Published July 14, 2020 | Version v1
Dataset Open

Mutual information for PP attachment (PP-PMI)

  • 1. ROR icon University of Tübingen

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)

Funding

Deutsche Forschungsgemeinschaft
SFB 833:  Bedeutungskonstitution - Dynamik und Adaptivität sprachlicher Strukturen 75650358

Data quality

Accuracy

Not specified.

Completeness

Not specified.

Conformity

Not specified.

Consistency

Not specified.

Credibility

Not specified.

Processability

Not specified.

Relevance

Not specified.

Timeliness

Not specified.

Understandability

Not specified.