Our ability to invariably identify words spoken with different accents and timbres requires 1 our perception to tolerate multiple acoustic modulations. However, the mechanisms by 2 which lexical information is perceived to be invariant is unknown. In this study, we 3 explored the ability of two trained rhesus monkeys to recognise many sounds, including 4 multisyllabic words presented in numerous variations. We found that the monkeys' lexical 5 representations are remarkably tolerant of dynamic acoustic changes. Furthermore, we 6 determined that the macaques invariably recognise sounds with formants at close Euclidean 7 distances from a learned category. Based on our results, we propose the possible existence 8 of neuronal circuits responsible for the invariant recognition of lexical representations in 9 macaques. 10 11 14 vital for communication in primates, the perceptual basis of invariant recognition of sounds 15has not extensively investigated. One possible reason for this is that non-human primates 16 may only demonstrate limited acoustic learning 4-6 , and that therefore their recognition 17 capability may depend on genetically-programmed circuits 7-9 . However, it is known that 18 macaques are capable of learning repertoires of visual categories 10 and even declare the 19 existence of objects with ambiguous or incomplete information 11 . Studies of the 20 inferotemporal and prefrontal cortices of monkeys showed neurons whose categorical 21 responses achieved the grouping of wide variations of images [12][13][14] , which is consistent with 22 perceptual reports 15,16 . Similarly, experiments in the prefrontal cortex and secondary 23 3 acoustic areas suggest invariant coding of acoustic categories [17][18][19][20][21][22][23][24][25] . In this paper, we sought 24 to determine what acoustic parameters stand for the invariant recognition of sounds (IRS) 25 in trained non-human primates 26 . We hypothesised that monkeys invariably recognise 26 sounds of frequency-saliences with dynamics similar to sounds the monkeys had 27 learned 27,28 . To test this, we designed a novel paradigm in which the macaques efficiently 28 reported target (T) sounds presented in sequences that included nontarget (N) sounds. We 29 found that the monkeys recognised sound fragments of frequency modulations in the range 30 of the prominences of the learned sounds as invariants. Our results allowed us to elucidate, 31 for the first time, the acoustic parameters 29-34 that lead to monkeys' IRS. We propose that 32 the ability to group complex stimuli into particular categories is served by chunks of 33 prominent features of the stimuli. 34 35
Results
36We trained two rhesus monkeys in an acoustic recognition task in order to study the 37 invariant recognition of sounds. During the task, the monkeys obtained a reward for 38 releasing a lever after identifying a T presented after zero, one or two Ns ( Fig. 1a-c; see 39 Methods). After two years of training, monkey V recognised seven Ts and twenty-one Ns, 40 and monkey X recognised eleven Ts a...