Information Extraction: Relations, Events, and Time

21.1 Relation extraction

1. What binary relations can be extracted from “American Airlines, a unit of AMR Corp., … spokesman Tim Wagner said”?
   Tim Wagner is a spokesman of (employed by) American Airlines; American Airlines is part of (a unit of) AMR Corp. These instantiate generic relations like part-of or employs.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 21, §21.1, p. 474
2. What is the ACE relation set?
   The 17 relations used in the ACE relation extraction evaluations, grouped into classes like physical (located, near), part-whole, person-social (family, business), org affiliation (employment, founder, membership…), agent-artifact, and gen-affiliation (citizen-resident-ethnicity-religion).
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 21, §21.1, p. 474
3. What is UMLS?
   The Unified Medical Language System of the US National Library of Medicine: a network of broad subject categories and entity types with relations between them, e.g. Doppler echocardiography Diagnoses acquired stenosis.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 21, §21.1, p. 475
4. What is an RDF triple?
   A subject–predicate–object tuple in the Resource Description Framework, e.g. (Golden Gate Park, location, San Francisco).
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 21, §21.1, p. 475
5. Which resources derived from Wikipedia provide relations?
   Infoboxes (structured facts like state = "California"), DBpedia (an ontology of RDF triples from Wikipedia), and Freebase (now part of Wikidata).
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 21, §21.1, p. 475
6. What ontological relations does WordNet offer?
   The is-a (hypernym) relation between classes (giraffe is-a ruminant is-a ungulate is-a mammal…) and the instance-of relation between individuals and classes (San Francisco instance-of city).
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 21, §21.1, p. 475
7. What datasets exist for training and testing relation extractors?
   TACRED (person/organization relation triples in news and web text with 41 relation types plus no_relation, mostly no_relation examples) and SemEval 2010 Task 8 (relations between nominals, e.g. product-producer, component-whole).
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 21, §21.1, pp. 475–476
8. Why does TACRED contain mostly no_relation examples?
   Sufficient negative data is important for training relation extractors.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 21, §21.1, p. 476