mdredze / speech_ner_entity_linking_data Goto Github PK

Authors:
adrian dot benton at gmail dot com
mark at dredze dot com

This distribution contains named entity tagged and entity linked speech data.

The named entity data was annotated as described in this paper:
Carolina Parada, Mark Dredze, Frederick Jelinek. OOV Sensitive Named-Entity
Recognition in Speech. International Speech Communication Association (INTERSPEECH), 2011.

A subset of the person entities in this dataset were annotated for
entity linking information as described in this paper:
Adrian Benton, Mark Dredze. Entity Linking for Spoken Language. 
North American Chapter of the Association for Computational Linguistics (NAACL), 2015.

The utterances were drawn from the HUB4
dataset https://catalog.ldc.upenn.edu/LDC98S71 , and the knowledge base is
the same Wikipedia dump used in the TAC 2009 KBP track.

This distribution includes the annotations only. The HUB4 data must be obtained
from the LDC.

The files contain the following:
- "folds.txt": Mapping from each entity linking query ID to fold

- "el_queries.txt": Entity linking queries.  Format:
    QUERY_ID ENTITY_MENTION HUB4_UTTERANCE_ID ENTITY_TYPE KB_ID

- "ne_el_labels.txt": All named entities annotated in the HUB4 transcripts.

Any entity that does not have a linking annotation to the Wikipedia KB
is marked as "NOT_ANNOTATED".  Format:
    HUB4_UTTERANCE_ID ENTITY_MENTION START_SPAN END_SPAN ENTITY_TYPE KB_ID