User:Gandreal/sandbox

Statistical Learning in Child Language Acquisition
Infants learn a language at an astounding rate. They go from not being able to speak to speaking their native language within a few years. Infants are exposed to diverse samples of utterances but converge on the same grammar. The way in which infants learn a language so quickly has stumped scientists for a long time. Scientists believe that infants learn a language rapidly through exposure to language as they combine pattern detection and computational abilities with social skills. These pattern detection and computational abilities are called statistical learning mechanisms. The use of statistical learning has helped scientists approach the question of how infants learn a language at such a rapid rate.

History
In humans, all the tools needed to learn a language seem to be innate and are developed as we interact with our surroundings. Scientists were concerned with how infants separate words when there is a continuous nature to speech. Harris, a structural linguist from the mid 20th century, noted the statistic of transitional probability. A transitional probability is the likelihood of a sound that can follow a word given the last sound of that word. Transitional probabilities help account for infants’ ability to differentiate from part-words in tone sequences. 8-month-old infants are able to discriminate between words and part words, which indicates that they are sensitive to conditional probabilities of consecutive sounds. There is definitely more than one mechanism, apart from statistical learning, that helps explain the acquisition in language structure in infants.

Experiments conducted by Marcus et. al (1999) set out to prove that infants are not limited to statistical learning in language acquisition, they have more mechanisms to use when learning a language. Marcus et. al (1999) conducted two experiments on 7-month old infants. The first experiment, infants were randomly assigned to one of two conditions: “ABA” or “ABB.” In the ABA condition, infants were habituated with a 2-minute speech sample that contained three word sentences that followed an ABA grammar (i.e. “ga ti ga” and “li na li”). In the ABB condition, infants were habituated in the same way expect that the sample followed an ABB grammar (i.e. “ga ti ti” and “li na na”). Half of the trials used consistent sentences that followed the grammar with which the infant was trained. The other half of the trials used inconsistent sentences, so that the ABA group would receive ABB sentences and vice versa (Marcus, 1999). Results of this first experiment showed that infants were surprised only when they heard an inconsistent sentence. To rule out any possibility that the infant used statistical learning rules, another experiment was conducted (Marcus, 1999). In the second experiment, the test words were distinguished from one another by phonetic features and were not distinguished in the habituation sentences. The results from the first experiment were the same in the second experiment. A third experiment was conducted to see if infants who were habituated with an ABB grammar were surprised when a sentence followed an AAB grammar. This experiment tested whether or not the infant could differentiate between the duplication in each of the grammars. The results show that the infants could not distinguish between the ABB and the AAB grammar. The results of the experiments conducted by Marcus et. al (1999) demonstrate that statistical learning mechanisms do not account for how humans generalize rules in conditions where the items shown in an experiment do not overlap with the items shown in the habituation. Thus, Marcus et. al (1999) propose that there is another tool, other than statistical, for learning language and that is one that manipulates variables which allows children to learn rules.

Current Research
Recent research indicates that neural commitment is involved in language acquisition. Infants commit their brain’s neural networks to code for the patterns in their native language, which interferes with learning new patterns (i.e. another language). Therefore, there is a sensitive period in which infants commit their neural networks which is the basis of future learning because future learning of a language conforms and builds on the patterns that the infant’s neural networks have committed to. Phoentic contrasts that infants discriminate well early on, remain discernible if they are consistent with their native language.

The nativist view of language acquisition views that language learning mechanisms are innate in humans. The empiricist view of language acquisition views humans as a blank slate with general language learning mechanisms.

Further Research
Future research should research the other sensory fields in order to verify that statistical learning is domain-specific.