The GREC Corpus

Download

The corpus in available for download in 2 formats:

• A standoff format, based on the BioNLP'09 Shared Task format
  • UPDATE 10/12/2009: Two files in the Human part of the corpus (IDs 8205615 and 9778250) were found to have character offset problems due to foreign characters. Offsets in these files were previously based on bytes rather than characters. The corpus has now been updated so that all offsets are based on characters.
• An XML format, based on the GENIA event annotation format

The annotation guidelines are also available to download.

Background

Information Extraction (IE) is a component of text mining that facilitates knowledge discovery by automatically locating instances of interesting biomedical events from huge document collections. Effective IE systems require training data or annotated corpora, in which instances of biomedical events are explicitly identified in texts. The trained IE systems can then recognise instances of new events in texts, facilitating a number of text mining applications, such as pathway maintenance and semantic searching.

The Corpus

The GREC corpus is a semantically annotated corpus of MEDLINE abstracts which is intended for training IE systems and/or resources which are used to extract events from biomedical literature.

The corpus has been manually annotated with events relating to gene regulation by biologists. Each event is centred on either a verb (e.g. transcribe) or nominalised verb (e.g. transcription) and annotation consists of identifying, as exhaustively as possible, the structually-related arguments of the verb or nominalised verb within the same sentence. Each event argument is then assigned the following information:

• A semantic role from a fixed set of 13 roles which are tailored to the biomedical domain.
• A biomedical concept type (where appropriate).

As a simple example, consider the following sentence:

The narL gene product activates the nitrate reductase operon

The sentence contains a single event, centred on the verb activates, with 2 arguments, i.e.:

1. The narL gene product
2. the nitrate reductase operon

The argument The narL gene product is assigned the semantic role AGENT and the biological concept Protein, whilst the argument the nitrate reductase operon is assigned the semantic role THEME and the biological concept Operon.

Other types of argument include:

• LOCATION, e.g. In Escherichia Coli, glnAP2 may be activated by NifA
• MANNER, e.g. cpxA gene increases the levels of csgA transcription by dephosphorylation of CpxR
• CONDITION, e.g. Strains carrying a mutation in the crp structural gene fail to repress ODC and ADC activities in response to increased cAMP

Full details of the annotation scheme can be found in the annotation guidelines.

Contact

For any queries relating to the corpus, please contact:
paul.thompson at manchester.ac.uk