This Title All WIREs
How to cite this WIREs title:
WIREs Cogn Sci
Impact Factor: 3.476

Words, rules, and mechanisms of language acquisition

Full article on Wiley Online Library:   HTML PDF

Can't access this content? Tell your librarian.

We review recent artificial language learning studies, especially those following Endress and Bonatti (Endress AD, Bonatti LL. Rapid learning of syllable classes from a perceptually continuous speech stream. Cognition 2007, 105:247–299), suggesting that humans can deploy a variety of learning mechanisms to acquire artificial languages. Several experiments provide evidence for multiple learning mechanisms that can be deployed in fluent speech: one mechanism encodes the positions of syllables within words and can be used to extract generalization, while the other registers co‐occurrence statistics of syllables and can be used to break a continuum into its components. We review dissociations between these mechanisms and their potential role in language acquisition. We then turn to recent criticisms of the multiple mechanisms hypothesis and show that they are inconsistent with the available data. Our results suggest that artificial and natural language learning is best understood by dissecting the underlying specialized learning abilities, and that these data provide a rare opportunity to link important language phenomena to basic psychological mechanisms. WIREs Cogn Sci 2016, 7:19–35. doi: 10.1002/wcs.1376 This article is categorized under: Psychology > Language Psychology > Theory and Methods Psychology > Learning
Effect sizes (Cohen's d) for the class‐word versus part‐word discrimination (dark bars) and the word versus rule‐word discrimination with segmented 2‐min streams in (a) humans and (b) networks. (a) In humans, the word versus rule‐word discrimination is numerically easier than the class‐words versus part‐word discrimination. Note that Endress and Bonatti used a different stimulus materials in Experiment 10 than in Experiments 3 and 8. (b) After the number of training cycles Laakso & Calvo propose to correspond to a 2‐min familiarization, the network performance on the class‐word versus part‐word discrimination is much better than on the word versus rule‐word discrimination, showing the opposite pattern from humans.
[ Normal View | Magnified View ]
We simulated experiments by recording the results of 20 simulations with different network initializations, representing 20 participants. One experiment was simulated for each set of network parameters. The networks did not reproduce the preference for class‐words over part‐words in the edge condition for any set of network parameters. However, for completeness, we report more detailed results. (a) Proportion of simulated experiments where the preference for class‐words over part‐words is (significantly or numerically) stronger in the edge condition or the middle condition, depending on whether, in the middle condition, the fourth or the fifth syllable is considered as the target syllable. For most simulated experiments, the preference for class‐words is stronger in the middle condition than in the edge condition (i.e., the preference for part‐words is weaker), suggesting that the network does not intrinsically account for Endress and Mehler's data. (b) F‐values associated with the interaction between the preference for class‐words over part‐words and the edge versus middle condition when the fourth syllable is considered the target syllable. When the preference for class‐words was stronger in the middle condition, the F‐values were multiplied by −1. (c) F‐values associated with the aforementioned interaction when the fifth syllable is considered the target syllable.
[ Normal View | Magnified View ]
Difference in predictions for the last syllable of words and part‐words, respectively, after (a) a familiarization with a segmented stream and (b) a familiarization with a continuous stream. The solid line shows the average difference between the cosine values between the predicted network output and the target ‘syllables.’ The dashed line shows the corresponding effect sizes (Cohen's d). Laakso and Calvo simulations show that it should be harder to recognize words when they are encountered more often.
[ Normal View | Magnified View ]

Browse by Topic

Psychology > Theory and Methods
Psychology > Learning
Psychology > Language

Access to this WIREs title is by subscription only.

Recommend to Your
Librarian Now!

The latest WIREs articles in your inbox

Sign Up for Article Alerts