Unit 1: Corpora
Target objectives for this unit
After going through the material in this unit you should
- know what data-driven learning is
- know what a corpus is
- know where corpora can be found online
- have a good idea of the diversity of corpora that are out there
What is Data-Driven Language Learning?
Read a mini-lecture on the subject: Unit 1 - mini-lecture
View part of an example data-driven English lesson: Unit 1 - example lesson
What is a corpus?
A language corpus (plural: corpora) is a collection of authentic instances of the use of the language. There are corpora of specific usage type (medical communications, newswire transcriptions), and general corpora, which try to be balanced with respect to as many types of language use as possible, from fiction to online records; from scientific journals to sitcom scripts.
A corpus usually provides the "data" in data-driven language learning. It serves as a representative sample of every use of every word in the language. For any given word, we can answer the question "How is this word used?", not by looking it up and trusting some authority, but by exposing ourselves to authentic instances of use.
It is recognized in language learning that authentic input is important and effective. A corpus is a huge quantity of authentic input.
So, a corpus has the following characteristics that make it an attractive source of input for language learning:
- Size - corpora are large enough collections of language use that it's possible to discover patterns of use.
- Authenticity - Corpora contain texts that are guaranteed to be authentic examples of language use.
- Searchability - Since corpora are controlled and studied, searches of them are more likely to turn up useful information than, for example, a web search. Metadata, for instance part-of-speech tagging, makes them useful in ways that uncontrolled language samples are not.
Unit 1 exercises
On to Unit 2: Concordancers
Back to the beginning: Data-Driven Language Learning Using Corpora and Concordancing