The longevity of statistical learning: When infant memory decays, isolated words come to the rescue

A changing role for transitional probabilities in word learning during the transition to toddlerhood?

Infants' sensitivity to transitional probabilities (TPs) supports language development by facilitating mapping high-TP (HTP) words to meaning, at least up to 18 months of age. Here we tested whether this HTP advantage holds as lexical development progresses, and infants become better at forming word-referent mappings. Two groups of 24-month-olds (N = 64 and all White, tested in the United States) first listened to Italian sentences containing HTP and low-TP (LTP) words. We then used HTP and LTP words, and sequences that violated these statistics, in a mapping task. Infants learned HTP and LTP words equally well. They also learned LTP violations as well as LTP words, but learned HTP words better than HTP violations. Thus, by 2 years of age sensitivity to TPs does not lead to an HTP advantage but rather to poor mapping of violations of HTP word forms. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Can Infants Retain Statistically Segmented Words and Mappings Across a Delay?

Infants are sensitive to statistics in spoken language that aid word-form segmentation and immediate mapping to referents. However, it is not clear whether this sensitivity influences the formation and retention of word-referent mappings across a delay, two real-world challenges that learners must overcome. We tested how the timing of referent training, relative to familiarization with transitional probabilities (TPs) in speech, impacts English-learning 23-month-olds' ability to form and retain word-referent mappings. In Experiment 1, we tested infants' ability to retain TP information across a 10-min delay and use it in the service of word learning. Infants successfully mapped high-TP but not low-TP words to referents. In Experiment 2, infants readily mapped the same words even when they were unfamiliar. In Experiment 3, high- and low-TP word-referent mappings were trained immediately after familiarization, and infants readily remembered these associations 10 min later. In sum, although 23-month-old infants do not need strong statistics to map word forms to referents immediately, or to remember those mappings across a delay, infants are nevertheless sensitive to these statistics in the speech stream, and they influence mapping after a delay. These findings suggest that, by 23 months of age, sensitivity to statistics in speech may impact infants' language development by leading word forms with low coherence to be poorly mapped following even a short period of consolidation.

Hebbian learning can explain rhythmic neural entrainment to statistical regularities

In many domains, learners extract recurring units from continuous sequences. For example, in unknown languages, fluent speech is perceived as a continuous signal. Learners need to extract the underlying words from this continuous signal and then memorize them. One prominent candidate mechanism is statistical learning, whereby learners track how predictive syllables (or other items) are of one another. Syllables within the same word predict each other better than syllables straddling word boundaries. But does statistical learning lead to memories of the underlying words-or just to pairwise associations among syllables? Electrophysiological results provide the strongest evidence for the memory view. Electrophysiological responses can be time-locked to statistical word boundaries (e.g., N400s) and show rhythmic activity with a periodicity of word durations. Here, I reproduce such results with a simple Hebbian network. When exposed to statistically structured syllable sequences (and when the underlying words are not excessively long), the network activation is rhythmic with the periodicity of a word duration and activation maxima on word-final syllables. This is because word-final syllables receive more excitation from earlier syllables with which they are associated than less predictable syllables that occur earlier in words. The network is also sensitive to information whose electrophysiological correlates were used to support the encoding of ordinal positions within words. Hebbian learning can thus explain rhythmic neural activity in statistical learning tasks without any memory representations of words. Learners might thus need to rely on cues beyond statistical associations to learn the words of their native language. RESEARCH HIGHLIGHTS: Statistical learning may be utilized to identify recurring units in continuous sequences (e.g., words in fluent speech) but may not generate explicit memory for words. Exposure to statistically structured sequences leads to rhythmic activity with a period of the duration of the underlying units (e.g., words). I show that a memory-less Hebbian network model can reproduce this rhythmic neural activity as well as putative encodings of ordinal positions observed in earlier research. Direct tests are needed to establish whether statistical learning leads to declarative memories for words.

Speech Segmentation and Cross-Situational Word Learning in Parallel

Language learners track conditional probabilities to find words in continuous speech and to map words and objects across ambiguous contexts. It remains unclear, however, whether learners can leverage the structure of the linguistic input to do both tasks at the same time. To explore this question, we combined speech segmentation and cross-situational word learning into a single task. In Experiment 1, when adults (N = 60) simultaneously segmented continuous speech and mapped the newly segmented words to objects, they demonstrated better performance than when either task was performed alone. However, when the speech stream had conflicting statistics, participants were able to correctly map words to objects, but were at chance level on speech segmentation. In Experiment 2, we used a more sensitive speech segmentation measure to find that adults (N = 35), exposed to the same conflicting speech stream, correctly identified non-words as such, but were still unable to discriminate between words and part-words. Again, mapping was above chance. Our study suggests that learners can track multiple sources of statistical information to find and map words to objects in noisy environments. It also prompts questions on how to effectively measure the knowledge arising from these learning experiences.

Statistical word segmentation: Anchoring learning across contexts

The present experiments were designed to assess infants' abilities to use syllable co-occurrence regularities to segment fluent speech across contexts. Specifically, we investigated whether 9-month-old infants could use statistical regularities in one speech context to support speech segmentation in a second context. Contexts were defined by different word sets representing contextual differences that might occur across conversations or utterances. This mimics the integration of information across multiple interactions within a single language, which is critical for language acquisition. In particular, we performed two experiments to assess whether a statistically segmented word could be used to anchor segmentation in a second, more challenging context, namely speech with variable word lengths. The results of Experiment 1 were consistent with past work suggesting that statistical learning may be hindered by speech with word-length variability, which is inherent to infants' natural speech environments. In Experiment 2, we found that infants could use a previously statistically segmented word to support word segmentation in a novel, challenging context. We also present findings suggesting that this ability was associated with infants' early word knowledge but not their performance on a cognitive development assessment.

Seeing a talking face matters: Infants' segmentation of continuous auditory-visual speech

Visual speech cues from a speaker's talking face aid speech segmentation in adults, but despite the importance of speech segmentation in language acquisition, little is known about the possible influence of visual speech on infants' speech segmentation. Here, to investigate whether there is facilitation of speech segmentation by visual information, two groups of English-learning 7-month-old infants were presented with continuous speech passages, one group with auditory-only (AO) speech and the other with auditory-visual (AV) speech. Additionally, the possible relation between infants' relative attention to the speaker's mouth versus eye regions and their segmentation performance was examined. Both the AO and the AV groups of infants successfully segmented words from the continuous speech stream, but segmentation performance persisted for longer for infants in the AV group. Interestingly, while AV group infants showed no significant relation between the relative amount of time spent fixating the speaker's mouth versus eyes and word segmentation, their attention to the mouth was greater than that of AO group infants, especially early in test trials. The results are discussed in relation to the possible pathways through which visual speech cues aid speech perception.

Hebbian, correlational learning provides a memory-less mechanism for Statistical Learning irrespective of implementational choices: Reply to Tovar and Westermann (2022)

Statistical learning relies on detecting the frequency of co-occurrences of items and has been proposed to be crucial for a variety of learning problems, notably to learn and memorize words from fluent speech. Endress and Johnson (2021) (hereafter EJ) recently showed that such results can be explained based on simple memory-less correlational learning mechanisms such as Hebbian Learning. Tovar and Westermann (2022) (hereafter TW) reproduced these results with a different Hebbian model. We show that the main differences between the models are whether temporal decay acts on both the connection weights and the activations (in TW) or only on the activations (in EJ), and whether interference affects weights (in TW) or activations (in EJ). Given that weights and activations are linked through the Hebbian learning rule, the networks behave similarly. However, in contrast to TW, we do not believe that neurophysiological data are relevant to adjudicate between abstract psychological models with little biological detail. Taken together, both models show that different memory-less correlational learning mechanisms provide a parsimonious account of Statistical Learning results. They are consistent with evidence that Statistical Learning might not allow learners to learn and retain words, and Statistical Learning might support predictive processing instead.

When statistics collide: The use of transitional and phonotactic probability cues to word boundaries

Statistical regularities in linguistic input, such as transitional probability and phonotactic probability, have been shown to promote speech segmentation. It remains unclear, however, whether or how the combination of transitional probabilities and subtle phonotactic probabilities influence segmentation. The present study provides a fine-grained investigation of the effects of such combined statistics. Adults (N = 81) were tested in one of two conditions. In the Anchor condition, they heard a continuous stream of words with small differences in phonotactic probabilities. In the Uniform condition, all words had comparable phonotactic probabilities. In both conditions, transitional probability was stronger in words than in part-words. Only participants from the Anchor condition preferred words at test, indicating that the combination of transitional probabilities and subtle phonotactic probabilities may facilitate speech segmentation. We discuss the methodological implications of our findings, which demonstrate that even small phonotactic variations should be accounted for when investigating statistical speech segmentation.

One language or two? Navigating cross-language conflict in statistical word segmentation

Bilingual infants must navigate the similarities and differences between their languages to achieve native proficiency in childhood. Bilinguals learning to find individual words in fluent speech face the possibility of conflicting cues to word boundaries across their languages. Despite this challenge, bilingual infants typically begin to segment and learn words in both languages around the same time as monolinguals. It is possible that early bilingual experience may support infants' abilities to track regularities relevant for word segmentation separately across their languages. In a dual speech stream statistical word segmentation task, we assessed whether 16-month-old infants could track syllable co-occurrence regularities in two artificial languages despite conflicting information across the languages. We found that bilingual, but not monolingual, infants were able to segment the dual speech streams using statistical regularities. Although the two language groups did not differ on secondary measures of cognitive and linguistic development, bilingual infants' real-world experience with bilingual speakers was predictive of their performance in the dual language statistical segmentation task.

Do Infants Learn Words From Statistics? Evidence From English-Learning Infants Hearing Italian

Infants are sensitive to statistical regularities (i.e., transitional probabilities, or TPs) relevant to segmenting words in fluent speech. However, there is debate about whether tracking TPs results in representations of possible words. Infants show preferential learning of sequences with high TPs (HTPs) as object labels relative to those with low TPs (LTPs). Such findings could mean that only the HTP sequences have a word-like status, and they are more readily mapped to a referent for that reason. But these findings could also suggest that HTP sequences are easier to encode, just like any other predictable sequence. Here we aimed to distinguish between these explanations. To do so, we built on findings that infants become resistant to learning labels that are not typical of their native language as they approach 2 years of age and add words to their lexicons. If tracking TPs in speech results in identifying candidate words, at this age TPs may have reduced power to confer lexical status when they yield a unit that is very dissimilar to word forms that are typical of infants' native language. Indeed, we found that at 20 months, English-learning infants with relatively small vocabularies learned HTP Italian words (but not LTP words) as object labels, while infants with larger vocabularies resisted learning HTP Italian words. These findings suggest that the HTP sequences may be represented as candidate words, and more broadly, that TP statistics are relevant to word learning.