PartML: meronymic relationship annotation and classification by machine learning
Specifications
The whole in a relationship could be divided into a number of parts, where each plays a structural or functional role to each other and to the whole. In this kind of relationship, the ’whole’ is always arranged in a patterned structure
The difference between member - collection relationship and part - whole relationship is that member entities are not required to have structural or funtional role of the collection. The collection is normally a group of homogenous members that are grouped together because of their spatial proximity or social connections.
The ’subtance’ is an inseparable portion of the ’object’. The ’subtance’ is normally a material or chemical subtance that the ’object’ contains. Inseparibility in this sense only regards the current context, it doesn’t mean that another object that have the same generic term as this ’object’ or a similar object must also compose of this ’subtance’
Annotations
The corpus was collected from a variety of English Wikipedia articles, stripped of formatting and non-textual elements. It is manually collected with the help of built-it random article feature.
Our corpus (counting only those documents which were annotated) contains 40 documents, with an average of roughly 425 words per document. Each document in our corpus (with a few exceptions) was annotated twice. We used these pairs of duplicate documents to calculate Cohen’s Kappa as a measure of inter-annotator agreement on our extents and attributes.
Resources
Document type definition (DTD) file
Gold annotated corpus
Specification
Report