Mutual information for PP attachment (PP-PMI)
Description
This dataset was extracted from a dependency corpus. Dependency parser dpar was used to parse articles from the German newspaper taz from 1986 to 2009. All the prepositional phrases were extracted and bi-lexical association scores were calculated for 3 types of token pairs: (candidate, object), (candidate, preposition) and (candidate, preposition+object). Tri-lexical association scores were also taken into account by using specific interaction information and total correlation. This dataset was initially used as auxiliary distributions to augment the data beyond the hand-annotated training set to help with prepositional phrase attachment.
Other (English)
Research carried out in work package A03 of the SFB 833.
Files
CMDI.xml
Files
(330.3 MB)
| Name | Size | Download all |
|---|---|---|
|
md5:2201e6d3c872b70db905d166d5eb27cf
|
16.6 kB | Preview Download |
|
md5:697b103995acb3abd1e3a1bfe8642ac3
|
330.3 MB | Preview Download |
Additional details
Related works
- Is described by
- Data paper: https://www.aclweb.org/anthology/E17-2050/ (URL)