Substrates - Princeton University

Snowden and Verstraten – Motion transparency

and Psychophysical Perspectives (Watanabe, T., ed.), pp. 187–211, MIT Press

Review

of Motion (Smith, A.T. and Snowden, R.J., eds), pp. 253–290, Academic Press 44 Koechlin, E., Anton, J.L. and Burnod, Y. (1999) Bayesian inference in

38 Knill, D.C. and Richards, W., eds (1996) Perception as Bayesian Inference, Cambridge University Press

populations of cortical neurons: a model of motion integration and segregation in area MT Biol. Cybern. 80, 25–44

39 Nowlan, S.J. and Sejnowski, T.J. (1994) Filter selection model for motion segmentation and velocity integration J. Opt. Soc. Am. Ser. A 11, 3177–3200

45 Snowden, R.J. and Braddick, O.J. (1989) The combination of motion signals over time Vis. Res. 29, 1621–1630 46 Smith, A.T., Curran, W. and Braddick, O.J. (1999) What motion

40 Nowlan, S.J. and Sejnowski, T.J. (1995) A selection model for motion processing in area MT of primates J. Neurosci. 15, 1195–1214

distributions yield global transparency and spatial segmentation? Vis. Res. 39, 1121–1132

41 Simoncelli, E.P. and Heeger, D.J. (1998) A model of neuronal responses in visual area MT Vis. Res. 38, 743–761

47 Braddick, O.J. et al. (1998) Brain areas differentially activated by coherent visual motion and dynamic noise NeuroImage 7, S322

42 Stoner, G.R., Albright, T.D. and Ramachandran, V.S. (1990) Transparency and coherence in human motion perception Nature 344, 153–155 43 Stoner, G.R. and Albright, T.D. (1994) Visual motion integration: a neurophysiological and psychophysical perspective, in Visual Detection

48 Marshak, W. and Sekuler, R. (1979) Mutual repulsion between moving visual targets Science 205, 1399–1401 49 Niemann, T., Ilg, U.J. and Hoffmann, K.P. (1994) Eye movements elicited by transparent stimuli Exp. Brain Res. 98, 314–322

The neuroethology of primate vocal communication: substrates for the evolution of speech Asif A. Ghazanfar and Marc D. Hauser In this article, we review behavioral and neurobiological studies of the perception and use of species-specific vocalizations by non-human primates. At the behavioral level, primate vocal perception shares many features with speech perception by humans. These features include a left-hemisphere bias towards conspecific vocalizations, the use of temporal features for identifying different calls, and the use of calls to refer to objects and events in the environment. The putative neural bases for some of these behaviors have been revealed by recent studies of the primate auditory and prefrontal cortices. These studies also suggest homologies with the human language circuitry. Thus, a synthesis of cognitive, ethological and neurobiological approaches to primate vocal behavior is likely to yield the richest understanding of the neural bases of speech perception, and might also shed light on the evolutionary precursors to language.

T

he species-specific vocalizations of non-human primates are crucial for their social interactions, reproductive success and survival1,2, and some have argued that speech has played a similar role in human history3. Investigating the perception and social use of vocalizations in extant non-human primates might be the most direct route to understanding the substrates underlying the evolution of speech and language. It follows, therefore, that investigating the neural processes underlying the vocal behavior of primates might yield important insights into the neurobiology of speech.

Neuroethological research has already added much to our understanding of how natural selection shapes brain-design for complex behaviors such as echolocation in bats4, song learning in birds5–7, and mate-choice in frogs8. Likewise, in the visual behavior of primates, faces are highly relevant stimuli in their day-to-day social interactions and specialized regions of the temporal lobes appear to be dedicated to face processing9. Based on the consistency with which behavioral adaptations are mediated by specialized neural systems in the animal kingdom, we hypothesize that the design of the

1364-6613/99/$ – see front matter © 1999 Elsevier Science Ltd. All rights reserved.

PII: S1364-6613(99)01379-0

Trends in Cognitive Sciences – Vol. 3, No. 10,

October 1999

A.A. Ghazanfar and M.D. Hauser are at the Primate Cognitive Neuroscience Laboratory, Department of Psychology, 33 Kirkland Street, Harvard University, Cambridge, MA 02138, USA. tel: +1 617 495 6575 fax: +1 617 496 7077 e-mail: aghazanf@ wjh.harvard.edu; [email protected]. edu

377

Review

Ghazanfar and Hauser – Primate vocal communication

primate auditory system should reflect the specialized functions that it has evolved to carry out: communication between conspecifics. It stands to reason that neurobiological studies of primate auditory function must adopt a cognitive ethological perspective2,10. Recent work in this area has revealed remarkable similarities between primate vocal behavior and human speech. Many primates have vocalizations that appear to be functionally referential in that they convey meaningful information about objects and events in their environments11–13. Some primate species show hemispheric biases in the perception of conspecific vocalizations14,15, attend to the temporal aspects of such calls16,17, and show plasticity in the usage and comprehension of their species-typical calls18. In this paper, we review current understanding of the vocal behavior of non-human primates (hereafter, ‘primates’), with a special focus on problems that can benefit from neurobiological investigations as well as inform cognitive ethologists about neurobiologically significant problems. We hope to convince the reader that primate vocal behavior represents an excellent model system for studying the substrates of speech perception in particular, and the neural basis of complex biological signal recognition in general. The ethology of primate vocal communication: problems for the neurosciences Temporal processing of vocal signals Humans use temporal cues such as the duration, interval and order of acoustic features to distinguish among categories of speech sounds19,20. For example, humans distinguish /pa/ from /ba/ on the basis of voice onset time, and /sa/ from /sta/ on the basis of the silent time between consonants and vowels. Based on these data and studies of language-impaired children, it has been suggested that speech perception is based on the rapid processing of temporal information21. Do primates perceive their own vocalizations in a categorical fashion, and if so, do they base their perceptual classifications on temporal features? In an early study of primate communication, Green characterized the vocal repertoire of the Japanese macaque (Macaca fuscata)22. He found that Japanese macaques used subtle differences in the acoustic structure of their calls to distinguish between types that covary with particular contexts. For example, among the affiliative ‘coo’ call, the ‘smooth early high’ (SE) type was given by young individuals isolated from companions, while the ‘smooth late high’ (SL) type was given by subordinate animals to more dominant troop members. Green’s diagnostic for distinguishing these types was the temporal position of the fundamental frequency peak, which in the SE coo occurred in the first two-thirds of the call, and in the SL coo occurred in the final third22. This temporal cue could be an acoustic feature that Japanese macaques use to distinguish between these two calls. As humans use specialized neural circuitry to parse speech sounds based on temporal cues21, perceptual experiments on Japanese macaques investigated whether they too are able to discriminate these two call types using temporal features of the signal17,23,24. Japanese macaques, as well as several closely related Old World monkeys, were trained on two discrimination tasks, one using peak fundamental frequency position and the other using initial frequency as the relevant parameter.

Results revealed that Japanese macaques, but not the other species, were better able to discriminate between the two coo types using peak position than using the initial frequency. In contrast, the other species performed better using the initial frequency of the coos than using peak position23,24. When individual acoustic elements of each coo vocalization were edited out, results showed that for Japanese macaques, the temporal position of the peak frequency was the most significant feature for discriminating between the coos17. Thus, at least one non-human primate species appears to have specialized neural mechanisms for categorizing conspecific vocalizations on the basis of temporal cues. Behavioral asymmetries in the processing of vocal signals A classic feature of language processing is its neural lateralization. Although there is no clear general dichotomy of function between the cerebral hemispheres, perceptual experiments, studies of brain-damaged patients and functional imaging studies have indicated that speech perception is usually lateralized to the left temporal lobe21. One indication of left-hemispheric specialization for speech processing in intact humans is the performance advantage exhibited by the right ear for the identification of speech sounds, and the lack of, or left-ear, advantage for non-speech sounds25. Behavioral experiments under laboratory and field conditions reveal that primates also exhibit similar asymmetries in the perception of their vocalizations. Data from the perceptual experiments on Japanese macaques described above, and from other studies, revealed a right-ear advantage when discriminating coo calls according to their species-specific, communicative relevance. No ear advantage was shown for discriminations based on initial frequency15,23. Based on the fact that the coos were of functional significance to the Japanese macaques, and that the other primate species tested showed no ear advantage in their performance (save for one vervet monkey, Cercopithecus aethiops), the observed asymmetry in perception could be attributed to the communicative valence of the signals and not particular acoustic characteristics. To explore further the problem of hemispheric asymmetries in acoustic perception, Hauser and colleagues ran a series of field experiments on the closely related rhesus monkey (M. mulatta). Playback experiments using 12 different call types associated with coarse-grained emotional states (affiliation, aggression and submission), and several socioecological contexts, revealed that most adult rhesus monkeys turned their heads right when orienting towards a vocalization emitted from a hidden speaker placed 180 degrees behind them14 (see Fig. 1). Thus, this species also shows a right-ear bias in the perception of conspecific vocalizations14,16. Hauser and colleagues suggested that this orienting bias is the result of left-hemisphere dominance for the processing of conspecific vocalizations. Interestingly, infant rhesus monkeys failed to show any orienting bias to conspecific calls (Fig. 1), similar to the absence of perceptual asymmetries of speech sounds in human children with specific language impairment21. The relationship between temporal cues and neural lateralization can be revealed by manipulating vocal signals and measuring behavioral performance. Using the dichotic listening paradigm, the magnitude of the right-ear advantage

378 Trends in Cognitive Sciences – Vol. 3, No. 10,

October 1999

Ghazanfar and Hauser – Primate vocal communication

Review

exhibited by humans can be altered by speeding up or slowing down formant transitions within a speech syllable26. Similarly, temporally manipulating rhesus-monkey calls can abolish the right-ear bias observed in the field14,16. By contracting or expanding the inter-pulse duration of three pulsatile vocalizations beyond the species-typical range, the right-ear orienting bias for conspecific vocalizations was abolished for two of the three vocalizations (‘grunt’ and ‘shrill bark’), resulting in a no-ear or left-ear bias16. These results support the hypothesis that, like human brains, some non-human primate brains are specialized to process conspecific vocalizations, and that efficiently parsing temporal cues might be one of the critical processing components. Referential communication by monkeys in their natural environments One of the most important features of human language is its ability to refer to objects and events in the external world. Until 1980, it had been assumed that pritrends in Cognitive Sciences mate vocalizations simply reflected the Fig. 1. Left-hemisphere bias towards conspecific vocalizations. Rhesus monkeys (M. mulatta) were tested when seated at one of three food dispensers, which provided a consistent context for testing. Vocalizations were caller’s emotional state and nothing played from a hidden speaker placed directly behind the subjects (A). The response assay was to score whether more27,28; different call types were associthe subjects turned the right or left ear in the direction of the speaker. The graph plots the proportion of adult ated with different emotional states (e.g. (B) and 4–12-month-old infant (C) rhesus monkeys turning the right ear towards the speaker in response to three screams for fear or barks for aggression). types of conspecific vocalizations (black bars) and one control heterospecific call (the turnstone’s alarm call, a sound the subjects were familiar with) (white bars). Adult, but not infant, monkeys showed a right-ear bias in reOver the last two decades, however, there sponse to conspecific vocalizations. (Data from Ref. 14.) has been an accumulation of data supporting the claim that many primate vocalizations are functionally referential, providing listeners with vocalizations to refer to different kinds of food, and indiinformation about food, predators, and social relationships. viduals respond to such calls as if in search of food, often The clearest example of functionally referential signals is calling back with comparable calls36,37. 11,29 the vervet monkey’s alarm-call system . Vervets produce While data on referential signaling in primates reveals some similarities to human words, there are also fundamental acoustically distinct alarm calls to their various predators differences between these two systems. Unlike human words, (snakes, eagles, leopards, baboons, humans and small carnithere is no evidence that primate calls can reference either vores)29. In response to such calls, vervets behave adaptively, the past or the future. Primate vocalizations typically refer to responding as a function of the predator’s species-typical events or objects that are in the present. Furthermore, there hunting strategy. For example, when one vervet emits a snake is no evidence that whole calls or parts of calls can be strung alarm call, other vervets immediately inspect the ground together to produce more complex utterances with different around them; following an eagle alarm call, they look up meanings in the way words or parts of words can2. For priand/or run into a dense bush, presumably to avoid the eagle’s stoop. Playback studies have shown that the acoustics of the mates, the call appears to be the primary unit of analysis. alarm call alone are sufficient to elicit predator-specific Nonetheless, the fact that some non-human primates have adaptive responses11. Thus, vervet alarm calls are functionthe capacity to produce a rudimentary form of referential signal provides an avenue for looking at this system from a ally referential signals that convey information about both cognitive neuroscience perspective, one aimed at revealing predator type and the caller’s affective state. the neural substrates underlying conceptual representations. These pioneering experiments prepared the way for several other examples of referential signaling by primates in their natural environments12,30–34. In rhesus and pig-tailed macaques, Perceptual versus conceptual mechanisms of classification In language, two words, such as ‘soda’ and ‘pop’, can mean the field observations and playback experiments have revealed same thing but have very different acoustic features. This is that individuals produce one of five acoustically distinct ‘reone sense in which the acoustic properties of a word are arcruitment screams’ in order to elicit aid from allies during bitrary relative to its meaning. Although humans can certainly agonistic encounters31,35. The particular call used specifies hear the difference between ‘soda’ and ‘pop’, they preferenthe particular class of opponent and the level of physical tially attend to the referential similarity between these words. aggression in the encounter. In a different context, toque Primates also exhibit this capacity12,32,38. (M. sinica) and rhesus macaques give acoustically distinct

379 Trends in Cognitive Sciences – Vol. 3, No. 10,

October 1999

Review

Ghazanfar and Hauser – Primate vocal communication

trends in Cognitive Sciences

Fig. 2. Functional referents and acoustic similarity. (A) Spectrograms of the calls used by rhesus monkeys that have discovered food. The ‘warble’ and the ‘harmonic arch’ are given by individuals that have found rare, high-quality foods. The ‘grunt’ is given by individuals who have found a common food of lower quality. (B–C) Within-referent habituation experiments: (B) subjects were habituated to ‘harmonic arches’ and then tested with a ‘warble’; (C) subjects were habituated to ‘warbles’ and then tested with a ‘harmonic arch’. (D–E) Between-referent habituation experiments: (D) subjects were habituated to ‘grunts’ and then tested with either a ‘warble’ or a ‘harmonic arch’; (E) subjects were habituated to either ‘warbles’ or ‘harmonic arches’ and then tested with a ‘grunt’. The results show that, after habituation, monkeys respond only to changes in referential features (between-referent conditions) of subsequent calls, and not to changes in acoustic features alone (within-referent conditions). Data represent the mean time individuals spent looking in the direction of the speaker (error bars represent 6SD). All dishabituation effects were statistically significant. (Data from Ref. 12.)

Rhesus monkeys produce two call types (‘harmonic arches’ and ‘warbles’) when they find high quality/rare food, and produce another call type (‘grunts’) when they find low quality/common food12 (Fig. 2A). To determine whether rhesus classify these calls as a function of their acoustic or referential similarities, an habituation–dishabituation procedure was used. If rhesus monkeys attend to acoustic features alone, then having been habituated to multiple exemplars of one call type, they should respond to a test playback involving either of the other two call types. In contrast, if they attend to referential features, then having been habituated to one call type they should respond to a call type from a different class or category, but not respond to a call type from the same class. Results provide strong support for the referential hypothesis. In particular, when rhesus monkeys were habituated to harmonic arches, they failed to respond to warbles (Fig. 2B). Similarly, when they were habituated to warbles, they failed to respond to harmonic arches (Fig. 2C). However, when they were habituated to either warbles or harmonic arches, they responded to grunts. Interestingly, when they were habituated to grunts, they also responded to

harmonic arches and warbles, but the magnitude of the response was substantially greater than in the opposite habituation–dishabituation order (Fig. 2D,E). This asymmetry in the pattern of response suggests that rhesus attend to the putative reference of the call when classifying stimuli. These general findings have been obtained for two other species (vervet and Diana monkeys) and three other contexts (social affiliation, aggressive inter-group encounters, and predator alarm)32,38. Together, these studies provide considerable support for the notion that, like humans, primates can use their vocalizations to refer functionally to objects or events in the external environment. Several primates respond to the call’s referent rather than to its acoustic morphology alone, suggesting that they have a representation of the external referent and not a simple conditioned response to the acoustic signal. Monkeys also store a representation of caller identity and use this information to guide their responses. The neural bases for these representations and decision processes have yet to be explored, but with the behavioral data in hand, such primates are ideally suited for neurobiological exploration.

380 Trends in Cognitive Sciences – Vol. 3, No. 10,

October 1999

Ghazanfar and Hauser – Primate vocal communication

Review

The neurobiology of primate vocal communication: a brief review and prospectus The wealth of data on the vocal behavior of primates provides an unprecedented foundation for investigating the neurobiological mechanisms underlying auditory processing. As illustrated by the neuroethological studies of birdsong, bat echolocation and face processing in primates, research on the neurobiology of primate auditory function might profit from using the animal’s species-specific vocalizations to explore the brain’s design features. Homologous substrates for homologous behaviors The region of neocortex that responds trends in Cognitive Sciences most robustly to auditory stimuli lies in Fig. 3. Organization of the auditory cortex. (A) Lateral view of the left hemisphere of the rhesus monkey cerand around the superior temporal plane ebral cortex. Shaded region corresponds to the approximate location of the auditory-related areas on the exposed and superior temporal gyrus of the temposurface of the superior temporal gyrus. (B) View of the superior temporal plane, which contains the ‘core’ and ral lobe (Fig. 3A,B). This region has been ‘belt’ areas of auditory cortex. In this representation, the overlying dorsal bank of the lateral sulcus is graphically broadly subdivided into three areas each reflected upwards to reveal the ventral bank. (C) Lateral view of the rhesus brain indicating some of the major connections between the rostral and caudal parabelt regions with subdivisions of the prefrontal cortex. representing a different level of cortical (Abbreviations: AS, arcuate sulcus; CS, central sulcus; CPB, caudal parabelt; IPS, intraparietal sulcus; LS, lateral sulprocessing: the core, belt and parabelt39 cus; LOS, lateral orbital sulcus; LuS, lunate sulcus; PS, principal sulcus, RPB, rostral parabelt; STG, superior temporal (Fig. 3B). Neurons in the core area regyrus; STS, superior temporal sulcus; wm, white matter.) (Adapted from Ref. 66.) spond best to simple acoustic stimuli such as tones, while belt and parabelt neurons respond best to more complex stimuli. Our anatomical the functional organization of auditory cortex, they nevertheknowledge of these areas derives mainly from data collected less provided substantial evidence that auditory neurons were on species of the Old World monkey genus, Macaca. tuned to species-specific vocalizations. More recent experiHowever, based on tonotopic organization and intracortical ments in identified subdivisions of auditory cortex of anesconnections, several New World monkey species40–43 appear thetized rhesus and marmoset monkeys largely support the results from squirrel monkeys, demonstrating that cortical to have at least a subset of core auditory cortical areas that neurons selectively respond to conspecific vocalizations with are homologous to those found in macaques. complex temporal patterns of firing47–49. In comparison with the human temporal lobe, there is cytoarchitectonic evidence that macaques and humans share Given that auditory cortical neurons can be call-seleca number of auditory cortical areas44. More recently, the use of tive, how is selectivity built up by neural circuits? One approach to answering these questions involves presenting multiple staining techniques to delineate and compare directly acoustically manipulated vocal stimuli. With the advent of the architecture of auditory cortex in macaques, chimpanzees sophisticated digital-signal technology for bioacousticians50, and humans has revealed similarities both in the architecture and shape of auditory areas in these primates. These data sugit is possible to alter systematically specific features of a call gest that at least some stages of auditory cortical processing and then use such perturbed signals to determine how commight be similar45. While these comparative neuroanatomical ponents of the call affect neural response patterns. In rhesus macaques, filtering certain frequencies of a call results in less data are first approximations at best, they provide a convincrobust responses from call-selective neurons when compared ing rationale for the application of knowledge from primate with responses to normal, intact vocalizations47. Similarly, in auditory behavioral and neurobiological studies to humans. the temporal domain, it has been shown that editing out parts Neural processing of spectro-temporally-manipulated of, or reversing, vocalizations used as stimuli results in a drop vocal signals in neuronal responsiveness for call-selective neurons in squirTo date, the squirrel monkey represents the most extensively rel monkeys46, marmosets49, and rhesus monkeys51. Together, studied mammalian model system for the auditory processthese data suggest that neurons in the auditory cortex of ing of species-specific vocalizations. Recordings of single-unit primates are ‘combination-sensitive’ (i.e. they respond nonactivity in the superior temporal gyrus of the awake squirrel linearly) to conspecific vocalizations in the same way that monkey revealed that more than 80–90% of the neurons in neurons in the songbird forebrain and bat auditory cortex this region responded differentially to species-specific vocalizare combination-sensitive to their own vocalizations4,6. 46 ations used as stimuli . Although the relative lack of inforSeveral questions remain concerning the behavioral relevance of the combination-selectivity of primate auditory neurmation regarding squirrel monkey cortical architectonic ons. For example, how do spectro-temporal manipulations boundaries in these studies limits what one can say about

381 Trends in Cognitive Sciences – Vol. 3, No. 10,

October 1999

Review

Ghazanfar and Hauser – Primate vocal communication

that affect responses at the neural level affect responses at the perceptual level, and vice versa. Does removing or extending particular portions of vocalizations affect how subjects respond to them? And how are the temporal manipulations of vocalizations that influence behavior16 processed by callselective neurons? A final point to consider here is that the relative importance of the temporal as opposed to the spectral domain might differ depending on call type or the information to be extracted, a possibility that can be explored both at the neural and behavioral levels. Neural correlates of behavioral asymmetries The behavioral asymmetries discussed earlier are supported by both neuroanatomical and experimental lesion studies. For human subjects that show functional left-hemispheric biases for language processing, it has been shown that the Sylvian fissure (bordering auditory cortex) is significantly longer in the left hemisphere than in the right52. It is assumed that the length of the fissure corresponds to the size of auditory cortex. Using this measurement, potential anatomical asymmetries have similarly been measured in several species of primates. Left Sylvian-fissure length was found to be significantly greater than in the right hemisphere in apes (Pan, Gorilla, and Pongo)53, Old World macaques (M. fascicularis and M. mulatta), and New World Callichtrids (cotton-top tamarins

Outstanding questions • Like adult humans, some species of adult non-human primates exhibit a right-ear bias for processing conspecific vocalizations. These asymmetries are matched by neuroanatomical asymmetries in the temporal lobe. Similar behavioral biases are absent in some children with specific language impairments as well as in some primate infants. A primate model, such as the rhesus monkey, might allow us to explore the development of these hemispheric biases at both the behavioral and neuronal level. With the advent of better staining techniques, the size of different auditory areas can be measured and a more detailed analysis of which particular areas are related to the developmental onset of orienting asymmetries carried out. This circuitry can also be explored with multi-electrode recording techniques, and recent advances in functional imaging techniques. • As in speech, primate calls are multidimensional: one call can represent many things at the same time, including a specific object or event (i.e. the call’s referent), the caller’s identity (e.g. sex, species), and the caller’s emotional state (e.g. aggressive, fearful). There is a need to investigate how the acoustic morphology of calls relates to these information channels. Using digital sound synthesis and manipulation, different acoustic features can be manipulated and then used to test behavioral responses. For example, an aggressive call from a large dominant male could be made to sound as if it were produced by a smaller individual by shifting formant frequencies. These manipulated calls could then be tested both in the field and the lab to measure the perceptual and neural correlates, respectively. With the current technology, such neuroethological experiments are quite feasible. • The functional organization of the primate auditory cortex has not been determined beyond the tonotopical mapping of the core areas. In the bat auditory cortex, neurons that are sensitive to distinct features carried in the biosonar signal are clustered together. An understanding of the information-bearing parameters of primate calls might lead to similar insights into the organization of primate auditory cortex. For example, what is the organization of neural responses to atonal versus tonal vocalizations in higher-order auditory cortical areas? Do the different categories of vocalizations, such as food calls versus alarm calls, map to different auditory cortical areas? For example, do food calls map to auditory cortical areas that project to the orbitofrontal cortex given this region’s apparent association with the reward system?

and marmosets)54. These cerebral asymmetries might represent the ‘specialized’ neural circuitry that mediates the behavioral asymmetries to conspecific vocalizations14,15. Further evidence of specialization comes from experiments on Japanese macaques. Following lesions of the left auditory cortex, subjects exhibited a selective impairment for discriminating species-specific vocalizations but not other types of auditory stimuli55. In particular, subjects’ performance on discriminating SE (‘smooth early high’) and SL (‘smooth late high’) coos was greatly impaired following lesions of the left superior temporal gyrus, but was unimpaired by similar lesions in the right hemisphere. Thus, the left auditory cortex of Japanese macaques appears to be specialized for processing communicative signals. Similar experiments on rhesus macaques, which also show a left-hemisphere bias to conspecific vocalizations, would be beneficial. For example, pharmacological inactivation of specific subdivisions of the auditory cortex (e.g. injections of muscimol) could be combined with playback experiments (e.g. monaural presentation or head-orienting task) to investigate which particular cortical areas are specialized for communication signals. Auditory–prefrontal cortical interactions When a non-human primate hears a call from a conspecific, its response will depend upon the identity of the caller, his distance, the current context, and the message conveyed. As in human interactions, there are times when it is appropriate for a primate to respond to a call and there are times when it must withhold a response. The prefrontal cortex has been implicated in such ‘response inhibition’56. In the auditory domain, rhesus monkeys can be trained to reach into a box for a food pellet after hearing one (positive) tone, or to withhold this response upon hearing a different (negative) tone. However, lesions of the inferior frontal convexity (see Fig. 3C), the region lying between the principal and lateral orbital sulci, result in perseverative interference in the performance of this auditory go/no-go task (i.e. they were unable to withhold their responses to the negative tone)57. The rostral ‘belt’ and ‘parabelt’ regions of macaque auditory cortex project to this region of the prefrontal cortex39 (Fig. 3C), and it seems likely that this auditory–prefrontal circuitry plays a pivotal role in controlling responses to behaviorally relevant signals. Indeed, recent neurophysiological experiments have shown that neurons in the inferior convexity are selective to faces58 and species-specific vocalizations59. Field observations and playback experiments have shown that when a call is heard, primates will almost invariably (unless habituated) look towards the source of the signal and/or its referent (e.g. a predator). Higher-order auditory areas of the caudal ‘belt’ and ‘parabelt’ regions, which are responsive to vocalizations47 and sound source location60, send projections to the periarcuate region of the prefrontal cortex39. This circuit might be involved in the decision process and the control of eye movements to auditory targets, such as a group member or a predator. Neurons in this area are involved in selective attention, are responsive to auditory stimuli, and have been shown to encode the association of visual and auditory stimuli (see, for example, Ref. 61). The frontal eye field (FEF) lies within this area and plays an important role in the selection and control of eye movements to particular targets in the environment62.

382 Trends in Cognitive Sciences – Vol. 3, No. 10,

October 1999

Ghazanfar and Hauser – Primate vocal communication

Moreover, it has been demonstrated that FEF neurons are active during eye movements to both aurally and visually guided eye movements63. It is likely that this neural circuitry underlies the eye movement responses of primates to the source of vocalizations and other relevant sounds.

Review

4 Suga, N. et al. (1987) The personalized auditory cortex of the mustached bat: adaptation for echolocation J. Neurophysiol. 58, 643–654 5 Marler, P. (1970) Birdsong and speech development: could there be parallels? Am. Sci. 58, 669–673 6 Margoliash, D. and Fortune, E.S. (1992) Temporal and harmonic combination-sensitive neurons in the zebra finch’s HVc J. Neurosci. 12, 4309–4326

Future directions: the neurobiology of call meaning Field experiments on several species of monkeys have provided evidence that individuals often appear to ignore large acoustic differences between two calls, focusing primarily on the call’s referent12,32,38. It might be possible to simulate the habituation–dishabituation paradigm used in the field to explore the underlying neurophysiology of the representation of call meaning in the laboratory – in other words, to measure the habituation–dishabituation of neural, as opposed to behavioral, responses to vocalizations. It remains unclear how auditory neurons will habituate to ethologically relevant stimuli such as vocalizations, and whether this varies according to the level of cortical processing. In this paradigm, neurons habituated to one vocalization might be dishabituated by another vocalization that differs with respect to acoustic features alone, or acoustic features and meaning. The hypothetical neurons that encode the call’s referent should not be dishabituated by a functionally similar call despite its different acoustic morphology. Such neurons are not likely to exist in lower-order core areas, such as primary auditory cortex, but might be present in higher-order areas. Alternatively, it is conceivable that the correlates of call-meaning might only be discerned by observing the collective action of many neurons. With the advent of functional magnetic resonance imaging (fMRI) technology suitable for smaller animals, experiments using both anesthetized and alert primates are now feasible64. Experiments such as the one just described could provide an excellent opportunity to explore the large-scale neural circuitry involved in the processing of vocalizations. At a basic level, it could assist in the localization of areas responsible for higher-level vocal processing to guide subsequent multi-area, multi-electrode neurophysiological experiments65. At a cognitive level, if fMRI responses to vocalizations are lateralized, then experiments parallel to those done in the field16 could be conducted during imaging to determine whether or not activity shifts to different cortical areas and/or cerebral hemispheres when acoustic features of vocalizations are manipulated within and beyond the species-typical range. Ultimately, both neurophysiology and neuroimaging experiments will shed light on the common neural circuitry underlying vocal behavior between primates and humans, and will pave the way for a deeper understanding of the evolution of human speech.

7 Nottebohm, F. (1989) From bird song to neurogenesis Sci. Am. 260, 74–79 8 Ryan, M.J., Perrill, S.A. and Wilczynksi, W. (1992) Auditory tuning and call frequency predict population-based mating preferences in the cricket frog, Acris crepitans Am. Nat. 139, 1370–1383 9 Perrett, D.I. et al. (1988) Specialized face processing and hemispheric asymmetry in man and monkey: evidence from single unit and reaction time studies Behav. Brain Res. 29, 245–258 10 Worden, F.G. and Galambos, R. (1972) Auditory processing of biologically significant sounds Neurosci. Res. Prog. Bull. 10, 1–119 11 Seyfarth, R.M., Cheney, D.L. and Marler, P. (1980) Vervet monkey alarm calls: semantic communication in a free-ranging primate Anim. Behav. 28, 1070–1094 12 Hauser, M.D. (1998) Functional referents and acoustic similarity: field playback experiments with rhesus monkeys Anim. Behav. 55, 1647–1658 13 Marler, P., Evans, C.S. and Hauser, M.D. (1992) Animal signals?: reference, motivation, or both?, in Nonverbal Vocal Communication: Comparative and Developmental Approaches (Papoucek, H., Jurgens, U. and Papoucek, M., eds), pp. 66–86, Cambridge University Press 14 Hauser, M.D. and Andersson, K. (1994) Left hemisphere dominance for processing vocalizations in adult, but not infant, rhesus monkeys: field experiments Proc. Natl. Acad. Sci. U. S. A. 91, 3946–3948 15 Petersen, M.R. et al. (1984) Neural lateralization of vocalizations by Japanese macaques: communicative significance is more important than acoustic structure Behav. Neurosci. 98, 779–790 16 Hauser, M.D., Agnetta, B. and Perez, C. (1998) Orienting asymmetries in rhesus monkeys: effect of time-domain changes on acoustic perception Anim. Behav. 56, 41–47 17 May, B., Moody, D.B. and Stebbins, W.C. (1988) The significant features of Japanese macaque coo sounds: a psychophysical study Anim. Behav. 36, 1432–1444 18 Cheney, D.L. and Seyfarth, R.M. (1997) Some general features of vocal development in nonhuman primates, in Social Influences on Vocal Development (Snowdon, C.T. and Hausberger, M., eds), pp. 249–273, Cambridge University Press 19 Harnad, S. (1987) Categorical Perception: The Groundwork of Cognition, Cambridge University Press 20 Liberman, A.M. et al. (1967) Perception of the speech code Psychol. Rev. 74, 431–461 21 Fitch, R.H., Miller, S. and Tallal, P. (1997) Neurobiology of speech perception Annu. Rev. Neurosci. 20, 331–353 22 Green, S. (1975) Variation of vocal pattern with social situation in the Japanese monkey (Macaca fuscata): a field study, in Primate Behavior (Rosenblum, L.A., ed.), pp. 1–102, Academic Press 23 Beecher, M.D. et al. (1979) Perception of conspecific vocalizations by Japanese macaques Brain Behav. Evol. 16, 443–460 24 Zoloth, S.R. et al. (1979) Species-specific perceptual processing of vocal sounds by monkeys Science 204, 870–873 25 Kimura, D. (1993) Neuromotor Mechanisms in Human Communication, Oxford University Press 26 Schwartz, J. and Tallal, P. (1980) Rate of acoustic change may underlie hemispheric specialization for speech perception Science 207, 1380–1381 27 Jurgens, U. (1979) Vocalizations as an emotional indicator: a neuroethological study in the squirrel monkey Behaviour 69, 88–117

Acknowledgements

28 Smith, W.J. (1977) The Behavior of Communicating, Harvard University

We thank Troy Hackett, Jon Kaas, Don Katz and Cory Miller for their helpful comments on this manuscript.

Press 29 Struhsaker, T.T. (1967) Auditory communication among vervet monkeys (Cercopithecus aethiops), in Social Communication Among Primates (Altmann, S.A., ed.), pp. 281–324, University of Chicago Press

References

30 Macedonia, J.M. (1991) What is communicated in the antipredator

1 Cheney, D.L. and Seyfarth, R.M. (1990) How Monkeys See the World, University of Chicago Press

and ruffed lemurs Ethology 93, 177–197

2 Hauser, M.D. (1996) The Evolution of Communication, MIT Press 3 Deacon, T.W. (1997) The Symbolic Species: The Co-Evolution of Language and the Brain, W.W. Norton

calls of lemurs?: evidence from playback experiments with ring-tailed 31 Gouzoules, S., Gouzoules, H. and Marler, P. (1984) Rhesus monkey (Macaca mulatta) screams: representation signalling in the recruitment of agonistic aid Anim. Behav. 32, 182–193

383 Trends in Cognitive Sciences – Vol. 3, No. 10,

October 1999

Review

Ghazanfar and Hauser – Primate vocal communication

32 Cheney, D.L. and Seyfarth, R.M. (1988) Assessment of meaning and the detection of unreliable signals by vervet monkeys Anim. Behav. 36, 477–486

22, 1623 49 Wang, X. et al. (1995) Representation of a species-specific vocalization in the primary auditory cortex of the common marmoset: temporal

33 Zuberbuhler, K., Noe, R. and Seyfarth, R.M. (1997) Diana monkey longdistance calls: messages for conspecifics and predators Anim. Behav. 53, 589–604

and spectral characteristics J. Neurophysiol. 74, 2685–2706 50 Beeman, K. (1998) Digital signal analysis, editing and synthesis, in Animal Acoustic Communication (Hopp, S.L., Owren, M.J. and Evans,

34 Rendall, D. et al. (1999) The meaning and function of grunt variants in baboons Anim. Behav. 57, 583–592

C.S., eds), pp. 59–104, Springer-Verlag 51 Rauschecker, J.P. (1998) Parallel processing in the auditory cortex of

35 Gouzoules, H. and Gouzoules, S. (1989) Design features and developmental modifications of pigtail macaque (Macaca nemestrina) agonistic screams Anim. Behav. 37, 383–401

primates Audiol. Neuro-otol. 3, 86–103 52 Geschwind, N. and Levitsky, W. (1968) Human brain: left–right asymmetries in temporal speech region Science 161, 186–187

36 Dittus, W. (1984) Toque macaque food calls: semantic communication concerning food distribution in the environment Anim. Behav. 32, 470–477

53 LeMay, M. and Geschwind, N. (1975) Hemispheric differences in the brains of great apes Brain Behav. Evol. 11, 48–52 54 Heilbroner, P.L. and Holloway, R.L. (1988) Anatomical brain asymmetries

37 Hauser, M.D. and Marler, P. (1993) Food-associated calls in rhesus macaques (Macaca mulatta): I. Socioecological factors Behav. Ecol. 4, 194–205

in New World and Old World monkeys: stages of temporal lobe development in primate evolution Am. J. Phys. Anthropol. 16, 38–48 55 Heffner, H.E. and Heffner, R.S. (1984) Temporal lobe lesions and

38 Zuberbuhler, K., Cheney, D.L. and Seyfarth, R.M. (1999) Conceptual semantics in a nonhuman primate J. Comp. Psychol. 113, 33–42

perception of species-specific vocalizations by macaques Science 226, 75–76

39 Kaas, J.H., Hackett, T.A. and Tramo, M.J. (1999) Auditory processing in primate cerebral cortex Curr. Opin. Neurobiol. 9, 164–170

56 Hauser, M.D. (1999) Perseveration, inhibition and the prefrontal cortex: a new look Curr. Opin. Neurobiol. 9, 214–222

40 Morel, A. and Kaas, J.H. (1992) Subdivisions and connections of auditory cortex in owl monkeys J. Comp. Neurol. 318, 27–63

57 Iversen, S.D. and Mishkin, M. (1970) Perseverative interference in monkeys following selective lesions of the inferior prefrontal convexity

41 Bieser, A. and Muller-Preuss, P. (1996) Auditory responsive cortex in the squirrel monkey: neural responses to amplitude-modulated sounds Exp. Brain Res. 108, 273–284

Exp. Brain Res. 11, 376–386 58 Scalaidhe, S.P.O., Wilson, F.A.W. and Goldman-Rakic, P.S. (1997) Areal segregation of face-processing neurons in prefrontal cortex Science

42 Luethke, L.E., Krubitzer, L.A. and Kaas, J.H. (1989) Connections of primary auditory cortex in the New World monkey, Saguinus J. Comp. Neurol. 285, 487–513

278, 1135–1138 59 Romanski, L. and Goldman-Rakic, P.S. Physiological identification of an auditory domain in the prefrontal cortex of an awake behaving

43 Aitkin, L.M., Kudo, M. and Irvine, D.R.F. (1988) Connections of the primary auditory cortex in the common marmoset, Callithrix jacchus jacchus J. Comp. Neurol. 269, 235–248

monkey Soc. Neurosci. Abstr. 25 (in press) 60 Leinonen, L., Hyvarinen, J. and Sovijarvi, A.R.A. (1980) Functional properties of neurons in the temporo-parietal association cortex of

44 Galaburda, A. and Sanides, F. (1980) Cytoarchitectonic organization of the human auditory cortex J. Comp. Neurol. 190, 597–610

awake monkey Exp. Brain Res. 38, 203–215 61 Watanabe, M. (1992) Frontal units of the monkey coding the associative

45 Hackett, T.A., Preuss, T.M. and Kaas, J.H. (1998) Architectonic identification of core and belt auditory cortex in macaque monkeys, chimpanzees, and humans Soc. Neurosci. Abstr. 24, 1880

significance of visual and auditory stimuli Exp. Brain Res. 89, 233–247 62 Schall, J.D. and Thompson, K.G. (1999) Neural selection and control of visually guided eye movements Annu. Rev. Neurosci. 22, 241–259

46 Wollberg, Z. and Newman, J.D. (1972) Auditory cortex of squirrel monkeys: response properties of single cells to species-specific vocalizations Science 175, 212–214

63 Russo, G.S. and Bruce, C.J. (1994) Frontal eye field activity preceding aurally-guided saccades J. Neurophysiol. 71, 1250–1253 64 Logothetis, N.K. et al. (1999) Functional imaging of the monkey brain

47 Rauschecker, J.P., Tian, B. and Hauser, M. (1995) Processing of complex sounds in the macaque nonprimary auditory cortex Science 268, 111–114 48 Tramo, M.J., Bellew, B.F. and Hauser, M.D. (1996) Discharge patterns of auditory cortical neurons evoked by species-specific vocalizations and synthetic complex signals in alert Macaca mulatta Soc. Neurosci. Abstr.

Nat. Neurosci. 2, 555–562 65 Nicolelis, M.A.L. et al. (1998) Simultaneous encoding of tactile information by three primate cortical areas Nat. Neurosci. 1, 621–630 66 Kaas, J.H. and Hackett, T.A. (1998) Subdivisions of auditory cortex and levels of processing in primates Audiol. Neuro-otol. 3, 73–85

TICS Editorial Policy Trends in Cognitive Sciences is indispensable reading for all interested in this diverse field. The journal carries a core of authoritative Reviews, written in an easily accessible style by leading authors, summarizing the exciting developments of the subjects you need to know about. Accompanying the Reviews are a range of other types of article. The Comment section offers short updates and discussions of one or two ground-breaking and important primary papers. The authors of the primary papers have the opportunity of replying to the Comments, thus extending the dialogue on their work. Monitor pieces summarize, in 100–200 words, recently published papers. Meeting reports convey the than more ts do Concepgorize excitement of recent conferences or seminars, focussing on the cate latest developments and the discussions surrounding them. Opinion articles are reviews of subjects written from a personal slant, and therefore highlight some of the controversies in cognition. Books etcetera features stimulating essay-reviews of recent publications, whether books, software, CD-ROMs, films, etcetera. And a list of books received for review will be published regularly. Letters stimulated by any of the articles published in TICS are welcome. Authors will be offered the chance to reply to any points raised.

Update Moni tor

Summ aries of scient ists. Readerecently publis identi rs hed paper fying appro who would conta s of priate ct the paper like to contri interest Editor s and to cognit . bute writin g short to this sectio ive summ aries, n, by should

Facing

up (or

down)

Music al brainwa

There is an active garding ongoin the role g debate of the of the reventral fusiform When faces, occipito The recent people region and two-ton tempor read incongr (high-c al cortex. has provide increase in uous sentencor hear syntact e Moone ontrast day, he y images d substan the use of black–w ically faces). es, such leading mow the fMRI The results hite images ce as: ‘Yester to the brain lawn’ hypoth hypoth for this debate the fusiform strongl activity their averag of esis of esis y suppor , recorde (eventa ‘face strating ed the percept gyrus (FG) that an area t the related d at that the area’ while in is speciali the scalp potent show a ion of strongl Kanwis FG respond demonzed for P600 wavefo ials, or human y to her et ERPs) ponent to inverte inverted graysca s more rm (a positive will al. reporte faces. Nancy tional of the d two-ton imagin le faces ERP) at word. d some Althou comg demon than the It gh a face-sp e Moone strate that data that seemedfuncwhethe is still a matterincongruous y images demon ecific a ‘face r there might strated . to of debate area’ restrict this brainwave by these area in the FG that there indeed thors gave in the FG ed compo is be (Ref. reflects to language-rela are other studies, it is nent subject areas yet 1). The ceptual more ted input is to be discove face-res likely tasks and s four differen augenera cesses. functio or cantly l cogniti Patel et t perred. Undouponsive nal imagin recorde greater dence ve proal. 1 provide d a signifito play respons the FG for an importag studies will btedly, e in when 15 musica the latter positio some evicontinu gation nt face stimuli the subject a region of role of the e n. They to sequen lly educated s were visual processin the investithe recogn compar had commo shown subject ed with n objects es underly ition of of these ces of musica s these in a stronge . The authors stimuli of l chords. listen chords turn could human faces, ing highly r respons were Some velopm also shown help to and incong modera ent of full front-v e when subject found guide ruous mony the next models the dewhen and key-rel in terms tely or s were iew face of brain genera shown preced of harphotos face-re atedne functio tion of houses. front-v ing cogniti iew photosthan Three-q listened ones. The same ss given the on mechann, incorporating uarterelicited to English subject isms. of view Referen a which s also senten ces of humangreater respons face photos include ces, some d words 1 Kanwish ungram e than hands, of greater that matica er, N. et and intact photos respons l or hard were either al. (1997) given the area: a tone faces. e than faces The fusiform module a scramb in human For both preceding syntactto integrate specializ Recentl face conduc extrastria ed for y, Kanwis led twoic structu ted the inconglinguistic and te cortex face perceptio 17, 4302–431 her et re. ing both an additional musica ruous al. n J. Neurosci 1 l upright compo study 2 Kanwish nent in targets elicited stimuli, . and inverte 2 involver, N. et the degree a P600 al. (1998) d graysca inversion of congru ERP. Moreo The effect le on the fects on ver, the ency had of face human Cognitio the P600 fusiform n 68, B1–B11 domain face area amplitu similar efs. de for P600 does These results both not reflect suggest that process the langua What structu es, but ge-spec gration rather ific the operati res and process process structu ral intees that specific on of es mediat tion in . The are not the human selective visual e author there was directe s point domainare cases sugges attenbrain? ts that out d One selectiv of eral stimuli. to one of hypoth object differe ely impairs brain damag that esis represe nt locatio The authors the four periphmusica finding the percep e that ing capacit l harmon ns compet ntations at s are argue that out affectin tion of ic relation that attenti consistent these that attentiy in the visual e for process ships on is protectwith the notion the opposi g syntactic system, withon works tation compet knowle of te is ing a represe ition in by biasing and these also true. dge, and tion. It the target item locatio deficits favour is also this from competnn. Record of Howev might terms possibl differe er, of awake be explain ings of the attended nce betwee e, though imonke knowle damage to a single success ys have with this n simulta , that the dge base cells domain ed in ive been consiste in idea 1. The main-u neous as oppose -specific from otherstimulation to an nspecif and might respons nt optima d to ic process factors have arisen e of a neuron ence in l stimulu dostantia , such es. Identif structural integra present lly s is lates of ying the two conditio ation rate as the differtion present when an irreleva reduced subharmon ed simulta betwee tactic ns. Howev ic process neural corrent stimulu experim n the process cation neously ing and ent, the er, in in the ing gration at anothe s is a control stimuli synthe animal receptive authors process independent r loat field. es should unders of present that the a constant the competdirects its attenti However, tanding ed allow a interate and respons of the domain better on to one if item present showed e to a field, the ing stimuli treatm -specifi single of ent of in the c disorde ed alone respon periphe it was the recepti the stimulu ses are rs. was ral presentand Edward Referen ve edited A. Anderson Rosenfeld, s is present as large ce ed simultalowere d whenMIT Press, 1998. Kastner as by Jamesthree other items. 1 Patel, neously ed alone. when and colleag differen A.D. et (xi + 434 pages) ISBN 0 262 01167with 0 This suggest that a Now, ues provide£31.95/$39.95 ce betwee al. (1998) similar relations s that the taneou Processin mechan evidenc in human in language s stimula n successive g syntactic ism 2 e related and simultion is to differen and music: cortical s . Using fMRI, might operate potentia an eventl study t present not simply related respon they examin 717–733 finding J. Cogn. jects present ses to four and Bertrand Russell is worth the price What is networks that ation s address Neurosci edit about neural rates. adjacen . 10, ed either These or one book (for which you get several makes peopleofsodirected excitable? directly the of the t obat a time simulta mechan attentio gation Althou neously isms The classic McCulloch and James Anderson, n. Furtherversions). in rapid willa noted neuralgh the , ion allels betweeexplore investitotal amoun success stimula Pitts papers were written when Pitts network researcher, and Edward the rizat ion. tego tion n single-c intriguing t of Rosenfeld, nd ca the same (integrated wasparstill in his teens and living with aimagin journalist, had the splenretinal ell and g results pts a over once in the functio reveale . – C ditions two experim time) nalMcCulloch and his wife. Pitts subsedidwas idea toReferen generate an oral history of d here. et al , mon cesby talking to most ental Solo Simulta cortical respon quently gravitated to a second fatherneural-net research conneous 1 Moran, ses differe present activity J. and in the field. proven mathematically, that neural figure, Norbert Wiener. Wiener and figures Desimon ation evokedof the d. pioneering than attention e, R. (1985) McCulloch had a falling out because this differe successive networks cannot ever do anything inThe Talking Nets, gates is an extralessresult, extrastria present Selective visual nce was ation,ordinary higher teresting.’ Widrow, in contrast, says, ‘I documentte cortex that is a processin page- g Mrs. Wiener didn’t like McCulloch; Pitts more pronou and cortical in the 2 Kastner, Science areas. duction nced 229, 782–784was couldn’t understand what the point of caught in the middle, had a nerturner Moreov De Weerd, in for cognitiveS.,scientists. with Subsc er, the was much simultaneous P., Desimon ribehimself [Perceptrons] was, why the hell they vous breakdown, and drank to Physics Ungerlei is a science. Entymology is a der, L.G. reto TICS e, R. and present less severe directed (1998) did it. I figured that they must have ation science. ‘Neural networks’ Mechani an early grave. And we haven’t even attention is a toolbox when sms of in the cortexof tools for attenti– a collection 50% gotten inspired to write that book regotten to at Minsky and Papert yet. the human simula90 as revealed on disco extrastria by function 1364-661 te The tale of how Minskyunt ally early on to squelch the field, to do 108–111 and using Papert tion and 282, analysis of complex systems. al MRI 3/99/$ – see front what they could to stick pins in the balTren slew the neural-net dragon in the the These tools were developed by people Science matter ds in © 1999 Cogn boun loon. But by the time the book came Elsevier 1960s is the most important in working in diverse fields ranging from itive d-in legend Science. Scien card in out, the field was already gone. There All rights ces – neural-net culture. The testimony theory and signal processing to V ocontrol reserved l. 3, . N o . 3 and psychology. was just about nobody doing it.’ Talking Nets suggests a more complineurobiology They are , Ma rch 1 of So what really happened? It seems 9 to cated story, however. It is true that being vigorously applied 9 9 an equally the study clear that people such as Widrow had Minsky and Papert come off badly in broad range of problems ranging from recently, only one become discouraged because they is this telling of the tale (of course, it isn’t understanding basic brain mechanisms Until ion sses. knew that multilayer nets were needed their side that is being told). The MIT to forecasting the fluctuations in the categorizat tive proce rstood l cogni but didn’t know how to train them. His ion. But people seem to have been bothered by bond market. This book recapitulates t be unde r-leve orizat ion, for highe testimony on this point is poignant: pts canno publicity over neural-network research much of the convoluted history of of categ in isolat rlie all t study that conce being conducted elsewhere (Widrow: neural-network research through the function, pts unde ct to affec argue been the Conce intera ‘We would have given our eye any other al. We ‘You know, we had a lot of controversy testimony of people who figured cenlargely which ion, or g sever ion pts has teeth to come up with somein the early days. It was due to publicity trally in its development. It is a remarkorizat amon conce in isolat functions of categ function thing like Backprop…Backprop that [Frank] Rosenblatt had in the able story, replete with drama, function multiple study ptual ion, but single conce gh the pts serve to me is almost miraculous. news media and publicity that I had…I tragedy, hubris, irony, humor, bitter ining a throu conce each funct d, study For First, iently The first exposure I had to found that this kind of publicity infuritellectual warfare, and a couple of ular to Secon suffic reasons. functions. ssing. Backprop was around 1985 at are partic xt ates colleagues…’). One has to imagine corpses. It is told in the words of a brilple rtant proce that multi sses the conte two impo a meeting at Snowbird, Utah… on to what transpired the day that liant and eccentric collection of people, ure and ed in struct tive proce are comm Someone gave a paper in the ptual Rosenblatt, the psychologist from including several certifiable geniuses d be studi see cogni conce sses that d instea first morning session and during Cornell, came down to MIT to tell an and a couple who are merely certifione to proce rages pts shoul very of the question period…someone audience that included Minsky, able, as Groucho would have said. (The encou conce s the disco st that got up and said, “You know, Shannon and McCulloch the news interviewees comprise Bernard Widrow, sugge discourage ns, we something like that was done about perceptrons. (Cowan: ‘It was a Carver Mead, Stephen Grossberg, functions. two reaso interrelations elated by Widrow back in the early terrible lecture. McCulloch didn’t say Michael Arbib, James Anderson, David these discuss section m of interr ‘60s.” They began to have this anything. Shannon said, ‘It’s worth Rumelhart, Geoff Hinton, Terry following of a syste represenin the big discussion about what looking at’… But by and large it was Sejnowski and nine others.) The acns. life, and conceptual an entity Widrow did and Widrow didclear that the perceptron wasn’t doing counts aren’t all consistent with each r these functio assume that How conamong ing whethe hers n’t do, and I’m just sitting the things that Frank claimed it could other and can’t all be equally true. thought. to as for identify referred blocks of therefore, central Most researc there, listening to all this do.’) Rosenblatt had the even greater Yet in reading them, one gets a vivid but building procedures a process often are is are the in itself, stuff…I was like a dead man. I updated concepts misfortune to die soon thereafter in a sense of where people came from, oncepts tations includeof a category, an end ennot used, and The literature on , r is novel was a man who’d died, who boating accident. what they thought they were doing, . formed recent edited is a membe . Categorization new: categorizing s ve science cepts are 4–7 for was sitting up on a cloud The ironies here are staggering. and how their ideas developed. It’s also t previou to review. in cogniti and Refs categorization to connect old somewhere, looking down on in a short bring relevan the novel Minsky’s reaction to the publicity over a lot of fun. questions for reviews it serves system to summarize namely, the idea Refs 1–3 ve tanding to the Earth, watching what hapAt one level, the history of neuralperceptrons has to be considered in (see rather ible – vast of unders the cogniti ‘toothbrush’ and imposs on a single issue interact to affect pened after he died.’ (p. 61) network research can be read as a light of his subsequent willingness to tities allows bear in the service l shape as a volumes) ns. A we focus to mulns which long-running soap opera with a cast of serve the mass media as a kind of allunusua reason, ts serve e functio their functio knowledge For this izing some People also had to wait for comcharacters that is a little strange even purpose, techno-egghead commentacategory parts and have multipl processing. Concep ns are not of its Recogn ts dge tand entity. and puter power to catch up with the ideas, for a group of academics. The aptor on the issues of the day (e.g. ‘The that concep r. For these functio t with and to unders ce: knowle structure will see, a point that Arbib emphasizes. Widrow proach originated with some papers by future of money’, Discover, 1998; ‘Is allows one is inferen ions about behavio conceptual ns, and, as we they interac physicians function and Rosenblatt were experimenting with Turing (on biological dynamic systems), the body obsolete?, Whole Earth interactions r; rather, ts predict related ies allow e. tiple functio of one anothe , suppor that these categor effectiv physical networks created out of elecization Wiener’s ‘cybernetics’, and the crucial Review, 1989). And if the perceptron be membership diagnostic We believe g categor nts will ng. independent tric motors and potentiometers (some articles by McCulloch and Pitts treating was not able to do all the things that medical each other. strategy of studyin ver, these interof treatme tion and reasoni example, good to three decimal places!). Comone neurons as computational devices, were being claimed, how would that influence what sorts popular n. Moreo in explana ‘football fan’, in the conpredict ine the puters were primitive, limiting the use crucial to which in turn greatly influenced von compare to the years of relentless hype in isolatio a n, studied as underm also street be the man functio ts are of simulation as a tool; luckily things Neumann, who was about to invent about artificial intelligence? ts should g down a young Concep or any other that concep ns. is walkin bellowns of categorized were to move ahead rapidly on that the modern computer. McCulloch was On the other hand, did it matter? why he suggest ated functio common functio on his face to inHaving actions front. In the meantime, some interestof interrel a neurophysiologist at the University of Interestingly, opinions vary. Robert able to explainand yellow paint strates used some be also system a demon outline are might that ing work was done and the people in l fan text of Chicago. Pitts was a prodigy and misfit Hecht-Nielsen thinks that Minsky and we first with blue Categories research another. our footbal this book managed to survive quite bare-chested l In this article, turn to recent an rouser. who hung around the university but Papert’s book1 Perceptrons mattered , t with one example, ted to a footbal the Michig K.O. Solomon g8 . For Then, we do interac how an integra well, ending up with positions at places because it created ‘a new conventional wasn’t actually enrolled. The story of ing out concepts. in plannin y of things to bring transistor of es ns indeed Medin and e goals D.L. like Berkeley, Stanford, and CalTech. wisdom that some MIT professors have e how Pitts met McCulloch via Carnap process functio and ve an exampl categor stantiat are at the that these de with an ad hoc seat-cushion ry of cogniti E.Lynch we conclu age the discove might create beer, binoculars, ent of Finally, of old, Departm ns. can encour in terms y, le functio game (e.g. approach Psycholog That is, understood across multip entities stern concepts. radio). r Trends in Cognitive Sciences – Vol. 3, No. 3, March 1999 Northwe that extend are new and update tering a membe in this Not only ty, 2029 However, also modify Universi ts entities Thus, encoun an electric toothRoad, of concep difficult to define. representation but new Sheridan learning. Functions y (e.g. rated very support , IL 60208, t as a mental ns. The most t can be novel propert being incorpo n is Evanston concepts a concep functio A concep y with a property refer to a prog functio cognitive we will USA. ization, of a categor in that novel variety of how The learnin

ves

Attent ional suppre extras triate ssion cortex in hum an

Books etcetera

Talking Nets: An Oral History of Neural Networks

Review

Students

ch

glas

mon, Dou

in L. Med

abeth Lyn

and Eliz

O. Solo Karen

C

paper, a . categor as result to meet has been ts) are used ns, such representation brush) can to that is used function s (or concep toward conceptual a number of questio old, when nly studied representation into the commo been a trend ce, and ation versus that inted with mental there has also associa to give new inform Moreover, given ns, inferen cess by which Recently, ies there on. entities. tual functio see, concepts are much weight gories, and so ping categor all to classify two other concep ries le overlap should modify as we shall of bounda set up subcate to multip ation the study ation. But, tasks, and the t and can belong r new inform combin ve ns interac te stances conceptual many cogniti blur as functio of whethe 288-7 perform 13(99)01 is the issue briefly illustra y functions used to we will PII: S1364-66 999 . rch 1 conceptual In this section in everyda reserved , Ma between are used All rights No. 3 l. 3, Science. each other. which concepts o V ce Elsevier s – influen in ience © 1999 ve Sc matter r of ways gniti – see front n Co a numbe 3/99/$ nds i 1364-661

384 Trends in Cognitive Sciences – Vol. 3, No. 10,

October 1999

Tre

467 6513 tel: +1 847 491 7859 847 fax: +1 n@ k-solomo e-mail: nwu.edu

99

121