Unsupervised statistical learning of higher-order spatial structures from visual scenes.

Three experiments investigated the ability of human observers to extract the joint and conditional probabilities of shape co-occurrences during passive viewing of complex visual scenes. Results indicated that statistical learning of shape conjunctions was both rapid and automatic, as subjects were not instructed to attend to any particularfeatures of the displays. Moreover, in addition to single-shape frequency, subjects acquired in parallel several different higher-order aspects of the statistical structure of the displays, including absolute shape-position relations in an array, shape-pair arrangements independent of position, and conditional probabilities of shape co-occurrences. Unsupervised learning of these higher-order statistics provides support for Barlow's theory of visual recognition, which posits that detecting "suspicious coincidences" of elements during recognition is a necessary prerequisite for efficient learning of new visual features.

Statistical learning in a serial reaction time task: access to separable statistical cues by individual learners.

The ability of adult learners to exploit the joint and conditional probabilities in a serial reaction time task containing both deterministic and probabilistic information was investigated. Learners used the statistical information embedded in a continuous input stream to improve their performance for certain transitions by simultaneously exploiting differences in the predictability of 2 or more underlying statistics. Analysis of individual learners revealed that although most acquired the underlying statistical structure veridically, others used an alternate strategy that was partially predictive of the sequences. The findings show that learners possess a robust learning device well suited to exploiting the relative predictability of more than I source of statistical information at the same time. This work expands on previous studies of statistical learning, as well as studies of artificial grammar learning and implicit sequence learning.

Segmentation of the speech stream in a non-human primate: statistical learning in cotton-top tamarins.

Previous work has shown that human adults, children, and infants can rapidly compute sequential statistics from a stream of speech and then use these statistics to determine which syllable sequences form potential words. In the present paper we ask whether this ability reflects a mechanism unique to humans, or might be used by other species as well, to acquire serially organized patterns. In a series of four experimental conditions, we exposed a New World monkey, the cotton-top tamarin (Saguinus oedipus), to the same speech streams used by Saffran, Aslin, and Newport (Science 274 (1996) 1926) with human infants, and then tested their learning using similar methods to those used with infants. Like humans, tamarins showed clear evidence of discriminating between sequences of syllables that differed only in the frequency or probability with which they occurred in the input streams. These results suggest that both humans and non-human primates possess mechanisms capable of computing these particular aspects of serial order. Future work must now show where humans' (adults and infants) and non-human primates' abilities in these tasks diverge.

Infants' perception of transparency.

Four- and 7-month-old infants' perception of transparency was investigated with computer-generated achromatic or color displays depicting a semitransparent box occluding the center of a rod. Following habituation, infants viewed test displays consisting of either a two-color rod (corresponding to the habituation display's proximal characteristics) or a solid rod (corresponding to the distal characteristics of the event depicted by the habituation display). Looking-time results from 4-month-olds suggested perception of transparency in color displays but not in an achromatic display. An additional condition indicated that transparency perception may rely on the visibility of background texture through the transparent surface. Seven-month-olds, in contrast, provided some evidence of transparency perception in the achromatic display. Implications for the development of infants' responses to object properties and perceptual segregation are discussed.

Motor timing learned without motor training.

Improvements due to perceptual training are often specific to the trained task and do not generalize to similar perceptual tasks. Surprisingly, given this history of highly constrained, context-specific perceptual learning, we found that training on a perceptual task showed significant transfer to a motor task. This result provides evidence for a common neural architecture underlying analysis of sensory input and control of motor output, and suggests a potential role for perception in motor development and rehabilitation.

Spoken word recognition and lexical representation in very young children.

Although children's knowledge of the sound patterns of words has been a focus of debate for many years, little is known about the lexical representations very young children use in word recognition. In particular, researchers have questioned the degree of specificity encoded in early lexical representations. The current study addressed this issue by presenting 18-23-month-olds with object labels that were either correctly pronounced, or mispronounced. Mispronunciations involved replacement of one segment with a similar segment, as in 'baby-vaby'. Children heard sentences containing these words while viewing two pictures, one of which was the referent of the sentence. Analyses of children's eye movements showed that children recognized the spoken words in both conditions, but that recognition was significantly poorer when words were mispronounced. The effects of mispronunciation on recognition were unrelated to age or to spoken vocabulary size. The results suggest that children's representations of familiar words are phonetically well-specified, and that this specification may not be a consequence of the need to differentiate similar words in production.

Statistical learning of tone sequences by human infants and adults.

Previous research suggests that language learners can detect and use the statistical properties of syllable sequences to discover words in continuous speech (e.g. Aslin, R.N., Saffran, J.R., Newport, E.L., 1998. Computation of conditional probability statistics by 8-month-old infants. Psychological Science 9, 321-324; Saffran, J.R., Aslin, R.N., Newport, E.L., 1996. Statistical learning by 8-month-old infants. Science 274, 1926-1928; Saffran, J., R., Newport, E.L., Aslin, R.N., (1996). Word segmentation: the role of distributional cues. Journal of Memory and Language 35, 606-621; Saffran, J.R., Newport, E.L., Aslin, R.N., Tunick, R.A., Barrueco, S., 1997. Incidental language learning: Listening (and learning) out of the corner of your ear. Psychological Science 8, 101-195). In the present research, we asked whether this statistical learning ability is uniquely tied to linguistic materials. Subjects were exposed to continuous non-linguistic auditory sequences whose elements were organized into 'tone words'. As in our previous studies, statistical information was the only word boundary cue available to learners. Both adults and 8-month-old infants succeeded at segmenting the tone stream, with performance indistinguishable from that obtained with syllable streams. These results suggest that a learning mechanism previously shown to be involved in word segmentation can also be used to segment sequences of non-linguistic stimuli.

Young infants' perception of illusory contours in dynamic displays.

Ninety-six 4-month-old infants were habituated to one of three computer-generated displays depicting two rod parts above and below an occluding box. In the first display, the surfaces and boundaries of the rod and box were specified by dense surface texture. Their depth segregation was specified by accretion and deletion of background texture and motion shear. In the second display, the unity of the rod parts and box, and their depth segregation, were specified only by illusory contours. In the third display, the boundaries of the rod and box were specified by illusory contours, perceptible only via spatiotemporal integration of accretion and deletion of sparse-background-texture elements. Infants appeared to perceive object unity, and segregate the rod and box surfaces, in all three displays, indicating use of illusory contours to perceive bounded surfaces in depth. The results suggest a cognitive contribution to perception of some illusory contours, abilities which seem to be present by at least 4 months of age.

Statistical learning by 8-month-old infants.

Learners rely on a combination of experience-independent and experience-dependent mechanisms to extract information from the environment. Language acquisition involves both types of mechanisms, but most theorists emphasize the relative importance of experience-independent mechanisms. The present study shows that a fundamental task of language acquisition, segmentation of words from fluent speech, can be accomplished by 8-month-old infants based solely on the statistical relationships between neighboring speech sounds. Moreover, this word segmentation was based on statistical learning from only 2 minutes of exposure, suggesting that infants have access to a powerful mechanism for the computation of statistical properties of the language input.

Infants' detection of the sound patterns of words in fluent speech.

A series of four experiments examined infants' capacities to detect repeated words in fluent speech. In Experiment 1, 7 1/2-month old American infants were familiarized with two different monosyllabic words and subsequently were presented with passages which either included or did not include the familiar target words embedded in sentences. The infants listened significantly longer to the passages containing the familiar target words than to passages containing unfamiliar words. A comparable experiment with 6-month-olds provided no indication that infants at this age detected the target words in the passages. In Experiment 3, a group of 7 1/2-month-olds was familiarized with two different non-word targets which differed in their initial phonetic segment by only one or two phonetic features from words presented in two of the passages. These infants showed no tendency to listen significantly longer to the passages with the similar sounding words, suggesting that the infants may be matching rather detailed information about the items in the familiarization period to words in the test passages. Finally, Experiment 4 demonstrated that even when the 7 1/2-month-olds were initially familiarized with target words in sentential contexts rather than in isolation, they still showed reliable evidence of recognizing these words during the test phase. Taken together, the results of these studies suggest that some ability to detect words in fluent speech contexts is present by 7 1/2 months of age.

Assessment of vernier acuity development using the "equivalent intrinsic blur" paradigm.

Vernier acuity in human infants is more than two orders of magnitude poorer than in adults and does not appear to reach adult levels until well beyond the first postnatal year. One source of these developmental differences in vernier acuity may be the presence of levels of intrinsic blur that are higher in infants than in adults. We investigated this hypothesis by measuring vernier acuity in 3-month-olds, 5-month-olds, and adults using stimuli blurred by two-dimensional Gaussian filters. Experiment 1 showed that more stimulus blur is required to degrade vernier acuity in infants than in adults. From these data we estimated that the level of equivalent intrinsic blur for this vernier acuity task decreased by approx. 1.5 log units between 3 months of age and adulthood. These results also suggested that this reduction of equivalent intrinsic blur can account entirely for the improvement in vernier acuity between 3 and 5 months postnatal. However, the large further improvement which occurs between 5 months of age and adulthood cannot be explained solely by equivalent intrinsic blur. In Expt 2, we measured vernier acuity in 3-month-olds and adults using a non-blurred stimulus with the same luminance contrast as the most-blurred stimulus used in Expt 1. Infants' and adults' thresholds were degraded slightly, relative to the non-blurred stimulus from Expt 1, but still were significantly better than the most blurred condition of Expt 1. This suggests that the results from the first experiment were not due simply to the reduction in overall luminance contrast which occurs when stimuli are blurred.

Three- and four-year-olds' perceptual confusions for spoken words.

Although infants have the ability to discriminate a variety of speech contrasts, young children cannot always use this ability in the service of spoken-word recognition. The research reported here asked whether the reason young children sometimes fail to discriminate minimal word pairs is that they are less efficient at word recognition than adults, or whether it is that they employ different lexical representations. In particular, the research evaluated the proposal that young children's lexical representations are more "holistic" than those of adults, and are based on overall acoustic-phonetic properties, as opposed to phonetic segments. Three- and four-year-olds were exposed initially to an invariant target word and were subsequently asked to determine whether a series of auditory stimuli matched or did not match the target. The critical test stimuli were nonwords that varied in their degree of phonetic featural overlap with the target, as well as in terms of the position(s) within the stimuli at which they differed from the target, and whether they differed from the target on one or two segments. Data from four experiments demonstrated that the frequency with which children mistook a nonword stimulus for the target was influenced by extent of featural overlap, but not by word position. The data also showed that, contrary to the predictions of the holistic hypothesis, stimuli differing from the target by two features on a single segment were confused with the target more often than were stimuli differing by a single feature on each of two segments. This finding suggests that children use both phonetic features and segments in accessing their mental lexicons, and that they are therefore much more similar to adults than is suggested by the holistic hypothesis.

Developmental differences in infant attention to the spectral properties of infant-directed speech.

Across several independent studies, infants from a few days to 9 months of age have shown preferences for infant-directed (ID) over adult-directed (AD) speech. Moreover, 4-month-olds have been shown to prefer sine-wave analogs of the fundamental frequency of ID speech, suggesting that exaggerated pitch contours are prepotent stimuli for infants. The possibility of similar preferences by 1-month-olds was examined in a series of experiments, using a fixation-based preference procedure. Results from the first 2 experiments showed that 1-month-olds did not prefer the lower-frequency pitch characteristics of ID speech, even though 1-month-olds were able to discriminate low-pass filtered ID and AD speech. Since low-pass filtering may have distorted the fundamental frequency characteristics of ID speech, 1-month-olds were also tested with sine-wave analogs of the fundamental frequencies of the ID utterances. Infants in this third experiment also showed no preference for ID pitch contours. In the fourth experiment, 1-month-olds preferred a natural recording of ID speech over a version which preserved only its lower frequency prosodic features. From these results, it is argued that, although young infants are similar to older infants in their attraction to ID speech, their preferences depend on a wider range of acoustic features (e.g., spectral structure). It is suggested that exaggerated pitch contours which characterize ID speech may become salient communicative signals for infants through language-rich, interactive experiences with caretakers and increased perceptual acuity over the first months after birth.

Spatiotemporal factors in infant position sensitivity: single bar stimuli.

The ability of 3-month-old human infants to detect the presence of spatial misalignments in single-bar stimuli was investigated in a series of spatiotemporal stimulus manipulations. Both discrete and sinusoidal positional offsets, either stationary or temporally modulated, were presented using the forced-choice preferential looking technique. When discrete offsets were presented in alteration with the absence of offsets, thresholds were a factor of two worse than previous estimates of infant vernier acuity using grating stimuli. However, when offsets (discrete or sinusoidal) were presented in continuous motion, mean threshold was 22-24 arcmin, comparable to previous estimates using gratings with moving discrete offsets. For stimuli containing continuous motion, lowering velocity reduced infants' positional sensitivity. However, when either velocity or temporal frequency was held constant, the most important determinant of positional sensitivity was the sharpness of the offsets. The results from the temporally-modulated stimuli suggest that a simple local flicker mechanism cannot account for sensitivity when continuous motion is used in the stimulus; rather, a local motion mechanism may govern sensitivity in these conditions. We characterize this mechanism as one of "local motion" because it does have a position-sensitive component. The two stationary stimulus conditions in the present study indicate that infants can use a position-sensitive mechanism when no temporal modulation is present.

Preference for infant-directed speech in the first month after birth.

2 experiments examined behavioral preferences for infant-directed (ID) speech over adult-directed (AD) speech in young infants. Using a modification of the visual-fixation-based auditory-preference procedure, Experiments 1 and 2 examined whether 12 1-month-old and 16 2-day-old infants looked longer at a visual stimulus when looking produced ID as opposed to AD speech. The results showed that both 1-month-olds and newborns preferred ID over AD speech. Although the absolute magnitude of the ID speech preference was significantly greater, with the older infants showing longer looking durations than the younger infants, subsequent analyses showed no significant difference in the relative magnitude of this effect. Differences in overall looking times between the 2 groups apparently reflect task variables rather than differences in speech processing. These results suggest that infants' preference for the exaggerated prosodic features of ID speech is present from birth and may not depend on any specific postnatal experience. However, the possible role of prenatal auditory experience with speech is considered.

Oculomotor responses to step-ramp targets by young human infants.

Previous research has shown that the smooth pursuit system in early infancy is quite immature. Infants' tracking of a single, small target moving at velocities greater than 10 deg/sec is almost entirely saccadic until the end of the second postnatal month. The emergence of smooth pursuit is characterized by low gain (less than 0.5) and frequent saccadic intrusions. To provide a quantitative description of pursuit to relatively slow target velocities, 10 infants ranging in age from 7 to 11 weeks viewed a 2 deg target that was stepped 5 to 10 deg from screen center and then ramped back to screen center and 10 deg beyond at a constant velocity of 3, 6 or 12 deg/sec. Smooth pursuit was observed even in the youngest infant whose segments of pursuit between saccades were up to 5 sec in duration. At the slowest target velocity, mean pursuit gain across infants was 0.50, while at 6 and 12 deg/sec mean pursuit gain was 0.25 and 0.11. This systematic decrease in pursuit gain with increasing target velocity implies that pursuit velocity was invariant across the three target velocities. These findings suggest that smooth pursuit can be generated consistently by the end of the second postnatal month, but that it is slow and uncalibrated to the velocity of the target.

Identification of vowels in "vowel-less" syllables by 3-year-olds.

The ability of 3-year-old children to perceive the identity of vowels in full-vowel and silent-center, consonant-vowel-consonant (CVC) syllables was investigated using a two-alternative pointing procedure. Silence replaced the middle 10%, 35%, 65%, or 90% of the steady-state formants of synthetic "bad" and "bud" syllables. Identification of the two full-vowel syllables was 87% correct, whereas performance for the silent-center syllables was somewhat lower (72%, 70%, 67%, and 66% correct for the 10%, 35%, 65%, and 90% deletion stimuli, respectively). The performance of individual children fell into two subgroups: (1) those who performed like adults by maintaining correct vowel identification for all of the silent-center syllables, and (2) those who identified the full-vowel syllables correctly but performed at chance for all of the silent-center syllables. Three additional experiments showed that none of the children performed poorly when noise replaced the gap in the silent-center syllables. These results demonstrate that many 3-year-olds can identify vowels correctly in CVC syllables in the absence of the full spectral properties of steady-state formants.

Discrimination of frequency transitions by human infants.

Difference limens (DLs) for linear frequency transitions using a 1.0-kHz pulsed-tone standard were obtained from 6- to 9-month-old human infants in a series of three experiments. A repeating standard "yes-no" operant headturning technique and an adaptive staircase (tracking) procedure were used to obtain difference limens from a total of 71 infants. The DLs for 300-ms upward and downward linear frequency sweeps were approximately 3%-4% when the repeating standard was an unmodulated 1.0-kHz pulsed tone of 300-ms duration. These DLs for frequency sweeps were not significantly different from DLs for frequency increments and decrements using 330-ms pulsed tones [J. M. Sinnott and R. N. Aslin, J. Acoust. Soc. Am. 78, 1986-1992 (1985)]. The DLs for frequency sweeps of 50 ms appended to the beginning or the end of a 250-ms unmodulated 1.0-kHz tone were approximately 6%-7%. This greater DL for brief frequency sweeps was confirmed by varying the duration but not the extent of the sweep. Finally, DLs were greater than 50% when the repeating standard was a 50-ms rising or falling frequency sweep appended to the beginning of a 250-ms unmodulated 1.0-kHz tone. These results suggest that rapid frequency transitions are much more difficult to discriminate from frequency transitions of the same category (rising or falling) than from either a frequency transition of the opposite category (falling or rising) or an unmodulated tone.(ABSTRACT TRUNCATED AT 250 WORDS)

The language environment of the young infant: implications for early perceptual development.

The purpose of this article is to review recent research on the young infant's use of voice-specific as well as voice-nonspecific auditory information during early language processing, to suggest a possible mechanism that biases the young infant towards this information, and to discuss potential implications of the early saliency of this information for later language development. Auditory preferences expressed by young infants are of interest because they demonstrate which properties of complex auditory events are effective in capturing the infant's attention. Moreover, the presence of such auditory preferences has led to speculations about their possible origins in both pre- and postnatal auditory experience. Research examining the role of early auditory experience in the formation of preferences is presented, along with a discussion of how the constrained nature of early auditory experience (particularly prenatal) may bias the young infant towards specific features of maternal as well as nonmaternal speech.