Statistical Learning

Learning unattested languages

by Sara Finley

Cogsci 2012 paper.

This paper demonstrates the role of morphological alternations in learning novel phonotactic patterns. In an... more This paper demonstrates the role of morphological alternations in learning novel phonotactic patterns. In an artificial grammar learning task, adult learners were exposed to a phonotactic pattern in which the first and last consonant agreed in voicing. Long-distance phonotactics encoded as strictly piecewise languages suggest that first-last phonotactic patterns should be unattested in natural language. However, recent theories of morphologically induced phonological patterns predict that long-distance agreement between the first and last consonant of a word can occur when the agreement is induced as a morphological alternation. The results of two experiments support the prediction that first-last harmony patterns are more easily learned when morphological cues to the pattern are present. Participants only learned the first-last pattern when presented as a morphological alternation.

Statistical Learning and Language: An Individual Differences Study

by Jennifer Misyak

Misyak, J. B. & Christiansen, M. H. (2012). Statistical learning and language: An individual differences study. Language Learning, 62, 302-331.

How Many Makes a Crowd? On the Evolution of Learning as a Factor of Community Coverage

by Yaniv Altshuler

As truly ubiquitous wearable computers, mobile phones are quickly becoming the primary source for social, behavioral... more As truly ubiquitous wearable computers, mobile phones are quickly becoming the primary source for social, behavioral and environmental sensing and data collection. Today’s smartphones are equipped with increasingly more sensors and accessible data types that enable the collection of literally dozens of signals related to the phone, its user, and its environment. A great deal of research effort in academia and industry is put into mining this raw data for higher level sense-making, such as understanding user context, inferring social networks, learning individual features, and so on. In many cases, this analysis work is the result of exploratory forays and trial-and-error. In this work we investigate the properties of learning and inferences of real world data collected via mobile phones for different sizes of analyzed networks. In particular, we examine how the ability to predict individual features and social links is incrementally enhanced with the accumulation of additional data. To accomplish this, we use the Friends and Family dataset, which contains rich data signals gathered from the smartphones of 130 adult members of a young-family residential community over the course of a year and consequently has become one of the most comprehensive mobile phone datasets gathered in academia to date. Our results show that features such as ethnicity, age and marital status can be detected by analyzing social and behavioral signals. We then investigate how the prediction accuracy is increased when the users sample set grows. Finally, we propose a method for advanced prediction of the maximal learning accuracy possible for the learning task at hand, based on an initial set of measurements. These predictions have practical implications, such as influencing the design of mobile data collection campaigns or evaluating analysis strategies.

Towards a Substantively Biased Theory of Learning

by Sara Finley

Right-to-Left Biases for Vowel Harmony: Evidence from Artificial Grammar

by Sara Finley

Repetition streaks increase perceptual sensitivity in visual search of brief displays

by Heida Maria Sigurdardottir

Frank, M., C, Tily, H., Arnon, I., & Goldwater, S. (2010). Beyond Transitional Probabilities: Human Learners Impose a Parsimony Bias in Statistical Word Segmentation. Proceedings of the 32nd Annual Conference of the Cognitive Science Society (pp. 000-000). Cognitive Science Society

by Inbal Arnon

CogSci proceedings

Human infants and adults are able to segment coherent sequences from unsegmented strings of auditory stimuli after... more Human infants and adults are able to segment coherent sequences from unsegmented strings of auditory stimuli after only a short exposure, an ability thought to be linked to early
language acquisition. Although some research has hypothesized that learners succeed in these tasks by computing transitional probabilities between syllables, current experimental results do not differentiate between a range of models of different computations that learners could perform. We created a set of stimuli that was consistent with two different lexicons—one consisting of two-syllable words and one of three-syllable words—but where transition probabilities would not lead learners to segment sentences consistently according to either lexicon. Participants’ responses formed a distribution over possible segmentations that included consistent segmentations into both two- and three-syllable words, suggesting that learners do not use pure transitional probabilities to segment but instead impose a bias towards parsimony on the lexicons they learn.

Language learning and adaptation: Getting explicit about implicit learning

by Hartmut Fitz

Co-authored with Franklin Chang and Marius Janciauskas, University of Liverpool.

Linguistic adaptation is a phenomenon where language representations change in response to linguistic input.... more Linguistic adaptation is a phenomenon where language representations change in response to linguistic input. Adaptation can occur on multiple linguistic levels such as phonology (tuning of phonotactic constraints), words (repetition priming), and syntax (structural priming). The persistent nature of these adaptations suggests that they may be a form of implicit learning and connectionist models have been developed which instantiate this hypothesis. Research on implicit learning, however, has also produced evidence that explicit chunk knowledge is involved in the performance of these tasks. In this review, we examine how these interacting implicit and explicit processes may change our understanding of language learning and processing.

Learning simple statistics for language comprehension and production: The CAPPUCCINO model

by Stewart McCauley

McCauley, S.M. & Christiansen, M.H. (2011). Learning simple statistics for language comprehension and production: The CAPPUCCINO model. In L. Carlson, C. Hölscher, & T. Shipley (Eds.), Proceedings of the 33rd Annual Conference of the Cognitive Science Society (pp. 1619-1624). Austin, TX: Cognitive Science Society.

Whether the input available to children is sufficient to explain their ability to use language has been the subject of... more Whether the input available to children is sufficient to explain their ability to use language has been the subject of much theoretical debate in cognitive science. Here, we present a simple, developmentally motivated computational model that learns to comprehend and produce language when exposed to child-directed speech. The model uses backward transitional probabilities to create an inventory of ‘chunks’ consisting of one or more words. Language comprehension is approximated in terms of shallow parsing of adult speech and production as the reconstruction of the child’s actual utterances. The model functions in a fully incremental, on-line fashion, has broad cross-linguistic coverage, and is able to fit child data from Saffran’s (2002) statistical learning study. Moreover, word-based distributional information is found to be more useful than statistics over word classes. Together, these results suggest that much of children’s early linguistic behavior can be accounted for in a usage-based manner using distributional statistics.

Statistical Learning of Complex Questions

by Hartmut Fitz

Proceedings of the 32nd Annual Conference of the Cognitive Science Society, 2010, pages 2692--2698.

The problem of auxiliary fronting in complex polar questions occupies a prominent position within the nature versus... more The problem of auxiliary fronting in complex polar questions occupies a prominent position within the nature versus nurture controversy in language acquisition. We employ a model of statistical learning which uses sequential and semantic information to produce utterances from a bag of words. This linear learner is capable of generating grammatical questions without exposure to these structures in its training environment. We show that the model performs superior to n-gram learners on this task. Implications for nativist theories of language acquisition are discussed.

A liquid-state model of variability effects in learning nonadjacent dependencies

by Hartmut Fitz

In L. Carlson, C. Hölscher and T. Shipley (Eds.), Proceedings of the 33rd Annual Conference of the Cognitive Science Society (pp. 897–903). Austin, TX: Cognitive Science Society.

Language acquisition involves learning nonadjacent dependencies that can exist between words in a sentence. Several... more Language acquisition involves learning nonadjacent dependencies that can exist between words in a sentence. Several artificial grammar learning studies have shown that the human ability to detect dependencies between A and B in sequences AXB is influenced by the amount of variation in the X element. This paper presents a model of statistical learning that displays similar behavior on this task and generalizes in a human-like way. The model was also used to predict human behavior for increased distance and more variation in dependencies. We compare this model-based approach with the standard invariance account of the variability effect.

The role of negative and positive evidence in adult phonological learning

by Sara Finley

Penn Linguistics Colloquium proceedings paper

One of the great mysteries of language development is how children acquire language so efficiently while adults are... more One of the great mysteries of language development is how children acquire language so efficiently while adults are never able to reach the same level of proficiency. Adding to this mystery is that child learners rarely receive negative evidence regarding the nature of the grammatical structure of their language, but adults are more likely to receive and use such evidence (in classes, corrections, etc.) (Baker, 1979). The present study tests the role of negative evidence in adult language learners, who were exposed to an artificial grammar characterized by round vowel harmony, a phonological process whereby vowels agree in the feature round. Participants were exposed to either positive evidence only (Positive Evidence Condition), or both negative and positive evidence (Positive Evidence Condition). In two experiments, participants in the Positive Evidence Condition outperformed participants in the Negative Evidence Condition, specifically for test items tat measured extension of learned items to novel items. These results suggest that negative evidence may hinder adult grammatical rule learning.

The Perception of Tonal Structure Through the Differentiation and Organization of Pitches.

by Mark Schmuckler

Morpheme segmentation in school-aged children

by Sara Finley

How Implicit is Statistical Learning?

by Phillip Hamrick

Co-authored with Patrick Rebuschat. To appear in P. Rebuschat & J. N. Williams (Eds.) Statistical learning and language acquisition. Berlin: Mouton de Gruyter.

An important difference between implicit learning and statistical learning research is that implicit learning studies... more An important difference between implicit learning and statistical learning research is that implicit learning studies contain measures of awareness, while statistical learning studies do not. As such, it is unclear whether statistical learning typically results in explicit (conscious) or implicit (unconscious) knowledge. To address this gap, the present study investigated whether a typical statistical learning experiment results in implicit knowledge, explicit knowledge, or both. The experiment combined the cross-situational word learning paradigm (Yu & Smith, 2007) and subjective measures of awareness (Dienes & Scott, 2005). Thirty subjects were exposed to an artificial vocabulary under incidental or intentional learning conditions. The results show a clear learning effect in both learning conditions, with a more robust performance in the intentional group. The subjective measures of awareness further indicate that subjects in the intentional group developed both implicit and explicit knowledge, while the subjects in the incidental group developed primarily implicit knowledge. The experiment illustrates the usefulness of including measures of awareness when researching statistical learning.

Expectancy modulates a Late Positive ERP in an artificial grammar task

by Angel Tabullo

Brain Res. 2011 Feb 10;1373:131-43

A wide range of studies have found late positive ERP components in response to anomalies during processing of... more A wide range of studies have found late positive ERP components in response to anomalies during processing of structured sequences. In language studies, this component is named Syntactic Positive Shift (SPS) or P600. It is characterized by an increase in potential peaking around 600 ms after the appearance of the syntactic anomaly and has a centroparietal topography. Similar late positive components were found more recently in non-linguistic contexts. These results have led to the hypothesis that these components index the detection of anomalies in rule-governed sequences, or the access to abstract rule representations, regardless of the nature of the stimuli. Additionally, there is evidence showing that the SPS/P600 is sensitive to probability manipulations, which affect the subjects’ expectancy of the stimuli. Our aim in the present work was to address the hypothesis that the late positive component is modulated by the subject’s expectancy of the stimuli. To do so, we employed an artificial grammar learning task, and controlled the frequency of presentation to different kind of sequences during training. Results showed that certain sequence types elicited a late positive component which was modulated by different factors in two distinct time-windows. In an earlier window, the component was higher for sequences which had a low or null probability of occurrence during training, while in a later window, the component was higher for incorrect than correct sequences. Furthermore, this late-window effect was absent in those subjects whose performance was not significantly above chance. Two possible explanations for this effect are suggested.

Multimodal Transfer of Repetition Patterns in Artificial Grammar Learning

by Sara Finley

2011 cogsci paper with Morten Chrristiansen

Extending learned patterns to previously unseen ones is a key
hallmark of complex cognition. This paper presents... more Extending learned patterns to previously unseen ones is a key
hallmark of complex cognition. This paper presents evidence
that learners are able to generalize learned patterns to novel
stimuli with very different surface properties within and
across modalities. Using a statistical learning paradigm, adult
learners were exposed to a repetition (reduplication) pattern in
which the first element of a three-element sequence repeated
(e.g., ABAAB). The pattern was presented as either
spoken repetition (e.g., bago, babago) or a non-linguistic
visual analogue (i.e., repetition of non-nameable shapes).
Learners showed significant transfer from a non-linguistic
repetition pattern to a linguistic reduplication pattern, and vice
versa. However, we found a small bias towards linguistic
reduplication, as responses to linguistic patterns were
numerically higher. This suggests that while learners are able
to extend learned patterns to novel patterns in other domains,
factors such as familiarity and naturalness may privilege
linguistic patterns over non-linguistic analogues.

Generalization to Novel Consonants in Artificial Grammar Learning

by Sara Finley

2011 cogsci paper

While theoretical phonologists rely on abstract phonetic
features to account for the variety of phonological... more While theoretical phonologists rely on abstract phonetic
features to account for the variety of phonological patterns
that exist in the world’s languages, it is unclear whether such
abstract representations bear psychological reality. Previous
research has shown that learners in artificial grammar
learning experiments are able to generalize a newly learned
phonological pattern to novel segments, suggesting that
learners are able to form abstract, feature-based
representations. However, conflicting results suggest that this
level of abstraction may be restricted to vowels, rather than
consonants. The present experiment extends previous findings
on generalization to novel segments in vowel harmony to an
analogous pattern, consonant harmony. We show that learners
fail to generalize to novel consonants in consonant harmony,
but succeed at generalization to novel consonants in a general,
consonant deletion pattern. Implications for the role of
distinctive