Search

To link to full-text access for this article, visit this link: http://dx.doi.org/10.1016/j.cogpsych.2007.04.002 Byline: Elizabeth Wonnacott (a), Elissa L. Newport (b), Michael K. Tanenhaus (b) Keywords: Language acquisition; Sentence processing; Verb argument structures; Eye-tracking; Artificial language learning Abstract: Adult knowledge of a language involves correctly balancing lexically-based and more language-general patterns. For example, verb argument structures may sometimes readily generalize to new verbs, yet with particular verbs may resist generalization. From the perspective of acquisition, this creates significant learnability problems, with some researchers claiming a crucial role for verb semantics in the determination of when generalization may and may not occur. Similarly, there has been debate regarding how verb-specific and more generalized constraints interact in sentence processing and on the role of semantics in this process. The current work explores these issues using artificial language learning. In three experiments using languages without semantic cues to verb distribution, we demonstrate that learners can acquire both verb-specific and verb-general patterns, based on distributional information in the linguistic input regarding each of the verbs as well as across the language as a whole. As with natural languages, these factors are shown to affect production, judgments and real-time processing. We demonstrate that learners apply a rational procedure in determining their usage of these different input statistics and conclude by suggesting that a Bayesian perspective on statistical learning may be an appropriate framework for capturing our findings. Author Affiliation: (a) Department of Experimental Psychology, University of Oxford, South Parks Road, Oxford OX1 3UD, UK (b) Department of Brain & Cognitive Sciences, University of Rochester, Rochester, NY 14627, USA Article History: Accepted 25 April 2007 Article Note: (footnote) [star] This research was supported by National Institute of Health research Grants DC-00167 and NIH HD-27206 awarded to Elissa Newport and Michael Tanenhaus, respectively, and by ESRC Grant PTA-026-1296 awarded to Elizabeth Wonnacott. Many thanks to Edward Longhurst, who wrote the ExBuilder software which ran these experiments, to Dr. Ted Supalla and Don Metlay for providing their time and resources to create the video stimuli, to those working in the lab who were involved in creating stimuli and running participants: Dana Subik, Whitney Hopfinger, Carol Faden, Maggie Chang, Catherine Krafft, Katie Schuler, Joyce Akwaa and Katie Dickerson, and to Dr. Jeff Runner and Dr. Joyce McDonough for their insightful comments on these topics.

Successful language acquisition requires both generalization and lexically based learning. Previous research suggests that this is achieved, at least in part, by tracking distributional statistics at and above the level of lexical items. We explored this learning using a semi-artificial language learning paradigm with 6-year-olds and adults, looking at learning of co-occurrence relationships between (meaningless) particles and English nouns. Both age groups showed stronger lexical learning (and less generalization) given “skewed” languages where a majority particle co-occurred with most nouns. In addition, adults, but not children, were affected by overall lexicality, showing weaker lexical learning (more generalization) when some input nouns were seen to alternate (i.e. occur with both particles). The results suggest that restricting generalization is affected by distributional statistics above the level of words/bigrams. Findings are discussed within the framework offered by models capturing generalization as rational inference, namely hierarchical-Bayesian and simplicity-based models. [ABSTRACT FROM AUTHOR]

We present a novel experimental technique using artificial language learning to investigate the relationship between structural priming during communicative interaction, and linguistic regularity. We use unpredictable variation as a test-case, because it is a well-established paradigm to study learners’ biases during acquisition, transmission and interaction. We trained participants on artificial languages exhibiting unpredictable variation in word order, and subsequently had them communicate using these artificial languages. We found evidence for structural priming in two different grammatical constructions and across human-human and human-computer interaction. Priming occurred regardless of behavioral convergence: communication led to shared word order use only in human-human interaction, but priming was observed in all conditions. Furthermore, interaction resulted in the reduction of unpredictable variation in all conditions, suggesting a role for communicative interaction in eliminating unpredictable variation. Regularisation was strongest in human-human interaction and in a condition where participants believed they were interacting with a human but were in fact interacting with a computer. We suggest that participants recognize the counter-functional nature of unpredictable variation and thus act to eliminate this variability during communication. Furthermore, reciprocal priming occurring in human-human interaction drove some pairs of participants to converge on maximally regular, highly predictable linguistic systems. Our method offers potential benefits to both the artificial language learning and the structural priming fields, and provides a useful tool to investigate communicative processes that lead to language change and ultimately language design. [ABSTRACT FROM AUTHOR]

Abstract: The present experiments demonstrate that children as young as five years old (M =5:2) generalize beyond their input on the basis of minimal exposure to a novel argument structure construction. The novel construction that was used involved a non-English phrasal pattern: VN1N2, paired with a novel abstract meaning: N2 approaches N1. At the same time, we find that children are keenly sensitive to the input: they show knowledge of the construction after a single day of exposure but this grows stronger after 3days; also, children generalize more readily to new verbs when the input contains more than one verb. [Copyright &y& Elsevier]

Abstract: Successful language acquisition involves generalization, but learners must balance this against the acquisition of lexical constraints. Such learning has been considered problematic for theories of acquisition: if learners generalize abstract patterns to new words, how do they learn lexically-based exceptions? One approach claims that learners use distributional statistics to make inferences about when generalization is appropriate, a hypothesis which has recently received support from Artificial Language Learning experiments with adult learners (). Since adult and child language learning may be different (), it is essential to extend these results to child learners. In the current work, four groups of children (6years) were each exposed to one of four semi-artificial languages. The results demonstrate that children are sensitive to linguistic distributions at and above the level of particular lexical items, and that these statistics influence the balance between generalization and lexical conservatism. The data are in line with an approach which models generalization as rational inference and in particular with the predictions of the domain general hierarchical Bayesian model developed in Kemp, Perfors & Tenenbaum, 2006. This suggests that such models have relevance for theories of language acquisition. [Copyright &y& Elsevier]

• Children are sensitive to untaught probabilistic orthographic patterns. • Patterns with phonological counterparts can be learned from artificial lexicons. • Graphotactic patterns unconfounded by correlated phonotactics can also be learned. • Homophonic CVC letter strings of varying complexity are learned implicitly. • Explicit instruction benefits generalisation and correlate with literacy attainment. Children are powerful statistical spellers, showing sensitivity to untaught orthographic patterns. They can also learn novel written patterns with phonological counterparts via statistical learning processes, akin to those established for spoken language acquisition. It is unclear whether children can learn written (graphotactic) patterns which are unconfounded from correlated phonotactics. We address this question by inducing novel graphotactic learning under incidental versus explicit conditions. Across three artificial lexicon experiments, we exposed children and adults to letter strings ending either in singlets or doublets (that share the same pronunciation, e.g., s vs. ss) depending on the preceding vowel. In post-tests, children and adults incidentally generalized over such context-based constraints that varied in complexity. Explicit instruction further benefitted pattern generalization, supporting the practice of teaching spelling patterns, and there was a relationship between explicit learning and literacy scores. We are first to demonstrate that statistical learning processes underlie graphotactic generalizations among developing spellers. [ABSTRACT FROM AUTHOR]

Discriminative theories frame language learning as a process of reducing uncertainty about the meaning of an utterance by discriminating informative from uninformative cues via the mechanisms of prediction error and cue competition. Previous work showed that discriminative learning is affected by the order in which information is presented during language learning. Specifically, learning suffixes, where complex stems precede affixes, promotes better generalization than prefixing, which tends to promote better item-learning instead. We explored this in two large-scale web-based artificial language learning experiments with adult learners (total N = 434), as well as two computational simulations implementing a discriminative learning model. While we did not find an overall benefit of suffixing over prefixing in generalization, consistent with our theoretical and computational predictions, we found that participants in the prefix condition were unable to discriminate between frequent, but uninformative cues and low-frequency, informative cues. This resulted in them being more likely to show incorrect overgeneralization of that feature for low frequency test items than participants in the suffix condition. We did not find a benefit of prefixing in item learning (although there was overall better item-learning of low type-frequency items), which we discuss in terms of the methodological limitations of our empirical paradigm. Taken together, these results underline the crucial role prediction error plays in learning linguistic generalization, and have implications for how generalization interacts with item-learning. [ABSTRACT FROM AUTHOR]

• Phonetic training can enable learners to distinguish between non-native phonemes. • Seminal studies propose that high talker variability (HV) is key to generalisation. • We conducted a replication (N = 166) of seminal studies to test for such a HV benefit. • All evidence for a HV benefit was ambiguous, contradicting original findings. • Our study raises questions about the true size and robustness of the HV benefit. Acquiring non-native speech contrasts can be difficult. A seminal study by Logan, Lively and Pisoni (1991) established the effectiveness of phonetic training for improving non-native speech perception: Japanese learners of English were trained to perceive /r/-/l/ using minimal pairs over 15 training sessions. A pre/post-test design established learning and generalisation. In a follow-up study, Lively, Logan and Pisoni (1993) presented further evidence which suggested that talker variability in training stimuli was crucial in leading to greater generalisation. These findings have been very influential and "high variability phonetic training" is now a standard methodology in the field. However, while the general benefit of phonetic training is well replicated, the evidence for an advantage of high over lower variability training remains mixed. In a large-scale replication of the original studies using updated statistical analyses we test whether learners generalise more after phonetic training using multiple talkers over a single talker. We find that listeners learn in both multiple and single talker conditions. However, in training, we find no difference in how well listeners learn for high vs low variability training. When comparing generalisation to novel talkers after training in relation to pre-training accuracy, we find ambiguous evidence for a high-variability benefit over low-variability training: This means that if a high-variability benefit exists, the effect is much smaller than originally thought, such that it cannot be detected in our sample of 166 listeners. [ABSTRACT FROM AUTHOR]