing line disappeared participants could see its full spatial extent, from end to end, seemingly at a glance. By contrast, the spatial extent of a moving dot's path could never be seen all at once, rather it had to be imagined: in order to compute the distance that a dot travelled, participants had to retrieve the dot's starting point from memory once its ending point was reached. The spatial and temporal dimensions of the dot stimulus had to be processed similarly in this regard: whenever we compute the extent of a temporal interval we must retrieve its starting point from memory once the end of the interval is reached. Results (figure 1e) replicated those of previous experiments. Experiment 6 investigated whether motion or speed affected participants' time estimates in Experiments 1-5, rather than stimulus displacement. Materials and 10 procedures were identical to those used in Experiment 2, with the following exception. Rather than growing lines, participants viewed stationary lines, and estimated either their duration or displacement. Results (figure 1f) replicate those of previous five experiments, indicating that stimulus displacement can strongly modulate time estimates even in the absence of stimulus motion. [INSERT FIGURE 1 ABOUT HERE] Results of all six experiments unequivocally support the hypothesis that people incorporate spatial information into their time judgments more than they incorporate temporal information into their spatial judgments. These findings converge with those of Over the past century of psychophysical experimentation on space and time judgments, two findings have emerged repeatedly: the Kappa effect and the Tau effect Although we describe time in terms of space almost obligatorily (Jackendoff, 1983; Pinker, 1997), we can also optionally describe space in terms of time. For example, in English we could say my brothers live 5 minutes apart to indicate that they live a short distance apart. Thus, the relation between time and space in linguistic metaphors is asymmetrical, but not unidirectional. Accordingly, we predicted asymmetrical cross-dimensional interference between space and time. This prediction does not entail that time can never affect spatial judgments: only that the effect of space on time estimation would be greater than the effect of time on space estimation in our tasks. We did not observe any significant effect of time on distance estimation, but such a finding would still be compatible with our hypothesis, so long as we also found a significantly greater effect of distance on time estimation. We show such a significant 11 difference between the cross-dimensional effects of space-on-time and time-on-space in Experiments 1-6. Although no theory on offer can fully explain Tau and Kappa effects There is no reason to suspect that participants imputed illusory speed to our stimuli. In Experiments 1-5, the actual speed was given by the stimuli, and in Experiment 6, there was no motion or speed information in stimuli at all, real or implied. Furthermore, there is no reason to believe that participants' expectations of constant velocity were violated, given that all moving stimuli moved at a constant velocity, all of our stimuli were spatiotemporally continuous, and none of our judgments required comparisons between successive intervals. If imputing speed to discontinuous successive intervals accounts for Tau effects, and our stimuli do not require participants to impute speed or judge successive intervals, then we should not expect to find Tau-like effects (which, indeed, we do not). By the same token, we should not expect to find Kappa-like effects in our studies if these effects were driven by imputed speed. The effects of distance on time estimation that we report are importantly different from Kappa effects, insomuch as they do not depend on speed --real or imputed. Rather, our data show that distance, per se, affects time estimation, independent of speed; partial correlations reveal that the effect of actual distance on reported time remains statistically significant even when the effect of actual speed is removed. The existence of the Tau effect in no way conflicts with our finding of a spacetime asymmetry. Both Tau and Kappa effects are extensively documented, but the relative strengths of these effects have never been explicitly compared in appropriate tests; whether these effects are symmetrical or asymmetrical remains an open question. A survey of the literature suggests that the Kappa effect may be more robust, as it has been demonstrated in a greater variety of tasks, and over a greater range of spatiotemporal intervals. Importantly, however, the question of symmetry for Tau and Kappa effects is not directly relevant to our studies if indeed Tau and Kappa effects reveal influences of imputed speed on time and space estimation, whereas the tasks described here test for interference between temporal and spatial information, per se. It is noteworthy that space influenced temporal judgments even for spatiotemporal stimuli that participants could experience directly. Growing lines are observable, and are arguably less abstract than entities like the 'moving meeting' described in section 1.1. Brief durations could, in principle, be mentally represented independently of space, by an interval-timer or pulse-accumulator (see 3: Does language shape the way we think about time? The first set of experiments supports the Deep View of language-thought relations by showing that temporal representations depend, in part, on spatial representations, as predicted by metaphors in English --even when people are performing low-level, nonlinguistic psychophysical tasks (see Are our own concepts of 'time,' 'space,' and 'matter' given in substantially the same form by experience to all men, or are they in part conditioned by the structure of particular languages?" (1939/2000, pg. 138.) This Whorfian question remains the subject of renewed interest and debate. Does language shape thought? The answer yes would call for a reexamination of the 'universalist' assumption that has guided Cognitive Science for decades, according to which nonlinguistic concepts are formed independently of the words that name them, and are invariant across languages and cultures Skepticism about some Whorfian claims has been well founded (see Pinker, 1994, ch. 3, for a review of evidence against the Whorfian hypothesis). A notorious fallacy, attributable in part to Whorf, illustrates the need for methodological rigor. Whorf (1939Whorf ( /2000 argued that Eskimos must conceive of snow differently than English speakers because the Eskimo lexicon contains multiple words that distinguish different types of snow, whereas English has only one word to describe all types. The exact number of snow words the Eskimos were purported to have is not clear. This number has now been inflated by the popular press to as many as four-hundred. According to a Western Greenlandic Eskimo dictionary published in Whorf's time, however, Eskimos may have had as few as two distinct words for snow Setting aside Whorf's imprecision and the media's exaggeration, there remains a critical missing link between Whorf's data and his conclusions: Whorf (like many researchers today) used purely linguistic data to support inferences about nonlinguistic mental representations. Steven Pinker illustrates the resulting circularity of Whorf's claim in this parody of his logic: 13 [They] speak differently so they must think differently. How do we know that they think differently? Just listen to the way they speak! (Pinker, 1994, pg. 61). Such circularity would be escaped if nonlinguistic evidence could be produced to show that two groups of speakers who talk differently also think differently in corresponding ways. We conducted a series of experiments to explore relations between spatiotemporal language and nonlinguistic mental representation of time. The first experiment, a corpus search, uncovered previously unexplored cross-linguistic differences in spatial metaphors for duration. Next, we tested whether these linguistic differences correlate with differences in speakers' low-level, nonlinguistic time representations. Experiment 7 compared the use of 'time as distance' and 'time as volume' metaphors across six languages. Every language we examined uses both distance and volume metaphors, but their relative prevalence and productivity appear to vary markedly. In English, it is natural to talk about a long time, borrowing the structure and vocabulary of a linear spatial expression like a long rope. Yet in Spanish, the direct translation of 'long time', largo tiempo, sounds awkward to speakers of most dialects. 6 Mucho tiempo, which means 'much time', is preferred. In Greek, the words makris and kontos are the literal equivalents of the English spatial terms long and short. They can be used in spatial contexts much the way long and short are used in English (e.g., ena makry skoini means 'a long rope'). In temporal contexts, however, makris and kontos are dispreferred in instances where long and short would be used naturally in English. It would be unnatural to translate a long meeting literally as mia makria synantisi. Rather than using distance terms, Greek speakers typically indicate that an event lasted a long time using megalos, which in spatial contexts means physically 'large' (e.g., a big building), or using poli, which in spatial contexts means 'much' (e.g., much water). Compare how English and Greek typically modify the duration of the following events (literal translations in parentheses): 1e. long night 1g. megali nychta (big night) 2e. long relationship 2g. megali schesi (big relationship) 3e. long party 3g. parti pou kratise poli (party that lasts much) 4e. long meeting 4g. synantisi pou diekese poli (meeting that lasts much) 14 In examples 1g. and 2g., the literal translations might surprise an English speaker, for whom big night is likely to mean 'an exciting night', and big relationship 'an important relationship'. For Greek speakers, however, these phrases communicate duration, expressing time not in terms of 1-dimensional linear space, but rather in terms of 3-dimensional volume. To quantify the relative prevalence of distance and volume metaphors for duration across languages, the most natural phrases expressing the ideas 'a long time' and 'much time' were elicited from native speakers of English (long time, much time), French (longtemps, beaucoup de temps), Greek (makry kroniko diastima, poli ora), Indonesian (waktu panjang, waktu banyak), Italian (lungo tempo, molto tempo), and Spanish (largo tiempo, mucho tiempo). The frequencies of these expressions were compared in a very large multilingual text corpus: www.google.com. Each expression was entered as a search term. Google's language tools were used to find exact matches for each expression, and to restrict the search to web pages written only in the appropriate languages. The number of google 'hits' for each expression was tabulated, and the proportion of distance hits and volume hits was calculated for each pair of expressions, as a measure of their relative frequency. Results showed that in English, French, and Indonesian, distance metaphors were dramatically more frequent than volume metaphors. The opposite pattern was found in Greek, Italian, and Spanish (figure 2). Although all languages surveyed use both distance and volume metaphors for duration, the relative strengths of these metaphors appears to vary across languages. This simple corpus search by no means captures all of the complexities of how time is metaphorized in terms of space within or between languages, but these findings corroborate native speakers' intuitions for each language, and provide a quantitative linguistic measure on which to base predictions about behavior in nonlinguistic tasks. [INSERT FIGURE 2 ABOUT HERE] 3.2 Do people who talk differently think differently? Do people who use different spatiotemporal metaphors think about time differentlyeven when they're not using language? Experiments 8 and 9 explored the possibility that speakers who preferentially use distance metaphors in language tend to co-opt linear spatial representations to understand duration, whereas speakers who preferentially use volume metaphors tend to co-opt 3-dimensional spatial representations. Speakers of four languages surveyed in Experiment 7 performed a pair of nonlinguistic psychophysical tasks, which required them to estimate duration while overcoming different kinds of spatial interference (i.e., distance or volume interference). If people's conceptions of time are substantially the same universally irrespective of the languages they speak, as suggested by the Shallow View, then performance on these tasks should not differ between language groups. On the Deep View, however, it was predicted that participants' performance should vary in ways that parallel the metaphors in their native languages. The 'distance interference' task was modeled on the 'growing line' task described in Experiment 2. English participants in the previous growing line studies may have suffered interference from distance during duration estimation, in part, because distance and duration are strongly conflated in the English lexicon. Would the same confusion be found in speakers of other languages? It was predicted that speakers of 'Distance Languages' (i.e., English and Indonesian) would show a strong effect of 15 distance on time estimation when performing the growing line task, whereas speakers of 'Volume Languages' (i.e., Spanish and Greek) would show a weaker effect. A complementary 'volume interference' task was developed, in which participants watched a schematically drawn container of water filling up, one row of pixels at a time, and estimated either how full it became or how much time it remained on the computer screen, using mouse clicks as in the growing line tasks. Spatial and temporal parameters of the stimuli were equated across tasks. Behavioral predictions for the Filling Tank task were the mirror image of predictions for the Growing Line task: speakers of Volume Languages should show a considerable influence of 'fullness' on time estimation, whereas speakers of Distance Languages should show a milder effect. Results showed that effects of spatial interference on duration estimation followed predictions based on the relative prevalence of distance and volume metaphors for time in speakers' native languages. English and Indonesian speakers showed a strong effect of distance but a weak effect of volume on duration estimation; Greek and Spanish speakers showed the opposite pattern of results (figure 3). A 4 x 2 factorial ANOVA compared the slopes of the effects of target distance and target fullness on time estimation, with Language (English, Indonesian, Greek, Spanish) and Task (Growing Lines, Filling Tanks) as between-subject factors. Results showed a highly significant Language by Task interaction (F (3,126) = 4.82, p<0.003), with no main effects, signaling a true crossover interaction. [INSERT FIGURE 3 ABOUT HERE] The observed differences in the effects of distance and volume on duration estimation cannot be attributed to overall differences in performance across tasks or across groups. Within-domain performance (i.e., the effect of target duration on estimated duration; the effect of target distance or fullness on estimated distance or fullness) was compared across tasks and across groups: no significant differences were found between correlations or slopes, even in pairwise comparisons. One difference between the Growing Line and Filling Tank tasks was that the lines grew horizontally, but the tanks filled vertically. To determine whether the spatial orientation of the stimuli and responses gave rise to the observed cross-linguistic differences in performance on the Growing Lines and Filling Tank tasks, an Upward Growing Lines task was administered to speakers of English, Indonesian, Greek, and Spanish. No significant difference was found in the effect of vertical displacement on time estimation across languages, suggesting that the orientation of stimuli cannot account for the between-group differences observed in Experiments 8 and 9. Two further analyses were conducted to explore the relation between language and time estimation. First, data from speakers of Distance Languages (English, Indonesian) were pooled, and compared with pooled data from speakers of Volume Languages (Greek, Spanish). This analysis was important for distinguishing effects of language from more general effects of culture. Based on the metaphors in their native languages, participants who presumably had different ethnic, educational, and socioeconomic profiles were placed in the same group, while participants with more similar profiles were placed in different groups. For example, based on the linguistic metaphors in their first languages, English speaking MIT students were placed in a different group from Spanish speaking MIT students, but in the same group as Indonesian speakers sampled from the general population in Jakarta. A 2 x 2 ANOVA compared the slopes of the effects of target distance and target fullness on time estimation, with Language (Distance Language, Volume Language) and Task (Growing 16 Lines, Filling Tanks) as between-subject factors. As before in the analysis of individual languages, results showed a highly significant Language by Task interaction Finally, an analysis was conducted to quantify the relation between linguistic metaphors and performance on the nonlinguistic time estimation tasks. Using the corpus data from Experiment 7, an asymmetry ratio (AR) was computed in order to express the relative prevalence of distance and volume metaphors in each language as a value on a scale from -1 to 1: (proportion distance metaphors -proportion volume metaphors) AR_metaphors = (proportion distance metaphors + proportion volume metaphors) A positive AR_metaphors indicated a preference for distance metaphors and a negative AR_metaphors indicated a preference for volume metaphors, according to the corpus data. Likewise, an asymmetry ratio was computed to express the relative effects of distance and volume interference on time estimation for speakers of each language as a value on a scale from -1 to 1: (effect of distance on time estimation -effect of volume on time estimation) AR_slopes = (effect of distance on time estimation + effect of volume on time estimation) A positive AR_slopes indicated a greater effect of distance interference on time estimation, and a negative AR_slopes indicated a greater effect of volume interference on time estimation, according to the data from the Growing Line and Filling Tank tasks (Experiments 8 and 9). The asymmetry of slopes for speakers of English, Indonesian, Spanish, and Greek was plotted as a function of the asymmetry of metaphors in these languages, and a nonparametric correlation was computed. Results showed a perfect rank order correlation (Kendall's Tau_b = 1.00, p<.02), demonstrating a strong positive association between linguistic metaphors for duration and nonlinguistic mental representations of time. Overall, Experiments 7-9 show that the way people talk about time correlates strongly with the way they think about it, even when they're performing simple nonlinguistic perceptuo-motor tasks, as predicted by the Deep View of languagethought relations. (See How might perceptual and linguistic experience shape abstract thought? How do people come to think about time in terms of space? How do speakers of different languages come to conceptualize time differently? Turning to the first question, some mappings from concrete to abstract domains of knowledge may be initially established pre-linguistically, based on interactions with the physical world If metaphorical mappings are experience-based, and are established prelinguistically, what role might language play in shaping abstract thought? Since the laws of physics are the same in all language communities, pre-linguistic children's conceptual mappings between time, distance, and volume could be the same universally. Later, as children acquire language, these mappings could be adjusted: each time we use a linguistic metaphor, we may invoke the corresponding conceptual mapping. Speakers of Distance Languages then would invoke the time-distance mapping frequently, eventually strengthening it at the expense of the time-volume mapping (and vice versa for speakers of Volume Languages). Did language experience give rise to the language-related differences in performance reported for the Growing Line and Filling Tank experiments? A perennial complaint about studies that purport to show effects of language on thought is that researchers mistake correlation for causation. Although it is difficult to imagine what nonlinguistic cultural or environmental factors could have caused performance on Experiments 8 and 9 in English, Indonesian, Greek, and Spanish speakers to align so uncannily with the metaphors in these languages, the data are nevertheless correlational. Using cross-linguistic data to test for a causal influence of language on thought is problematic, since experimenters cannot randomly assign subjects to have one first language or another. For Experiment 10, a pair of training tasks was conducted to provide an in principle demonstration that language can influence even the kinds of low-level mental representations that people construct while performing psychophysical tasks, and to test the hypothesis that language shapes time representations in natural settings by adjusting the strengths of cross-domain mappings. Native English speakers were randomly assigned to perform either a Distance Training or Volume Training task. Participants completed 192 fill-in-the-blank sentences using the words longer or shorter for Distance Training, and more or less for the Volume Training task. Half of the sentences compared the length or capacity of physical objects (e.g., An alley is longer / shorter 18 than a clothesline; A teaspoon is more / less than an ocean), the other half compared the duration of events (e.g., A sneeze is longer / shorter than a vacation; A sneeze is more / less than a vacation). By using distance terms to compare event durations, English speakers were reinforcing the already preferred source-target mapping between distance and time. By using volume terms, English speakers were describing event durations similarly to speakers of a Volume Language (see Greek examples in section 3.1), and by hypothesis, they were invoking the dispreferred volume-time mapping. After this linguistic training, all participants performed the nonlinguistic Filling Tank task from Experiment 9. We predicted that if using a linguistic metaphor invokes the corresponding conceptual mapping between source and target domains, then repeatedly using volume metaphors during training should (transiently) strengthen participants' nonlinguistic volume-time mapping. Consistent with this prediction, the slope of the effect of volume on time estimation was significantly greater after volume training than after distance training (difference of slopes = 0.89, t(28) = 1.73, p<.05; These findings help to resolve apparent tensions between the proposal that perceptuo-motor image schemas underlie our abstract concepts and the notion of linguistic relativity. Johnson (2005) defines an image schema as "a dynamic recurring pattern of organism-environment interactions" (pg. 19). Presumably, people from all language communities inhabit the same physical world and interact with their environment using the same perceptuo-motor capacities, therefore the image schemas they develop should be universal. Yet, even if we all develop similar image schemas based on our physical experiences, Experiments 8-10 suggest the concepts that these image schemas undergird are not immutable. Duration can be understood both in terms of distance and in terms of volume. The extent to which one of these conceptual spacetime mappings is invoked in a given speaker or community of speakers varies with the strength of the corresponding linguistic metaphor. The structure of abstract concepts like duration appears to be shaped both by perceptuo-motor experience (which is plausibly universal) and by language use (which is culture-specific). [INSERT FIGURE 4 ABOUT HERE] Beyond space and time Time and space provide a model system for exploring connections between abstract and concrete mental representations, but time is just one among many domains that we spatialize in language; time may be one of many abstract domains that import their structure or content, in part, from the domain of space. In Experiment 11, the psychophysical tasks we developed to investigate space and time were adapted to explore relations between space and musical pitch. Like time, pitch is often described in English using linear spatial terms. Unlike time, pitch tends to be described using vertical rather than horizontal metaphors. Pitches can be high or low, and can rise, fall, soar, or dip below the staff. Yet, the fact that we talk about pitch in terms of vertical space doesn't necessarily mean that we think about it that way. One possibility is that pitch is mentally represented on its own terms, and is only coded into the same words that we use to describe space as a matter of 19 convenience: domains that share structural similarities may be amenable to common linguistic description, obviating multiple domain-specific vocabularies. Alternatively, the spatialization of pitch in language may serve as a clue that leads us to a fuller understanding of how pitch is mentally represented. The 'growing line' task described in Experiment 2 was modified for a nonlinguistic test of the hypothesis that our mental representations of musical pitch depend, in part, on spatial representations. Nine displacements ranging from 100 to 500 pixels (in 50 pixel increments) were fully crossed with nine different pitches ranging from middle C4 to G#4 (in semitone increments). For each trial, participants heard a constant pitch while watching a line grow up the screen from bottom to top (for half of the subjects) or across the screen from left to right (for the other half of the subjects). Before each stimulus, participants were informed whether they would need to estimate distance or pitch, to encourage them to attend to the trial-relevant stimulus dimension and, if possible, to ignore the trial-irrelevant dimension. Participants estimated line displacements using mouse clicks, as in previous experiments. To estimate pitch, participants used the mouse to adjust a probe tone until it matched the remembered target pitch. Watching vertical lines significantly modulated subjects' pitch estimates: tones of the same average frequency were judged to be higher in pitch if they accompanied lines that grew higher on the screen (effect of actual distance on estimated pitch: slope=.37; r 2 =.77, p<.003). By contrast, watching horizontal lines did not significantly modulate pitch estimates. This finding is consistent with the occurrence of vertical but not horizontal metaphors for pitch in English. Further analyses showed that whereas vertical displacement affected estimates of pitch, pitch did not significantly influence estimates of vertical displacement. Thus, the relation between nonlinguistic mental representations of space and pitch appears asymmetrical, as predicted by the directionality of space-pitch metaphors in language. While these results support the claim that musical pitch is mentally represented in part metaphorically, in terms of vertical space, they are agnostic as to the direction of causation between language and thought. Further studies (such as those described in sections 3.1-3.3) are needed to investigate whether linguistic metaphors merely reflect the spatial schemas underlying pitch representations, or whether the way we talk about pitch can also shape the way we think about it. Conclusions Together, the experiments described in this chapter suggest that people not only talk about abstract domains using spatial words, they also think about them using spatial representations. Results are incompatible with the Shallow View of language-thought relations, and provide some of the first evidence for the view that language has Deep influences on nonlinguistic mental representation (see These findings are difficult to reconcile with a universalist position according to which language calls upon nonlinguistic concepts that are presumed to be "universal" (Pinker, 1994, pg. 82) and "immutable" (Papafragou, Massey, & Gleitman, 2002, pg. 216). Beyond influencing thinking for speaking Direct evidence that spatial cognition supported the evolution of abstract concepts may forever elude us, because human history cannot be recreated in the laboratory, and the mind leaves no fossil record. However, the studies reported here demonstrate the importance of spatial representations for abstract thinking in the mind that evolution produced. Previously, inferences about the perceptual foundations of abstract thought have rested principally on linguistic and psycholinguistic data. The psychophysical experiments presented here show that even nonlinguistic representations in the domains of space and time are related as predicted by linguistic metaphors, and as expected if the more abstract domain arose as an exaptation from the more concrete. The structure of abstract domains like time appears to depend, in part, on both perceptuo-motor and linguistic experience.