Children are extremely flexible with respect to language acquisition - any child can learn any of the world's approximately 7000 languages. While psycholinguistic research has made considerable progress with regard to specific learning mechanisms in specific languages, our knowledge is still poor when it comes to general mechanisms that would apply in all languages. In addition, little is known about language-specific differences in acquisition.
The classical answer to the question how language acquisition is possible in spite of substantial linguistic diversity is Principle and Parameters Theory, which claims that children's brains are equipped with a hard-wired grammar module called Universal Grammar, from which all human languages can be derived by setting certain parameters. However, this theory has recently been challenged by research in linguistic theory, typology, and psycholinguistics. In the meantime it seems more likely that language acquisition is based on general cognitive mechanisms which are not specific to language.
The central goals of ACQDIV are to add further, decisive support to this hypothesis, to establish the precise set of cognitive mechanisms, and to determine conditions on variation in the acquisition of individual languages.
Universal cognitive mechanisms in acquisition
The main candidates for mechanisms enabling language acquisition that will be examined in ACQDIV are distributional learning, generalisation-based learning, and interactional learning:
- Distributional learning: Children are known to be able to recognise and reproduce distributional patterns across various cognitive domains. What role does this mechanism play precisely in the acquisition of language? How is distributional learning facilitated by patterns in the input?
- Generalisation-based learning: Apart from reproducing distributions, there is no doubt that children must also be able to extract patterns in order to reach full proficiency. But how abstract are these patterns? Is generalisation better characterised as based on rules or as based on items?
- Interactional learning: Successful language acquisition depends on active conversational interaction - passive "intake" is not sufficient. How exactly do various patterns of interaction relate to language acquisition, and to what extent are they independent of different cultures of linguistic socialisation?
Patterns in the input
The core hypothesis of ACQDIV that language acquisition is based on general cognitive mechanisms relies crucially on patterns in the input children receive - only if such patterns exist can children learn language without an innate Universal Grammar. While Principles and Parameters Theory has traditionally stressed the "poverty of the stimulus", there is now growing evidence that the input is indeed not chaotic but structured. ACQDIV will also explore the question what structures exist and how children exploit them by focussing on two research areas:
- Repetitions: Repetitions and frequency are known to play an important role in language acquisition. However, there are many open questions - for instance, in which positions in an utterance and in which time intervals are repetitions usually placed and/or easiest to perceive? What is the role of cumulated repetitions and burst effects?
- Directedness: Input patterns are most prominent in child-directed speech, which is to some degree adapted to the child's requirements. However, children also receive a lot of non-directed (child-surrounding) input. How different are these two speech varieties, and how relevant are the differences for acquisition?
Grammatical phenomena of interest
Apart from the theoretical problems lined out above, ACQDIV will also concentrate on a few grammatical phenomena in order to make more specific questions possible. The two central phenomena are aspect and negation. Both phenomena are present in all languages of the sample (see below), and negation is even universal to human language. On the other hand, they are expressed by very different morphological and syntactic means, thus making them ideal candidates for comparing the effects of linguistic diversity on acquisition.
Each language in the ACQDIV sample also features individual areas of interest where children are confronted with particularly complex structures, for instance, polysynthetic verb morphology in Dene, Cree, and Chintang, or noun classes and gender in Sesotho and Cree. For a more detailed overview, see our list of languages.
Data and methods
In order to observe language acquisition in a natural environment and in its development over time, ACQDIV makes use of longitudinal corpus data featuring transcriptions, translations, and morphosyntactic annotation (glosses). While most of the corpora are already completed, a few will be enhanced with additional material and/or annotations during the project. One - the Dene corpus - does not exist yet but will be compiled during the project (see this page).
In order to represent the typological diversity found in the languages of the world in a satisfying way, five clusters of maximally diverse languages were calculated from a subset of the AUTOTYP database based on a fuzzy clustering algorithm. The language sample was then built by choosing two languages from each cluster for which sufficient resources were available.
In order to analyse the corpus data, to find patterns both in children's and adults' speech, and to track developments in children's speech, ACQDIV will make use of and develop new quantitative methods such as similarity measures, clustering and pattern recognition algorithms, or various kinds of mixed regression models.