Editing Language acquisition (section)

==== Chunking ====

The central idea of these theories is that language development occurs through the incremental acquisition of meaningful [[chunking (psychology)#Chunking as the learning of long-term memory structures|chunks]] of elementary [[constituent (linguistics)|constituents]], which can be words, phonemes, or syllables. Recently, this approach has been highly successful in simulating several phenomena in the acquisition of [[syntactic category|syntactic categories]]<ref>{{cite journal|last=Freudenthal|first=Daniel|author2=J.M. Pine|author3=F. Gobet|year=2005|title=Modelling the development of children's use of optional infinitives in English and Dutch using MOSAIC|url=http://bura.brunel.ac.uk/bitstream/2438/731/1/oi-paper-all.pdf|journal=Cognitive Science|volume=30|issue=2|pages=277–310|doi=10.1207/s15516709cog0000_47|pmid=21702816|access-date=2 April 2009|author3-link=Fernand Gobet|doi-access=free}}</ref> and the acquisition of phonological knowledge.<ref>{{cite journal|last=Jones|first=Gary|author2=F. Gobet|author3=J.M. Pine|year=2007|title=Linking working memory and long-term memory: A computational model of the learning of new words|url=http://bura.brunel.ac.uk/bitstream/2438/618/1/DevSci_revised-final.pdf|journal=Developmental Science|volume=10|issue=6|pages=853–873|doi=10.1111/j.1467-7687.2007.00638.x|pmid=17973801|access-date=2 April 2009|author2-link=Fernand Gobet}}</ref>

Chunking theories of language acquisition constitute a group of theories related to statistical learning theories, in that they assume that the input from the environment plays an essential role; however, they postulate different learning mechanisms.{{clarify|reason=Different than what?|date=January 2020}}

Researchers at the [[Max Planck Institute for Evolutionary Anthropology]] have developed a computer model analyzing early toddler conversations to predict the structure of later conversations. They showed that toddlers develop their own individual rules for speaking, with 'slots' into which they put certain kinds of words. A significant outcome of this research is that rules inferred from toddler speech were better predictors of subsequent speech than traditional grammars.<ref>{{cite journal|vauthors=Bannard C, Lieven E, Tomasello M|date=October 2009|title=Modeling children's early grammatical knowledge|journal=Proc. Natl. Acad. Sci. U.S.A.|volume=106|issue=41|pages=17284–9|bibcode=2009PNAS..10617284B|doi=10.1073/pnas.0905638106|pmc=2765208|pmid=19805057|doi-access=free}}</ref>

This approach has several features that make it unique: the models are implemented as computer programs, which enables clear-cut and quantitative predictions to be made; they learn from naturalistic input—actual child-directed utterances; and attempt to create their own utterances, the model was tested in languages including English, Spanish, and German. Chunking for this model was shown to be most effective in learning a first language but was able to create utterances learning a second language.<ref>{{cite journal |last1=McCauley |first1=Stewart M. |last2=Christiansen |first2=Morten H. |title=Computational Investigations of Multiword Chunks in Language Learning |journal=Topics in Cognitive Science |date=July 2017 |volume=9 |issue=3 |pages=637–652 |doi=10.1111/tops.12258 |pmid=28481476 |doi-access=free }}</ref>