One of the major challenges in language acquisition is the learning of non-adjacent dependencies in syntax. For example, in the sentence The books on the shelf are dusty, the plural verb form are corresponds to the plural of books occurring several words earlier. Such learning can be difficult because adjacent forms, such as the singular shelf, must be ignored.
Research exposing both adults and infants to sentences in an artificial language has discovered that the learning of such non-adjacent dependencies is facilitated by a high degree of variability in words occurring between the dependent forms (Gómez 2002). Such experiments with infants may uncover the fundamental properties of the human language-learning system (Gómez & Gerken, 2000). The present project attempts to simulate these results in neural networks, allowing the predictions of new effects to be tested with adults and infants.
The simulations might well involve coding sentences in terms of sonority (or vowel likeness), a realistic coding technique that has worked well in other simulations of infant exposure to artificial languages (Shultz & Bale, 2001).
Some knowledge of syntax. Ability to learn and use a neural network simulator, or alternatively the ability to design and run experiments with infants or adults. Familiarity with statistical software for ANOVA and possibly PCA.
Gómez, R. L. (2002). Variability and detection of invariant structure. Psychological Science, 13, 431-436.
Gómez, R. L., & Gerken, L. (2000). Infant artificial language learning and language acquisition. Trends in Cognitive Sciences, 4, 178-186.
Shultz, T. R., & Bale, A. C. (2001). Neural network simulation of infant familiarization to artificial sentences: Rule-like behavior without explicit rules and variables. Infancy, 2, 501-536.