Observed effects of "distributional learning" may not relate to the number of peaks. A test of "dispersion" as a confounding factor

What do you learn from a single cue? Dimensional reweighting and cue reassociation from experience with a newly unreliable phonetic cue

In language comprehension, we use perceptual cues to infer meanings. Some of these cues reside on perceptual dimensions. For example, the difference between bear and pear is cued by a difference in voice onset time (VOT), which is a continuous perceptual dimension. The present paper asks whether, and when, experience with a single value on a dimension behaving unexpectedly is used by the learner to reweight the whole dimension. We show that learners reweight the whole VOT dimension when exposed to a single VOT value (e.g., 45 ms) and provided with feedback indicating that the speaker intended to produce a /b/ 50% of the time and a /p/ the other 50% of the time. Importantly, dimensional reweighting occurs only if 1) the 50/50 feedback is unexpected for the VOT value, and 2) there is another dimension that is predictive of feedback. When no predictive dimension is available, listeners reassociate the experienced VOT value with the more surprising outcome but do not downweight the entire VOT dimension. These results provide support for perceptual representations of speech sounds that combine cues and dimensions, for viewing perceptual learning in speech as a combination of error-driven cue reassociation and dimensional reweighting, and for considering dimensional reweighting to be reallocation of attention that occurs only when there is evidence that reallocating attention would improve prediction accuracy (Harmon, Z., Idemaru, K., & Kapatsinski, V. 2019. Learning mechanisms in cue reweighting. Cognition, 189, 76-88.).

School-Aged Children Learn Novel Categories on the Basis of Distributional Information

Categorization of sensory stimuli is a vital process in understanding the world. In this paper we show that distributional learning plays a role in learning novel object categories in school-aged children. An 11-step continuum was constructed based on two novel animate objects by morphing one object into the other in 11 equal steps. Forty-nine children (7–9 years old) were subjected to one of two familiarization conditions during which they saw tokens from the continuum. The conditions differed in the position of the distributional peaks along the continuum. After familiarization it was tested how the children categorized the stimuli. Results show that, in line with our expectations, familiarization condition influenced categorization during the test phase, indicating that the frequency distribution of tokens in the input had induced novel object category formation. These results suggest that distributional learning could play an important role in categorizing sensory stimuli throughout life.

Distributional learning of speech sound categories is gated by sensitive periods

Perceptual attunement to the native phonetic repertoire occurs over the first year of life: an infant's discrimination of non-native phonetic contrasts declines while their discrimination of native phonetic contrasts improves, with the timing of change consistent with sensitive periods. The statistics of speech sound distributions is one source of input used to collapse non-native phonetic category boundaries, while sharpening native ones. Distributional learning can be a domain-general mechanism, yet given the timing of perceptual attunement, we hypothesized that this learning mechanism may be maturationally delimited in the content domain of phonetic categories. Here, we assessed whether sensitivity to the distribution of speech sounds in the environment declines as the period of perceptual attunement closes. We used electroencephalography (EEG) to investigate whether neuronal responses to native 'ra' and 'la' phones are modulated differently in older vs young infants by exposure to either a bimodal or unimodal sound distribution spanning the [r] ~ [l] phoneme space. The native contrast, ra-la, is discriminable at all three ages, ensuring that we were testing the distributional learning mechanism, rather than confounding it with a decline in discrimination to a non-native distinction. English monolingual infants (n = 131) at 5-, 9- and 12-months-old were familiarized to either a unimodal or bimodal distribution of /ra/-/la/ speech sounds. Immediately following familiarization, an ERP oddball task was used to assess discrimination. Results showed that brief exposure to a bi- vs uni-modal distribution is sufficient to alter neuronal responses to subsequent /ra/ vs /la/ speech sounds at 5-months and 9-months, but not at 12-months. These results are the first to capture a progressive decline in sensitivity to distributional statistics in the environment. A potential mechanistic explanation based on critical period biology is discussed.

Statistical learning of unbalanced exclusive-or temporal sequences in humans

A pervasive issue in statistical learning has been to determine the parameters of regularity extraction. Our hypothesis was that the extraction of transitional probabilities can prevail over frequency if the task involves prediction. Participants were exposed to four repeated sequences of three stimuli (XYZ) with each stimulus corresponding to the position of a red dot on a touch screen that participants were required to touch sequentially. The temporal and spatial structure of the positions corresponded to a serial version of the exclusive-or (XOR) that allowed testing of the respective effect of frequency and first- and second-order transitional probabilities. The XOR allowed the first-order transitional probability to vary while being not completely related to frequency and to vary while the second-order transitional probability was fixed (p(Z|X, Y) = 1). The findings show that first-order transitional probability prevails over frequency to predict the second stimulus from the first and that it also influences the prediction of the third item despite the presence of second-order transitional probability that could have offered a certain prediction of the third item. These results are particularly informative in light of statistical learning models.

Of mice and men: Speech sound acquisition as discriminative learning from prediction error, not just statistical tracking

Despite burgeoning evidence that listeners are highly sensitive to statistical distributions of speech cues, the mechanism underlying learning may not be purely statistical tracking. Decades of research in animal learning suggest that learning results from prediction and prediction error. Two artificial language learning experiments test two predictions that distinguish error-driven from purely statistical models; namely, cue competition - specifically, Kamin's (1968) 'blocking' effect (Experiment 1) - and the predictive structure of learning events (Experiment 2). In Experiment 1, prior knowledge of an informative cue blocked learning of a second cue. This finding may help explain second language learners' difficulty in acquiring native-level perception of non-native speech cues. In Experiment 2, learning was better with a discriminative (cue-outcome) order compared to a non-discriminative (outcome-cue) order. Experiment 2 suggests that learning speech cues, including reversing effects of blocking, depends on (un)learning from prediction error and depends on the temporal order of auditory cues versus semantic outcomes. Together, these results show that (a) existing knowledge of acoustic cues can block later learning of new cues, and (b) speech sound acquisition depends on the predictive structure of learning events. When feedback from prediction error is available, this drives learners to ignore salient non-discriminative cues and effectively learn to use target cue dimensions. These findings may have considerable implications for the field of speech acquisition.

Training a non-native vowel contrast with a distributional learning paradigm results in improved perception and production

Previous distributional learning research suggests that adults can improve perception of a non-native contrast more efficiently when exposed to a bimodal than a unimodal distribution. Studies have also suggested that perceptual learning can transfer to production. The current study tested whether the addition of visual images to reinforce the contrast and active learning with feedback would result in lcearning in both conditions and would transfer to gains in production. Native English-speaking adults heard stimuli from a bimodal or unimodal /o/-/œ/ continuum. No group differences were found on a discrimination task, possibly suggesting that the supports eliminated previously documented group differences. On an identification task, listeners in the bimodal group showed better performance than the unimodal group on the endpoint stimuli. Production results indicated that both groups showed increased Euclidean distance between the target vowels after training, suggesting that perceptual training improved production skills in both conditions. Contrary to expectations, degree of perception and production learning were not correlated. Together, these results suggest that a bimodal distribution may aid learning, but that adding images to reinforce the contrast and active learning to the training paradigm could mitigate disadvantages found previously for participants exposed to a unimodal distribution.

Learning mechanisms in cue reweighting

Feedback has been shown to be effective in shifting attention across perceptual cues to a phonological contrast in speech perception (Francis, Baldwin & Nusbaum, 2000). However, the learning mechanisms behind this process remain obscure. We compare the predictions of supervised error-driven learning (Rescorla & Wagner, 1972) and reinforcement learning (Sutton & Barto, 1998) using computational simulations. Supervised learning predicts downweighting of an informative cue when the learner receives evidence that it is no longer informative. In contrast, reinforcement learning suggests that a reduction in cue weight requires positive evidence for the informativeness of an alternative cue. Experimental evidence supports the latter prediction, implicating reinforcement learning as the mechanism behind the effect of feedback on cue weighting in speech perception. Native English listeners were exposed to either bimodal or unimodal VOT distributions spanning the unaspirated/aspirated boundary (bear/pear). VOT is the primary cue to initial stop voicing in English. However, lexical feedback in training indicated that VOT was no longer predictive of voicing. Reduction in the weight of VOT was observed only when participants could use an alternative cue, F0, to predict voicing. Frequency distributions had no effect on learning. Overall, the results suggest that attention shifting in learning the phonetic cues to phonological categories is accomplished using simple reinforcement learning principles that also guide the choice of actions in other domains.