Learning languages

How children learn maximally diverse languages



What general cognitive mechanisms enable children to learn every language in the world?

Children are so flexible in learning languages because their learning is based on a few universal cognitive mechanisms that structure their learning at different times in different ways.

A child can learn any language, no matter how "difficult" it is. In view of the huge differences between languages across the world, this is astonishing. The main question behind the ACQDIV project (Acquisition processes in maximally diverse languages) therefore is: Which universal mechanisms enable an equally rapid acquisition of all known languages?

To answer this question, we are analyzing data from eleven languages (Chintang, Cree, Dene, Indonesian, Inuktitut, Japanese, Romansh, Russian, Sesotho, Turkish, and Yucatec). The random sample was selected to ensure maximal diversity, achieved by a cluster analysis based on a large typological database. It is no coincidence that our selection incorporates many small, hardly known languages - only through their inclusion can the whole range of human language be covered.

ACQDIV not only uses existing data, but also builds language corpora itself. The above-mentioned corpora of Chintang and Russian have already been completed as "in-house corpora". The Dene and Romansh coporara are now being compiled in La Loche (Saskatchewan, Canada) and Sedrun (Graubünden, Switzerland), with the participation of numerous children and adult speakers

This is where Citizen Science comes in. Our primary data comprise multimedia recordings of natural situations in which children learn language. In order to make these data usable, they must be transcribed, translated, and annotated on a word-for-word basis. This process is difficult for two reasons: one, because we do not speak most languages in the sample, so we have to rely on the help of native speakers; and secondly, recordings of a single minute can require up to an hour’s work which, on the basis of the hundreds of hours of recording needed today for a typical corpus, means many years of work from numerous people.

Our work with our native speaker Citizen Scientists covers the following specific areas: For the Chintang corpus, some recordings as well as all transcriptions and translations have been produced by native speakers. From their remuneration throughout the project, 15 young members of the native speaker community have been able to finance their university studies.

In building the Dene and Romansh corpora, almost all recordings are also being conducted by native speakers. Most of these are the mothers of the children being observed, who receive not only financial support, but also the opportunity to pursue further training or education (for example with a view to language teaching).

In order to encourage more people to participate in the particularly time-consuming process of transcribing, we are currently planning to develop a serious game with students of ZHdK (Zurich University of the Arts). Their contribution could, for example – subject to the clarification of ethical questions – be tested in the Romansh part of the project.

ACQDIV and speakers of the researched languages are working actively together on a number of projects to collect, transcribe and translate data. As a result, a bridge is built between the two groups, which ultimately enables knowledge and views to travel in both directions.


Project members:

Department of Comparative Linguistics, Psycholinguistics Laboratory, University of Zurich

Project Manager: Prof. Dr. Sabine Stoll

Project staff at UZH: Dr. Dagmar Jung, Dr. Steven Moran, Dr. Robert Schikowski, Damián Blasi, Jekaterina Mažara, Dr. Géraldine Walther, Dr. Claudia Cathomas, temporary student assistants.

Cooperation partners and affiliated institutions around the world.

Funded by ERC grant and SNF project.


Graphic artists:

Lisa Senn und Michael Koller